4 The first step in getting fio to simulate a desired I/O workload, is writing a
5 job file describing that specific setup. A job file may contain any number of
6 threads and/or files -- the typical contents of the job file is a *global*
7 section defining shared parameters, and one or more job sections describing the
8 jobs involved. When run, fio parses this file and sets everything up as
9 described. If we break down a job from top to bottom, it contains the following
14 Defines the I/O pattern issued to the file(s). We may only be reading
15 sequentially from this file(s), or we may be writing randomly. Or even
16 mixing reads and writes, sequentially or randomly.
17 Should we be doing buffered I/O, or direct/raw I/O?
21 In how large chunks are we issuing I/O? This may be a single value,
22 or it may describe a range of block sizes.
26 How much data are we going to be reading/writing.
30 How do we issue I/O? We could be memory mapping the file, we could be
31 using regular read/write, we could be using splice, async I/O, or even
36 If the I/O engine is async, how large a queuing depth do we want to
42 How many files are we spreading the workload over.
44 `Threads, processes and job synchronization`_
46 How many threads or processes should we spread this workload over.
48 The above are the basic parameters defined for a workload, in addition there's a
49 multitude of parameters that modify other aspects of how this job behaves.
55 .. option:: --debug=type
57 Enable verbose tracing `type` of various fio actions. May be ``all`` for all types
58 or individual types separated by a comma (e.g. ``--debug=file,mem`` will
59 enable file and memory debugging). Currently, additional logging is
63 Dump info related to processes.
65 Dump info related to file actions.
67 Dump info related to I/O queuing.
69 Dump info related to memory allocations.
71 Dump info related to blktrace setup.
73 Dump info related to I/O verification.
75 Enable all debug options.
77 Dump info related to random offset generation.
79 Dump info related to option matching and parsing.
81 Dump info related to disk utilization updates.
83 Dump info only related to job number x.
85 Dump info only related to mutex up/down ops.
87 Dump info related to profile extensions.
89 Dump info related to internal time keeping.
91 Dump info related to networking connections.
93 Dump info related to I/O rate switching.
95 Dump info related to log compress/decompress.
97 Dump info related to steadystate detection.
99 Dump info related to the helper thread.
101 Dump info related to support for zoned block devices.
103 Show available debug options.
105 .. option:: --parse-only
107 Parse options only, don't start any I/O.
109 .. option:: --merge-blktrace-only
111 Merge blktraces only, don't start any I/O.
113 .. option:: --output=filename
115 Write output to file `filename`.
117 .. option:: --output-format=format
119 Set the reporting `format` to `normal`, `terse`, `json`, or `json+`. Multiple
120 formats can be selected, separated by a comma. `terse` is a CSV based
121 format. `json+` is like `json`, except it adds a full dump of the latency
124 .. option:: --bandwidth-log
126 Generate aggregate bandwidth logs.
128 .. option:: --minimal
130 Print statistics in a terse, semicolon-delimited format.
132 .. option:: --append-terse
134 Print statistics in selected mode AND terse, semicolon-delimited format.
135 **Deprecated**, use :option:`--output-format` instead to select multiple
138 .. option:: --terse-version=version
140 Set terse `version` output format (default 3, or 2 or 4 or 5).
142 .. option:: --version
144 Print version information and exit.
148 Print a summary of the command line options and exit.
150 .. option:: --cpuclock-test
152 Perform test and validation of internal CPU clock.
154 .. option:: --crctest=[test]
156 Test the speed of the built-in checksumming functions. If no argument is
157 given, all of them are tested. Alternatively, a comma separated list can
158 be passed, in which case the given ones are tested.
160 .. option:: --cmdhelp=command
162 Print help information for `command`. May be ``all`` for all commands.
164 .. option:: --enghelp=[ioengine[,command]]
166 List all commands defined by `ioengine`, or print help for `command`
167 defined by `ioengine`. If no `ioengine` is given, list all
170 .. option:: --showcmd
172 Convert given job files to a set of command-line options.
174 .. option:: --readonly
176 Turn on safety read-only checks, preventing writes and trims. The
177 ``--readonly`` option is an extra safety guard to prevent users from
178 accidentally starting a write or trim workload when that is not desired.
179 Fio will only modify the device under test if
180 `rw=write/randwrite/rw/randrw/trim/randtrim/trimwrite` is given. This
181 safety net can be used as an extra precaution.
183 .. option:: --eta=when
185 Specifies when real-time ETA estimate should be printed. `when` may be
186 `always`, `never` or `auto`. `auto` is the default, it prints ETA
187 when requested if the output is a TTY. `always` disregards the output
188 type, and prints ETA when requested. `never` never prints ETA.
190 .. option:: --eta-interval=time
192 By default, fio requests client ETA status roughly every second. With
193 this option, the interval is configurable. Fio imposes a minimum
194 allowed time to avoid flooding the console, less than 250 msec is
197 .. option:: --eta-newline=time
199 Force a new line for every `time` period passed. When the unit is omitted,
200 the value is interpreted in seconds.
202 .. option:: --status-interval=time
204 Force a full status dump of cumulative (from job start) values at `time`
205 intervals. This option does *not* provide per-period measurements. So
206 values such as bandwidth are running averages. When the time unit is omitted,
207 `time` is interpreted in seconds. Note that using this option with
208 ``--output-format=json`` will yield output that technically isn't valid
209 json, since the output will be collated sets of valid json. It will need
210 to be split into valid sets of json after the run.
212 .. option:: --section=name
214 Only run specified section `name` in job file. Multiple sections can be specified.
215 The ``--section`` option allows one to combine related jobs into one file.
216 E.g. one job file could define light, moderate, and heavy sections. Tell
217 fio to run only the "heavy" section by giving ``--section=heavy``
218 command line option. One can also specify the "write" operations in one
219 section and "verify" operation in another section. The ``--section`` option
220 only applies to job sections. The reserved *global* section is always
223 .. option:: --alloc-size=kb
225 Allocate additional internal smalloc pools of size `kb` in KiB. The
226 ``--alloc-size`` option increases shared memory set aside for use by fio.
227 If running large jobs with randommap enabled, fio can run out of memory.
228 Smalloc is an internal allocator for shared structures from a fixed size
229 memory pool and can grow to 16 pools. The pool size defaults to 16MiB.
231 NOTE: While running :file:`.fio_smalloc.*` backing store files are visible
234 .. option:: --warnings-fatal
236 All fio parser warnings are fatal, causing fio to exit with an
239 .. option:: --max-jobs=nr
241 Set the maximum number of threads/processes to support to `nr`.
242 NOTE: On Linux, it may be necessary to increase the shared-memory
243 limit (:file:`/proc/sys/kernel/shmmax`) if fio runs into errors while
246 .. option:: --server=args
248 Start a backend server, with `args` specifying what to listen to.
249 See `Client/Server`_ section.
251 .. option:: --daemonize=pidfile
253 Background a fio server, writing the pid to the given `pidfile` file.
255 .. option:: --client=hostname
257 Instead of running the jobs locally, send and run them on the given `hostname`
258 or set of `hostname`\s. See `Client/Server`_ section.
260 .. option:: --remote-config=file
262 Tell fio server to load this local `file`.
264 .. option:: --idle-prof=option
266 Report CPU idleness. `option` is one of the following:
269 Run unit work calibration only and exit.
272 Show aggregate system idleness and unit work.
275 As **system** but also show per CPU idleness.
277 .. option:: --inflate-log=log
279 Inflate and output compressed `log`.
281 .. option:: --trigger-file=file
283 Execute trigger command when `file` exists.
285 .. option:: --trigger-timeout=time
287 Execute trigger at this `time`.
289 .. option:: --trigger=command
291 Set this `command` as local trigger.
293 .. option:: --trigger-remote=command
295 Set this `command` as remote trigger.
297 .. option:: --aux-path=path
299 Use the directory specified by `path` for generated state files instead
300 of the current working directory.
302 Any parameters following the options will be assumed to be job files, unless
303 they match a job file parameter. Multiple job files can be listed and each job
304 file will be regarded as a separate group. Fio will :option:`stonewall`
305 execution between each group.
311 As previously described, fio accepts one or more job files describing what it is
312 supposed to do. The job file format is the classic ini file, where the names
313 enclosed in [] brackets define the job name. You are free to use any ASCII name
314 you want, except *global* which has special meaning. Following the job name is
315 a sequence of zero or more parameters, one per line, that define the behavior of
316 the job. If the first character in a line is a ';' or a '#', the entire line is
317 discarded as a comment.
319 A *global* section sets defaults for the jobs described in that file. A job may
320 override a *global* section parameter, and a job file may even have several
321 *global* sections if so desired. A job is only affected by a *global* section
324 The :option:`--cmdhelp` option also lists all options. If used with a `command`
325 argument, :option:`--cmdhelp` will detail the given `command`.
327 See the `examples/` directory for inspiration on how to write job files. Note
328 the copyright and license requirements currently apply to `examples/` files.
330 So let's look at a really simple job file that defines two processes, each
331 randomly reading from a 128MiB file:
335 ; -- start job file --
346 As you can see, the job file sections themselves are empty as all the described
347 parameters are shared. As no :option:`filename` option is given, fio makes up a
348 `filename` for each of the jobs as it sees fit. On the command line, this job
349 would look as follows::
351 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
354 Let's look at an example that has a number of processes writing randomly to
359 ; -- start job file --
370 Here we have no *global* section, as we only have one job defined anyway. We
371 want to use async I/O here, with a depth of 4 for each file. We also increased
372 the buffer size used to 32KiB and define numjobs to 4 to fork 4 identical
373 jobs. The result is 4 processes each randomly writing to their own 64MiB
374 file. Instead of using the above job file, you could have given the parameters
375 on the command line. For this case, you would specify::
377 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
379 When fio is utilized as a basis of any reasonably large test suite, it might be
380 desirable to share a set of standardized settings across multiple job files.
381 Instead of copy/pasting such settings, any section may pull in an external
382 :file:`filename.fio` file with *include filename* directive, as in the following
385 ; -- start job file including.fio --
389 include glob-include.fio
396 include test-include.fio
397 ; -- end job file including.fio --
401 ; -- start job file glob-include.fio --
404 ; -- end job file glob-include.fio --
408 ; -- start job file test-include.fio --
411 ; -- end job file test-include.fio --
413 Settings pulled into a section apply to that section only (except *global*
414 section). Include directives may be nested in that any included file may contain
415 further include directive(s). Include files may not contain [] sections.
418 Environment variables
419 ~~~~~~~~~~~~~~~~~~~~~
421 Fio also supports environment variable expansion in job files. Any sub-string of
422 the form ``${VARNAME}`` as part of an option value (in other words, on the right
423 of the '='), will be expanded to the value of the environment variable called
424 `VARNAME`. If no such environment variable is defined, or `VARNAME` is the
425 empty string, the empty string will be substituted.
427 As an example, let's look at a sample fio invocation and job file::
429 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
433 ; -- start job file --
440 This will expand to the following equivalent job file at runtime:
444 ; -- start job file --
451 Fio ships with a few example job files, you can also look there for inspiration.
456 Additionally, fio has a set of reserved keywords that will be replaced
457 internally with the appropriate value. Those keywords are:
461 The architecture page size of the running system.
465 Megabytes of total memory in the system.
469 Number of online available CPUs.
471 These can be used on the command line or in the job file, and will be
472 automatically substituted with the current system values when the job is
473 run. Simple math is also supported on these keywords, so you can perform actions
478 and get that properly expanded to 8 times the size of memory in the machine.
484 This section describes in details each parameter associated with a job. Some
485 parameters take an option of a given type, such as an integer or a
486 string. Anywhere a numeric value is required, an arithmetic expression may be
487 used, provided it is surrounded by parentheses. Supported operators are:
496 For time values in expressions, units are microseconds by default. This is
497 different than for time values not in expressions (not enclosed in
498 parentheses). The following types are used:
505 String: A sequence of alphanumeric characters.
508 Integer with possible time suffix. Without a unit value is interpreted as
509 seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
510 hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and
511 'us' (or 'usec') for microseconds. For example, use 10m for 10 minutes.
516 Integer. A whole number value, which may contain an integer prefix
517 and an integer suffix:
519 [*integer prefix*] **number** [*integer suffix*]
521 The optional *integer prefix* specifies the number's base. The default
522 is decimal. *0x* specifies hexadecimal.
524 The optional *integer suffix* specifies the number's units, and includes an
525 optional unit prefix and an optional unit. For quantities of data, the
526 default unit is bytes. For quantities of time, the default unit is seconds
527 unless otherwise specified.
529 With :option:`kb_base`\=1000, fio follows international standards for unit
530 prefixes. To specify power-of-10 decimal values defined in the
531 International System of Units (SI):
533 * *K* -- means kilo (K) or 1000
534 * *M* -- means mega (M) or 1000**2
535 * *G* -- means giga (G) or 1000**3
536 * *T* -- means tera (T) or 1000**4
537 * *P* -- means peta (P) or 1000**5
539 To specify power-of-2 binary values defined in IEC 80000-13:
541 * *Ki* -- means kibi (Ki) or 1024
542 * *Mi* -- means mebi (Mi) or 1024**2
543 * *Gi* -- means gibi (Gi) or 1024**3
544 * *Ti* -- means tebi (Ti) or 1024**4
545 * *Pi* -- means pebi (Pi) or 1024**5
547 For Zone Block Device Mode:
550 With :option:`kb_base`\=1024 (the default), the unit prefixes are opposite
551 from those specified in the SI and IEC 80000-13 standards to provide
552 compatibility with old scripts. For example, 4k means 4096.
554 For quantities of data, an optional unit of 'B' may be included
555 (e.g., 'kB' is the same as 'k').
557 The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
558 not milli). 'b' and 'B' both mean byte, not bit.
560 Examples with :option:`kb_base`\=1000:
562 * *4 KiB*: 4096, 4096b, 4096B, 4ki, 4kib, 4kiB, 4Ki, 4KiB
563 * *1 MiB*: 1048576, 1mi, 1024ki
564 * *1 MB*: 1000000, 1m, 1000k
565 * *1 TiB*: 1099511627776, 1ti, 1024gi, 1048576mi
566 * *1 TB*: 1000000000, 1t, 1000m, 1000000k
568 Examples with :option:`kb_base`\=1024 (default):
570 * *4 KiB*: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
571 * *1 MiB*: 1048576, 1m, 1024k
572 * *1 MB*: 1000000, 1mi, 1000ki
573 * *1 TiB*: 1099511627776, 1t, 1024g, 1048576m
574 * *1 TB*: 1000000000, 1ti, 1000mi, 1000000ki
576 To specify times (units are not case sensitive):
580 * *M* -- means minutes
581 * *s* -- or sec means seconds (default)
582 * *ms* -- or *msec* means milliseconds
583 * *us* -- or *usec* means microseconds
585 If the option accepts an upper and lower range, use a colon ':' or
586 minus '-' to separate such values. See :ref:`irange <irange>`.
587 If the lower value specified happens to be larger than the upper value
588 the two values are swapped.
593 Boolean. Usually parsed as an integer, however only defined for
594 true and false (1 and 0).
599 Integer range with suffix. Allows value range to be given, such as
600 1024-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
601 option allows two sets of ranges, they can be specified with a ',' or '/'
602 delimiter: 1k-4k/8k-32k. Also see :ref:`int <int>`.
605 A list of floating point numbers, separated by a ':' character.
607 With the above in mind, here follows the complete list of fio job parameters.
613 .. option:: kb_base=int
615 Select the interpretation of unit prefixes in input parameters.
618 Inputs comply with IEC 80000-13 and the International
619 System of Units (SI). Use:
621 - power-of-2 values with IEC prefixes (e.g., KiB)
622 - power-of-10 values with SI prefixes (e.g., kB)
625 Compatibility mode (default). To avoid breaking old scripts:
627 - power-of-2 values with SI prefixes
628 - power-of-10 values with IEC prefixes
630 See :option:`bs` for more details on input parameters.
632 Outputs always use correct prefixes. Most outputs include both
635 bw=2383.3kB/s (2327.4KiB/s)
637 If only one value is reported, then kb_base selects the one to use:
639 **1000** -- SI prefixes
641 **1024** -- IEC prefixes
643 .. option:: unit_base=int
645 Base unit for reporting. Allowed values are:
648 Use auto-detection (default).
660 ASCII name of the job. This may be used to override the name printed by fio
661 for this job. Otherwise the job name is used. On the command line this
662 parameter has the special purpose of also signaling the start of a new job.
664 .. option:: description=str
666 Text description of the job. Doesn't do anything except dump this text
667 description when this job is run. It's not parsed.
669 .. option:: loops=int
671 Run the specified number of iterations of this job. Used to repeat the same
672 workload a given number of times. Defaults to 1.
674 .. option:: numjobs=int
676 Create the specified number of clones of this job. Each clone of job
677 is spawned as an independent thread or process. May be used to setup a
678 larger number of threads/processes doing the same thing. Each thread is
679 reported separately; to see statistics for all clones as a whole, use
680 :option:`group_reporting` in conjunction with :option:`new_group`.
681 See :option:`--max-jobs`. Default: 1.
684 Time related parameters
685 ~~~~~~~~~~~~~~~~~~~~~~~
687 .. option:: runtime=time
689 Tell fio to terminate processing after the specified period of time. It
690 can be quite hard to determine for how long a specified job will run, so
691 this parameter is handy to cap the total runtime to a given time. When
692 the unit is omitted, the value is interpreted in seconds.
694 .. option:: time_based
696 If set, fio will run for the duration of the :option:`runtime` specified
697 even if the file(s) are completely read or written. It will simply loop over
698 the same workload as many times as the :option:`runtime` allows.
700 .. option:: startdelay=irange(time)
702 Delay the start of job for the specified amount of time. Can be a single
703 value or a range. When given as a range, each thread will choose a value
704 randomly from within the range. Value is in seconds if a unit is omitted.
706 .. option:: ramp_time=time
708 If set, fio will run the specified workload for this amount of time before
709 logging any performance numbers. Useful for letting performance settle
710 before logging results, thus minimizing the runtime required for stable
711 results. Note that the ``ramp_time`` is considered lead in time for a job,
712 thus it will increase the total runtime if a special timeout or
713 :option:`runtime` is specified. When the unit is omitted, the value is
716 .. option:: clocksource=str
718 Use the given clocksource as the base of timing. The supported options are:
721 :manpage:`gettimeofday(2)`
724 :manpage:`clock_gettime(2)`
727 Internal CPU clock source
729 cpu is the preferred clocksource if it is reliable, as it is very fast (and
730 fio is heavy on time calls). Fio will automatically use this clocksource if
731 it's supported and considered reliable on the system it is running on,
732 unless another clocksource is specifically set. For x86/x86-64 CPUs, this
733 means supporting TSC Invariant.
735 .. option:: gtod_reduce=bool
737 Enable all of the :manpage:`gettimeofday(2)` reducing options
738 (:option:`disable_clat`, :option:`disable_slat`, :option:`disable_bw_measurement`) plus
739 reduce precision of the timeout somewhat to really shrink the
740 :manpage:`gettimeofday(2)` call count. With this option enabled, we only do
741 about 0.4% of the :manpage:`gettimeofday(2)` calls we would have done if all
742 time keeping was enabled.
744 .. option:: gtod_cpu=int
746 Sometimes it's cheaper to dedicate a single thread of execution to just
747 getting the current time. Fio (and databases, for instance) are very
748 intensive on :manpage:`gettimeofday(2)` calls. With this option, you can set
749 one CPU aside for doing nothing but logging current time to a shared memory
750 location. Then the other threads/processes that run I/O workloads need only
751 copy that segment, instead of entering the kernel with a
752 :manpage:`gettimeofday(2)` call. The CPU set aside for doing these time
753 calls will be excluded from other uses. Fio will manually clear it from the
754 CPU mask of other jobs.
760 .. option:: directory=str
762 Prefix filenames with this directory. Used to place files in a different
763 location than :file:`./`. You can specify a number of directories by
764 separating the names with a ':' character. These directories will be
765 assigned equally distributed to job clones created by :option:`numjobs` as
766 long as they are using generated filenames. If specific `filename(s)` are
767 set fio will use the first listed directory, and thereby matching the
768 `filename` semantic (which generates a file for each clone if not
769 specified, but lets all clones use the same file if set).
771 See the :option:`filename` option for information on how to escape "``:``"
772 characters within the directory path itself.
774 Note: To control the directory fio will use for internal state files
775 use :option:`--aux-path`.
777 .. option:: filename=str
779 Fio normally makes up a `filename` based on the job name, thread number, and
780 file number (see :option:`filename_format`). If you want to share files
781 between threads in a job or several
782 jobs with fixed file paths, specify a `filename` for each of them to override
783 the default. If the ioengine is file based, you can specify a number of files
784 by separating the names with a ':' colon. So if you wanted a job to open
785 :file:`/dev/sda` and :file:`/dev/sdb` as the two working files, you would use
786 ``filename=/dev/sda:/dev/sdb``. This also means that whenever this option is
787 specified, :option:`nrfiles` is ignored. The size of regular files specified
788 by this option will be :option:`size` divided by number of files unless an
789 explicit size is specified by :option:`filesize`.
791 Each colon in the wanted path must be escaped with a ``\``
792 character. For instance, if the path is :file:`/dev/dsk/foo@3,0:c` then you
793 would use ``filename=/dev/dsk/foo@3,0\:c`` and if the path is
794 :file:`F:\\filename` then you would use ``filename=F\:\filename``.
796 On Windows, disk devices are accessed as :file:`\\\\.\\PhysicalDrive0` for
797 the first device, :file:`\\\\.\\PhysicalDrive1` for the second etc.
798 Note: Windows and FreeBSD prevent write access to areas
799 of the disk containing in-use data (e.g. filesystems).
801 The filename "`-`" is a reserved name, meaning *stdin* or *stdout*. Which
802 of the two depends on the read/write direction set.
804 .. option:: filename_format=str
806 If sharing multiple files between jobs, it is usually necessary to have fio
807 generate the exact names that you want. By default, fio will name a file
808 based on the default file format specification of
809 :file:`jobname.jobnumber.filenumber`. With this option, that can be
810 customized. Fio will recognize and replace the following keywords in this
814 The name of the worker thread or process.
816 IP of the fio process when using client/server mode.
818 The incremental number of the worker thread or process.
820 The incremental number of the file for that worker thread or
823 To have dependent jobs share a set of files, this option can be set to have
824 fio generate filenames that are shared between the two. For instance, if
825 :file:`testfiles.$filenum` is specified, file number 4 for any job will be
826 named :file:`testfiles.4`. The default of :file:`$jobname.$jobnum.$filenum`
827 will be used if no other format specifier is given.
829 If you specify a path then the directories will be created up to the
830 main directory for the file. So for example if you specify
831 ``filename_format=a/b/c/$jobnum`` then the directories a/b/c will be
832 created before the file setup part of the job. If you specify
833 :option:`directory` then the path will be relative that directory,
834 otherwise it is treated as the absolute path.
836 .. option:: unique_filename=bool
838 To avoid collisions between networked clients, fio defaults to prefixing any
839 generated filenames (with a directory specified) with the source of the
840 client connecting. To disable this behavior, set this option to 0.
842 .. option:: opendir=str
844 Recursively open any files below directory `str`.
846 .. option:: lockfile=str
848 Fio defaults to not locking any files before it does I/O to them. If a file
849 or file descriptor is shared, fio can serialize I/O to that file to make the
850 end result consistent. This is usual for emulating real workloads that share
851 files. The lock modes are:
854 No locking. The default.
856 Only one thread or process may do I/O at a time, excluding all
859 Read-write locking on the file. Many readers may
860 access the file at the same time, but writes get exclusive access.
862 .. option:: nrfiles=int
864 Number of files to use for this job. Defaults to 1. The size of files
865 will be :option:`size` divided by this unless explicit size is specified by
866 :option:`filesize`. Files are created for each thread separately, and each
867 file will have a file number within its name by default, as explained in
868 :option:`filename` section.
871 .. option:: openfiles=int
873 Number of files to keep open at the same time. Defaults to the same as
874 :option:`nrfiles`, can be set smaller to limit the number simultaneous
877 .. option:: file_service_type=str
879 Defines how fio decides which file from a job to service next. The following
883 Choose a file at random.
886 Round robin over opened files. This is the default.
889 Finish one file before moving on to the next. Multiple files can
890 still be open depending on :option:`openfiles`.
893 Use a *Zipf* distribution to decide what file to access.
896 Use a *Pareto* distribution to decide what file to access.
899 Use a *Gaussian* (normal) distribution to decide what file to
905 For *random*, *roundrobin*, and *sequential*, a postfix can be appended to
906 tell fio how many I/Os to issue before switching to a new file. For example,
907 specifying ``file_service_type=random:8`` would cause fio to issue
908 8 I/Os before selecting a new file at random. For the non-uniform
909 distributions, a floating point postfix can be given to influence how the
910 distribution is skewed. See :option:`random_distribution` for a description
911 of how that would work.
913 .. option:: ioscheduler=str
915 Attempt to switch the device hosting the file to the specified I/O scheduler
918 .. option:: create_serialize=bool
920 If true, serialize the file creation for the jobs. This may be handy to
921 avoid interleaving of data files, which may greatly depend on the filesystem
922 used and even the number of processors in the system. Default: true.
924 .. option:: create_fsync=bool
926 :manpage:`fsync(2)` the data file after creation. This is the default.
928 .. option:: create_on_open=bool
930 If true, don't pre-create files but allow the job's open() to create a file
931 when it's time to do I/O. Default: false -- pre-create all necessary files
934 .. option:: create_only=bool
936 If true, fio will only run the setup phase of the job. If files need to be
937 laid out or updated on disk, only that will be done -- the actual job contents
938 are not executed. Default: false.
940 .. option:: allow_file_create=bool
942 If true, fio is permitted to create files as part of its workload. If this
943 option is false, then fio will error out if
944 the files it needs to use don't already exist. Default: true.
946 .. option:: allow_mounted_write=bool
948 If this isn't set, fio will abort jobs that are destructive (e.g. that write)
949 to what appears to be a mounted device or partition. This should help catch
950 creating inadvertently destructive tests, not realizing that the test will
951 destroy data on the mounted file system. Note that some platforms don't allow
952 writing against a mounted device regardless of this option. Default: false.
954 .. option:: pre_read=bool
956 If this is given, files will be pre-read into memory before starting the
957 given I/O operation. This will also clear the :option:`invalidate` flag,
958 since it is pointless to pre-read and then drop the cache. This will only
959 work for I/O engines that are seek-able, since they allow you to read the
960 same data multiple times. Thus it will not work on non-seekable I/O engines
961 (e.g. network, splice). Default: false.
963 .. option:: unlink=bool
965 Unlink the job files when done. Not the default, as repeated runs of that
966 job would then waste time recreating the file set again and again. Default:
969 .. option:: unlink_each_loop=bool
971 Unlink job files after each iteration or loop. Default: false.
973 .. option:: zonemode=str
978 The :option:`zonerange`, :option:`zonesize`,
979 :option `zonecapacity` and option:`zoneskip`
980 parameters are ignored.
982 I/O happens in a single zone until
983 :option:`zonesize` bytes have been transferred.
984 After that number of bytes has been
985 transferred processing of the next zone
986 starts. :option `zonecapacity` is ignored.
988 Zoned block device mode. I/O happens
989 sequentially in each zone, even if random I/O
990 has been selected. Random I/O happens across
991 all zones instead of being restricted to a
992 single zone. The :option:`zoneskip` parameter
993 is ignored. :option:`zonerange` and
994 :option:`zonesize` must be identical.
995 Trim is handled using a zone reset operation.
996 Trim only considers non-empty sequential write
997 required and sequential write preferred zones.
999 .. option:: zonerange=int
1001 Size of a single zone. See also :option:`zonesize` and
1004 .. option:: zonesize=int
1006 For :option:`zonemode` =strided, this is the number of bytes to
1007 transfer before skipping :option:`zoneskip` bytes. If this parameter
1008 is smaller than :option:`zonerange` then only a fraction of each zone
1009 with :option:`zonerange` bytes will be accessed. If this parameter is
1010 larger than :option:`zonerange` then each zone will be accessed
1011 multiple times before skipping to the next zone.
1013 For :option:`zonemode` =zbd, this is the size of a single zone. The
1014 :option:`zonerange` parameter is ignored in this mode.
1017 .. option:: zonecapacity=int
1019 For :option:`zonemode` =zbd, this defines the capacity of a single zone,
1020 which is the accessible area starting from the zone start address.
1021 This parameter only applies when using :option:`zonemode` =zbd in
1022 combination with regular block devices. If not specified it defaults to
1023 the zone size. If the target device is a zoned block device, the zone
1024 capacity is obtained from the device information and this option is
1027 .. option:: zoneskip=int
1029 For :option:`zonemode` =strided, the number of bytes to skip after
1030 :option:`zonesize` bytes of data have been transferred. This parameter
1031 must be zero for :option:`zonemode` =zbd.
1033 .. option:: read_beyond_wp=bool
1035 This parameter applies to :option:`zonemode` =zbd only.
1037 Zoned block devices are block devices that consist of multiple zones.
1038 Each zone has a type, e.g. conventional or sequential. A conventional
1039 zone can be written at any offset that is a multiple of the block
1040 size. Sequential zones must be written sequentially. The position at
1041 which a write must occur is called the write pointer. A zoned block
1042 device can be either drive managed, host managed or host aware. For
1043 host managed devices the host must ensure that writes happen
1044 sequentially. Fio recognizes host managed devices and serializes
1045 writes to sequential zones for these devices.
1047 If a read occurs in a sequential zone beyond the write pointer then
1048 the zoned block device will complete the read without reading any data
1049 from the storage medium. Since such reads lead to unrealistically high
1050 bandwidth and IOPS numbers fio only reads beyond the write pointer if
1051 explicitly told to do so. Default: false.
1053 .. option:: max_open_zones=int
1055 When running a random write test across an entire drive many more
1056 zones will be open than in a typical application workload. Hence this
1057 command line option that allows to limit the number of open zones. The
1058 number of open zones is defined as the number of zones to which write
1059 commands are issued.
1061 .. option:: job_max_open_zones=int
1063 Limit on the number of simultaneously opened zones per single
1066 .. option:: ignore_zone_limits=bool
1068 If this option is used, fio will ignore the maximum number of open
1069 zones limit of the zoned block device in use, thus allowing the
1070 option :option:`max_open_zones` value to be larger than the device
1071 reported limit. Default: false.
1073 .. option:: zone_reset_threshold=float
1075 A number between zero and one that indicates the ratio of logical
1076 blocks with data to the total number of logical blocks in the test
1077 above which zones should be reset periodically.
1079 .. option:: zone_reset_frequency=float
1081 A number between zero and one that indicates how often a zone reset
1082 should be issued if the zone reset threshold has been exceeded. A zone
1083 reset is submitted after each (1 / zone_reset_frequency) write
1084 requests. This and the previous parameter can be used to simulate
1085 garbage collection activity.
1091 .. option:: direct=bool
1093 If value is true, use non-buffered I/O. This is usually O_DIRECT. Note that
1094 OpenBSD and ZFS on Solaris don't support direct I/O. On Windows the synchronous
1095 ioengines don't support direct I/O. Default: false.
1097 .. option:: atomic=bool
1099 If value is true, attempt to use atomic direct I/O. Atomic writes are
1100 guaranteed to be stable once acknowledged by the operating system. Only
1101 Linux supports O_ATOMIC right now.
1103 .. option:: buffered=bool
1105 If value is true, use buffered I/O. This is the opposite of the
1106 :option:`direct` option. Defaults to true.
1108 .. option:: readwrite=str, rw=str
1110 Type of I/O pattern. Accepted values are:
1117 Sequential trims (Linux block devices and SCSI
1118 character devices only).
1124 Random trims (Linux block devices and SCSI
1125 character devices only).
1127 Sequential mixed reads and writes.
1129 Random mixed reads and writes.
1131 Sequential trim+write sequences. Blocks will be trimmed first,
1132 then the same blocks will be written to. So if ``io_size=64K``
1133 is specified, Fio will trim a total of 64K bytes and also
1134 write 64K bytes on the same trimmed blocks. This behaviour
1135 will be consistent with ``number_ios`` or other Fio options
1136 limiting the total bytes or number of I/O's.
1138 Like trimwrite, but uses random offsets rather
1139 than sequential writes.
1141 Fio defaults to read if the option is not specified. For the mixed I/O
1142 types, the default is to split them 50/50. For certain types of I/O the
1143 result may still be skewed a bit, since the speed may be different.
1145 It is possible to specify the number of I/Os to do before getting a new
1146 offset by appending ``:<nr>`` to the end of the string given. For a
1147 random read, it would look like ``rw=randread:8`` for passing in an offset
1148 modifier with a value of 8. If the suffix is used with a sequential I/O
1149 pattern, then the *<nr>* value specified will be **added** to the generated
1150 offset for each I/O turning sequential I/O into sequential I/O with holes.
1151 For instance, using ``rw=write:4k`` will skip 4k for every write. Also see
1152 the :option:`rw_sequencer` option.
1154 .. option:: rw_sequencer=str
1156 If an offset modifier is given by appending a number to the ``rw=<str>``
1157 line, then this option controls how that number modifies the I/O offset
1158 being generated. Accepted values are:
1161 Generate sequential offset.
1163 Generate the same offset.
1165 ``sequential`` is only useful for random I/O, where fio would normally
1166 generate a new random offset for every I/O. If you append e.g. 8 to randread,
1167 you would get a new random offset for every 8 I/Os. The result would be a
1168 seek for only every 8 I/Os, instead of for every I/O. Use ``rw=randread:8``
1169 to specify that. As sequential I/O is already sequential, setting
1170 ``sequential`` for that would not result in any differences. ``identical``
1171 behaves in a similar fashion, except it sends the same offset 8 number of
1172 times before generating a new offset.
1174 .. option:: unified_rw_reporting=str
1176 Fio normally reports statistics on a per data direction basis, meaning that
1177 reads, writes, and trims are accounted and reported separately. This option
1178 determines whether fio reports the results normally, summed together, or as
1180 Accepted values are:
1183 Normal statistics reporting.
1186 Statistics are summed per data direction and reported together.
1189 Statistics are reported normally, followed by the mixed statistics.
1192 Backward-compatible alias for **none**.
1195 Backward-compatible alias for **mixed**.
1200 .. option:: randrepeat=bool
1202 Seed the random number generator used for random I/O patterns in a
1203 predictable way so the pattern is repeatable across runs. Default: true.
1205 .. option:: allrandrepeat=bool
1207 Seed all random number generators in a predictable way so results are
1208 repeatable across runs. Default: false.
1210 .. option:: randseed=int
1212 Seed the random number generators based on this seed value, to be able to
1213 control what sequence of output is being generated. If not set, the random
1214 sequence depends on the :option:`randrepeat` setting.
1216 .. option:: fallocate=str
1218 Whether pre-allocation is performed when laying down files.
1219 Accepted values are:
1222 Do not pre-allocate space.
1225 Use a platform's native pre-allocation call but fall back to
1226 **none** behavior if it fails/is not implemented.
1229 Pre-allocate via :manpage:`posix_fallocate(3)`.
1232 Pre-allocate via :manpage:`fallocate(2)` with
1233 FALLOC_FL_KEEP_SIZE set.
1236 Extend file to final size via :manpage:`ftruncate(2)`
1237 instead of allocating.
1240 Backward-compatible alias for **none**.
1243 Backward-compatible alias for **posix**.
1245 May not be available on all supported platforms. **keep** is only available
1246 on Linux. If using ZFS on Solaris this cannot be set to **posix**
1247 because ZFS doesn't support pre-allocation. Default: **native** if any
1248 pre-allocation methods except **truncate** are available, **none** if not.
1250 Note that using **truncate** on Windows will interact surprisingly
1251 with non-sequential write patterns. When writing to a file that has
1252 been extended by setting the end-of-file information, Windows will
1253 backfill the unwritten portion of the file up to that offset with
1254 zeroes before issuing the new write. This means that a single small
1255 write to the end of an extended file will stall until the entire
1256 file has been filled with zeroes.
1258 .. option:: fadvise_hint=str
1260 Use :manpage:`posix_fadvise(2)` or :manpage:`posix_fadvise(2)` to
1261 advise the kernel on what I/O patterns are likely to be issued.
1262 Accepted values are:
1265 Backwards-compatible hint for "no hint".
1268 Backwards compatible hint for "advise with fio workload type". This
1269 uses **FADV_RANDOM** for a random workload, and **FADV_SEQUENTIAL**
1270 for a sequential workload.
1273 Advise using **FADV_SEQUENTIAL**.
1276 Advise using **FADV_RANDOM**.
1278 .. option:: write_hint=str
1280 Use :manpage:`fcntl(2)` to advise the kernel what life time to expect
1281 from a write. Only supported on Linux, as of version 4.13. Accepted
1285 No particular life time associated with this file.
1288 Data written to this file has a short life time.
1291 Data written to this file has a medium life time.
1294 Data written to this file has a long life time.
1297 Data written to this file has a very long life time.
1299 The values are all relative to each other, and no absolute meaning
1300 should be associated with them.
1302 .. option:: offset=int
1304 Start I/O at the provided offset in the file, given as either a fixed size in
1305 bytes, zones or a percentage. If a percentage is given, the generated offset will be
1306 aligned to the minimum ``blocksize`` or to the value of ``offset_align`` if
1307 provided. Data before the given offset will not be touched. This
1308 effectively caps the file size at `real_size - offset`. Can be combined with
1309 :option:`size` to constrain the start and end range of the I/O workload.
1310 A percentage can be specified by a number between 1 and 100 followed by '%',
1311 for example, ``offset=20%`` to specify 20%. In ZBD mode, value can be set as
1312 number of zones using 'z'.
1314 .. option:: offset_align=int
1316 If set to non-zero value, the byte offset generated by a percentage ``offset``
1317 is aligned upwards to this value. Defaults to 0 meaning that a percentage
1318 offset is aligned to the minimum block size.
1320 .. option:: offset_increment=int
1322 If this is provided, then the real offset becomes `offset + offset_increment
1323 * thread_number`, where the thread number is a counter that starts at 0 and
1324 is incremented for each sub-job (i.e. when :option:`numjobs` option is
1325 specified). This option is useful if there are several jobs which are
1326 intended to operate on a file in parallel disjoint segments, with even
1327 spacing between the starting points. Percentages can be used for this option.
1328 If a percentage is given, the generated offset will be aligned to the minimum
1329 ``blocksize`` or to the value of ``offset_align`` if provided. In ZBD mode, value can
1330 also be set as number of zones using 'z'.
1332 .. option:: number_ios=int
1334 Fio will normally perform I/Os until it has exhausted the size of the region
1335 set by :option:`size`, or if it exhaust the allocated time (or hits an error
1336 condition). With this setting, the range/size can be set independently of
1337 the number of I/Os to perform. When fio reaches this number, it will exit
1338 normally and report status. Note that this does not extend the amount of I/O
1339 that will be done, it will only stop fio if this condition is met before
1340 other end-of-job criteria.
1342 .. option:: fsync=int
1344 If writing to a file, issue an :manpage:`fsync(2)` (or its equivalent) of
1345 the dirty data for every number of blocks given. For example, if you give 32
1346 as a parameter, fio will sync the file after every 32 writes issued. If fio is
1347 using non-buffered I/O, we may not sync the file. The exception is the sg
1348 I/O engine, which synchronizes the disk cache anyway. Defaults to 0, which
1349 means fio does not periodically issue and wait for a sync to complete. Also
1350 see :option:`end_fsync` and :option:`fsync_on_close`.
1352 .. option:: fdatasync=int
1354 Like :option:`fsync` but uses :manpage:`fdatasync(2)` to only sync data and
1355 not metadata blocks. In Windows, DragonFlyBSD or OSX there is no
1356 :manpage:`fdatasync(2)` so this falls back to using :manpage:`fsync(2)`.
1357 Defaults to 0, which means fio does not periodically issue and wait for a
1358 data-only sync to complete.
1360 .. option:: write_barrier=int
1362 Make every `N-th` write a barrier write.
1364 .. option:: sync_file_range=str:int
1366 Use :manpage:`sync_file_range(2)` for every `int` number of write
1367 operations. Fio will track range of writes that have happened since the last
1368 :manpage:`sync_file_range(2)` call. `str` can currently be one or more of:
1371 SYNC_FILE_RANGE_WAIT_BEFORE
1373 SYNC_FILE_RANGE_WRITE
1375 SYNC_FILE_RANGE_WAIT_AFTER
1377 So if you do ``sync_file_range=wait_before,write:8``, fio would use
1378 ``SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE`` for every 8
1379 writes. Also see the :manpage:`sync_file_range(2)` man page. This option is
1382 .. option:: overwrite=bool
1384 If true, writes to a file will always overwrite existing data. If the file
1385 doesn't already exist, it will be created before the write phase begins. If
1386 the file exists and is large enough for the specified write phase, nothing
1387 will be done. Default: false.
1389 .. option:: end_fsync=bool
1391 If true, :manpage:`fsync(2)` file contents when a write stage has completed.
1394 .. option:: fsync_on_close=bool
1396 If true, fio will :manpage:`fsync(2)` a dirty file on close. This differs
1397 from :option:`end_fsync` in that it will happen on every file close, not
1398 just at the end of the job. Default: false.
1400 .. option:: rwmixread=int
1402 Percentage of a mixed workload that should be reads. Default: 50.
1404 .. option:: rwmixwrite=int
1406 Percentage of a mixed workload that should be writes. If both
1407 :option:`rwmixread` and :option:`rwmixwrite` is given and the values do not
1408 add up to 100%, the latter of the two will be used to override the
1409 first. This may interfere with a given rate setting, if fio is asked to
1410 limit reads or writes to a certain rate. If that is the case, then the
1411 distribution may be skewed. Default: 50.
1413 .. option:: random_distribution=str:float[:float][,str:float][,str:float]
1415 By default, fio will use a completely uniform random distribution when asked
1416 to perform random I/O. Sometimes it is useful to skew the distribution in
1417 specific ways, ensuring that some parts of the data is more hot than others.
1418 fio includes the following distribution models:
1421 Uniform random distribution
1430 Normal (Gaussian) distribution
1433 Zoned random distribution
1436 Zone absolute random distribution
1438 When using a **zipf** or **pareto** distribution, an input value is also
1439 needed to define the access pattern. For **zipf**, this is the `Zipf
1440 theta`. For **pareto**, it's the `Pareto power`. Fio includes a test
1441 program, :command:`fio-genzipf`, that can be used visualize what the given input
1442 values will yield in terms of hit rates. If you wanted to use **zipf** with
1443 a `theta` of 1.2, you would use ``random_distribution=zipf:1.2`` as the
1444 option. If a non-uniform model is used, fio will disable use of the random
1445 map. For the **normal** distribution, a normal (Gaussian) deviation is
1446 supplied as a value between 0 and 100.
1448 The second, optional float is allowed for **pareto**, **zipf** and **normal** distributions.
1449 It allows to set base of distribution in non-default place, giving more control
1450 over most probable outcome. This value is in range [0-1] which maps linearly to
1451 range of possible random values.
1452 Defaults are: random for **pareto** and **zipf**, and 0.5 for **normal**.
1453 If you wanted to use **zipf** with a `theta` of 1.2 centered on 1/4 of allowed value range,
1454 you would use ``random_distribution=zipf:1.2:0.25``.
1456 For a **zoned** distribution, fio supports specifying percentages of I/O
1457 access that should fall within what range of the file or device. For
1458 example, given a criteria of:
1460 * 60% of accesses should be to the first 10%
1461 * 30% of accesses should be to the next 20%
1462 * 8% of accesses should be to the next 30%
1463 * 2% of accesses should be to the next 40%
1465 we can define that through zoning of the random accesses. For the above
1466 example, the user would do::
1468 random_distribution=zoned:60/10:30/20:8/30:2/40
1470 A **zoned_abs** distribution works exactly like the **zoned**, except
1471 that it takes absolute sizes. For example, let's say you wanted to
1472 define access according to the following criteria:
1474 * 60% of accesses should be to the first 20G
1475 * 30% of accesses should be to the next 100G
1476 * 10% of accesses should be to the next 500G
1478 we can define an absolute zoning distribution with:
1480 random_distribution=zoned_abs=60/20G:30/100G:10/500g
1482 For both **zoned** and **zoned_abs**, fio supports defining up to
1485 Similarly to how :option:`bssplit` works for setting ranges and
1486 percentages of block sizes. Like :option:`bssplit`, it's possible to
1487 specify separate zones for reads, writes, and trims. If just one set
1488 is given, it'll apply to all of them. This goes for both **zoned**
1489 **zoned_abs** distributions.
1491 .. option:: percentage_random=int[,int][,int]
1493 For a random workload, set how big a percentage should be random. This
1494 defaults to 100%, in which case the workload is fully random. It can be set
1495 from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1496 sequential. Any setting in between will result in a random mix of sequential
1497 and random I/O, at the given percentages. Comma-separated values may be
1498 specified for reads, writes, and trims as described in :option:`blocksize`.
1500 .. option:: norandommap
1502 Normally fio will cover every block of the file when doing random I/O. If
1503 this option is given, fio will just get a new random offset without looking
1504 at past I/O history. This means that some blocks may not be read or written,
1505 and that some blocks may be read/written more than once. If this option is
1506 used with :option:`verify` and multiple blocksizes (via :option:`bsrange`),
1507 only intact blocks are verified, i.e., partially-overwritten blocks are
1508 ignored. With an async I/O engine and an I/O depth > 1, it is possible for
1509 the same block to be overwritten, which can cause verification errors. Either
1510 do not use norandommap in this case, or also use the lfsr random generator.
1512 .. option:: softrandommap=bool
1514 See :option:`norandommap`. If fio runs with the random block map enabled and
1515 it fails to allocate the map, if this option is set it will continue without
1516 a random block map. As coverage will not be as complete as with random maps,
1517 this option is disabled by default.
1519 .. option:: random_generator=str
1521 Fio supports the following engines for generating I/O offsets for random I/O:
1524 Strong 2^88 cycle random number generator.
1526 Linear feedback shift register generator.
1528 Strong 64-bit 2^258 cycle random number generator.
1530 **tausworthe** is a strong random number generator, but it requires tracking
1531 on the side if we want to ensure that blocks are only read or written
1532 once. **lfsr** guarantees that we never generate the same offset twice, and
1533 it's also less computationally expensive. It's not a true random generator,
1534 however, though for I/O purposes it's typically good enough. **lfsr** only
1535 works with single block sizes, not with workloads that use multiple block
1536 sizes. If used with such a workload, fio may read or write some blocks
1537 multiple times. The default value is **tausworthe**, unless the required
1538 space exceeds 2^32 blocks. If it does, then **tausworthe64** is
1539 selected automatically.
1545 .. option:: blocksize=int[,int][,int], bs=int[,int][,int]
1547 The block size in bytes used for I/O units. Default: 4096. A single value
1548 applies to reads, writes, and trims. Comma-separated values may be
1549 specified for reads, writes, and trims. A value not terminated in a comma
1550 applies to subsequent types.
1555 means 256k for reads, writes and trims.
1558 means 8k for reads, 32k for writes and trims.
1561 means 8k for reads, 32k for writes, and default for trims.
1564 means default for reads, 8k for writes and trims.
1567 means default for reads, 8k for writes, and default for trims.
1569 .. option:: blocksize_range=irange[,irange][,irange], bsrange=irange[,irange][,irange]
1571 A range of block sizes in bytes for I/O units. The issued I/O unit will
1572 always be a multiple of the minimum size, unless
1573 :option:`blocksize_unaligned` is set.
1575 Comma-separated ranges may be specified for reads, writes, and trims as
1576 described in :option:`blocksize`.
1578 Example: ``bsrange=1k-4k,2k-8k``.
1580 .. option:: bssplit=str[,str][,str]
1582 Sometimes you want even finer grained control of the block sizes
1583 issued, not just an even split between them. This option allows you to
1584 weight various block sizes, so that you are able to define a specific
1585 amount of block sizes issued. The format for this option is::
1587 bssplit=blocksize/percentage:blocksize/percentage
1589 for as many block sizes as needed. So if you want to define a workload
1590 that has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would
1593 bssplit=4k/10:64k/50:32k/40
1595 Ordering does not matter. If the percentage is left blank, fio will
1596 fill in the remaining values evenly. So a bssplit option like this one::
1598 bssplit=4k/50:1k/:32k/
1600 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always
1601 add up to 100, if bssplit is given a range that adds up to more, it
1604 Comma-separated values may be specified for reads, writes, and trims as
1605 described in :option:`blocksize`.
1607 If you want a workload that has 50% 2k reads and 50% 4k reads, while
1608 having 90% 4k writes and 10% 8k writes, you would specify::
1610 bssplit=2k/50:4k/50,4k/90:8k/10
1612 Fio supports defining up to 64 different weights for each data
1615 .. option:: blocksize_unaligned, bs_unaligned
1617 If set, fio will issue I/O units with any size within
1618 :option:`blocksize_range`, not just multiples of the minimum size. This
1619 typically won't work with direct I/O, as that normally requires sector
1622 .. option:: bs_is_seq_rand=bool
1624 If this option is set, fio will use the normal read,write blocksize settings
1625 as sequential,random blocksize settings instead. Any random read or write
1626 will use the WRITE blocksize settings, and any sequential read or write will
1627 use the READ blocksize settings.
1629 .. option:: blockalign=int[,int][,int], ba=int[,int][,int]
1631 Boundary to which fio will align random I/O units. Default:
1632 :option:`blocksize`. Minimum alignment is typically 512b for using direct
1633 I/O, though it usually depends on the hardware block size. This option is
1634 mutually exclusive with using a random map for files, so it will turn off
1635 that option. Comma-separated values may be specified for reads, writes, and
1636 trims as described in :option:`blocksize`.
1642 .. option:: zero_buffers
1644 Initialize buffers with all zeros. Default: fill buffers with random data.
1646 .. option:: refill_buffers
1648 If this option is given, fio will refill the I/O buffers on every
1649 submit. Only makes sense if :option:`zero_buffers` isn't specified,
1650 naturally. Defaults to being unset i.e., the buffer is only filled at
1651 init time and the data in it is reused when possible but if any of
1652 :option:`verify`, :option:`buffer_compress_percentage` or
1653 :option:`dedupe_percentage` are enabled then `refill_buffers` is also
1654 automatically enabled.
1656 .. option:: scramble_buffers=bool
1658 If :option:`refill_buffers` is too costly and the target is using data
1659 deduplication, then setting this option will slightly modify the I/O buffer
1660 contents to defeat normal de-dupe attempts. This is not enough to defeat
1661 more clever block compression attempts, but it will stop naive dedupe of
1662 blocks. Default: true.
1664 .. option:: buffer_compress_percentage=int
1666 If this is set, then fio will attempt to provide I/O buffer content
1667 (on WRITEs) that compresses to the specified level. Fio does this by
1668 providing a mix of random data followed by fixed pattern data. The
1669 fixed pattern is either zeros, or the pattern specified by
1670 :option:`buffer_pattern`. If the `buffer_pattern` option is used, it
1671 might skew the compression ratio slightly. Setting
1672 `buffer_compress_percentage` to a value other than 100 will also
1673 enable :option:`refill_buffers` in order to reduce the likelihood that
1674 adjacent blocks are so similar that they over compress when seen
1675 together. See :option:`buffer_compress_chunk` for how to set a finer or
1676 coarser granularity for the random/fixed data region. Defaults to unset
1677 i.e., buffer data will not adhere to any compression level.
1679 .. option:: buffer_compress_chunk=int
1681 This setting allows fio to manage how big the random/fixed data region
1682 is when using :option:`buffer_compress_percentage`. When
1683 `buffer_compress_chunk` is set to some non-zero value smaller than the
1684 block size, fio can repeat the random/fixed region throughout the I/O
1685 buffer at the specified interval (which particularly useful when
1686 bigger block sizes are used for a job). When set to 0, fio will use a
1687 chunk size that matches the block size resulting in a single
1688 random/fixed region within the I/O buffer. Defaults to 512. When the
1689 unit is omitted, the value is interpreted in bytes.
1691 .. option:: buffer_pattern=str
1693 If set, fio will fill the I/O buffers with this pattern or with the contents
1694 of a file. If not set, the contents of I/O buffers are defined by the other
1695 options related to buffer contents. The setting can be any pattern of bytes,
1696 and can be prefixed with 0x for hex values. It may also be a string, where
1697 the string must then be wrapped with ``""``. Or it may also be a filename,
1698 where the filename must be wrapped with ``''`` in which case the file is
1699 opened and read. Note that not all the file contents will be read if that
1700 would cause the buffers to overflow. So, for example::
1702 buffer_pattern='filename'
1706 buffer_pattern="abcd"
1714 buffer_pattern=0xdeadface
1716 Also you can combine everything together in any order::
1718 buffer_pattern=0xdeadface"abcd"-12'filename'
1720 .. option:: dedupe_percentage=int
1722 If set, fio will generate this percentage of identical buffers when
1723 writing. These buffers will be naturally dedupable. The contents of the
1724 buffers depend on what other buffer compression settings have been set. It's
1725 possible to have the individual buffers either fully compressible, or not at
1726 all -- this option only controls the distribution of unique buffers. Setting
1727 this option will also enable :option:`refill_buffers` to prevent every buffer
1730 .. option:: dedupe_mode=str
1732 If ``dedupe_percentage=<int>`` is given, then this option controls how fio
1733 generates the dedupe buffers.
1736 Generate dedupe buffers by repeating previous writes
1738 Generate dedupe buffers from working set
1740 ``repeat`` is the default option for fio. Dedupe buffers are generated
1741 by repeating previous unique write.
1743 ``working_set`` is a more realistic workload.
1744 With ``working_set``, ``dedupe_working_set_percentage=<int>`` should be provided.
1745 Given that, fio will use the initial unique write buffers as its working set.
1746 Upon deciding to dedupe, fio will randomly choose a buffer from the working set.
1747 Note that by using ``working_set`` the dedupe percentage will converge
1748 to the desired over time while ``repeat`` maintains the desired percentage
1751 .. option:: dedupe_working_set_percentage=int
1753 If ``dedupe_mode=<str>`` is set to ``working_set``, then this controls
1754 the percentage of size of the file or device used as the buffers
1755 fio will choose to generate the dedupe buffers from
1757 Note that size needs to be explicitly provided and only 1 file per
1760 .. option:: dedupe_global=bool
1762 This controls whether the deduplication buffers will be shared amongst
1763 all jobs that have this option set. The buffers are spread evenly between
1766 .. option:: invalidate=bool
1768 Invalidate the buffer/page cache parts of the files to be used prior to
1769 starting I/O if the platform and file type support it. Defaults to true.
1770 This will be ignored if :option:`pre_read` is also specified for the
1773 .. option:: sync=str
1775 Whether, and what type, of synchronous I/O to use for writes. The allowed
1779 Do not use synchronous IO, the default.
1785 Use synchronous file IO. For the majority of I/O engines,
1786 this means using O_SYNC.
1792 Use synchronous data IO. For the majority of I/O engines,
1793 this means using O_DSYNC.
1796 .. option:: iomem=str, mem=str
1798 Fio can use various types of memory as the I/O unit buffer. The allowed
1802 Use memory from :manpage:`malloc(3)` as the buffers. Default memory
1806 Use shared memory as the buffers. Allocated through
1807 :manpage:`shmget(2)`.
1810 Same as shm, but use huge pages as backing.
1813 Use :manpage:`mmap(2)` to allocate buffers. May either be anonymous memory, or can
1814 be file backed if a filename is given after the option. The format
1815 is `mem=mmap:/path/to/file`.
1818 Use a memory mapped huge file as the buffer backing. Append filename
1819 after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file`.
1822 Same as mmap, but use a MMAP_SHARED mapping.
1825 Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1826 The :option:`ioengine` must be `rdma`.
1828 The area allocated is a function of the maximum allowed bs size for the job,
1829 multiplied by the I/O depth given. Note that for **shmhuge** and
1830 **mmaphuge** to work, the system must have free huge pages allocated. This
1831 can normally be checked and set by reading/writing
1832 :file:`/proc/sys/vm/nr_hugepages` on a Linux system. Fio assumes a huge page
1833 is 2 or 4MiB in size depending on the platform. So to calculate the
1834 number of huge pages you need for a given job file, add up the I/O
1835 depth of all jobs (normally one unless :option:`iodepth` is used) and
1836 multiply by the maximum bs set. Then divide that number by the huge
1837 page size. You can see the size of the huge pages in
1838 :file:`/proc/meminfo`. If no huge pages are allocated by having a
1839 non-zero number in `nr_hugepages`, using **mmaphuge** or **shmhuge**
1840 will fail. Also see :option:`hugepage-size`.
1842 **mmaphuge** also needs to have hugetlbfs mounted and the file location
1843 should point there. So if it's mounted in :file:`/huge`, you would use
1844 `mem=mmaphuge:/huge/somefile`.
1846 .. option:: iomem_align=int, mem_align=int
1848 This indicates the memory alignment of the I/O memory buffers. Note that
1849 the given alignment is applied to the first I/O unit buffer, if using
1850 :option:`iodepth` the alignment of the following buffers are given by the
1851 :option:`bs` used. In other words, if using a :option:`bs` that is a
1852 multiple of the page sized in the system, all buffers will be aligned to
1853 this value. If using a :option:`bs` that is not page aligned, the alignment
1854 of subsequent I/O memory buffers is the sum of the :option:`iomem_align` and
1857 .. option:: hugepage-size=int
1859 Defines the size of a huge page. Must at least be equal to the system
1860 setting, see :file:`/proc/meminfo` and
1861 :file:`/sys/kernel/mm/hugepages/`. Defaults to 2 or 4MiB depending on
1862 the platform. Should probably always be a multiple of megabytes, so
1863 using ``hugepage-size=Xm`` is the preferred way to set this to avoid
1864 setting a non-pow-2 bad value.
1866 .. option:: lockmem=int
1868 Pin the specified amount of memory with :manpage:`mlock(2)`. Can be used to
1869 simulate a smaller amount of memory. The amount specified is per worker.
1875 .. option:: size=int
1877 The total size of file I/O for each thread of this job. Fio will run until
1878 this many bytes has been transferred, unless runtime is limited by other options
1879 (such as :option:`runtime`, for instance, or increased/decreased by :option:`io_size`).
1880 Fio will divide this size between the available files determined by options
1881 such as :option:`nrfiles`, :option:`filename`, unless :option:`filesize` is
1882 specified by the job. If the result of division happens to be 0, the size is
1883 set to the physical size of the given files or devices if they exist.
1884 If this option is not specified, fio will use the full size of the given
1885 files or devices. If the files do not exist, size must be given. It is also
1886 possible to give size as a percentage between 1 and 100. If ``size=20%`` is
1887 given, fio will use 20% of the full size of the given files or devices.
1888 In ZBD mode, value can also be set as number of zones using 'z'.
1889 Can be combined with :option:`offset` to constrain the start and end range
1890 that I/O will be done within.
1892 .. option:: io_size=int, io_limit=int
1894 Normally fio operates within the region set by :option:`size`, which means
1895 that the :option:`size` option sets both the region and size of I/O to be
1896 performed. Sometimes that is not what you want. With this option, it is
1897 possible to define just the amount of I/O that fio should do. For instance,
1898 if :option:`size` is set to 20GiB and :option:`io_size` is set to 5GiB, fio
1899 will perform I/O within the first 20GiB but exit when 5GiB have been
1900 done. The opposite is also possible -- if :option:`size` is set to 20GiB,
1901 and :option:`io_size` is set to 40GiB, then fio will do 40GiB of I/O within
1902 the 0..20GiB region.
1904 .. option:: filesize=irange(int)
1906 Individual file sizes. May be a range, in which case fio will select sizes for
1907 files at random within the given range. If not given, each created file is the
1908 same size. This option overrides :option:`size` in terms of file size, i.e. if
1909 :option:`filesize` is specified then :option:`size` becomes merely the default
1910 for :option:`io_size` and has no effect at all if :option:`io_size` is set
1913 .. option:: file_append=bool
1915 Perform I/O after the end of the file. Normally fio will operate within the
1916 size of a file. If this option is set, then fio will append to the file
1917 instead. This has identical behavior to setting :option:`offset` to the size
1918 of a file. This option is ignored on non-regular files.
1920 .. option:: fill_device=bool, fill_fs=bool
1922 Sets size to something really large and waits for ENOSPC (no space left on
1923 device) or EDQUOT (disk quota exceeded)
1924 as the terminating condition. Only makes sense with sequential
1925 write. For a read workload, the mount point will be filled first then I/O
1926 started on the result. This option doesn't make sense if operating on a raw
1927 device node, since the size of that is already known by the file system.
1928 Additionally, writing beyond end-of-device will not return ENOSPC there.
1934 .. option:: ioengine=str
1936 Defines how the job issues I/O to the file. The following types are defined:
1939 Basic :manpage:`read(2)` or :manpage:`write(2)`
1940 I/O. :manpage:`lseek(2)` is used to position the I/O location.
1941 See :option:`fsync` and :option:`fdatasync` for syncing write I/Os.
1944 Basic :manpage:`pread(2)` or :manpage:`pwrite(2)` I/O. Default on
1945 all supported operating systems except for Windows.
1948 Basic :manpage:`readv(2)` or :manpage:`writev(2)` I/O. Will emulate
1949 queuing by coalescing adjacent I/Os into a single submission.
1952 Basic :manpage:`preadv(2)` or :manpage:`pwritev(2)` I/O.
1955 Basic :manpage:`preadv2(2)` or :manpage:`pwritev2(2)` I/O.
1958 Fast Linux native asynchronous I/O. Supports async IO
1959 for both direct and buffered IO.
1960 This engine defines engine specific options.
1963 Fast Linux native asynchronous I/O for pass through commands.
1964 This engine defines engine specific options.
1967 Linux native asynchronous I/O. Note that Linux may only support
1968 queued behavior with non-buffered I/O (set ``direct=1`` or
1970 This engine defines engine specific options.
1973 POSIX asynchronous I/O using :manpage:`aio_read(3)` and
1974 :manpage:`aio_write(3)`.
1977 Solaris native asynchronous I/O.
1980 Windows native asynchronous I/O. Default on Windows.
1983 File is memory mapped with :manpage:`mmap(2)` and data copied
1984 to/from using :manpage:`memcpy(3)`.
1987 :manpage:`splice(2)` is used to transfer the data and
1988 :manpage:`vmsplice(2)` to transfer data from user space to the
1992 SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1993 ioctl, or if the target is an sg character device we use
1994 :manpage:`read(2)` and :manpage:`write(2)` for asynchronous
1995 I/O. Requires :option:`filename` option to specify either block or
1996 character devices. This engine supports trim operations.
1997 The sg engine includes engine specific options.
2000 Read, write, trim and ZBC/ZAC operations to a zoned
2001 block device using libzbc library. The target can be
2002 either an SG character device or a block device file.
2005 Doesn't transfer any data, just pretends to. This is mainly used to
2006 exercise fio itself and for debugging/testing purposes.
2009 Transfer over the network to given ``host:port``. Depending on the
2010 :option:`protocol` used, the :option:`hostname`, :option:`port`,
2011 :option:`listen` and :option:`filename` options are used to specify
2012 what sort of connection to make, while the :option:`protocol` option
2013 determines which protocol will be used. This engine defines engine
2017 Like **net**, but uses :manpage:`splice(2)` and
2018 :manpage:`vmsplice(2)` to map data and send/receive.
2019 This engine defines engine specific options.
2022 Doesn't transfer any data, but burns CPU cycles according to the
2023 :option:`cpuload`, :option:`cpuchunks` and :option:`cpumode` options.
2024 Setting :option:`cpuload`\=85 will cause that job to do nothing but burn 85%
2025 of the CPU. In case of SMP machines, use :option:`numjobs`\=<nr_of_cpu>
2026 to get desired CPU usage, as the cpuload only loads a
2027 single CPU at the desired rate. A job never finishes unless there is
2028 at least one non-cpuio job.
2029 Setting :option:`cpumode`\=qsort replace the default noop instructions loop
2030 by a qsort algorithm to consume more energy.
2033 The RDMA I/O engine supports both RDMA memory semantics
2034 (RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
2035 InfiniBand, RoCE and iWARP protocols. This engine defines engine
2039 I/O engine that does regular fallocate to simulate data transfer as
2043 does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
2046 does fallocate(,mode = 0).
2049 does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
2052 I/O engine that sends :manpage:`ftruncate(2)` operations in response
2053 to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
2054 size to the current block offset. :option:`blocksize` is ignored.
2057 I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
2058 defragment activity in request to DDIR_WRITE event.
2061 I/O engine supporting direct access to Ceph Reliable Autonomic
2062 Distributed Object Store (RADOS) via librados. This ioengine
2063 defines engine specific options.
2066 I/O engine supporting direct access to Ceph Rados Block Devices
2067 (RBD) via librbd without the need to use the kernel rbd driver. This
2068 ioengine defines engine specific options.
2071 I/O engine supporting GET/PUT requests over HTTP(S) with libcurl to
2072 a WebDAV or S3 endpoint. This ioengine defines engine specific options.
2074 This engine only supports direct IO of iodepth=1; you need to scale this
2075 via numjobs. blocksize defines the size of the objects to be created.
2077 TRIM is translated to object deletion.
2080 Using GlusterFS libgfapi sync interface to direct access to
2081 GlusterFS volumes without having to go through FUSE. This ioengine
2082 defines engine specific options.
2085 Using GlusterFS libgfapi async interface to direct access to
2086 GlusterFS volumes without having to go through FUSE. This ioengine
2087 defines engine specific options.
2090 Read and write through Hadoop (HDFS). The :option:`filename` option
2091 is used to specify host,port of the hdfs name-node to connect. This
2092 engine interprets offsets a little differently. In HDFS, files once
2093 created cannot be modified so random writes are not possible. To
2094 imitate this the libhdfs engine expects a bunch of small files to be
2095 created over HDFS and will randomly pick a file from them
2096 based on the offset generated by fio backend (see the example
2097 job file to create such files, use ``rw=write`` option). Please
2098 note, it may be necessary to set environment variables to work
2099 with HDFS/libhdfs properly. Each job uses its own connection to
2103 Read, write and erase an MTD character device (e.g.,
2104 :file:`/dev/mtd0`). Discards are treated as erases. Depending on the
2105 underlying device type, the I/O may have to go in a certain pattern,
2106 e.g., on NAND, writing sequentially to erase blocks and discarding
2107 before overwriting. The `trimwrite` mode works well for this
2111 Read and write using filesystem DAX to a file on a filesystem
2112 mounted with DAX on a persistent memory device through the PMDK
2116 Read and write using device DAX to a persistent memory device (e.g.,
2117 /dev/dax0.0) through the PMDK libpmem library.
2120 Prefix to specify loading an external I/O engine object file. Append
2121 the engine filename, e.g. ``ioengine=external:/tmp/foo.o`` to load
2122 ioengine :file:`foo.o` in :file:`/tmp`. The path can be either
2123 absolute or relative. See :file:`engines/skeleton_external.c` for
2124 details of writing an external I/O engine.
2127 Simply create the files and do no I/O to them. You still need to
2128 set `filesize` so that all the accounting still occurs, but no
2129 actual I/O will be done other than creating the file.
2132 Simply do stat() and do no I/O to the file. You need to set 'filesize'
2133 and 'nrfiles', so that files will be created.
2134 This engine is to measure file lookup and meta data access.
2137 Simply delete the files by unlink() and do no I/O to them. You need to set 'filesize'
2138 and 'nrfiles', so that the files will be created.
2139 This engine is to measure file delete.
2142 Read and write using mmap I/O to a file on a filesystem
2143 mounted with DAX on a persistent memory device through the PMDK
2147 Synchronous read and write using DDN's Infinite Memory Engine (IME).
2148 This engine is very basic and issues calls to IME whenever an IO is
2152 Synchronous read and write using DDN's Infinite Memory Engine (IME).
2153 This engine uses iovecs and will try to stack as much IOs as possible
2154 (if the IOs are "contiguous" and the IO depth is not exceeded)
2155 before issuing a call to IME.
2158 Asynchronous read and write using DDN's Infinite Memory Engine (IME).
2159 This engine will try to stack as much IOs as possible by creating
2160 requests for IME. FIO will then decide when to commit these requests.
2163 Read and write iscsi lun with libiscsi.
2166 Read and write a Network Block Device (NBD).
2169 I/O engine supporting libcufile synchronous access to nvidia-fs and a
2170 GPUDirect Storage-supported filesystem. This engine performs
2171 I/O without transferring buffers between user-space and the kernel,
2172 unless :option:`verify` is set or :option:`cuda_io` is `posix`.
2173 :option:`iomem` must not be `cudamalloc`. This ioengine defines
2174 engine specific options.
2177 I/O engine supporting asynchronous read and write operations to the
2178 DAOS File System (DFS) via libdfs.
2181 I/O engine supporting asynchronous read and write operations to
2182 NFS filesystems from userspace via libnfs. This is useful for
2183 achieving higher concurrency and thus throughput than is possible
2187 Execute 3rd party tools. Could be used to perform monitoring during jobs runtime.
2190 I/O engine using the xNVMe C API, for NVMe devices. The xnvme engine provides
2191 flexibility to access GNU/Linux Kernel NVMe driver via libaio, IOCTLs, io_uring,
2192 the SPDK NVMe driver, or your own custom NVMe driver. The xnvme engine includes
2193 engine specific options. (See https://xnvme.io).
2195 I/O engine specific parameters
2196 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2198 In addition, there are some parameters which are only valid when a specific
2199 :option:`ioengine` is in use. These are used identically to normal parameters,
2200 with the caveat that when used on the command line, they must come after the
2201 :option:`ioengine` that defines them is selected.
2203 .. option:: cmdprio_percentage=int[,int] : [io_uring] [libaio]
2205 Set the percentage of I/O that will be issued with the highest priority.
2206 Default: 0. A single value applies to reads and writes. Comma-separated
2207 values may be specified for reads and writes. For this option to be
2208 effective, NCQ priority must be supported and enabled, and the :option:`direct`
2209 option must be set. fio must also be run as the root user. Unlike
2210 slat/clat/lat stats, which can be tracked and reported independently, per
2211 priority stats only track and report a single type of latency. By default,
2212 completion latency (clat) will be reported, if :option:`lat_percentiles` is
2213 set, total latency (lat) will be reported.
2215 .. option:: cmdprio_class=int[,int] : [io_uring] [libaio]
2217 Set the I/O priority class to use for I/Os that must be issued with
2218 a priority when :option:`cmdprio_percentage` or
2219 :option:`cmdprio_bssplit` is set. If not specified when
2220 :option:`cmdprio_percentage` or :option:`cmdprio_bssplit` is set,
2221 this defaults to the highest priority class. A single value applies
2222 to reads and writes. Comma-separated values may be specified for
2223 reads and writes. See :manpage:`ionice(1)`. See also the
2224 :option:`prioclass` option.
2226 .. option:: cmdprio=int[,int] : [io_uring] [libaio]
2228 Set the I/O priority value to use for I/Os that must be issued with
2229 a priority when :option:`cmdprio_percentage` or
2230 :option:`cmdprio_bssplit` is set. If not specified when
2231 :option:`cmdprio_percentage` or :option:`cmdprio_bssplit` is set,
2233 Linux limits us to a positive value between 0 and 7, with 0 being the
2234 highest. A single value applies to reads and writes. Comma-separated
2235 values may be specified for reads and writes. See :manpage:`ionice(1)`.
2236 Refer to an appropriate manpage for other operating systems since
2237 meaning of priority may differ. See also the :option:`prio` option.
2239 .. option:: cmdprio_bssplit=str[,str] : [io_uring] [libaio]
2241 To get a finer control over I/O priority, this option allows
2242 specifying the percentage of IOs that must have a priority set
2243 depending on the block size of the IO. This option is useful only
2244 when used together with the :option:`bssplit` option, that is,
2245 multiple different block sizes are used for reads and writes.
2247 The first accepted format for this option is the same as the format of
2248 the :option:`bssplit` option:
2250 cmdprio_bssplit=blocksize/percentage:blocksize/percentage
2252 In this case, each entry will use the priority class and priority
2253 level defined by the options :option:`cmdprio_class` and
2254 :option:`cmdprio` respectively.
2256 The second accepted format for this option is:
2258 cmdprio_bssplit=blocksize/percentage/class/level:blocksize/percentage/class/level
2260 In this case, the priority class and priority level is defined inside
2261 each entry. In comparison with the first accepted format, the second
2262 accepted format does not restrict all entries to have the same priority
2263 class and priority level.
2265 For both formats, only the read and write data directions are supported,
2266 values for trim IOs are ignored. This option is mutually exclusive with
2267 the :option:`cmdprio_percentage` option.
2269 .. option:: fixedbufs : [io_uring] [io_uring_cmd]
2271 If fio is asked to do direct IO, then Linux will map pages for each
2272 IO call, and release them when IO is done. If this option is set, the
2273 pages are pre-mapped before IO is started. This eliminates the need to
2274 map and release for each IO. This is more efficient, and reduces the
2277 .. option:: nonvectored=int : [io_uring] [io_uring_cmd]
2279 With this option, fio will use non-vectored read/write commands, where
2280 address must contain the address directly. Default is -1.
2282 .. option:: force_async=int : [io_uring] [io_uring_cmd]
2284 Normal operation for io_uring is to try and issue an sqe as
2285 non-blocking first, and if that fails, execute it in an async manner.
2286 With this option set to N, then every N request fio will ask sqe to
2287 be issued in an async manner. Default is 0.
2289 .. option:: registerfiles : [io_uring] [io_uring_cmd]
2291 With this option, fio registers the set of files being used with the
2292 kernel. This avoids the overhead of managing file counts in the kernel,
2293 making the submission and completion part more lightweight. Required
2294 for the below :option:`sqthread_poll` option.
2296 .. option:: sqthread_poll : [io_uring] [io_uring_cmd] [xnvme]
2298 Normally fio will submit IO by issuing a system call to notify the
2299 kernel of available items in the SQ ring. If this option is set, the
2300 act of submitting IO will be done by a polling thread in the kernel.
2301 This frees up cycles for fio, at the cost of using more CPU in the
2302 system. As submission is just the time it takes to fill in the sqe
2303 entries and any syscall required to wake up the idle kernel thread,
2304 fio will not report submission latencies.
2306 .. option:: sqthread_poll_cpu=int : [io_uring] [io_uring_cmd]
2308 When :option:`sqthread_poll` is set, this option provides a way to
2309 define which CPU should be used for the polling thread.
2311 .. option:: cmd_type=str : [io_uring_cmd]
2313 Specifies the type of uring passthrough command to be used. Supported
2314 value is nvme. Default is nvme.
2318 [io_uring] [io_uring_cmd] [xnvme]
2320 If this option is set, fio will attempt to use polled IO completions.
2321 Normal IO completions generate interrupts to signal the completion of
2322 IO, polled completions do not. Hence they are require active reaping
2323 by the application. The benefits are more efficient IO for high IOPS
2324 scenarios, and lower latencies for low queue depth IO.
2328 Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
2333 If this option is set, fio will attempt to use polled IO completions.
2334 This will have a similar effect as (io_uring)hipri. Only SCSI READ and
2335 WRITE commands will have the SGV4_FLAG_HIPRI set (not UNMAP (trim) nor
2336 VERIFY). Older versions of the Linux sg driver that do not support
2337 hipri will simply ignore this flag and do normal IO. The Linux SCSI
2338 Low Level Driver (LLD) that "owns" the device also needs to support
2339 hipri (also known as iopoll and mq_poll). The MegaRAID driver is an
2340 example of a SCSI LLD. Default: clear (0) which does normal
2341 (interrupted based) IO.
2343 .. option:: userspace_reap : [libaio]
2345 Normally, with the libaio engine in use, fio will use the
2346 :manpage:`io_getevents(2)` system call to reap newly returned events. With
2347 this flag turned on, the AIO ring will be read directly from user-space to
2348 reap events. The reaping mode is only enabled when polling for a minimum of
2349 0 events (e.g. when :option:`iodepth_batch_complete` `=0`).
2351 .. option:: hipri_percentage : [pvsync2]
2353 When hipri is set this determines the probability of a pvsync2 I/O being high
2354 priority. The default is 100%.
2356 .. option:: nowait=bool : [pvsync2] [libaio] [io_uring] [io_uring_cmd]
2358 By default if a request cannot be executed immediately (e.g. resource starvation,
2359 waiting on locks) it is queued and the initiating process will be blocked until
2360 the required resource becomes free.
2362 This option sets the RWF_NOWAIT flag (supported from the 4.14 Linux kernel) and
2363 the call will return instantly with EAGAIN or a partial result rather than waiting.
2365 It is useful to also use ignore_error=EAGAIN when using this option.
2367 Note: glibc 2.27, 2.28 have a bug in syscall wrappers preadv2, pwritev2.
2368 They return EOPNOTSUP instead of EAGAIN.
2370 For cached I/O, using this option usually means a request operates only with
2371 cached data. Currently the RWF_NOWAIT flag does not supported for cached write.
2373 For direct I/O, requests will only succeed if cache invalidation isn't required,
2374 file blocks are fully allocated and the disk request could be issued immediately.
2376 .. option:: cpuload=int : [cpuio]
2378 Attempt to use the specified percentage of CPU cycles. This is a mandatory
2379 option when using cpuio I/O engine.
2381 .. option:: cpuchunks=int : [cpuio]
2383 Split the load into cycles of the given time. In microseconds.
2385 .. option:: cpumode=str : [cpuio]
2387 Specify how to stress the CPU. It can take these two values:
2390 This is the default where the CPU executes noop instructions.
2392 Replace the default noop instructions loop with a qsort algorithm to
2393 consume more energy.
2395 .. option:: exit_on_io_done=bool : [cpuio]
2397 Detect when I/O threads are done, then exit.
2399 .. option:: namenode=str : [libhdfs]
2401 The hostname or IP address of a HDFS cluster namenode to contact.
2403 .. option:: port=int
2407 The listening port of the HFDS cluster namenode.
2411 The TCP or UDP port to bind to or connect to. If this is used with
2412 :option:`numjobs` to spawn multiple instances of the same job type, then
2413 this will be the starting port number since fio will use a range of
2418 The port to use for RDMA-CM communication. This should be the same value
2419 on the client and the server side.
2421 .. option:: hostname=str : [netsplice] [net] [rdma]
2423 The hostname or IP address to use for TCP, UDP or RDMA-CM based I/O. If the job
2424 is a TCP listener or UDP reader, the hostname is not used and must be omitted
2425 unless it is a valid UDP multicast address.
2427 .. option:: serverip=str : [librpma_*]
2429 The IP address to be used for RDMA-CM based I/O.
2431 .. option:: direct_write_to_pmem=bool : [librpma_*]
2433 Set to 1 only when Direct Write to PMem from the remote host is possible.
2434 Otherwise, set to 0.
2436 .. option:: busy_wait_polling=bool : [librpma_*_server]
2438 Set to 0 to wait for completion instead of busy-wait polling completion.
2441 .. option:: interface=str : [netsplice] [net]
2443 The IP address of the network interface used to send or receive UDP
2446 .. option:: ttl=int : [netsplice] [net]
2448 Time-to-live value for outgoing UDP multicast packets. Default: 1.
2450 .. option:: nodelay=bool : [netsplice] [net]
2452 Set TCP_NODELAY on TCP connections.
2454 .. option:: protocol=str, proto=str : [netsplice] [net]
2456 The network protocol to use. Accepted values are:
2459 Transmission control protocol.
2461 Transmission control protocol V6.
2463 User datagram protocol.
2465 User datagram protocol V6.
2469 When the protocol is TCP or UDP, the port must also be given, as well as the
2470 hostname if the job is a TCP listener or UDP reader. For unix sockets, the
2471 normal :option:`filename` option should be used and the port is invalid.
2473 .. option:: listen : [netsplice] [net]
2475 For TCP network connections, tell fio to listen for incoming connections
2476 rather than initiating an outgoing connection. The :option:`hostname` must
2477 be omitted if this option is used.
2479 .. option:: pingpong : [netsplice] [net]
2481 Normally a network writer will just continue writing data, and a network
2482 reader will just consume packages. If ``pingpong=1`` is set, a writer will
2483 send its normal payload to the reader, then wait for the reader to send the
2484 same payload back. This allows fio to measure network latencies. The
2485 submission and completion latencies then measure local time spent sending or
2486 receiving, and the completion latency measures how long it took for the
2487 other end to receive and send back. For UDP multicast traffic
2488 ``pingpong=1`` should only be set for a single reader when multiple readers
2489 are listening to the same address.
2491 .. option:: window_size : [netsplice] [net]
2493 Set the desired socket buffer size for the connection.
2495 .. option:: mss : [netsplice] [net]
2497 Set the TCP maximum segment size (TCP_MAXSEG).
2499 .. option:: donorname=str : [e4defrag]
2501 File will be used as a block donor (swap extents between files).
2503 .. option:: inplace=int : [e4defrag]
2505 Configure donor file blocks allocation strategy:
2508 Default. Preallocate donor's file on init.
2510 Allocate space immediately inside defragment event, and free right
2513 .. option:: clustername=str : [rbd,rados]
2515 Specifies the name of the Ceph cluster.
2517 .. option:: rbdname=str : [rbd]
2519 Specifies the name of the RBD.
2521 .. option:: clientname=str : [rbd,rados]
2523 Specifies the username (without the 'client.' prefix) used to access the
2524 Ceph cluster. If the *clustername* is specified, the *clientname* shall be
2525 the full *type.id* string. If no type. prefix is given, fio will add
2526 'client.' by default.
2528 .. option:: conf=str : [rados]
2530 Specifies the configuration path of ceph cluster, so conf file does not
2531 have to be /etc/ceph/ceph.conf.
2533 .. option:: busy_poll=bool : [rbd,rados]
2535 Poll store instead of waiting for completion. Usually this provides better
2536 throughput at cost of higher(up to 100%) CPU utilization.
2538 .. option:: touch_objects=bool : [rados]
2540 During initialization, touch (create if do not exist) all objects (files).
2541 Touching all objects affects ceph caches and likely impacts test results.
2544 .. option:: pool=str :
2548 Specifies the name of the Ceph pool containing RBD or RADOS data.
2552 Specify the label or UUID of the DAOS pool to connect to.
2554 .. option:: cont=str : [dfs]
2556 Specify the label or UUID of the DAOS container to open.
2558 .. option:: chunk_size=int
2562 Specify a different chunk size (in bytes) for the dfs file.
2563 Use DAOS container's chunk size by default.
2567 The size of the chunk to use for each file.
2569 .. option:: object_class=str : [dfs]
2571 Specify a different object class for the dfs file.
2572 Use DAOS container's object class by default.
2574 .. option:: skip_bad=bool : [mtd]
2576 Skip operations against known bad blocks.
2578 .. option:: hdfsdirectory : [libhdfs]
2580 libhdfs will create chunk in this HDFS directory.
2582 .. option:: verb=str : [rdma]
2584 The RDMA verb to use on this side of the RDMA ioengine connection. Valid
2585 values are write, read, send and recv. These correspond to the equivalent
2586 RDMA verbs (e.g. write = rdma_write etc.). Note that this only needs to be
2587 specified on the client side of the connection. See the examples folder.
2589 .. option:: bindname=str : [rdma]
2591 The name to use to bind the local RDMA-CM connection to a local RDMA device.
2592 This could be a hostname or an IPv4 or IPv6 address. On the server side this
2593 will be passed into the rdma_bind_addr() function and on the client site it
2594 will be used in the rdma_resolve_add() function. This can be useful when
2595 multiple paths exist between the client and the server or in certain loopback
2598 .. option:: stat_type=str : [filestat]
2600 Specify stat system call type to measure lookup/getattr performance.
2601 Default is **stat** for :manpage:`stat(2)`.
2603 .. option:: readfua=bool : [sg]
2605 With readfua option set to 1, read operations include
2606 the force unit access (fua) flag. Default is 0.
2608 .. option:: writefua=bool : [sg]
2610 With writefua option set to 1, write operations include
2611 the force unit access (fua) flag. Default is 0.
2613 .. option:: sg_write_mode=str : [sg]
2615 Specify the type of write commands to issue. This option can take three values:
2618 This is the default where write opcodes are issued as usual.
2619 **write_and_verify**
2620 Issue WRITE AND VERIFY commands. The BYTCHK bit is set to 0. This
2621 directs the device to carry out a medium verification with no data
2622 comparison. The writefua option is ignored with this selection.
2624 This option is deprecated. Use write_and_verify instead.
2626 Issue WRITE SAME commands. This transfers a single block to the device
2627 and writes this same block of data to a contiguous sequence of LBAs
2628 beginning at the specified offset. fio's block size parameter specifies
2629 the amount of data written with each command. However, the amount of data
2630 actually transferred to the device is equal to the device's block
2631 (sector) size. For a device with 512 byte sectors, blocksize=8k will
2632 write 16 sectors with each command. fio will still generate 8k of data
2633 for each command but only the first 512 bytes will be used and
2634 transferred to the device. The writefua option is ignored with this
2637 This option is deprecated. Use write_same instead.
2639 Issue WRITE SAME(16) commands as above but with the No Data Output
2640 Buffer (NDOB) bit set. No data will be transferred to the device with
2641 this bit set. Data written will be a pre-determined pattern such as
2644 Issue WRITE STREAM(16) commands. Use the **stream_id** option to specify
2645 the stream identifier.
2646 **verify_bytchk_00**
2647 Issue VERIFY commands with BYTCHK set to 00. This directs the
2648 device to carry out a medium verification with no data comparison.
2649 **verify_bytchk_01**
2650 Issue VERIFY commands with BYTCHK set to 01. This directs the device to
2651 compare the data on the device with the data transferred to the device.
2652 **verify_bytchk_11**
2653 Issue VERIFY commands with BYTCHK set to 11. This transfers a
2654 single block to the device and compares the contents of this block with the
2655 data on the device beginning at the specified offset. fio's block size
2656 parameter specifies the total amount of data compared with this command.
2657 However, only one block (sector) worth of data is transferred to the device.
2658 This is similar to the WRITE SAME command except that data is compared instead
2661 .. option:: stream_id=int : [sg]
2663 Set the stream identifier for WRITE STREAM commands. If this is set to 0 (which is not
2664 a valid stream identifier) fio will open a stream and then close it when done. Default
2667 .. option:: http_host=str : [http]
2669 Hostname to connect to. For S3, this could be the bucket hostname.
2670 Default is **localhost**
2672 .. option:: http_user=str : [http]
2674 Username for HTTP authentication.
2676 .. option:: http_pass=str : [http]
2678 Password for HTTP authentication.
2680 .. option:: https=str : [http]
2682 Enable HTTPS instead of http. *on* enables HTTPS; *insecure*
2683 will enable HTTPS, but disable SSL peer verification (use with
2684 caution!). Default is **off**
2686 .. option:: http_mode=str : [http]
2688 Which HTTP access mode to use: *webdav*, *swift*, or *s3*.
2689 Default is **webdav**
2691 .. option:: http_s3_region=str : [http]
2693 The S3 region/zone string.
2694 Default is **us-east-1**
2696 .. option:: http_s3_key=str : [http]
2700 .. option:: http_s3_keyid=str : [http]
2702 The S3 key/access id.
2704 .. option:: http_s3_sse_customer_key=str : [http]
2706 The encryption customer key in SSE server side.
2708 .. option:: http_s3_sse_customer_algorithm=str : [http]
2710 The encryption customer algorithm in SSE server side.
2711 Default is **AES256**
2713 .. option:: http_s3_storage_class=str : [http]
2715 Which storage class to access. User-customizable settings.
2716 Default is **STANDARD**
2718 .. option:: http_swift_auth_token=str : [http]
2720 The Swift auth token. See the example configuration file on how
2723 .. option:: http_verbose=int : [http]
2725 Enable verbose requests from libcurl. Useful for debugging. 1
2726 turns on verbose logging from libcurl, 2 additionally enables
2727 HTTP IO tracing. Default is **0**
2729 .. option:: uri=str : [nbd]
2731 Specify the NBD URI of the server to test. The string
2732 is a standard NBD URI
2733 (see https://github.com/NetworkBlockDevice/nbd/tree/master/doc).
2734 Example URIs: nbd://localhost:10809
2735 nbd+unix:///?socket=/tmp/socket
2736 nbds://tlshost/exportname
2738 .. option:: gpu_dev_ids=str : [libcufile]
2740 Specify the GPU IDs to use with CUDA. This is a colon-separated list of
2741 int. GPUs are assigned to workers roundrobin. Default is 0.
2743 .. option:: cuda_io=str : [libcufile]
2745 Specify the type of I/O to use with CUDA. Default is **cufile**.
2748 Use libcufile and nvidia-fs. This option performs I/O directly
2749 between a GPUDirect Storage filesystem and GPU buffers,
2750 avoiding use of a bounce buffer. If :option:`verify` is set,
2751 cudaMemcpy is used to copy verificaton data between RAM and GPU.
2752 Verification data is copied from RAM to GPU before a write
2753 and from GPU to RAM after a read. :option:`direct` must be 1.
2755 Use POSIX to perform I/O with a RAM buffer, and use cudaMemcpy
2756 to transfer data between RAM and the GPUs. Data is copied from
2757 GPU to RAM before a write and copied from RAM to GPU after a
2758 read. :option:`verify` does not affect use of cudaMemcpy.
2760 .. option:: nfs_url=str : [nfs]
2762 URL in libnfs format, eg nfs://<server|ipv4|ipv6>/path[?arg=val[&arg=val]*]
2763 Refer to the libnfs README for more details.
2765 .. option:: program=str : [exec]
2767 Specify the program to execute.
2769 .. option:: arguments=str : [exec]
2771 Specify arguments to pass to program.
2772 Some special variables can be expanded to pass fio's job details to the program.
2775 Replaced by the duration of the job in seconds.
2777 Replaced by the name of the job.
2779 .. option:: grace_time=int : [exec]
2781 Specify the time between the SIGTERM and SIGKILL signals. Default is 1 second.
2783 .. option:: std_redirect=bool : [exec]
2785 If set, stdout and stderr streams are redirected to files named from the job name. Default is true.
2787 .. option:: xnvme_async=str : [xnvme]
2789 Select the xnvme async command interface. This can take these values.
2792 This is default and use to emulate asynchronous I/O by using a
2793 single thread to create a queue pair on top of a synchronous
2794 I/O interface using the NVMe driver IOCTL.
2796 Emulate an asynchronous I/O interface with a pool of userspace
2797 threads on top of a synchronous I/O interface using the NVMe
2798 driver IOCTL. By default four threads are used.
2800 Linux native asynchronous I/O interface which supports both
2801 direct and buffered I/O.
2803 Fast Linux native asynchronous I/O interface for NVMe pass
2804 through commands. This only works with NVMe character device
2807 Use Linux aio for Asynchronous I/O.
2809 Use the posix asynchronous I/O interface to perform one or
2810 more I/O operations asynchronously.
2812 Do not transfer any data; just pretend to. This is mainly used
2813 for introspective performance evaluation.
2815 .. option:: xnvme_sync=str : [xnvme]
2817 Select the xnvme synchronous command interface. This can take these values.
2820 This is default and uses Linux NVMe Driver ioctl() for
2823 This supports regular as well as vectored pread() and pwrite()
2826 This is the same as psync except that it also supports zone
2827 management commands using Linux block layer IOCTLs.
2829 .. option:: xnvme_admin=str : [xnvme]
2831 Select the xnvme admin command interface. This can take these values.
2834 This is default and uses linux NVMe Driver ioctl() for admin
2837 Use Linux Block Layer ioctl() and sysfs for admin commands.
2839 .. option:: xnvme_dev_nsid=int : [xnvme]
2841 xnvme namespace identifier for userspace NVMe driver, such as SPDK.
2843 .. option:: xnvme_iovec=int : [xnvme]
2845 If this option is set. xnvme will use vectored read/write commands.
2850 .. option:: iodepth=int
2852 Number of I/O units to keep in flight against the file. Note that
2853 increasing *iodepth* beyond 1 will not affect synchronous ioengines (except
2854 for small degrees when :option:`verify_async` is in use). Even async
2855 engines may impose OS restrictions causing the desired depth not to be
2856 achieved. This may happen on Linux when using libaio and not setting
2857 :option:`direct`\=1, since buffered I/O is not async on that OS. Keep an
2858 eye on the I/O depth distribution in the fio output to verify that the
2859 achieved depth is as expected. Default: 1.
2861 .. option:: iodepth_batch_submit=int, iodepth_batch=int
2863 This defines how many pieces of I/O to submit at once. It defaults to 1
2864 which means that we submit each I/O as soon as it is available, but can be
2865 raised to submit bigger batches of I/O at the time. If it is set to 0 the
2866 :option:`iodepth` value will be used.
2868 .. option:: iodepth_batch_complete_min=int, iodepth_batch_complete=int
2870 This defines how many pieces of I/O to retrieve at once. It defaults to 1
2871 which means that we'll ask for a minimum of 1 I/O in the retrieval process
2872 from the kernel. The I/O retrieval will go on until we hit the limit set by
2873 :option:`iodepth_low`. If this variable is set to 0, then fio will always
2874 check for completed events before queuing more I/O. This helps reduce I/O
2875 latency, at the cost of more retrieval system calls.
2877 .. option:: iodepth_batch_complete_max=int
2879 This defines maximum pieces of I/O to retrieve at once. This variable should
2880 be used along with :option:`iodepth_batch_complete_min`\=int variable,
2881 specifying the range of min and max amount of I/O which should be
2882 retrieved. By default it is equal to the :option:`iodepth_batch_complete_min`
2887 iodepth_batch_complete_min=1
2888 iodepth_batch_complete_max=<iodepth>
2890 which means that we will retrieve at least 1 I/O and up to the whole
2891 submitted queue depth. If none of I/O has been completed yet, we will wait.
2895 iodepth_batch_complete_min=0
2896 iodepth_batch_complete_max=<iodepth>
2898 which means that we can retrieve up to the whole submitted queue depth, but
2899 if none of I/O has been completed yet, we will NOT wait and immediately exit
2900 the system call. In this example we simply do polling.
2902 .. option:: iodepth_low=int
2904 The low water mark indicating when to start filling the queue
2905 again. Defaults to the same as :option:`iodepth`, meaning that fio will
2906 attempt to keep the queue full at all times. If :option:`iodepth` is set to
2907 e.g. 16 and *iodepth_low* is set to 4, then after fio has filled the queue of
2908 16 requests, it will let the depth drain down to 4 before starting to fill
2911 .. option:: serialize_overlap=bool
2913 Serialize in-flight I/Os that might otherwise cause or suffer from data races.
2914 When two or more I/Os are submitted simultaneously, there is no guarantee that
2915 the I/Os will be processed or completed in the submitted order. Further, if
2916 two or more of those I/Os are writes, any overlapping region between them can
2917 become indeterminate/undefined on certain storage. These issues can cause
2918 verification to fail erratically when at least one of the racing I/Os is
2919 changing data and the overlapping region has a non-zero size. Setting
2920 ``serialize_overlap`` tells fio to avoid provoking this behavior by explicitly
2921 serializing in-flight I/Os that have a non-zero overlap. Note that setting
2922 this option can reduce both performance and the :option:`iodepth` achieved.
2924 This option only applies to I/Os issued for a single job except when it is
2925 enabled along with :option:`io_submit_mode`\=offload. In offload mode, fio
2926 will check for overlap among all I/Os submitted by offload jobs with :option:`serialize_overlap`
2931 .. option:: io_submit_mode=str
2933 This option controls how fio submits the I/O to the I/O engine. The default
2934 is `inline`, which means that the fio job threads submit and reap I/O
2935 directly. If set to `offload`, the job threads will offload I/O submission
2936 to a dedicated pool of I/O threads. This requires some coordination and thus
2937 has a bit of extra overhead, especially for lower queue depth I/O where it
2938 can increase latencies. The benefit is that fio can manage submission rates
2939 independently of the device completion rates. This avoids skewed latency
2940 reporting if I/O gets backed up on the device side (the coordinated omission
2941 problem). Note that this option cannot reliably be used with async IO
2948 .. option:: thinktime=time
2950 Stall the job for the specified period of time after an I/O has completed before issuing the
2951 next. May be used to simulate processing being done by an application.
2952 When the unit is omitted, the value is interpreted in microseconds. See
2953 :option:`thinktime_blocks`, :option:`thinktime_iotime` and :option:`thinktime_spin`.
2955 .. option:: thinktime_spin=time
2957 Only valid if :option:`thinktime` is set - pretend to spend CPU time doing
2958 something with the data received, before falling back to sleeping for the
2959 rest of the period specified by :option:`thinktime`. When the unit is
2960 omitted, the value is interpreted in microseconds.
2962 .. option:: thinktime_blocks=int
2964 Only valid if :option:`thinktime` is set - control how many blocks to issue,
2965 before waiting :option:`thinktime` usecs. If not set, defaults to 1 which will make
2966 fio wait :option:`thinktime` usecs after every block. This effectively makes any
2967 queue depth setting redundant, since no more than 1 I/O will be queued
2968 before we have to complete it and do our :option:`thinktime`. In other words, this
2969 setting effectively caps the queue depth if the latter is larger.
2971 .. option:: thinktime_blocks_type=str
2973 Only valid if :option:`thinktime` is set - control how :option:`thinktime_blocks`
2974 triggers. The default is `complete`, which triggers thinktime when fio completes
2975 :option:`thinktime_blocks` blocks. If this is set to `issue`, then the trigger happens
2978 .. option:: thinktime_iotime=time
2980 Only valid if :option:`thinktime` is set - control :option:`thinktime`
2981 interval by time. The :option:`thinktime` stall is repeated after IOs
2982 are executed for :option:`thinktime_iotime`. For example,
2983 ``--thinktime_iotime=9s --thinktime=1s`` repeat 10-second cycle with IOs
2984 for 9 seconds and stall for 1 second. When the unit is omitted,
2985 :option:`thinktime_iotime` is interpreted as a number of seconds. If
2986 this option is used together with :option:`thinktime_blocks`, the
2987 :option:`thinktime` stall is repeated after :option:`thinktime_iotime`
2988 or after :option:`thinktime_blocks` IOs, whichever happens first.
2990 .. option:: rate=int[,int][,int]
2992 Cap the bandwidth used by this job. The number is in bytes/sec, the normal
2993 suffix rules apply. Comma-separated values may be specified for reads,
2994 writes, and trims as described in :option:`blocksize`.
2996 For example, using `rate=1m,500k` would limit reads to 1MiB/sec and writes to
2997 500KiB/sec. Capping only reads or writes can be done with `rate=,500k` or
2998 `rate=500k,` where the former will only limit writes (to 500KiB/sec) and the
2999 latter will only limit reads.
3001 .. option:: rate_min=int[,int][,int]
3003 Tell fio to do whatever it can to maintain at least this bandwidth. Failing
3004 to meet this requirement will cause the job to exit. Comma-separated values
3005 may be specified for reads, writes, and trims as described in
3006 :option:`blocksize`.
3008 .. option:: rate_iops=int[,int][,int]
3010 Cap the bandwidth to this number of IOPS. Basically the same as
3011 :option:`rate`, just specified independently of bandwidth. If the job is
3012 given a block size range instead of a fixed value, the smallest block size
3013 is used as the metric. Comma-separated values may be specified for reads,
3014 writes, and trims as described in :option:`blocksize`.
3016 .. option:: rate_iops_min=int[,int][,int]
3018 If fio doesn't meet this rate of I/O, it will cause the job to exit.
3019 Comma-separated values may be specified for reads, writes, and trims as
3020 described in :option:`blocksize`.
3022 .. option:: rate_process=str
3024 This option controls how fio manages rated I/O submissions. The default is
3025 `linear`, which submits I/O in a linear fashion with fixed delays between
3026 I/Os that gets adjusted based on I/O completion rates. If this is set to
3027 `poisson`, fio will submit I/O based on a more real world random request
3028 flow, known as the Poisson process
3029 (https://en.wikipedia.org/wiki/Poisson_point_process). The lambda will be
3030 10^6 / IOPS for the given workload.
3032 .. option:: rate_ignore_thinktime=bool
3034 By default, fio will attempt to catch up to the specified rate setting,
3035 if any kind of thinktime setting was used. If this option is set, then
3036 fio will ignore the thinktime and continue doing IO at the specified
3037 rate, instead of entering a catch-up mode after thinktime is done.
3043 .. option:: latency_target=time
3045 If set, fio will attempt to find the max performance point that the given
3046 workload will run at while maintaining a latency below this target. When
3047 the unit is omitted, the value is interpreted in microseconds. See
3048 :option:`latency_window` and :option:`latency_percentile`.
3050 .. option:: latency_window=time
3052 Used with :option:`latency_target` to specify the sample window that the job
3053 is run at varying queue depths to test the performance. When the unit is
3054 omitted, the value is interpreted in microseconds.
3056 .. option:: latency_percentile=float
3058 The percentage of I/Os that must fall within the criteria specified by
3059 :option:`latency_target` and :option:`latency_window`. If not set, this
3060 defaults to 100.0, meaning that all I/Os must be equal or below to the value
3061 set by :option:`latency_target`.
3063 .. option:: latency_run=bool
3065 Used with :option:`latency_target`. If false (default), fio will find
3066 the highest queue depth that meets :option:`latency_target` and exit. If
3067 true, fio will continue running and try to meet :option:`latency_target`
3068 by adjusting queue depth.
3070 .. option:: max_latency=time[,time][,time]
3072 If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
3073 maximum latency. When the unit is omitted, the value is interpreted in
3074 microseconds. Comma-separated values may be specified for reads, writes,
3075 and trims as described in :option:`blocksize`.
3077 .. option:: rate_cycle=int
3079 Average bandwidth for :option:`rate` and :option:`rate_min` over this number
3080 of milliseconds. Defaults to 1000.
3086 .. option:: write_iolog=str
3088 Write the issued I/O patterns to the specified file. See
3089 :option:`read_iolog`. Specify a separate file for each job, otherwise the
3090 iologs will be interspersed and the file may be corrupt. This file will
3091 be opened in append mode.
3093 .. option:: read_iolog=str
3095 Open an iolog with the specified filename and replay the I/O patterns it
3096 contains. This can be used to store a workload and replay it sometime
3097 later. The iolog given may also be a blktrace binary file, which allows fio
3098 to replay a workload captured by :command:`blktrace`. See
3099 :manpage:`blktrace(8)` for how to capture such logging data. For blktrace
3100 replay, the file needs to be turned into a blkparse binary data file first
3101 (``blkparse <device> -o /dev/null -d file_for_fio.bin``).
3102 You can specify a number of files by separating the names with a ':'
3103 character. See the :option:`filename` option for information on how to
3104 escape ':' characters within the file names. These files will
3105 be sequentially assigned to job clones created by :option:`numjobs`.
3106 '-' is a reserved name, meaning read from stdin, notably if
3107 :option:`filename` is set to '-' which means stdin as well, then
3108 this flag can't be set to '-'.
3110 .. option:: read_iolog_chunked=bool
3112 Determines how iolog is read. If false(default) entire :option:`read_iolog`
3113 will be read at once. If selected true, input from iolog will be read
3114 gradually. Useful when iolog is very large, or it is generated.
3116 .. option:: merge_blktrace_file=str
3118 When specified, rather than replaying the logs passed to :option:`read_iolog`,
3119 the logs go through a merge phase which aggregates them into a single
3120 blktrace. The resulting file is then passed on as the :option:`read_iolog`
3121 parameter. The intention here is to make the order of events consistent.
3122 This limits the influence of the scheduler compared to replaying multiple
3123 blktraces via concurrent jobs.
3125 .. option:: merge_blktrace_scalars=float_list
3127 This is a percentage based option that is index paired with the list of
3128 files passed to :option:`read_iolog`. When merging is performed, scale
3129 the time of each event by the corresponding amount. For example,
3130 ``--merge_blktrace_scalars="50:100"`` runs the first trace in halftime
3131 and the second trace in realtime. This knob is separately tunable from
3132 :option:`replay_time_scale` which scales the trace during runtime and
3133 does not change the output of the merge unlike this option.
3135 .. option:: merge_blktrace_iters=float_list
3137 This is a whole number option that is index paired with the list of files
3138 passed to :option:`read_iolog`. When merging is performed, run each trace
3139 for the specified number of iterations. For example,
3140 ``--merge_blktrace_iters="2:1"`` runs the first trace for two iterations
3141 and the second trace for one iteration.
3143 .. option:: replay_no_stall=bool
3145 When replaying I/O with :option:`read_iolog` the default behavior is to
3146 attempt to respect the timestamps within the log and replay them with the
3147 appropriate delay between IOPS. By setting this variable fio will not
3148 respect the timestamps and attempt to replay them as fast as possible while
3149 still respecting ordering. The result is the same I/O pattern to a given
3150 device, but different timings.
3152 .. option:: replay_time_scale=int
3154 When replaying I/O with :option:`read_iolog`, fio will honor the
3155 original timing in the trace. With this option, it's possible to scale
3156 the time. It's a percentage option, if set to 50 it means run at 50%
3157 the original IO rate in the trace. If set to 200, run at twice the
3158 original IO rate. Defaults to 100.
3160 .. option:: replay_redirect=str
3162 While replaying I/O patterns using :option:`read_iolog` the default behavior
3163 is to replay the IOPS onto the major/minor device that each IOP was recorded
3164 from. This is sometimes undesirable because on a different machine those
3165 major/minor numbers can map to a different device. Changing hardware on the
3166 same system can also result in a different major/minor mapping.
3167 ``replay_redirect`` causes all I/Os to be replayed onto the single specified
3168 device regardless of the device it was recorded
3169 from. i.e. :option:`replay_redirect`\= :file:`/dev/sdc` would cause all I/O
3170 in the blktrace or iolog to be replayed onto :file:`/dev/sdc`. This means
3171 multiple devices will be replayed onto a single device, if the trace
3172 contains multiple devices. If you want multiple devices to be replayed
3173 concurrently to multiple redirected devices you must blkparse your trace
3174 into separate traces and replay them with independent fio invocations.
3175 Unfortunately this also breaks the strict time ordering between multiple
3178 .. option:: replay_align=int
3180 Force alignment of the byte offsets in a trace to this value. The value
3181 must be a power of 2.
3183 .. option:: replay_scale=int
3185 Scale byte offsets down by this factor when replaying traces. Should most
3186 likely use :option:`replay_align` as well.
3188 .. option:: replay_skip=str
3190 Sometimes it's useful to skip certain IO types in a replay trace.
3191 This could be, for instance, eliminating the writes in the trace.
3192 Or not replaying the trims/discards, if you are redirecting to
3193 a device that doesn't support them. This option takes a comma
3194 separated list of read, write, trim, sync.
3197 Threads, processes and job synchronization
3198 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3202 Fio defaults to creating jobs by using fork, however if this option is
3203 given, fio will create jobs by using POSIX Threads' function
3204 :manpage:`pthread_create(3)` to create threads instead.
3206 .. option:: wait_for=str
3208 If set, the current job won't be started until all workers of the specified
3209 waitee job are done.
3211 ``wait_for`` operates on the job name basis, so there are a few
3212 limitations. First, the waitee must be defined prior to the waiter job
3213 (meaning no forward references). Second, if a job is being referenced as a
3214 waitee, it must have a unique name (no duplicate waitees).
3216 .. option:: nice=int
3218 Run the job with the given nice value. See man :manpage:`nice(2)`.
3220 On Windows, values less than -15 set the process class to "High"; -1 through
3221 -15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
3224 .. option:: prio=int
3226 Set the I/O priority value of this job. Linux limits us to a positive value
3227 between 0 and 7, with 0 being the highest. See man
3228 :manpage:`ionice(1)`. Refer to an appropriate manpage for other operating
3229 systems since meaning of priority may differ. For per-command priority
3230 setting, see I/O engine specific :option:`cmdprio_percentage` and
3231 :option:`cmdprio` options.
3233 .. option:: prioclass=int
3235 Set the I/O priority class. See man :manpage:`ionice(1)`. For per-command
3236 priority setting, see I/O engine specific :option:`cmdprio_percentage`
3237 and :option:`cmdprio_class` options.
3239 .. option:: cpus_allowed=str
3241 Controls the same options as :option:`cpumask`, but accepts a textual
3242 specification of the permitted CPUs instead and CPUs are indexed from 0. So
3243 to use CPUs 0 and 5 you would specify ``cpus_allowed=0,5``. This option also
3244 allows a range of CPUs to be specified -- say you wanted a binding to CPUs
3245 0, 5, and 8 to 15, you would set ``cpus_allowed=0,5,8-15``.
3247 On Windows, when ``cpus_allowed`` is unset only CPUs from fio's current
3248 processor group will be used and affinity settings are inherited from the
3249 system. An fio build configured to target Windows 7 makes options that set
3250 CPUs processor group aware and values will set both the processor group
3251 and a CPU from within that group. For example, on a system where processor
3252 group 0 has 40 CPUs and processor group 1 has 32 CPUs, ``cpus_allowed``
3253 values between 0 and 39 will bind CPUs from processor group 0 and
3254 ``cpus_allowed`` values between 40 and 71 will bind CPUs from processor
3255 group 1. When using ``cpus_allowed_policy=shared`` all CPUs specified by a
3256 single ``cpus_allowed`` option must be from the same processor group. For
3257 Windows fio builds not built for Windows 7, CPUs will only be selected from
3258 (and be relative to) whatever processor group fio happens to be running in
3259 and CPUs from other processor groups cannot be used.
3261 .. option:: cpus_allowed_policy=str
3263 Set the policy of how fio distributes the CPUs specified by
3264 :option:`cpus_allowed` or :option:`cpumask`. Two policies are supported:
3267 All jobs will share the CPU set specified.
3269 Each job will get a unique CPU from the CPU set.
3271 **shared** is the default behavior, if the option isn't specified. If
3272 **split** is specified, then fio will assign one cpu per job. If not
3273 enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
3276 .. option:: cpumask=int
3278 Set the CPU affinity of this job. The parameter given is a bit mask of
3279 allowed CPUs the job may run on. So if you want the allowed CPUs to be 1
3280 and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
3281 :manpage:`sched_setaffinity(2)`. This may not work on all supported
3282 operating systems or kernel versions. This option doesn't work well for a
3283 higher CPU count than what you can store in an integer mask, so it can only
3284 control cpus 1-32. For boxes with larger CPU counts, use
3285 :option:`cpus_allowed`.
3287 .. option:: numa_cpu_nodes=str
3289 Set this job running on specified NUMA nodes' CPUs. The arguments allow
3290 comma delimited list of cpu numbers, A-B ranges, or `all`. Note, to enable
3291 NUMA options support, fio must be built on a system with libnuma-dev(el)
3294 .. option:: numa_mem_policy=str
3296 Set this job's memory policy and corresponding NUMA nodes. Format of the
3301 ``mode`` is one of the following memory policies: ``default``, ``prefer``,
3302 ``bind``, ``interleave`` or ``local``. For ``default`` and ``local`` memory
3303 policies, no node needs to be specified. For ``prefer``, only one node is
3304 allowed. For ``bind`` and ``interleave`` the ``nodelist`` may be as
3305 follows: a comma delimited list of numbers, A-B ranges, or `all`.
3307 .. option:: cgroup=str
3309 Add job to this control group. If it doesn't exist, it will be created. The
3310 system must have a mounted cgroup blkio mount point for this to work. If
3311 your system doesn't have it mounted, you can do so with::
3313 # mount -t cgroup -o blkio none /cgroup
3315 .. option:: cgroup_weight=int
3317 Set the weight of the cgroup to this value. See the documentation that comes
3318 with the kernel, allowed values are in the range of 100..1000.
3320 .. option:: cgroup_nodelete=bool
3322 Normally fio will delete the cgroups it has created after the job
3323 completion. To override this behavior and to leave cgroups around after the
3324 job completion, set ``cgroup_nodelete=1``. This can be useful if one wants
3325 to inspect various cgroup files after job completion. Default: false.
3327 .. option:: flow_id=int
3329 The ID of the flow. If not specified, it defaults to being a global
3330 flow. See :option:`flow`.
3332 .. option:: flow=int
3334 Weight in token-based flow control. If this value is used, then fio
3335 regulates the activity between two or more jobs sharing the same
3336 flow_id. Fio attempts to keep each job activity proportional to other
3337 jobs' activities in the same flow_id group, with respect to requested
3338 weight per job. That is, if one job has `flow=3', another job has
3339 `flow=2' and another with `flow=1`, then there will be a roughly 3:2:1
3340 ratio in how much one runs vs the others.
3342 .. option:: flow_sleep=int
3344 The period of time, in microseconds, to wait after the flow counter
3345 has exceeded its proportion before retrying operations.
3347 .. option:: stonewall, wait_for_previous
3349 Wait for preceding jobs in the job file to exit, before starting this
3350 one. Can be used to insert serialization points in the job file. A stone
3351 wall also implies starting a new reporting group, see
3352 :option:`group_reporting`.
3356 By default, fio will continue running all other jobs when one job finishes.
3357 Sometimes this is not the desired action. Setting ``exitall`` will instead
3358 make fio terminate all jobs in the same group, as soon as one job of that
3361 .. option:: exit_what=str
3363 By default, fio will continue running all other jobs when one job finishes.
3364 Sometimes this is not the desired action. Setting ``exitall`` will
3365 instead make fio terminate all jobs in the same group. The option
3366 ``exit_what`` allows to control which jobs get terminated when ``exitall`` is
3367 enabled. The default is ``group`` and does not change the behaviour of
3368 ``exitall``. The setting ``all`` terminates all jobs. The setting ``stonewall``
3369 terminates all currently running jobs across all groups and continues execution
3370 with the next stonewalled group.
3372 .. option:: exec_prerun=str
3374 Before running this job, issue the command specified through
3375 :manpage:`system(3)`. Output is redirected in a file called
3376 :file:`jobname.prerun.txt`.
3378 .. option:: exec_postrun=str
3380 After the job completes, issue the command specified though
3381 :manpage:`system(3)`. Output is redirected in a file called
3382 :file:`jobname.postrun.txt`.
3386 Instead of running as the invoking user, set the user ID to this value
3387 before the thread/process does any work.
3391 Set group ID, see :option:`uid`.
3397 .. option:: verify_only
3399 Do not perform specified workload, only verify data still matches previous
3400 invocation of this workload. This option allows one to check data multiple
3401 times at a later date without overwriting it. This option makes sense only
3402 for workloads that write data, and does not support workloads with the
3403 :option:`time_based` option set.
3405 .. option:: do_verify=bool
3407 Run the verify phase after a write phase. Only valid if :option:`verify` is
3410 .. option:: verify=str
3412 If writing to a file, fio can verify the file contents after each iteration
3413 of the job. Each verification method also implies verification of special
3414 header, which is written to the beginning of each block. This header also
3415 includes meta information, like offset of the block, block number, timestamp
3416 when block was written, etc. :option:`verify` can be combined with
3417 :option:`verify_pattern` option. The allowed values are:
3420 Use an md5 sum of the data area and store it in the header of
3424 Use an experimental crc64 sum of the data area and store it in the
3425 header of each block.
3428 Use a crc32c sum of the data area and store it in the header of
3429 each block. This will automatically use hardware acceleration
3430 (e.g. SSE4.2 on an x86 or CRC crypto extensions on ARM64) but will
3431 fall back to software crc32c if none is found. Generally the
3432 fastest checksum fio supports when hardware accelerated.
3438 Use a crc32 sum of the data area and store it in the header of each
3442 Use a crc16 sum of the data area and store it in the header of each
3446 Use a crc7 sum of the data area and store it in the header of each
3450 Use xxhash as the checksum function. Generally the fastest software
3451 checksum that fio supports.
3454 Use sha512 as the checksum function.
3457 Use sha256 as the checksum function.
3460 Use optimized sha1 as the checksum function.
3463 Use optimized sha3-224 as the checksum function.
3466 Use optimized sha3-256 as the checksum function.
3469 Use optimized sha3-384 as the checksum function.
3472 Use optimized sha3-512 as the checksum function.
3475 This option is deprecated, since now meta information is included in
3476 generic verification header and meta verification happens by
3477 default. For detailed information see the description of the
3478 :option:`verify` setting. This option is kept because of
3479 compatibility's sake with old configurations. Do not use it.
3482 Verify a strict pattern. Normally fio includes a header with some
3483 basic information and checksumming, but if this option is set, only
3484 the specific pattern set with :option:`verify_pattern` is verified.
3487 Only pretend to verify. Useful for testing internals with
3488 :option:`ioengine`\=null, not for much else.
3490 This option can be used for repeated burn-in tests of a system to make sure
3491 that the written data is also correctly read back. If the data direction
3492 given is a read or random read, fio will assume that it should verify a
3493 previously written file. If the data direction includes any form of write,
3494 the verify will be of the newly written data.
3496 To avoid false verification errors, do not use the norandommap option when
3497 verifying data with async I/O engines and I/O depths > 1. Or use the
3498 norandommap and the lfsr random generator together to avoid writing to the
3499 same offset with multiple outstanding I/Os.
3501 .. option:: verify_offset=int
3503 Swap the verification header with data somewhere else in the block before
3504 writing. It is swapped back before verifying.
3506 .. option:: verify_interval=int
3508 Write the verification header at a finer granularity than the
3509 :option:`blocksize`. It will be written for chunks the size of
3510 ``verify_interval``. :option:`blocksize` should divide this evenly.
3512 .. option:: verify_pattern=str
3514 If set, fio will fill the I/O buffers with this pattern. Fio defaults to
3515 filling with totally random bytes, but sometimes it's interesting to fill
3516 with a known pattern for I/O verification purposes. Depending on the width
3517 of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time (it can
3518 be either a decimal or a hex number). The ``verify_pattern`` if larger than
3519 a 32-bit quantity has to be a hex number that starts with either "0x" or
3520 "0X". Use with :option:`verify`. Also, ``verify_pattern`` supports %o
3521 format, which means that for each block offset will be written and then
3522 verified back, e.g.::
3526 Or use combination of everything::
3528 verify_pattern=0xff%o"abcd"-12
3530 .. option:: verify_fatal=bool
3532 Normally fio will keep checking the entire contents before quitting on a
3533 block verification failure. If this option is set, fio will exit the job on
3534 the first observed failure. Default: false.
3536 .. option:: verify_dump=bool
3538 If set, dump the contents of both the original data block and the data block
3539 we read off disk to files. This allows later analysis to inspect just what
3540 kind of data corruption occurred. Off by default.
3542 .. option:: verify_async=int
3544 Fio will normally verify I/O inline from the submitting thread. This option
3545 takes an integer describing how many async offload threads to create for I/O
3546 verification instead, causing fio to offload the duty of verifying I/O
3547 contents to one or more separate threads. If using this offload option, even
3548 sync I/O engines can benefit from using an :option:`iodepth` setting higher
3549 than 1, as it allows them to have I/O in flight while verifies are running.
3550 Defaults to 0 async threads, i.e. verification is not asynchronous.
3552 .. option:: verify_async_cpus=str
3554 Tell fio to set the given CPU affinity on the async I/O verification
3555 threads. See :option:`cpus_allowed` for the format used.
3557 .. option:: verify_backlog=int
3559 Fio will normally verify the written contents of a job that utilizes verify
3560 once that job has completed. In other words, everything is written then
3561 everything is read back and verified. You may want to verify continually
3562 instead for a variety of reasons. Fio stores the meta data associated with
3563 an I/O block in memory, so for large verify workloads, quite a bit of memory
3564 would be used up holding this meta data. If this option is enabled, fio will
3565 write only N blocks before verifying these blocks.
3567 .. option:: verify_backlog_batch=int
3569 Control how many blocks fio will verify if :option:`verify_backlog` is
3570 set. If not set, will default to the value of :option:`verify_backlog`
3571 (meaning the entire queue is read back and verified). If
3572 ``verify_backlog_batch`` is less than :option:`verify_backlog` then not all
3573 blocks will be verified, if ``verify_backlog_batch`` is larger than
3574 :option:`verify_backlog`, some blocks will be verified more than once.
3576 .. option:: verify_state_save=bool
3578 When a job exits during the write phase of a verify workload, save its
3579 current state. This allows fio to replay up until that point, if the verify
3580 state is loaded for the verify read phase. The format of the filename is,
3583 <type>-<jobname>-<jobindex>-verify.state.
3585 <type> is "local" for a local run, "sock" for a client/server socket
3586 connection, and "ip" (192.168.0.1, for instance) for a networked
3587 client/server connection. Defaults to true.
3589 .. option:: verify_state_load=bool
3591 If a verify termination trigger was used, fio stores the current write state
3592 of each thread. This can be used at verification time so that fio knows how
3593 far it should verify. Without this information, fio will run a full
3594 verification pass, according to the settings in the job file used. Default
3597 .. option:: trim_percentage=int
3599 Number of verify blocks to discard/trim.
3601 .. option:: trim_verify_zero=bool
3603 Verify that trim/discarded blocks are returned as zeros.
3605 .. option:: trim_backlog=int
3607 Trim after this number of blocks are written.
3609 .. option:: trim_backlog_batch=int
3611 Trim this number of I/O blocks.
3613 .. option:: experimental_verify=bool
3615 Enable experimental verification.
3620 .. option:: steadystate=str:float, ss=str:float
3622 Define the criterion and limit for assessing steady state performance. The
3623 first parameter designates the criterion whereas the second parameter sets
3624 the threshold. When the criterion falls below the threshold for the
3625 specified duration, the job will stop. For example, `iops_slope:0.1%` will
3626 direct fio to terminate the job when the least squares regression slope
3627 falls below 0.1% of the mean IOPS. If :option:`group_reporting` is enabled
3628 this will apply to all jobs in the group. Below is the list of available
3629 steady state assessment criteria. All assessments are carried out using only
3630 data from the rolling collection window. Threshold limits can be expressed
3631 as a fixed value or as a percentage of the mean in the collection window.
3633 When using this feature, most jobs should include the :option:`time_based`
3634 and :option:`runtime` options or the :option:`loops` option so that fio does not
3635 stop running after it has covered the full size of the specified file(s) or device(s).
3638 Collect IOPS data. Stop the job if all individual IOPS measurements
3639 are within the specified limit of the mean IOPS (e.g., ``iops:2``
3640 means that all individual IOPS values must be within 2 of the mean,
3641 whereas ``iops:0.2%`` means that all individual IOPS values must be
3642 within 0.2% of the mean IOPS to terminate the job).
3645 Collect IOPS data and calculate the least squares regression
3646 slope. Stop the job if the slope falls below the specified limit.
3649 Collect bandwidth data. Stop the job if all individual bandwidth
3650 measurements are within the specified limit of the mean bandwidth.
3653 Collect bandwidth data and calculate the least squares regression
3654 slope. Stop the job if the slope falls below the specified limit.
3656 .. option:: steadystate_duration=time, ss_dur=time
3658 A rolling window of this duration will be used to judge whether steady state
3659 has been reached. Data will be collected once per second. The default is 0
3660 which disables steady state detection. When the unit is omitted, the
3661 value is interpreted in seconds.
3663 .. option:: steadystate_ramp_time=time, ss_ramp=time
3665 Allow the job to run for the specified duration before beginning data
3666 collection for checking the steady state job termination criterion. The
3667 default is 0. When the unit is omitted, the value is interpreted in seconds.
3670 Measurements and reporting
3671 ~~~~~~~~~~~~~~~~~~~~~~~~~~
3673 .. option:: per_job_logs=bool
3675 If set, this generates bw/clat/iops log with per file private filenames. If
3676 not set, jobs with identical names will share the log filename. Default:
3679 .. option:: group_reporting
3681 It may sometimes be interesting to display statistics for groups of jobs as
3682 a whole instead of for each individual job. This is especially true if
3683 :option:`numjobs` is used; looking at individual thread/process output
3684 quickly becomes unwieldy. To see the final report per-group instead of
3685 per-job, use :option:`group_reporting`. Jobs in a file will be part of the
3686 same reporting group, unless if separated by a :option:`stonewall`, or by
3687 using :option:`new_group`.
3689 .. option:: new_group
3691 Start a new reporting group. See: :option:`group_reporting`. If not given,
3692 all jobs in a file will be part of the same reporting group, unless
3693 separated by a :option:`stonewall`.
3695 .. option:: stats=bool
3697 By default, fio collects and shows final output results for all jobs
3698 that run. If this option is set to 0, then fio will ignore it in
3699 the final stat output.
3701 .. option:: write_bw_log=str
3703 If given, write a bandwidth log for this job. Can be used to store data of
3704 the bandwidth of the jobs in their lifetime.
3706 If no str argument is given, the default filename of
3707 :file:`jobname_type.x.log` is used. Even when the argument is given, fio
3708 will still append the type of log. So if one specifies::
3712 The actual log name will be :file:`foo_bw.x.log` where `x` is the index
3713 of the job (`1..N`, where `N` is the number of jobs). If
3714 :option:`per_job_logs` is false, then the filename will not include the
3717 The included :command:`fio_generate_plots` script uses :command:`gnuplot` to turn these
3718 text files into nice graphs. See `Log File Formats`_ for how data is
3719 structured within the file.
3721 .. option:: write_lat_log=str
3723 Same as :option:`write_bw_log`, except this option creates I/O
3724 submission (e.g., :file:`name_slat.x.log`), completion (e.g.,
3725 :file:`name_clat.x.log`), and total (e.g., :file:`name_lat.x.log`)
3726 latency files instead. See :option:`write_bw_log` for details about
3727 the filename format and `Log File Formats`_ for how data is structured
3730 .. option:: write_hist_log=str
3732 Same as :option:`write_bw_log` but writes an I/O completion latency
3733 histogram file (e.g., :file:`name_hist.x.log`) instead. Note that this
3734 file will be empty unless :option:`log_hist_msec` has also been set.
3735 See :option:`write_bw_log` for details about the filename format and
3736 `Log File Formats`_ for how data is structured within the file.
3738 .. option:: write_iops_log=str
3740 Same as :option:`write_bw_log`, but writes an IOPS file (e.g.
3741 :file:`name_iops.x.log`) instead. Because fio defaults to individual
3742 I/O logging, the value entry in the IOPS log will be 1 unless windowed
3743 logging (see :option:`log_avg_msec`) has been enabled. See
3744 :option:`write_bw_log` for details about the filename format and `Log
3745 File Formats`_ for how data is structured within the file.
3747 .. option:: log_entries=int
3749 By default, fio will log an entry in the iops, latency, or bw log for
3750 every I/O that completes. The initial number of I/O log entries is 1024.
3751 When the log entries are all used, new log entries are dynamically
3752 allocated. This dynamic log entry allocation may negatively impact
3753 time-related statistics such as I/O tail latencies (e.g. 99.9th percentile
3754 completion latency). This option allows specifying a larger initial
3755 number of log entries to avoid run-time allocations of new log entries,
3756 resulting in more precise time-related I/O statistics.
3757 Also see :option:`log_avg_msec`. Defaults to 1024.
3759 .. option:: log_avg_msec=int
3761 By default, fio will log an entry in the iops, latency, or bw log for every
3762 I/O that completes. When writing to the disk log, that can quickly grow to a
3763 very large size. Setting this option makes fio average the each log entry
3764 over the specified period of time, reducing the resolution of the log. See
3765 :option:`log_max_value` as well. Defaults to 0, logging all entries.
3766 Also see `Log File Formats`_.
3768 .. option:: log_hist_msec=int
3770 Same as :option:`log_avg_msec`, but logs entries for completion latency
3771 histograms. Computing latency percentiles from averages of intervals using
3772 :option:`log_avg_msec` is inaccurate. Setting this option makes fio log
3773 histogram entries over the specified period of time, reducing log sizes for
3774 high IOPS devices while retaining percentile accuracy. See
3775 :option:`log_hist_coarseness` and :option:`write_hist_log` as well.
3776 Defaults to 0, meaning histogram logging is disabled.
3778 .. option:: log_hist_coarseness=int
3780 Integer ranging from 0 to 6, defining the coarseness of the resolution of
3781 the histogram logs enabled with :option:`log_hist_msec`. For each increment
3782 in coarseness, fio outputs half as many bins. Defaults to 0, for which
3783 histogram logs contain 1216 latency bins. See :option:`write_hist_log`
3784 and `Log File Formats`_.
3786 .. option:: log_max_value=bool
3788 If :option:`log_avg_msec` is set, fio logs the average over that window. If
3789 you instead want to log the maximum value, set this option to 1. Defaults to
3790 0, meaning that averaged values are logged.
3792 .. option:: log_offset=bool
3794 If this is set, the iolog options will include the byte offset for the I/O
3795 entry as well as the other data values. Defaults to 0 meaning that
3796 offsets are not present in logs. Also see `Log File Formats`_.
3798 .. option:: log_compression=int
3800 If this is set, fio will compress the I/O logs as it goes, to keep the
3801 memory footprint lower. When a log reaches the specified size, that chunk is
3802 removed and compressed in the background. Given that I/O logs are fairly
3803 highly compressible, this yields a nice memory savings for longer runs. The
3804 downside is that the compression will consume some background CPU cycles, so
3805 it may impact the run. This, however, is also true if the logging ends up
3806 consuming most of the system memory. So pick your poison. The I/O logs are
3807 saved normally at the end of a run, by decompressing the chunks and storing
3808 them in the specified log file. This feature depends on the availability of
3811 .. option:: log_compression_cpus=str
3813 Define the set of CPUs that are allowed to handle online log compression for
3814 the I/O jobs. This can provide better isolation between performance
3815 sensitive jobs, and background compression work. See
3816 :option:`cpus_allowed` for the format used.
3818 .. option:: log_store_compressed=bool
3820 If set, fio will store the log files in a compressed format. They can be
3821 decompressed with fio, using the :option:`--inflate-log` command line
3822 parameter. The files will be stored with a :file:`.fz` suffix.
3824 .. option:: log_unix_epoch=bool
3826 If set, fio will log Unix timestamps to the log files produced by enabling
3827 write_type_log for each log type, instead of the default zero-based
3830 .. option:: log_alternate_epoch=bool
3832 If set, fio will log timestamps based on the epoch used by the clock specified
3833 in the log_alternate_epoch_clock_id option, to the log files produced by
3834 enabling write_type_log for each log type, instead of the default zero-based
3837 .. option:: log_alternate_epoch_clock_id=int
3839 Specifies the clock_id to be used by clock_gettime to obtain the alternate epoch
3840 if either log_unix_epoch or log_alternate_epoch are true. Otherwise has no
3841 effect. Default value is 0, or CLOCK_REALTIME.
3843 .. option:: block_error_percentiles=bool
3845 If set, record errors in trim block-sized units from writes and trims and
3846 output a histogram of how many trims it took to get to errors, and what kind
3847 of error was encountered.
3849 .. option:: bwavgtime=int
3851 Average the calculated bandwidth over the given time. Value is specified in
3852 milliseconds. If the job also does bandwidth logging through
3853 :option:`write_bw_log`, then the minimum of this option and
3854 :option:`log_avg_msec` will be used. Default: 500ms.
3856 .. option:: iopsavgtime=int
3858 Average the calculated IOPS over the given time. Value is specified in
3859 milliseconds. If the job also does IOPS logging through
3860 :option:`write_iops_log`, then the minimum of this option and
3861 :option:`log_avg_msec` will be used. Default: 500ms.
3863 .. option:: disk_util=bool
3865 Generate disk utilization statistics, if the platform supports it.
3868 .. option:: disable_lat=bool
3870 Disable measurements of total latency numbers. Useful only for cutting back
3871 the number of calls to :manpage:`gettimeofday(2)`, as that does impact
3872 performance at really high IOPS rates. Note that to really get rid of a
3873 large amount of these calls, this option must be used with
3874 :option:`disable_slat` and :option:`disable_bw_measurement` as well.
3876 .. option:: disable_clat=bool
3878 Disable measurements of completion latency numbers. See
3879 :option:`disable_lat`.
3881 .. option:: disable_slat=bool
3883 Disable measurements of submission latency numbers. See
3884 :option:`disable_lat`.
3886 .. option:: disable_bw_measurement=bool, disable_bw=bool
3888 Disable measurements of throughput/bandwidth numbers. See
3889 :option:`disable_lat`.
3891 .. option:: slat_percentiles=bool
3893 Report submission latency percentiles. Submission latency is not recorded
3894 for synchronous ioengines.
3896 .. option:: clat_percentiles=bool
3898 Report completion latency percentiles.
3900 .. option:: lat_percentiles=bool
3902 Report total latency percentiles. Total latency is the sum of submission
3903 latency and completion latency.
3905 .. option:: percentile_list=float_list
3907 Overwrite the default list of percentiles for latencies and the block error
3908 histogram. Each number is a floating point number in the range (0,100], and
3909 the maximum length of the list is 20. Use ``:`` to separate the numbers. For
3910 example, ``--percentile_list=99.5:99.9`` will cause fio to report the
3911 latency durations below which 99.5% and 99.9% of the observed latencies fell,
3914 .. option:: significant_figures=int
3916 If using :option:`--output-format` of `normal`, set the significant
3917 figures to this value. Higher values will yield more precise IOPS and
3918 throughput units, while lower values will round. Requires a minimum
3919 value of 1 and a maximum value of 10. Defaults to 4.
3925 .. option:: exitall_on_error
3927 When one job finishes in error, terminate the rest. The default is to wait
3928 for each job to finish.
3930 .. option:: continue_on_error=str
3932 Normally fio will exit the job on the first observed failure. If this option
3933 is set, fio will continue the job when there is a 'non-fatal error' (EIO or
3934 EILSEQ) until the runtime is exceeded or the I/O size specified is
3935 completed. If this option is used, there are two more stats that are
3936 appended, the total error count and the first error. The error field given
3937 in the stats is the first error that was hit during the run.
3939 Note: a write error from the device may go unnoticed by fio when using
3940 buffered IO, as the write() (or similar) system call merely dirties the
3941 kernel pages, unless :option:`sync` or :option:`direct` is used. Device IO
3942 errors occur when the dirty data is actually written out to disk. If fully
3943 sync writes aren't desirable, :option:`fsync` or :option:`fdatasync` can be
3944 used as well. This is specific to writes, as reads are always synchronous.
3946 The allowed values are:
3949 Exit on any I/O or verify errors.
3952 Continue on read errors, exit on all others.
3955 Continue on write errors, exit on all others.
3958 Continue on any I/O error, exit on all others.
3961 Continue on verify errors, exit on all others.
3964 Continue on all errors.
3967 Backward-compatible alias for 'none'.
3970 Backward-compatible alias for 'all'.
3972 .. option:: ignore_error=str
3974 Sometimes you want to ignore some errors during test in that case you can
3975 specify error list for each error type, instead of only being able to
3976 ignore the default 'non-fatal error' using :option:`continue_on_error`.
3977 ``ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST`` errors for
3978 given error type is separated with ':'. Error may be symbol ('ENOSPC',
3979 'ENOMEM') or integer. Example::
3981 ignore_error=EAGAIN,ENOSPC:122
3983 This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
3984 WRITE. This option works by overriding :option:`continue_on_error` with
3985 the list of errors for each error type if any.
3987 .. option:: error_dump=bool
3989 If set dump every error even if it is non fatal, true by default. If
3990 disabled only fatal error will be dumped.
3992 Running predefined workloads
3993 ----------------------------
3995 Fio includes predefined profiles that mimic the I/O workloads generated by
3998 .. option:: profile=str
4000 The predefined workload to run. Current profiles are:
4003 Threaded I/O bench (tiotest/tiobench) like workload.
4006 Aerospike Certification Tool (ACT) like workload.
4008 To view a profile's additional options use :option:`--cmdhelp` after specifying
4009 the profile. For example::
4011 $ fio --profile=act --cmdhelp
4016 .. option:: device-names=str
4021 .. option:: load=int
4024 ACT load multiplier. Default: 1.
4026 .. option:: test-duration=time
4029 How long the entire test takes to run. When the unit is omitted, the value
4030 is given in seconds. Default: 24h.
4032 .. option:: threads-per-queue=int
4035 Number of read I/O threads per device. Default: 8.
4037 .. option:: read-req-num-512-blocks=int
4040 Number of 512B blocks to read at the time. Default: 3.
4042 .. option:: large-block-op-kbytes=int
4045 Size of large block ops in KiB (writes). Default: 131072.
4050 Set to run ACT prep phase.
4052 Tiobench profile options
4053 ~~~~~~~~~~~~~~~~~~~~~~~~
4055 .. option:: size=str
4060 .. option:: block=int
4063 Block size in bytes. Default: 4096.
4065 .. option:: numruns=int
4075 .. option:: threads=int
4080 Interpreting the output
4081 -----------------------
4084 Example output was based on the following:
4085 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --time_based \
4086 --rate=1256k --bs=14K --name=quick --runtime=1s --name=mixed \
4087 --runtime=2m --rw=rw
4089 Fio spits out a lot of output. While running, fio will display the status of the
4090 jobs created. An example of that would be::
4092 Jobs: 1 (f=1): [_(1),M(1)][24.8%][r=20.5MiB/s,w=23.5MiB/s][r=82,w=94 IOPS][eta 01m:31s]
4094 The characters inside the first set of square brackets denote the current status of
4095 each thread. The first character is the first job defined in the job file, and so
4096 forth. The possible values (in typical life cycle order) are:
4098 +------+-----+-----------------------------------------------------------+
4100 +======+=====+===========================================================+
4101 | P | | Thread setup, but not started. |
4102 +------+-----+-----------------------------------------------------------+
4103 | C | | Thread created. |
4104 +------+-----+-----------------------------------------------------------+
4105 | I | | Thread initialized, waiting or generating necessary data. |
4106 +------+-----+-----------------------------------------------------------+
4107 | | p | Thread running pre-reading file(s). |
4108 +------+-----+-----------------------------------------------------------+
4109 | | / | Thread is in ramp period. |
4110 +------+-----+-----------------------------------------------------------+
4111 | | R | Running, doing sequential reads. |
4112 +------+-----+-----------------------------------------------------------+
4113 | | r | Running, doing random reads. |
4114 +------+-----+-----------------------------------------------------------+
4115 | | W | Running, doing sequential writes. |
4116 +------+-----+-----------------------------------------------------------+
4117 | | w | Running, doing random writes. |
4118 +------+-----+-----------------------------------------------------------+
4119 | | M | Running, doing mixed sequential reads/writes. |
4120 +------+-----+-----------------------------------------------------------+
4121 | | m | Running, doing mixed random reads/writes. |
4122 +------+-----+-----------------------------------------------------------+
4123 | | D | Running, doing sequential trims. |
4124 +------+-----+-----------------------------------------------------------+
4125 | | d | Running, doing random trims. |
4126 +------+-----+-----------------------------------------------------------+
4127 | | F | Running, currently waiting for :manpage:`fsync(2)`. |
4128 +------+-----+-----------------------------------------------------------+
4129 | | V | Running, doing verification of written data. |
4130 +------+-----+-----------------------------------------------------------+
4131 | f | | Thread finishing. |
4132 +------+-----+-----------------------------------------------------------+
4133 | E | | Thread exited, not reaped by main thread yet. |
4134 +------+-----+-----------------------------------------------------------+
4135 | _ | | Thread reaped. |
4136 +------+-----+-----------------------------------------------------------+
4137 | X | | Thread reaped, exited with an error. |
4138 +------+-----+-----------------------------------------------------------+
4139 | K | | Thread reaped, exited due to signal. |
4140 +------+-----+-----------------------------------------------------------+
4143 Example output was based on the following:
4144 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --runtime=58m \
4145 --time_based --rate=2512k --bs=256K --numjobs=10 \
4146 --name=readers --rw=read --name=writers --rw=write
4148 Fio will condense the thread string as not to take up more space on the command
4149 line than needed. For instance, if you have 10 readers and 10 writers running,
4150 the output would look like this::
4152 Jobs: 20 (f=20): [R(10),W(10)][4.0%][r=20.5MiB/s,w=23.5MiB/s][r=82,w=94 IOPS][eta 57m:36s]
4154 Note that the status string is displayed in order, so it's possible to tell which of
4155 the jobs are currently doing what. In the example above this means that jobs 1--10
4156 are readers and 11--20 are writers.
4158 The other values are fairly self explanatory -- number of threads currently
4159 running and doing I/O, the number of currently open files (f=), the estimated
4160 completion percentage, the rate of I/O since last check (read speed listed first,
4161 then write speed and optionally trim speed) in terms of bandwidth and IOPS,
4162 and time to completion for the current running group. It's impossible to estimate
4163 runtime of the following groups (if any).
4166 Example output was based on the following:
4167 TZ=UTC fio --iodepth=16 --ioengine=posixaio --filename=/tmp/fiofile \
4168 --direct=1 --size=100M --time_based --runtime=50s --rate_iops=89 \
4169 --bs=7K --name=Client1 --rw=write
4171 When fio is done (or interrupted by :kbd:`Ctrl-C`), it will show the data for
4172 each thread, group of threads, and disks in that order. For each overall thread (or
4173 group) the output looks like::
4175 Client1: (groupid=0, jobs=1): err= 0: pid=16109: Sat Jun 24 12:07:54 2017
4176 write: IOPS=88, BW=623KiB/s (638kB/s)(30.4MiB/50032msec)
4177 slat (nsec): min=500, max=145500, avg=8318.00, stdev=4781.50
4178 clat (usec): min=170, max=78367, avg=4019.02, stdev=8293.31
4179 lat (usec): min=174, max=78375, avg=4027.34, stdev=8291.79
4180 clat percentiles (usec):
4181 | 1.00th=[ 302], 5.00th=[ 326], 10.00th=[ 343], 20.00th=[ 363],
4182 | 30.00th=[ 392], 40.00th=[ 404], 50.00th=[ 416], 60.00th=[ 445],
4183 | 70.00th=[ 816], 80.00th=[ 6718], 90.00th=[12911], 95.00th=[21627],
4184 | 99.00th=[43779], 99.50th=[51643], 99.90th=[68682], 99.95th=[72877],
4186 bw ( KiB/s): min= 532, max= 686, per=0.10%, avg=622.87, stdev=24.82, samples= 100
4187 iops : min= 76, max= 98, avg=88.98, stdev= 3.54, samples= 100
4188 lat (usec) : 250=0.04%, 500=64.11%, 750=4.81%, 1000=2.79%
4189 lat (msec) : 2=4.16%, 4=1.84%, 10=4.90%, 20=11.33%, 50=5.37%
4190 lat (msec) : 100=0.65%
4191 cpu : usr=0.27%, sys=0.18%, ctx=12072, majf=0, minf=21
4192 IO depths : 1=85.0%, 2=13.1%, 4=1.8%, 8=0.1%, 16=0.0%, 32=0.0%, >=64=0.0%
4193 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
4194 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
4195 issued rwt: total=0,4450,0, short=0,0,0, dropped=0,0,0
4196 latency : target=0, window=0, percentile=100.00%, depth=8
4198 The job name (or first job's name when using :option:`group_reporting`) is printed,
4199 along with the group id, count of jobs being aggregated, last error id seen (which
4200 is 0 when there are no errors), pid/tid of that thread and the time the job/group
4201 completed. Below are the I/O statistics for each data direction performed (showing
4202 writes in the example above). In the order listed, they denote:
4205 The string before the colon shows the I/O direction the statistics
4206 are for. **IOPS** is the average I/Os performed per second. **BW**
4207 is the average bandwidth rate shown as: value in power of 2 format
4208 (value in power of 10 format). The last two values show: (**total
4209 I/O performed** in power of 2 format / **runtime** of that thread).
4212 Submission latency (**min** being the minimum, **max** being the
4213 maximum, **avg** being the average, **stdev** being the standard
4214 deviation). This is the time from when fio initialized the I/O
4215 to submission. For synchronous ioengines this includes the time
4216 up until just before the ioengine's queue function is called.
4217 For asynchronous ioengines this includes the time up through the
4218 completion of the ioengine's queue function (and commit function
4219 if it is defined). For sync I/O this row is not displayed as the
4220 slat is negligible. This value can be in nanoseconds,
4221 microseconds or milliseconds --- fio will choose the most
4222 appropriate base and print that (in the example above
4223 nanoseconds was the best scale). Note: in :option:`--minimal`
4224 mode latencies are always expressed in microseconds.
4227 Completion latency. Same names as slat, this denotes the time from
4228 submission to completion of the I/O pieces. For sync I/O, this
4229 represents the time from when the I/O was submitted to the
4230 operating system to when it was completed. For asynchronous
4231 ioengines this is the time from when the ioengine's queue (and
4232 commit if available) functions were completed to when the I/O's
4233 completion was reaped by fio.
4236 Total latency. Same names as slat and clat, this denotes the time from
4237 when fio created the I/O unit to completion of the I/O operation.
4238 It is the sum of submission and completion latency.
4241 Bandwidth statistics based on samples. Same names as the xlat stats,
4242 but also includes the number of samples taken (**samples**) and an
4243 approximate percentage of total aggregate bandwidth this thread
4244 received in its group (**per**). This last value is only really
4245 useful if the threads in this group are on the same disk, since they
4246 are then competing for disk access.
4249 IOPS statistics based on samples. Same names as bw.
4251 **lat (nsec/usec/msec)**
4252 The distribution of I/O completion latencies. This is the time from when
4253 I/O leaves fio and when it gets completed. Unlike the separate
4254 read/write/trim sections above, the data here and in the remaining
4255 sections apply to all I/Os for the reporting group. 250=0.04% means that
4256 0.04% of the I/Os completed in under 250us. 500=64.11% means that 64.11%
4257 of the I/Os required 250 to 499us for completion.
4260 CPU usage. User and system time, along with the number of context
4261 switches this thread went through, usage of system and user time, and
4262 finally the number of major and minor page faults. The CPU utilization
4263 numbers are averages for the jobs in that reporting group, while the
4264 context and fault counters are summed.
4267 The distribution of I/O depths over the job lifetime. The numbers are
4268 divided into powers of 2 and each entry covers depths from that value
4269 up to those that are lower than the next entry -- e.g., 16= covers
4270 depths from 16 to 31. Note that the range covered by a depth
4271 distribution entry can be different to the range covered by the
4272 equivalent submit/complete distribution entry.
4275 How many pieces of I/O were submitting in a single submit call. Each
4276 entry denotes that amount and below, until the previous entry -- e.g.,
4277 16=100% means that we submitted anywhere between 9 to 16 I/Os per submit
4278 call. Note that the range covered by a submit distribution entry can
4279 be different to the range covered by the equivalent depth distribution
4283 Like the above submit number, but for completions instead.
4286 The number of read/write/trim requests issued, and how many of them were
4290 These values are for :option:`latency_target` and related options. When
4291 these options are engaged, this section describes the I/O depth required
4292 to meet the specified latency target.
4295 Example output was based on the following:
4296 TZ=UTC fio --ioengine=null --iodepth=2 --size=100M --numjobs=2 \
4297 --rate_process=poisson --io_limit=32M --name=read --bs=128k \
4298 --rate=11M --name=write --rw=write --bs=2k --rate=700k
4300 After each client has been listed, the group statistics are printed. They
4301 will look like this::
4303 Run status group 0 (all jobs):
4304 READ: bw=20.9MiB/s (21.9MB/s), 10.4MiB/s-10.8MiB/s (10.9MB/s-11.3MB/s), io=64.0MiB (67.1MB), run=2973-3069msec
4305 WRITE: bw=1231KiB/s (1261kB/s), 616KiB/s-621KiB/s (630kB/s-636kB/s), io=64.0MiB (67.1MB), run=52747-53223msec
4307 For each data direction it prints:
4310 Aggregate bandwidth of threads in this group followed by the
4311 minimum and maximum bandwidth of all the threads in this group.
4312 Values outside of brackets are power-of-2 format and those
4313 within are the equivalent value in a power-of-10 format.
4315 Aggregate I/O performed of all threads in this group. The
4316 format is the same as bw.
4318 The smallest and longest runtimes of the threads in this group.
4320 And finally, the disk statistics are printed. This is Linux specific. They will look like this::
4322 Disk stats (read/write):
4323 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
4325 Each value is printed for both reads and writes, with reads first. The
4329 Number of I/Os performed by all groups.
4331 Number of merges performed by the I/O scheduler.
4333 Number of ticks we kept the disk busy.
4335 Total time spent in the disk queue.
4337 The disk utilization. A value of 100% means we kept the disk
4338 busy constantly, 50% would be a disk idling half of the time.
4340 It is also possible to get fio to dump the current output while it is running,
4341 without terminating the job. To do that, send fio the **USR1** signal. You can
4342 also get regularly timed dumps by using the :option:`--status-interval`
4343 parameter, or by creating a file in :file:`/tmp` named
4344 :file:`fio-dump-status`. If fio sees this file, it will unlink it and dump the
4345 current output status.
4351 For scripted usage where you typically want to generate tables or graphs of the
4352 results, fio can output the results in a semicolon separated format. The format
4353 is one long line of values, such as::
4355 2;card0;0;0;7139336;121836;60004;1;10109;27.932460;116.933948;220;126861;3495.446807;1085.368601;226;126864;3523.635629;1089.012448;24063;99944;50.275485%;59818.274627;5540.657370;7155060;122104;60004;1;8338;29.086342;117.839068;388;128077;5032.488518;1234.785715;391;128085;5061.839412;1236.909129;23436;100928;50.287926%;59964.832030;5644.844189;14.595833%;19.394167%;123706;0;7313;0.1%;0.1%;0.1%;0.1%;0.1%;0.1%;100.0%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.01%;0.02%;0.05%;0.16%;6.04%;40.40%;52.68%;0.64%;0.01%;0.00%;0.01%;0.00%;0.00%;0.00%;0.00%;0.00%
4356 A description of this job goes here.
4358 The job description (if provided) follows on a second line for terse v2.
4359 It appears on the same line for other terse versions.
4361 To enable terse output, use the :option:`--minimal` or
4362 :option:`--output-format`\=terse command line options. The
4363 first value is the version of the terse output format. If the output has to be
4364 changed for some reason, this number will be incremented by 1 to signify that
4367 Split up, the format is as follows (comments in brackets denote when a
4368 field was introduced or whether it's specific to some terse version):
4372 terse version, fio version [v3], jobname, groupid, error
4376 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
4377 Submission latency: min, max, mean, stdev (usec)
4378 Completion latency: min, max, mean, stdev (usec)
4379 Completion latency percentiles: 20 fields (see below)
4380 Total latency: min, max, mean, stdev (usec)
4381 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
4382 IOPS [v5]: min, max, mean, stdev, number of samples
4388 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
4389 Submission latency: min, max, mean, stdev (usec)
4390 Completion latency: min, max, mean, stdev (usec)
4391 Completion latency percentiles: 20 fields (see below)
4392 Total latency: min, max, mean, stdev (usec)
4393 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
4394 IOPS [v5]: min, max, mean, stdev, number of samples
4396 TRIM status [all but version 3]:
4398 Fields are similar to READ/WRITE status.
4402 user, system, context switches, major faults, minor faults
4406 <=1, 2, 4, 8, 16, 32, >=64
4408 I/O latencies microseconds::
4410 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
4412 I/O latencies milliseconds::
4414 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
4416 Disk utilization [v3]::
4418 disk name, read ios, write ios, read merges, write merges, read ticks, write ticks,
4419 time spent in queue, disk utilization percentage
4421 Additional Info (dependent on continue_on_error, default off)::
4423 total # errors, first error code
4425 Additional Info (dependent on description being set)::
4429 Completion latency percentiles can be a grouping of up to 20 sets, so for the
4430 terse output fio writes all of them. Each field will look like this::
4434 which is the Xth percentile, and the `usec` latency associated with it.
4436 For `Disk utilization`, all disks used by fio are shown. So for each disk there
4437 will be a disk utilization section.
4439 Below is a single line containing short names for each of the fields in the
4440 minimal output v3, separated by semicolons::
4442 terse_version_3;fio_version;jobname;groupid;error;read_kb;read_bandwidth_kb;read_iops;read_runtime_ms;read_slat_min_us;read_slat_max_us;read_slat_mean_us;read_slat_dev_us;read_clat_min_us;read_clat_max_us;read_clat_mean_us;read_clat_dev_us;read_clat_pct01;read_clat_pct02;read_clat_pct03;read_clat_pct04;read_clat_pct05;read_clat_pct06;read_clat_pct07;read_clat_pct08;read_clat_pct09;read_clat_pct10;read_clat_pct11;read_clat_pct12;read_clat_pct13;read_clat_pct14;read_clat_pct15;read_clat_pct16;read_clat_pct17;read_clat_pct18;read_clat_pct19;read_clat_pct20;read_tlat_min_us;read_lat_max_us;read_lat_mean_us;read_lat_dev_us;read_bw_min_kb;read_bw_max_kb;read_bw_agg_pct;read_bw_mean_kb;read_bw_dev_kb;write_kb;write_bandwidth_kb;write_iops;write_runtime_ms;write_slat_min_us;write_slat_max_us;write_slat_mean_us;write_slat_dev_us;write_clat_min_us;write_clat_max_us;write_clat_mean_us;write_clat_dev_us;write_clat_pct01;write_clat_pct02;write_clat_pct03;write_clat_pct04;write_clat_pct05;write_clat_pct06;write_clat_pct07;write_clat_pct08;write_clat_pct09;write_clat_pct10;write_clat_pct11;write_clat_pct12;write_clat_pct13;write_clat_pct14;write_clat_pct15;write_clat_pct16;write_clat_pct17;write_clat_pct18;write_clat_pct19;write_clat_pct20;write_tlat_min_us;write_lat_max_us;write_lat_mean_us;write_lat_dev_us;write_bw_min_kb;write_bw_max_kb;write_bw_agg_pct;write_bw_mean_kb;write_bw_dev_kb;cpu_user;cpu_sys;cpu_csw;cpu_mjf;cpu_minf;iodepth_1;iodepth_2;iodepth_4;iodepth_8;iodepth_16;iodepth_32;iodepth_64;lat_2us;lat_4us;lat_10us;lat_20us;lat_50us;lat_100us;lat_250us;lat_500us;lat_750us;lat_1000us;lat_2ms;lat_4ms;lat_10ms;lat_20ms;lat_50ms;lat_100ms;lat_250ms;lat_500ms;lat_750ms;lat_1000ms;lat_2000ms;lat_over_2000ms;disk_name;disk_read_iops;disk_write_iops;disk_read_merges;disk_write_merges;disk_read_ticks;write_ticks;disk_queue_time;disk_util
4444 In client/server mode terse output differs from what appears when jobs are run
4445 locally. Disk utilization data is omitted from the standard terse output and
4446 for v3 and later appears on its own separate line at the end of each terse
4453 The `json` output format is intended to be both human readable and convenient
4454 for automated parsing. For the most part its sections mirror those of the
4455 `normal` output. The `runtime` value is reported in msec and the `bw` value is
4456 reported in 1024 bytes per second units.
4462 The `json+` output format is identical to the `json` output format except that it
4463 adds a full dump of the completion latency bins. Each `bins` object contains a
4464 set of (key, value) pairs where keys are latency durations and values count how
4465 many I/Os had completion latencies of the corresponding duration. For example,
4468 "bins" : { "87552" : 1, "89600" : 1, "94720" : 1, "96768" : 1, "97792" : 1, "99840" : 1, "100864" : 2, "103936" : 6, "104960" : 534, "105984" : 5995, "107008" : 7529, ... }
4470 This data indicates that one I/O required 87,552ns to complete, two I/Os required
4471 100,864ns to complete, and 7529 I/Os required 107,008ns to complete.
4473 Also included with fio is a Python script `fio_jsonplus_clat2csv` that takes
4474 json+ output and generates CSV-formatted latency data suitable for plotting.
4476 The latency durations actually represent the midpoints of latency intervals.
4477 For details refer to :file:`stat.h`.
4483 There are two trace file format that you can encounter. The older (v1) format is
4484 unsupported since version 1.20-rc3 (March 2008). It will still be described
4485 below in case that you get an old trace and want to understand it.
4487 In any case the trace is a simple text file with a single action per line.
4490 Trace file format v1
4491 ~~~~~~~~~~~~~~~~~~~~
4493 Each line represents a single I/O action in the following format::
4497 where `rw=0/1` for read/write, and the `offset` and `length` entries being in bytes.
4499 This format is not supported in fio versions >= 1.20-rc3.
4502 Trace file format v2
4503 ~~~~~~~~~~~~~~~~~~~~
4505 The second version of the trace file format was added in fio version 1.17. It
4506 allows to access more than one file per trace and has a bigger set of possible
4509 The first line of the trace file has to be::
4513 Following this can be lines in two different formats, which are described below.
4515 The file management format::
4519 The `filename` is given as an absolute path. The `action` can be one of these:
4522 Add the given `filename` to the trace.
4524 Open the file with the given `filename`. The `filename` has to have
4525 been added with the **add** action before.
4527 Close the file with the given `filename`. The file has to have been
4531 The file I/O action format::
4533 filename action offset length
4535 The `filename` is given as an absolute path, and has to have been added and
4536 opened before it can be used with this format. The `offset` and `length` are
4537 given in bytes. The `action` can be one of these:
4540 Wait for `offset` microseconds. Everything below 100 is discarded.
4541 The time is relative to the previous `wait` statement. Note that
4542 action `wait` is not allowed as of version 3, as the same behavior
4543 can be achieved using timestamps.
4545 Read `length` bytes beginning from `offset`.
4547 Write `length` bytes beginning from `offset`.
4549 :manpage:`fsync(2)` the file.
4551 :manpage:`fdatasync(2)` the file.
4553 Trim the given file from the given `offset` for `length` bytes.
4556 Trace file format v3
4557 ~~~~~~~~~~~~~~~~~~~~
4559 The third version of the trace file format was added in fio version 3.31. It
4560 forces each action to have a timestamp associated with it.
4562 The first line of the trace file has to be::
4566 Following this can be lines in two different formats, which are described below.
4568 The file management format::
4570 timestamp filename action
4572 The file I/O action format::
4574 timestamp filename action offset length
4576 The `timestamp` is relative to the beginning of the run (ie starts at 0). The
4577 `filename`, `action`, `offset` and `length` are identical to version 2, except
4578 that version 3 does not allow the `wait` action.
4581 I/O Replay - Merging Traces
4582 ---------------------------
4584 Colocation is a common practice used to get the most out of a machine.
4585 Knowing which workloads play nicely with each other and which ones don't is
4586 a much harder task. While fio can replay workloads concurrently via multiple
4587 jobs, it leaves some variability up to the scheduler making results harder to
4588 reproduce. Merging is a way to make the order of events consistent.
4590 Merging is integrated into I/O replay and done when a
4591 :option:`merge_blktrace_file` is specified. The list of files passed to
4592 :option:`read_iolog` go through the merge process and output a single file
4593 stored to the specified file. The output file is passed on as if it were the
4594 only file passed to :option:`read_iolog`. An example would look like::
4596 $ fio --read_iolog="<file1>:<file2>" --merge_blktrace_file="<output_file>"
4598 Creating only the merged file can be done by passing the command line argument
4599 :option:`--merge-blktrace-only`.
4601 Scaling traces can be done to see the relative impact of any particular trace
4602 being slowed down or sped up. :option:`merge_blktrace_scalars` takes in a colon
4603 separated list of percentage scalars. It is index paired with the files passed
4604 to :option:`read_iolog`.
4606 With scaling, it may be desirable to match the running time of all traces.
4607 This can be done with :option:`merge_blktrace_iters`. It is index paired with
4608 :option:`read_iolog` just like :option:`merge_blktrace_scalars`.
4610 In an example, given two traces, A and B, each 60s long. If we want to see
4611 the impact of trace A issuing IOs twice as fast and repeat trace A over the
4612 runtime of trace B, the following can be done::
4614 $ fio --read_iolog="<trace_a>:"<trace_b>" --merge_blktrace_file"<output_file>" --merge_blktrace_scalars="50:100" --merge_blktrace_iters="2:1"
4616 This runs trace A at 2x the speed twice for approximately the same runtime as
4617 a single run of trace B.
4620 CPU idleness profiling
4621 ----------------------
4623 In some cases, we want to understand CPU overhead in a test. For example, we
4624 test patches for the specific goodness of whether they reduce CPU usage.
4625 Fio implements a balloon approach to create a thread per CPU that runs at idle
4626 priority, meaning that it only runs when nobody else needs the cpu.
4627 By measuring the amount of work completed by the thread, idleness of each CPU
4628 can be derived accordingly.
4630 An unit work is defined as touching a full page of unsigned characters. Mean and
4631 standard deviation of time to complete an unit work is reported in "unit work"
4632 section. Options can be chosen to report detailed percpu idleness or overall
4633 system idleness by aggregating percpu stats.
4636 Verification and triggers
4637 -------------------------
4639 Fio is usually run in one of two ways, when data verification is done. The first
4640 is a normal write job of some sort with verify enabled. When the write phase has
4641 completed, fio switches to reads and verifies everything it wrote. The second
4642 model is running just the write phase, and then later on running the same job
4643 (but with reads instead of writes) to repeat the same I/O patterns and verify
4644 the contents. Both of these methods depend on the write phase being completed,
4645 as fio otherwise has no idea how much data was written.
4647 With verification triggers, fio supports dumping the current write state to
4648 local files. Then a subsequent read verify workload can load this state and know
4649 exactly where to stop. This is useful for testing cases where power is cut to a
4650 server in a managed fashion, for instance.
4652 A verification trigger consists of two things:
4654 1) Storing the write state of each job.
4655 2) Executing a trigger command.
4657 The write state is relatively small, on the order of hundreds of bytes to single
4658 kilobytes. It contains information on the number of completions done, the last X
4661 A trigger is invoked either through creation ('touch') of a specified file in
4662 the system, or through a timeout setting. If fio is run with
4663 :option:`--trigger-file`\= :file:`/tmp/trigger-file`, then it will continually
4664 check for the existence of :file:`/tmp/trigger-file`. When it sees this file, it
4665 will fire off the trigger (thus saving state, and executing the trigger
4668 For client/server runs, there's both a local and remote trigger. If fio is
4669 running as a server backend, it will send the job states back to the client for
4670 safe storage, then execute the remote trigger, if specified. If a local trigger
4671 is specified, the server will still send back the write state, but the client
4672 will then execute the trigger.
4674 Verification trigger example
4675 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
4677 Let's say we want to run a powercut test on the remote Linux machine 'server'.
4678 Our write workload is in :file:`write-test.fio`. We want to cut power to 'server' at
4679 some point during the run, and we'll run this test from the safety or our local
4680 machine, 'localbox'. On the server, we'll start the fio backend normally::
4682 server# fio --server
4684 and on the client, we'll fire off the workload::
4686 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
4688 We set :file:`/tmp/my-trigger` as the trigger file, and we tell fio to execute::
4690 echo b > /proc/sysrq-trigger
4692 on the server once it has received the trigger and sent us the write state. This
4693 will work, but it's not **really** cutting power to the server, it's merely
4694 abruptly rebooting it. If we have a remote way of cutting power to the server
4695 through IPMI or similar, we could do that through a local trigger command
4696 instead. Let's assume we have a script that does IPMI reboot of a given hostname,
4697 ipmi-reboot. On localbox, we could then have run fio with a local trigger
4700 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
4702 For this case, fio would wait for the server to send us the write state, then
4703 execute ``ipmi-reboot server`` when that happened.
4705 Loading verify state
4706 ~~~~~~~~~~~~~~~~~~~~
4708 To load stored write state, a read verification job file must contain the
4709 :option:`verify_state_load` option. If that is set, fio will load the previously
4710 stored state. For a local fio run this is done by loading the files directly,
4711 and on a client/server run, the server backend will ask the client to send the
4712 files over and load them from there.
4718 Fio supports a variety of log file formats, for logging latencies, bandwidth,
4719 and IOPS. The logs share a common format, which looks like this:
4721 *time* (`msec`), *value*, *data direction*, *block size* (`bytes`),
4722 *offset* (`bytes`), *command priority*
4724 *Time* for the log entry is always in milliseconds. The *value* logged depends
4725 on the type of log, it will be one of the following:
4728 Value is latency in nsecs
4734 *Data direction* is one of the following:
4743 The entry's *block size* is always in bytes. The *offset* is the position in bytes
4744 from the start of the file for that particular I/O. The logging of the offset can be
4745 toggled with :option:`log_offset`.
4747 *Command priority* is 0 for normal priority and 1 for high priority. This is controlled
4748 by the ioengine specific :option:`cmdprio_percentage`.
4750 Fio defaults to logging every individual I/O but when windowed logging is set
4751 through :option:`log_avg_msec`, either the average (by default) or the maximum
4752 (:option:`log_max_value` is set) *value* seen over the specified period of time
4753 is recorded. Each *data direction* seen within the window period will aggregate
4754 its values in a separate row. Further, when using windowed logging the *block
4755 size* and *offset* entries will always contain 0.
4761 Normally fio is invoked as a stand-alone application on the machine where the
4762 I/O workload should be generated. However, the backend and frontend of fio can
4763 be run separately i.e., the fio server can generate an I/O workload on the "Device
4764 Under Test" while being controlled by a client on another machine.
4766 Start the server on the machine which has access to the storage DUT::
4770 where `args` defines what fio listens to. The arguments are of the form
4771 ``type,hostname`` or ``IP,port``. *type* is either ``ip`` (or ip4) for TCP/IP
4772 v4, ``ip6`` for TCP/IP v6, or ``sock`` for a local unix domain socket.
4773 *hostname* is either a hostname or IP address, and *port* is the port to listen
4774 to (only valid for TCP/IP, not a local socket). Some examples:
4778 Start a fio server, listening on all interfaces on the default port (8765).
4780 2) ``fio --server=ip:hostname,4444``
4782 Start a fio server, listening on IP belonging to hostname and on port 4444.
4784 3) ``fio --server=ip6:::1,4444``
4786 Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
4788 4) ``fio --server=,4444``
4790 Start a fio server, listening on all interfaces on port 4444.
4792 5) ``fio --server=1.2.3.4``
4794 Start a fio server, listening on IP 1.2.3.4 on the default port.
4796 6) ``fio --server=sock:/tmp/fio.sock``
4798 Start a fio server, listening on the local socket :file:`/tmp/fio.sock`.
4800 Once a server is running, a "client" can connect to the fio server with::
4802 fio <local-args> --client=<server> <remote-args> <job file(s)>
4804 where `local-args` are arguments for the client where it is running, `server`
4805 is the connect string, and `remote-args` and `job file(s)` are sent to the
4806 server. The `server` string follows the same format as it does on the server
4807 side, to allow IP/hostname/socket and port strings.
4809 Fio can connect to multiple servers this way::
4811 fio --client=<server1> <job file(s)> --client=<server2> <job file(s)>
4813 If the job file is located on the fio server, then you can tell the server to
4814 load a local file as well. This is done by using :option:`--remote-config` ::
4816 fio --client=server --remote-config /path/to/file.fio
4818 Then fio will open this local (to the server) job file instead of being passed
4819 one from the client.
4821 If you have many servers (example: 100 VMs/containers), you can input a pathname
4822 of a file containing host IPs/names as the parameter value for the
4823 :option:`--client` option. For example, here is an example :file:`host.list`
4824 file containing 2 hostnames::
4826 host1.your.dns.domain
4827 host2.your.dns.domain
4829 The fio command would then be::
4831 fio --client=host.list <job file(s)>
4833 In this mode, you cannot input server-specific parameters or job files -- all
4834 servers receive the same job file.
4836 In order to let ``fio --client`` runs use a shared filesystem from multiple
4837 hosts, ``fio --client`` now prepends the IP address of the server to the
4838 filename. For example, if fio is using the directory :file:`/mnt/nfs/fio` and is
4839 writing filename :file:`fileio.tmp`, with a :option:`--client` `hostfile`
4840 containing two hostnames ``h1`` and ``h2`` with IP addresses 192.168.10.120 and
4841 192.168.10.121, then fio will create two files::
4843 /mnt/nfs/fio/192.168.10.120.fileio.tmp
4844 /mnt/nfs/fio/192.168.10.121.fileio.tmp
4846 Terse output in client/server mode will differ slightly from what is produced
4847 when fio is run in stand-alone mode. See the terse output section for details.