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 Show available debug options.
99 .. option:: --parse-only
101 Parse options only, don't start any I/O.
103 .. option:: --output=filename
105 Write output to file `filename`.
107 .. option:: --output-format=format
109 Set the reporting `format` to `normal`, `terse`, `json`, or `json+`. Multiple
110 formats can be selected, separated by a comma. `terse` is a CSV based
111 format. `json+` is like `json`, except it adds a full dump of the latency
114 .. option:: --bandwidth-log
116 Generate aggregate bandwidth logs.
118 .. option:: --minimal
120 Print statistics in a terse, semicolon-delimited format.
122 .. option:: --append-terse
124 Print statistics in selected mode AND terse, semicolon-delimited format.
125 **Deprecated**, use :option:`--output-format` instead to select multiple
128 .. option:: --terse-version=version
130 Set terse `version` output format (default 3, or 2 or 4 or 5).
132 .. option:: --version
134 Print version information and exit.
138 Print a summary of the command line options and exit.
140 .. option:: --cpuclock-test
142 Perform test and validation of internal CPU clock.
144 .. option:: --crctest=[test]
146 Test the speed of the built-in checksumming functions. If no argument is
147 given, all of them are tested. Alternatively, a comma separated list can
148 be passed, in which case the given ones are tested.
150 .. option:: --cmdhelp=command
152 Print help information for `command`. May be ``all`` for all commands.
154 .. option:: --enghelp=[ioengine[,command]]
156 List all commands defined by `ioengine`, or print help for `command`
157 defined by `ioengine`. If no `ioengine` is given, list all
160 .. option:: --showcmd=jobfile
162 Convert `jobfile` to a set of command-line options.
164 .. option:: --readonly
166 Turn on safety read-only checks, preventing writes and trims. The
167 ``--readonly`` option is an extra safety guard to prevent users from
168 accidentally starting a write or trim workload when that is not desired.
169 Fio will only modify the device under test if
170 `rw=write/randwrite/rw/randrw/trim/randtrim/trimwrite` is given. This
171 safety net can be used as an extra precaution.
173 .. option:: --eta=when
175 Specifies when real-time ETA estimate should be printed. `when` may be
176 `always`, `never` or `auto`. `auto` is the default, it prints ETA
177 when requested if the output is a TTY. `always` disregards the output
178 type, and prints ETA when requested. `never` never prints ETA.
180 .. option:: --eta-interval=time
182 By default, fio requests client ETA status roughly every second. With
183 this option, the interval is configurable. Fio imposes a minimum
184 allowed time to avoid flooding the console, less than 250 msec is
187 .. option:: --eta-newline=time
189 Force a new line for every `time` period passed. When the unit is omitted,
190 the value is interpreted in seconds.
192 .. option:: --status-interval=time
194 Force a full status dump of cumulative (from job start) values at `time`
195 intervals. This option does *not* provide per-period measurements. So
196 values such as bandwidth are running averages. When the time unit is omitted,
197 `time` is interpreted in seconds. Note that using this option with
198 ``--output-format=json`` will yield output that technically isn't valid
199 json, since the output will be collated sets of valid json. It will need
200 to be split into valid sets of json after the run.
202 .. option:: --section=name
204 Only run specified section `name` in job file. Multiple sections can be specified.
205 The ``--section`` option allows one to combine related jobs into one file.
206 E.g. one job file could define light, moderate, and heavy sections. Tell
207 fio to run only the "heavy" section by giving ``--section=heavy``
208 command line option. One can also specify the "write" operations in one
209 section and "verify" operation in another section. The ``--section`` option
210 only applies to job sections. The reserved *global* section is always
213 .. option:: --alloc-size=kb
215 Set the internal smalloc pool size to `kb` in KiB. The
216 ``--alloc-size`` switch allows one to use a larger pool size for smalloc.
217 If running large jobs with randommap enabled, fio can run out of memory.
218 Smalloc is an internal allocator for shared structures from a fixed size
219 memory pool and can grow to 16 pools. The pool size defaults to 16MiB.
221 NOTE: While running :file:`.fio_smalloc.*` backing store files are visible
224 .. option:: --warnings-fatal
226 All fio parser warnings are fatal, causing fio to exit with an
229 .. option:: --max-jobs=nr
231 Set the maximum number of threads/processes to support to `nr`.
232 NOTE: On Linux, it may be necessary to increase the shared-memory
233 limit (:file:`/proc/sys/kernel/shmmax`) if fio runs into errors while
236 .. option:: --server=args
238 Start a backend server, with `args` specifying what to listen to.
239 See `Client/Server`_ section.
241 .. option:: --daemonize=pidfile
243 Background a fio server, writing the pid to the given `pidfile` file.
245 .. option:: --client=hostname
247 Instead of running the jobs locally, send and run them on the given `hostname`
248 or set of `hostname`\s. See `Client/Server`_ section.
250 .. option:: --remote-config=file
252 Tell fio server to load this local `file`.
254 .. option:: --idle-prof=option
256 Report CPU idleness. `option` is one of the following:
259 Run unit work calibration only and exit.
262 Show aggregate system idleness and unit work.
265 As **system** but also show per CPU idleness.
267 .. option:: --inflate-log=log
269 Inflate and output compressed `log`.
271 .. option:: --trigger-file=file
273 Execute trigger command when `file` exists.
275 .. option:: --trigger-timeout=time
277 Execute trigger at this `time`.
279 .. option:: --trigger=command
281 Set this `command` as local trigger.
283 .. option:: --trigger-remote=command
285 Set this `command` as remote trigger.
287 .. option:: --aux-path=path
289 Use the directory specified by `path` for generated state files instead
290 of the current working directory.
292 Any parameters following the options will be assumed to be job files, unless
293 they match a job file parameter. Multiple job files can be listed and each job
294 file will be regarded as a separate group. Fio will :option:`stonewall`
295 execution between each group.
301 As previously described, fio accepts one or more job files describing what it is
302 supposed to do. The job file format is the classic ini file, where the names
303 enclosed in [] brackets define the job name. You are free to use any ASCII name
304 you want, except *global* which has special meaning. Following the job name is
305 a sequence of zero or more parameters, one per line, that define the behavior of
306 the job. If the first character in a line is a ';' or a '#', the entire line is
307 discarded as a comment.
309 A *global* section sets defaults for the jobs described in that file. A job may
310 override a *global* section parameter, and a job file may even have several
311 *global* sections if so desired. A job is only affected by a *global* section
314 The :option:`--cmdhelp` option also lists all options. If used with a `command`
315 argument, :option:`--cmdhelp` will detail the given `command`.
317 See the `examples/` directory for inspiration on how to write job files. Note
318 the copyright and license requirements currently apply to `examples/` files.
320 So let's look at a really simple job file that defines two processes, each
321 randomly reading from a 128MiB file:
325 ; -- start job file --
336 As you can see, the job file sections themselves are empty as all the described
337 parameters are shared. As no :option:`filename` option is given, fio makes up a
338 `filename` for each of the jobs as it sees fit. On the command line, this job
339 would look as follows::
341 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
344 Let's look at an example that has a number of processes writing randomly to
349 ; -- start job file --
360 Here we have no *global* section, as we only have one job defined anyway. We
361 want to use async I/O here, with a depth of 4 for each file. We also increased
362 the buffer size used to 32KiB and define numjobs to 4 to fork 4 identical
363 jobs. The result is 4 processes each randomly writing to their own 64MiB
364 file. Instead of using the above job file, you could have given the parameters
365 on the command line. For this case, you would specify::
367 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
369 When fio is utilized as a basis of any reasonably large test suite, it might be
370 desirable to share a set of standardized settings across multiple job files.
371 Instead of copy/pasting such settings, any section may pull in an external
372 :file:`filename.fio` file with *include filename* directive, as in the following
375 ; -- start job file including.fio --
379 include glob-include.fio
386 include test-include.fio
387 ; -- end job file including.fio --
391 ; -- start job file glob-include.fio --
394 ; -- end job file glob-include.fio --
398 ; -- start job file test-include.fio --
401 ; -- end job file test-include.fio --
403 Settings pulled into a section apply to that section only (except *global*
404 section). Include directives may be nested in that any included file may contain
405 further include directive(s). Include files may not contain [] sections.
408 Environment variables
409 ~~~~~~~~~~~~~~~~~~~~~
411 Fio also supports environment variable expansion in job files. Any sub-string of
412 the form ``${VARNAME}`` as part of an option value (in other words, on the right
413 of the '='), will be expanded to the value of the environment variable called
414 `VARNAME`. If no such environment variable is defined, or `VARNAME` is the
415 empty string, the empty string will be substituted.
417 As an example, let's look at a sample fio invocation and job file::
419 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
423 ; -- start job file --
430 This will expand to the following equivalent job file at runtime:
434 ; -- start job file --
441 Fio ships with a few example job files, you can also look there for inspiration.
446 Additionally, fio has a set of reserved keywords that will be replaced
447 internally with the appropriate value. Those keywords are:
451 The architecture page size of the running system.
455 Megabytes of total memory in the system.
459 Number of online available CPUs.
461 These can be used on the command line or in the job file, and will be
462 automatically substituted with the current system values when the job is
463 run. Simple math is also supported on these keywords, so you can perform actions
468 and get that properly expanded to 8 times the size of memory in the machine.
474 This section describes in details each parameter associated with a job. Some
475 parameters take an option of a given type, such as an integer or a
476 string. Anywhere a numeric value is required, an arithmetic expression may be
477 used, provided it is surrounded by parentheses. Supported operators are:
486 For time values in expressions, units are microseconds by default. This is
487 different than for time values not in expressions (not enclosed in
488 parentheses). The following types are used:
495 String: A sequence of alphanumeric characters.
498 Integer with possible time suffix. Without a unit value is interpreted as
499 seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
500 hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and
501 'us' (or 'usec') for microseconds. For example, use 10m for 10 minutes.
506 Integer. A whole number value, which may contain an integer prefix
507 and an integer suffix:
509 [*integer prefix*] **number** [*integer suffix*]
511 The optional *integer prefix* specifies the number's base. The default
512 is decimal. *0x* specifies hexadecimal.
514 The optional *integer suffix* specifies the number's units, and includes an
515 optional unit prefix and an optional unit. For quantities of data, the
516 default unit is bytes. For quantities of time, the default unit is seconds
517 unless otherwise specified.
519 With :option:`kb_base`\=1000, fio follows international standards for unit
520 prefixes. To specify power-of-10 decimal values defined in the
521 International System of Units (SI):
523 * *K* -- means kilo (K) or 1000
524 * *M* -- means mega (M) or 1000**2
525 * *G* -- means giga (G) or 1000**3
526 * *T* -- means tera (T) or 1000**4
527 * *P* -- means peta (P) or 1000**5
529 To specify power-of-2 binary values defined in IEC 80000-13:
531 * *Ki* -- means kibi (Ki) or 1024
532 * *Mi* -- means mebi (Mi) or 1024**2
533 * *Gi* -- means gibi (Gi) or 1024**3
534 * *Ti* -- means tebi (Ti) or 1024**4
535 * *Pi* -- means pebi (Pi) or 1024**5
537 With :option:`kb_base`\=1024 (the default), the unit prefixes are opposite
538 from those specified in the SI and IEC 80000-13 standards to provide
539 compatibility with old scripts. For example, 4k means 4096.
541 For quantities of data, an optional unit of 'B' may be included
542 (e.g., 'kB' is the same as 'k').
544 The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
545 not milli). 'b' and 'B' both mean byte, not bit.
547 Examples with :option:`kb_base`\=1000:
549 * *4 KiB*: 4096, 4096b, 4096B, 4ki, 4kib, 4kiB, 4Ki, 4KiB
550 * *1 MiB*: 1048576, 1mi, 1024ki
551 * *1 MB*: 1000000, 1m, 1000k
552 * *1 TiB*: 1099511627776, 1ti, 1024gi, 1048576mi
553 * *1 TB*: 1000000000, 1t, 1000m, 1000000k
555 Examples with :option:`kb_base`\=1024 (default):
557 * *4 KiB*: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
558 * *1 MiB*: 1048576, 1m, 1024k
559 * *1 MB*: 1000000, 1mi, 1000ki
560 * *1 TiB*: 1099511627776, 1t, 1024g, 1048576m
561 * *1 TB*: 1000000000, 1ti, 1000mi, 1000000ki
563 To specify times (units are not case sensitive):
567 * *M* -- means minutes
568 * *s* -- or sec means seconds (default)
569 * *ms* -- or *msec* means milliseconds
570 * *us* -- or *usec* means microseconds
572 If the option accepts an upper and lower range, use a colon ':' or
573 minus '-' to separate such values. See :ref:`irange <irange>`.
574 If the lower value specified happens to be larger than the upper value
575 the two values are swapped.
580 Boolean. Usually parsed as an integer, however only defined for
581 true and false (1 and 0).
586 Integer range with suffix. Allows value range to be given, such as
587 1024-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
588 option allows two sets of ranges, they can be specified with a ',' or '/'
589 delimiter: 1k-4k/8k-32k. Also see :ref:`int <int>`.
592 A list of floating point numbers, separated by a ':' character.
594 With the above in mind, here follows the complete list of fio job parameters.
600 .. option:: kb_base=int
602 Select the interpretation of unit prefixes in input parameters.
605 Inputs comply with IEC 80000-13 and the International
606 System of Units (SI). Use:
608 - power-of-2 values with IEC prefixes (e.g., KiB)
609 - power-of-10 values with SI prefixes (e.g., kB)
612 Compatibility mode (default). To avoid breaking old scripts:
614 - power-of-2 values with SI prefixes
615 - power-of-10 values with IEC prefixes
617 See :option:`bs` for more details on input parameters.
619 Outputs always use correct prefixes. Most outputs include both
622 bw=2383.3kB/s (2327.4KiB/s)
624 If only one value is reported, then kb_base selects the one to use:
626 **1000** -- SI prefixes
628 **1024** -- IEC prefixes
630 .. option:: unit_base=int
632 Base unit for reporting. Allowed values are:
635 Use auto-detection (default).
647 ASCII name of the job. This may be used to override the name printed by fio
648 for this job. Otherwise the job name is used. On the command line this
649 parameter has the special purpose of also signaling the start of a new job.
651 .. option:: description=str
653 Text description of the job. Doesn't do anything except dump this text
654 description when this job is run. It's not parsed.
656 .. option:: loops=int
658 Run the specified number of iterations of this job. Used to repeat the same
659 workload a given number of times. Defaults to 1.
661 .. option:: numjobs=int
663 Create the specified number of clones of this job. Each clone of job
664 is spawned as an independent thread or process. May be used to setup a
665 larger number of threads/processes doing the same thing. Each thread is
666 reported separately; to see statistics for all clones as a whole, use
667 :option:`group_reporting` in conjunction with :option:`new_group`.
668 See :option:`--max-jobs`. Default: 1.
671 Time related parameters
672 ~~~~~~~~~~~~~~~~~~~~~~~
674 .. option:: runtime=time
676 Tell fio to terminate processing after the specified period of time. It
677 can be quite hard to determine for how long a specified job will run, so
678 this parameter is handy to cap the total runtime to a given time. When
679 the unit is omitted, the value is interpreted in seconds.
681 .. option:: time_based
683 If set, fio will run for the duration of the :option:`runtime` specified
684 even if the file(s) are completely read or written. It will simply loop over
685 the same workload as many times as the :option:`runtime` allows.
687 .. option:: startdelay=irange(time)
689 Delay the start of job for the specified amount of time. Can be a single
690 value or a range. When given as a range, each thread will choose a value
691 randomly from within the range. Value is in seconds if a unit is omitted.
693 .. option:: ramp_time=time
695 If set, fio will run the specified workload for this amount of time before
696 logging any performance numbers. Useful for letting performance settle
697 before logging results, thus minimizing the runtime required for stable
698 results. Note that the ``ramp_time`` is considered lead in time for a job,
699 thus it will increase the total runtime if a special timeout or
700 :option:`runtime` is specified. When the unit is omitted, the value is
703 .. option:: clocksource=str
705 Use the given clocksource as the base of timing. The supported options are:
708 :manpage:`gettimeofday(2)`
711 :manpage:`clock_gettime(2)`
714 Internal CPU clock source
716 cpu is the preferred clocksource if it is reliable, as it is very fast (and
717 fio is heavy on time calls). Fio will automatically use this clocksource if
718 it's supported and considered reliable on the system it is running on,
719 unless another clocksource is specifically set. For x86/x86-64 CPUs, this
720 means supporting TSC Invariant.
722 .. option:: gtod_reduce=bool
724 Enable all of the :manpage:`gettimeofday(2)` reducing options
725 (:option:`disable_clat`, :option:`disable_slat`, :option:`disable_bw_measurement`) plus
726 reduce precision of the timeout somewhat to really shrink the
727 :manpage:`gettimeofday(2)` call count. With this option enabled, we only do
728 about 0.4% of the :manpage:`gettimeofday(2)` calls we would have done if all
729 time keeping was enabled.
731 .. option:: gtod_cpu=int
733 Sometimes it's cheaper to dedicate a single thread of execution to just
734 getting the current time. Fio (and databases, for instance) are very
735 intensive on :manpage:`gettimeofday(2)` calls. With this option, you can set
736 one CPU aside for doing nothing but logging current time to a shared memory
737 location. Then the other threads/processes that run I/O workloads need only
738 copy that segment, instead of entering the kernel with a
739 :manpage:`gettimeofday(2)` call. The CPU set aside for doing these time
740 calls will be excluded from other uses. Fio will manually clear it from the
741 CPU mask of other jobs.
747 .. option:: directory=str
749 Prefix filenames with this directory. Used to place files in a different
750 location than :file:`./`. You can specify a number of directories by
751 separating the names with a ':' character. These directories will be
752 assigned equally distributed to job clones created by :option:`numjobs` as
753 long as they are using generated filenames. If specific `filename(s)` are
754 set fio will use the first listed directory, and thereby matching the
755 `filename` semantic (which generates a file for each clone if not
756 specified, but lets all clones use the same file if set).
758 See the :option:`filename` option for information on how to escape "``:``" and
759 "``\``" characters within the directory path itself.
761 Note: To control the directory fio will use for internal state files
762 use :option:`--aux-path`.
764 .. option:: filename=str
766 Fio normally makes up a `filename` based on the job name, thread number, and
767 file number (see :option:`filename_format`). If you want to share files
768 between threads in a job or several
769 jobs with fixed file paths, specify a `filename` for each of them to override
770 the default. If the ioengine is file based, you can specify a number of files
771 by separating the names with a ':' colon. So if you wanted a job to open
772 :file:`/dev/sda` and :file:`/dev/sdb` as the two working files, you would use
773 ``filename=/dev/sda:/dev/sdb``. This also means that whenever this option is
774 specified, :option:`nrfiles` is ignored. The size of regular files specified
775 by this option will be :option:`size` divided by number of files unless an
776 explicit size is specified by :option:`filesize`.
778 Each colon and backslash in the wanted path must be escaped with a ``\``
779 character. For instance, if the path is :file:`/dev/dsk/foo@3,0:c` then you
780 would use ``filename=/dev/dsk/foo@3,0\:c`` and if the path is
781 :file:`F:\\filename` then you would use ``filename=F\:\\filename``.
783 On Windows, disk devices are accessed as :file:`\\\\.\\PhysicalDrive0` for
784 the first device, :file:`\\\\.\\PhysicalDrive1` for the second etc.
785 Note: Windows and FreeBSD prevent write access to areas
786 of the disk containing in-use data (e.g. filesystems).
788 The filename "`-`" is a reserved name, meaning *stdin* or *stdout*. Which
789 of the two depends on the read/write direction set.
791 .. option:: filename_format=str
793 If sharing multiple files between jobs, it is usually necessary to have fio
794 generate the exact names that you want. By default, fio will name a file
795 based on the default file format specification of
796 :file:`jobname.jobnumber.filenumber`. With this option, that can be
797 customized. Fio will recognize and replace the following keywords in this
801 The name of the worker thread or process.
803 The incremental number of the worker thread or process.
805 The incremental number of the file for that worker thread or
808 To have dependent jobs share a set of files, this option can be set to have
809 fio generate filenames that are shared between the two. For instance, if
810 :file:`testfiles.$filenum` is specified, file number 4 for any job will be
811 named :file:`testfiles.4`. The default of :file:`$jobname.$jobnum.$filenum`
812 will be used if no other format specifier is given.
814 If you specify a path then the directories will be created up to the
815 main directory for the file. So for example if you specify
816 ``filename_format=a/b/c/$jobnum`` then the directories a/b/c will be
817 created before the file setup part of the job. If you specify
818 :option:`directory` then the path will be relative that directory,
819 otherwise it is treated as the absolute path.
821 .. option:: unique_filename=bool
823 To avoid collisions between networked clients, fio defaults to prefixing any
824 generated filenames (with a directory specified) with the source of the
825 client connecting. To disable this behavior, set this option to 0.
827 .. option:: opendir=str
829 Recursively open any files below directory `str`.
831 .. option:: lockfile=str
833 Fio defaults to not locking any files before it does I/O to them. If a file
834 or file descriptor is shared, fio can serialize I/O to that file to make the
835 end result consistent. This is usual for emulating real workloads that share
836 files. The lock modes are:
839 No locking. The default.
841 Only one thread or process may do I/O at a time, excluding all
844 Read-write locking on the file. Many readers may
845 access the file at the same time, but writes get exclusive access.
847 .. option:: nrfiles=int
849 Number of files to use for this job. Defaults to 1. The size of files
850 will be :option:`size` divided by this unless explicit size is specified by
851 :option:`filesize`. Files are created for each thread separately, and each
852 file will have a file number within its name by default, as explained in
853 :option:`filename` section.
856 .. option:: openfiles=int
858 Number of files to keep open at the same time. Defaults to the same as
859 :option:`nrfiles`, can be set smaller to limit the number simultaneous
862 .. option:: file_service_type=str
864 Defines how fio decides which file from a job to service next. The following
868 Choose a file at random.
871 Round robin over opened files. This is the default.
874 Finish one file before moving on to the next. Multiple files can
875 still be open depending on :option:`openfiles`.
878 Use a *Zipf* distribution to decide what file to access.
881 Use a *Pareto* distribution to decide what file to access.
884 Use a *Gaussian* (normal) distribution to decide what file to
890 For *random*, *roundrobin*, and *sequential*, a postfix can be appended to
891 tell fio how many I/Os to issue before switching to a new file. For example,
892 specifying ``file_service_type=random:8`` would cause fio to issue
893 8 I/Os before selecting a new file at random. For the non-uniform
894 distributions, a floating point postfix can be given to influence how the
895 distribution is skewed. See :option:`random_distribution` for a description
896 of how that would work.
898 .. option:: ioscheduler=str
900 Attempt to switch the device hosting the file to the specified I/O scheduler
903 .. option:: create_serialize=bool
905 If true, serialize the file creation for the jobs. This may be handy to
906 avoid interleaving of data files, which may greatly depend on the filesystem
907 used and even the number of processors in the system. Default: true.
909 .. option:: create_fsync=bool
911 :manpage:`fsync(2)` the data file after creation. This is the default.
913 .. option:: create_on_open=bool
915 If true, don't pre-create files but allow the job's open() to create a file
916 when it's time to do I/O. Default: false -- pre-create all necessary files
919 .. option:: create_only=bool
921 If true, fio will only run the setup phase of the job. If files need to be
922 laid out or updated on disk, only that will be done -- the actual job contents
923 are not executed. Default: false.
925 .. option:: allow_file_create=bool
927 If true, fio is permitted to create files as part of its workload. If this
928 option is false, then fio will error out if
929 the files it needs to use don't already exist. Default: true.
931 .. option:: allow_mounted_write=bool
933 If this isn't set, fio will abort jobs that are destructive (e.g. that write)
934 to what appears to be a mounted device or partition. This should help catch
935 creating inadvertently destructive tests, not realizing that the test will
936 destroy data on the mounted file system. Note that some platforms don't allow
937 writing against a mounted device regardless of this option. Default: false.
939 .. option:: pre_read=bool
941 If this is given, files will be pre-read into memory before starting the
942 given I/O operation. This will also clear the :option:`invalidate` flag,
943 since it is pointless to pre-read and then drop the cache. This will only
944 work for I/O engines that are seek-able, since they allow you to read the
945 same data multiple times. Thus it will not work on non-seekable I/O engines
946 (e.g. network, splice). Default: false.
948 .. option:: unlink=bool
950 Unlink the job files when done. Not the default, as repeated runs of that
951 job would then waste time recreating the file set again and again. Default:
954 .. option:: unlink_each_loop=bool
956 Unlink job files after each iteration or loop. Default: false.
958 .. option:: zonemode=str
963 The :option:`zonerange`, :option:`zonesize` and
964 :option:`zoneskip` parameters are ignored.
966 I/O happens in a single zone until
967 :option:`zonesize` bytes have been transferred.
968 After that number of bytes has been
969 transferred processing of the next zone
972 Zoned block device mode. I/O happens
973 sequentially in each zone, even if random I/O
974 has been selected. Random I/O happens across
975 all zones instead of being restricted to a
976 single zone. The :option:`zoneskip` parameter
977 is ignored. :option:`zonerange` and
978 :option:`zonesize` must be identical.
980 .. option:: zonerange=int
982 Size of a single zone. See also :option:`zonesize` and
985 .. option:: zonesize=int
987 For :option:`zonemode` =strided, this is the number of bytes to
988 transfer before skipping :option:`zoneskip` bytes. If this parameter
989 is smaller than :option:`zonerange` then only a fraction of each zone
990 with :option:`zonerange` bytes will be accessed. If this parameter is
991 larger than :option:`zonerange` then each zone will be accessed
992 multiple times before skipping to the next zone.
994 For :option:`zonemode` =zbd, this is the size of a single zone. The
995 :option:`zonerange` parameter is ignored in this mode.
997 .. option:: zoneskip=int
999 For :option:`zonemode` =strided, the number of bytes to skip after
1000 :option:`zonesize` bytes of data have been transferred. This parameter
1001 must be zero for :option:`zonemode` =zbd.
1003 .. option:: read_beyond_wp=bool
1005 This parameter applies to :option:`zonemode` =zbd only.
1007 Zoned block devices are block devices that consist of multiple zones.
1008 Each zone has a type, e.g. conventional or sequential. A conventional
1009 zone can be written at any offset that is a multiple of the block
1010 size. Sequential zones must be written sequentially. The position at
1011 which a write must occur is called the write pointer. A zoned block
1012 device can be either drive managed, host managed or host aware. For
1013 host managed devices the host must ensure that writes happen
1014 sequentially. Fio recognizes host managed devices and serializes
1015 writes to sequential zones for these devices.
1017 If a read occurs in a sequential zone beyond the write pointer then
1018 the zoned block device will complete the read without reading any data
1019 from the storage medium. Since such reads lead to unrealistically high
1020 bandwidth and IOPS numbers fio only reads beyond the write pointer if
1021 explicitly told to do so. Default: false.
1023 .. option:: max_open_zones=int
1025 When running a random write test across an entire drive many more
1026 zones will be open than in a typical application workload. Hence this
1027 command line option that allows to limit the number of open zones. The
1028 number of open zones is defined as the number of zones to which write
1029 commands are issued.
1031 .. option:: zone_reset_threshold=float
1033 A number between zero and one that indicates the ratio of logical
1034 blocks with data to the total number of logical blocks in the test
1035 above which zones should be reset periodically.
1037 .. option:: zone_reset_frequency=float
1039 A number between zero and one that indicates how often a zone reset
1040 should be issued if the zone reset threshold has been exceeded. A zone
1041 reset is submitted after each (1 / zone_reset_frequency) write
1042 requests. This and the previous parameter can be used to simulate
1043 garbage collection activity.
1049 .. option:: direct=bool
1051 If value is true, use non-buffered I/O. This is usually O_DIRECT. Note that
1052 OpenBSD and ZFS on Solaris don't support direct I/O. On Windows the synchronous
1053 ioengines don't support direct I/O. Default: false.
1055 .. option:: atomic=bool
1057 If value is true, attempt to use atomic direct I/O. Atomic writes are
1058 guaranteed to be stable once acknowledged by the operating system. Only
1059 Linux supports O_ATOMIC right now.
1061 .. option:: buffered=bool
1063 If value is true, use buffered I/O. This is the opposite of the
1064 :option:`direct` option. Defaults to true.
1066 .. option:: readwrite=str, rw=str
1068 Type of I/O pattern. Accepted values are:
1075 Sequential trims (Linux block devices and SCSI
1076 character devices only).
1082 Random trims (Linux block devices and SCSI
1083 character devices only).
1085 Sequential mixed reads and writes.
1087 Random mixed reads and writes.
1089 Sequential trim+write sequences. Blocks will be trimmed first,
1090 then the same blocks will be written to.
1092 Fio defaults to read if the option is not specified. For the mixed I/O
1093 types, the default is to split them 50/50. For certain types of I/O the
1094 result may still be skewed a bit, since the speed may be different.
1096 It is possible to specify the number of I/Os to do before getting a new
1097 offset by appending ``:<nr>`` to the end of the string given. For a
1098 random read, it would look like ``rw=randread:8`` for passing in an offset
1099 modifier with a value of 8. If the suffix is used with a sequential I/O
1100 pattern, then the *<nr>* value specified will be **added** to the generated
1101 offset for each I/O turning sequential I/O into sequential I/O with holes.
1102 For instance, using ``rw=write:4k`` will skip 4k for every write. Also see
1103 the :option:`rw_sequencer` option.
1105 .. option:: rw_sequencer=str
1107 If an offset modifier is given by appending a number to the ``rw=<str>``
1108 line, then this option controls how that number modifies the I/O offset
1109 being generated. Accepted values are:
1112 Generate sequential offset.
1114 Generate the same offset.
1116 ``sequential`` is only useful for random I/O, where fio would normally
1117 generate a new random offset for every I/O. If you append e.g. 8 to randread,
1118 you would get a new random offset for every 8 I/Os. The result would be a
1119 seek for only every 8 I/Os, instead of for every I/O. Use ``rw=randread:8``
1120 to specify that. As sequential I/O is already sequential, setting
1121 ``sequential`` for that would not result in any differences. ``identical``
1122 behaves in a similar fashion, except it sends the same offset 8 number of
1123 times before generating a new offset.
1125 .. option:: unified_rw_reporting=bool
1127 Fio normally reports statistics on a per data direction basis, meaning that
1128 reads, writes, and trims are accounted and reported separately. If this
1129 option is set fio sums the results and report them as "mixed" instead.
1131 .. option:: randrepeat=bool
1133 Seed the random number generator used for random I/O patterns in a
1134 predictable way so the pattern is repeatable across runs. Default: true.
1136 .. option:: allrandrepeat=bool
1138 Seed all random number generators in a predictable way so results are
1139 repeatable across runs. Default: false.
1141 .. option:: randseed=int
1143 Seed the random number generators based on this seed value, to be able to
1144 control what sequence of output is being generated. If not set, the random
1145 sequence depends on the :option:`randrepeat` setting.
1147 .. option:: fallocate=str
1149 Whether pre-allocation is performed when laying down files.
1150 Accepted values are:
1153 Do not pre-allocate space.
1156 Use a platform's native pre-allocation call but fall back to
1157 **none** behavior if it fails/is not implemented.
1160 Pre-allocate via :manpage:`posix_fallocate(3)`.
1163 Pre-allocate via :manpage:`fallocate(2)` with
1164 FALLOC_FL_KEEP_SIZE set.
1167 Backward-compatible alias for **none**.
1170 Backward-compatible alias for **posix**.
1172 May not be available on all supported platforms. **keep** is only available
1173 on Linux. If using ZFS on Solaris this cannot be set to **posix**
1174 because ZFS doesn't support pre-allocation. Default: **native** if any
1175 pre-allocation methods are available, **none** if not.
1177 .. option:: fadvise_hint=str
1179 Use :manpage:`posix_fadvise(2)` or :manpage:`posix_fadvise(2)` to
1180 advise the kernel on what I/O patterns are likely to be issued.
1181 Accepted values are:
1184 Backwards-compatible hint for "no hint".
1187 Backwards compatible hint for "advise with fio workload type". This
1188 uses **FADV_RANDOM** for a random workload, and **FADV_SEQUENTIAL**
1189 for a sequential workload.
1192 Advise using **FADV_SEQUENTIAL**.
1195 Advise using **FADV_RANDOM**.
1197 .. option:: write_hint=str
1199 Use :manpage:`fcntl(2)` to advise the kernel what life time to expect
1200 from a write. Only supported on Linux, as of version 4.13. Accepted
1204 No particular life time associated with this file.
1207 Data written to this file has a short life time.
1210 Data written to this file has a medium life time.
1213 Data written to this file has a long life time.
1216 Data written to this file has a very long life time.
1218 The values are all relative to each other, and no absolute meaning
1219 should be associated with them.
1221 .. option:: offset=int
1223 Start I/O at the provided offset in the file, given as either a fixed size in
1224 bytes or a percentage. If a percentage is given, the generated offset will be
1225 aligned to the minimum ``blocksize`` or to the value of ``offset_align`` if
1226 provided. Data before the given offset will not be touched. This
1227 effectively caps the file size at `real_size - offset`. Can be combined with
1228 :option:`size` to constrain the start and end range of the I/O workload.
1229 A percentage can be specified by a number between 1 and 100 followed by '%',
1230 for example, ``offset=20%`` to specify 20%.
1232 .. option:: offset_align=int
1234 If set to non-zero value, the byte offset generated by a percentage ``offset``
1235 is aligned upwards to this value. Defaults to 0 meaning that a percentage
1236 offset is aligned to the minimum block size.
1238 .. option:: offset_increment=int
1240 If this is provided, then the real offset becomes `offset + offset_increment
1241 * thread_number`, where the thread number is a counter that starts at 0 and
1242 is incremented for each sub-job (i.e. when :option:`numjobs` option is
1243 specified). This option is useful if there are several jobs which are
1244 intended to operate on a file in parallel disjoint segments, with even
1245 spacing between the starting points.
1247 .. option:: number_ios=int
1249 Fio will normally perform I/Os until it has exhausted the size of the region
1250 set by :option:`size`, or if it exhaust the allocated time (or hits an error
1251 condition). With this setting, the range/size can be set independently of
1252 the number of I/Os to perform. When fio reaches this number, it will exit
1253 normally and report status. Note that this does not extend the amount of I/O
1254 that will be done, it will only stop fio if this condition is met before
1255 other end-of-job criteria.
1257 .. option:: fsync=int
1259 If writing to a file, issue an :manpage:`fsync(2)` (or its equivalent) of
1260 the dirty data for every number of blocks given. For example, if you give 32
1261 as a parameter, fio will sync the file after every 32 writes issued. If fio is
1262 using non-buffered I/O, we may not sync the file. The exception is the sg
1263 I/O engine, which synchronizes the disk cache anyway. Defaults to 0, which
1264 means fio does not periodically issue and wait for a sync to complete. Also
1265 see :option:`end_fsync` and :option:`fsync_on_close`.
1267 .. option:: fdatasync=int
1269 Like :option:`fsync` but uses :manpage:`fdatasync(2)` to only sync data and
1270 not metadata blocks. In Windows, FreeBSD, and DragonFlyBSD there is no
1271 :manpage:`fdatasync(2)` so this falls back to using :manpage:`fsync(2)`.
1272 Defaults to 0, which means fio does not periodically issue and wait for a
1273 data-only sync to complete.
1275 .. option:: write_barrier=int
1277 Make every `N-th` write a barrier write.
1279 .. option:: sync_file_range=str:int
1281 Use :manpage:`sync_file_range(2)` for every `int` number of write
1282 operations. Fio will track range of writes that have happened since the last
1283 :manpage:`sync_file_range(2)` call. `str` can currently be one or more of:
1286 SYNC_FILE_RANGE_WAIT_BEFORE
1288 SYNC_FILE_RANGE_WRITE
1290 SYNC_FILE_RANGE_WAIT_AFTER
1292 So if you do ``sync_file_range=wait_before,write:8``, fio would use
1293 ``SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE`` for every 8
1294 writes. Also see the :manpage:`sync_file_range(2)` man page. This option is
1297 .. option:: overwrite=bool
1299 If true, writes to a file will always overwrite existing data. If the file
1300 doesn't already exist, it will be created before the write phase begins. If
1301 the file exists and is large enough for the specified write phase, nothing
1302 will be done. Default: false.
1304 .. option:: end_fsync=bool
1306 If true, :manpage:`fsync(2)` file contents when a write stage has completed.
1309 .. option:: fsync_on_close=bool
1311 If true, fio will :manpage:`fsync(2)` a dirty file on close. This differs
1312 from :option:`end_fsync` in that it will happen on every file close, not
1313 just at the end of the job. Default: false.
1315 .. option:: rwmixread=int
1317 Percentage of a mixed workload that should be reads. Default: 50.
1319 .. option:: rwmixwrite=int
1321 Percentage of a mixed workload that should be writes. If both
1322 :option:`rwmixread` and :option:`rwmixwrite` is given and the values do not
1323 add up to 100%, the latter of the two will be used to override the
1324 first. This may interfere with a given rate setting, if fio is asked to
1325 limit reads or writes to a certain rate. If that is the case, then the
1326 distribution may be skewed. Default: 50.
1328 .. option:: random_distribution=str:float[,str:float][,str:float]
1330 By default, fio will use a completely uniform random distribution when asked
1331 to perform random I/O. Sometimes it is useful to skew the distribution in
1332 specific ways, ensuring that some parts of the data is more hot than others.
1333 fio includes the following distribution models:
1336 Uniform random distribution
1345 Normal (Gaussian) distribution
1348 Zoned random distribution
1351 Zone absolute random distribution
1353 When using a **zipf** or **pareto** distribution, an input value is also
1354 needed to define the access pattern. For **zipf**, this is the `Zipf
1355 theta`. For **pareto**, it's the `Pareto power`. Fio includes a test
1356 program, :command:`fio-genzipf`, that can be used visualize what the given input
1357 values will yield in terms of hit rates. If you wanted to use **zipf** with
1358 a `theta` of 1.2, you would use ``random_distribution=zipf:1.2`` as the
1359 option. If a non-uniform model is used, fio will disable use of the random
1360 map. For the **normal** distribution, a normal (Gaussian) deviation is
1361 supplied as a value between 0 and 100.
1363 For a **zoned** distribution, fio supports specifying percentages of I/O
1364 access that should fall within what range of the file or device. For
1365 example, given a criteria of:
1367 * 60% of accesses should be to the first 10%
1368 * 30% of accesses should be to the next 20%
1369 * 8% of accesses should be to the next 30%
1370 * 2% of accesses should be to the next 40%
1372 we can define that through zoning of the random accesses. For the above
1373 example, the user would do::
1375 random_distribution=zoned:60/10:30/20:8/30:2/40
1377 A **zoned_abs** distribution works exactly like the **zoned**, except
1378 that it takes absolute sizes. For example, let's say you wanted to
1379 define access according to the following criteria:
1381 * 60% of accesses should be to the first 20G
1382 * 30% of accesses should be to the next 100G
1383 * 10% of accesses should be to the next 500G
1385 we can define an absolute zoning distribution with:
1387 random_distribution=zoned_abs=60/20G:30/100G:10/500g
1389 For both **zoned** and **zoned_abs**, fio supports defining up to
1392 Similarly to how :option:`bssplit` works for setting ranges and
1393 percentages of block sizes. Like :option:`bssplit`, it's possible to
1394 specify separate zones for reads, writes, and trims. If just one set
1395 is given, it'll apply to all of them. This goes for both **zoned**
1396 **zoned_abs** distributions.
1398 .. option:: percentage_random=int[,int][,int]
1400 For a random workload, set how big a percentage should be random. This
1401 defaults to 100%, in which case the workload is fully random. It can be set
1402 from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1403 sequential. Any setting in between will result in a random mix of sequential
1404 and random I/O, at the given percentages. Comma-separated values may be
1405 specified for reads, writes, and trims as described in :option:`blocksize`.
1407 .. option:: norandommap
1409 Normally fio will cover every block of the file when doing random I/O. If
1410 this option is given, fio will just get a new random offset without looking
1411 at past I/O history. This means that some blocks may not be read or written,
1412 and that some blocks may be read/written more than once. If this option is
1413 used with :option:`verify` and multiple blocksizes (via :option:`bsrange`),
1414 only intact blocks are verified, i.e., partially-overwritten blocks are
1415 ignored. With an async I/O engine and an I/O depth > 1, it is possible for
1416 the same block to be overwritten, which can cause verification errors. Either
1417 do not use norandommap in this case, or also use the lfsr random generator.
1419 .. option:: softrandommap=bool
1421 See :option:`norandommap`. If fio runs with the random block map enabled and
1422 it fails to allocate the map, if this option is set it will continue without
1423 a random block map. As coverage will not be as complete as with random maps,
1424 this option is disabled by default.
1426 .. option:: random_generator=str
1428 Fio supports the following engines for generating I/O offsets for random I/O:
1431 Strong 2^88 cycle random number generator.
1433 Linear feedback shift register generator.
1435 Strong 64-bit 2^258 cycle random number generator.
1437 **tausworthe** is a strong random number generator, but it requires tracking
1438 on the side if we want to ensure that blocks are only read or written
1439 once. **lfsr** guarantees that we never generate the same offset twice, and
1440 it's also less computationally expensive. It's not a true random generator,
1441 however, though for I/O purposes it's typically good enough. **lfsr** only
1442 works with single block sizes, not with workloads that use multiple block
1443 sizes. If used with such a workload, fio may read or write some blocks
1444 multiple times. The default value is **tausworthe**, unless the required
1445 space exceeds 2^32 blocks. If it does, then **tausworthe64** is
1446 selected automatically.
1452 .. option:: blocksize=int[,int][,int], bs=int[,int][,int]
1454 The block size in bytes used for I/O units. Default: 4096. A single value
1455 applies to reads, writes, and trims. Comma-separated values may be
1456 specified for reads, writes, and trims. A value not terminated in a comma
1457 applies to subsequent types.
1462 means 256k for reads, writes and trims.
1465 means 8k for reads, 32k for writes and trims.
1468 means 8k for reads, 32k for writes, and default for trims.
1471 means default for reads, 8k for writes and trims.
1474 means default for reads, 8k for writes, and default for trims.
1476 .. option:: blocksize_range=irange[,irange][,irange], bsrange=irange[,irange][,irange]
1478 A range of block sizes in bytes for I/O units. The issued I/O unit will
1479 always be a multiple of the minimum size, unless
1480 :option:`blocksize_unaligned` is set.
1482 Comma-separated ranges may be specified for reads, writes, and trims as
1483 described in :option:`blocksize`.
1485 Example: ``bsrange=1k-4k,2k-8k``.
1487 .. option:: bssplit=str[,str][,str]
1489 Sometimes you want even finer grained control of the block sizes
1490 issued, not just an even split between them. This option allows you to
1491 weight various block sizes, so that you are able to define a specific
1492 amount of block sizes issued. The format for this option is::
1494 bssplit=blocksize/percentage:blocksize/percentage
1496 for as many block sizes as needed. So if you want to define a workload
1497 that has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would
1500 bssplit=4k/10:64k/50:32k/40
1502 Ordering does not matter. If the percentage is left blank, fio will
1503 fill in the remaining values evenly. So a bssplit option like this one::
1505 bssplit=4k/50:1k/:32k/
1507 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always
1508 add up to 100, if bssplit is given a range that adds up to more, it
1511 Comma-separated values may be specified for reads, writes, and trims as
1512 described in :option:`blocksize`.
1514 If you want a workload that has 50% 2k reads and 50% 4k reads, while
1515 having 90% 4k writes and 10% 8k writes, you would specify::
1517 bssplit=2k/50:4k/50,4k/90:8k/10
1519 Fio supports defining up to 64 different weights for each data
1522 .. option:: blocksize_unaligned, bs_unaligned
1524 If set, fio will issue I/O units with any size within
1525 :option:`blocksize_range`, not just multiples of the minimum size. This
1526 typically won't work with direct I/O, as that normally requires sector
1529 .. option:: bs_is_seq_rand=bool
1531 If this option is set, fio will use the normal read,write blocksize settings
1532 as sequential,random blocksize settings instead. Any random read or write
1533 will use the WRITE blocksize settings, and any sequential read or write will
1534 use the READ blocksize settings.
1536 .. option:: blockalign=int[,int][,int], ba=int[,int][,int]
1538 Boundary to which fio will align random I/O units. Default:
1539 :option:`blocksize`. Minimum alignment is typically 512b for using direct
1540 I/O, though it usually depends on the hardware block size. This option is
1541 mutually exclusive with using a random map for files, so it will turn off
1542 that option. Comma-separated values may be specified for reads, writes, and
1543 trims as described in :option:`blocksize`.
1549 .. option:: zero_buffers
1551 Initialize buffers with all zeros. Default: fill buffers with random data.
1553 .. option:: refill_buffers
1555 If this option is given, fio will refill the I/O buffers on every
1556 submit. Only makes sense if :option:`zero_buffers` isn't specified,
1557 naturally. Defaults to being unset i.e., the buffer is only filled at
1558 init time and the data in it is reused when possible but if any of
1559 :option:`verify`, :option:`buffer_compress_percentage` or
1560 :option:`dedupe_percentage` are enabled then `refill_buffers` is also
1561 automatically enabled.
1563 .. option:: scramble_buffers=bool
1565 If :option:`refill_buffers` is too costly and the target is using data
1566 deduplication, then setting this option will slightly modify the I/O buffer
1567 contents to defeat normal de-dupe attempts. This is not enough to defeat
1568 more clever block compression attempts, but it will stop naive dedupe of
1569 blocks. Default: true.
1571 .. option:: buffer_compress_percentage=int
1573 If this is set, then fio will attempt to provide I/O buffer content
1574 (on WRITEs) that compresses to the specified level. Fio does this by
1575 providing a mix of random data followed by fixed pattern data. The
1576 fixed pattern is either zeros, or the pattern specified by
1577 :option:`buffer_pattern`. If the `buffer_pattern` option is used, it
1578 might skew the compression ratio slightly. Setting
1579 `buffer_compress_percentage` to a value other than 100 will also
1580 enable :option:`refill_buffers` in order to reduce the likelihood that
1581 adjacent blocks are so similar that they over compress when seen
1582 together. See :option:`buffer_compress_chunk` for how to set a finer or
1583 coarser granularity for the random/fixed data region. Defaults to unset
1584 i.e., buffer data will not adhere to any compression level.
1586 .. option:: buffer_compress_chunk=int
1588 This setting allows fio to manage how big the random/fixed data region
1589 is when using :option:`buffer_compress_percentage`. When
1590 `buffer_compress_chunk` is set to some non-zero value smaller than the
1591 block size, fio can repeat the random/fixed region throughout the I/O
1592 buffer at the specified interval (which particularly useful when
1593 bigger block sizes are used for a job). When set to 0, fio will use a
1594 chunk size that matches the block size resulting in a single
1595 random/fixed region within the I/O buffer. Defaults to 512. When the
1596 unit is omitted, the value is interpreted in bytes.
1598 .. option:: buffer_pattern=str
1600 If set, fio will fill the I/O buffers with this pattern or with the contents
1601 of a file. If not set, the contents of I/O buffers are defined by the other
1602 options related to buffer contents. The setting can be any pattern of bytes,
1603 and can be prefixed with 0x for hex values. It may also be a string, where
1604 the string must then be wrapped with ``""``. Or it may also be a filename,
1605 where the filename must be wrapped with ``''`` in which case the file is
1606 opened and read. Note that not all the file contents will be read if that
1607 would cause the buffers to overflow. So, for example::
1609 buffer_pattern='filename'
1613 buffer_pattern="abcd"
1621 buffer_pattern=0xdeadface
1623 Also you can combine everything together in any order::
1625 buffer_pattern=0xdeadface"abcd"-12'filename'
1627 .. option:: dedupe_percentage=int
1629 If set, fio will generate this percentage of identical buffers when
1630 writing. These buffers will be naturally dedupable. The contents of the
1631 buffers depend on what other buffer compression settings have been set. It's
1632 possible to have the individual buffers either fully compressible, or not at
1633 all -- this option only controls the distribution of unique buffers. Setting
1634 this option will also enable :option:`refill_buffers` to prevent every buffer
1637 .. option:: invalidate=bool
1639 Invalidate the buffer/page cache parts of the files to be used prior to
1640 starting I/O if the platform and file type support it. Defaults to true.
1641 This will be ignored if :option:`pre_read` is also specified for the
1644 .. option:: sync=bool
1646 Use synchronous I/O for buffered writes. For the majority of I/O engines,
1647 this means using O_SYNC. Default: false.
1649 .. option:: iomem=str, mem=str
1651 Fio can use various types of memory as the I/O unit buffer. The allowed
1655 Use memory from :manpage:`malloc(3)` as the buffers. Default memory
1659 Use shared memory as the buffers. Allocated through
1660 :manpage:`shmget(2)`.
1663 Same as shm, but use huge pages as backing.
1666 Use :manpage:`mmap(2)` to allocate buffers. May either be anonymous memory, or can
1667 be file backed if a filename is given after the option. The format
1668 is `mem=mmap:/path/to/file`.
1671 Use a memory mapped huge file as the buffer backing. Append filename
1672 after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file`.
1675 Same as mmap, but use a MMAP_SHARED mapping.
1678 Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1679 The :option:`ioengine` must be `rdma`.
1681 The area allocated is a function of the maximum allowed bs size for the job,
1682 multiplied by the I/O depth given. Note that for **shmhuge** and
1683 **mmaphuge** to work, the system must have free huge pages allocated. This
1684 can normally be checked and set by reading/writing
1685 :file:`/proc/sys/vm/nr_hugepages` on a Linux system. Fio assumes a huge page
1686 is 4MiB in size. So to calculate the number of huge pages you need for a
1687 given job file, add up the I/O depth of all jobs (normally one unless
1688 :option:`iodepth` is used) and multiply by the maximum bs set. Then divide
1689 that number by the huge page size. You can see the size of the huge pages in
1690 :file:`/proc/meminfo`. If no huge pages are allocated by having a non-zero
1691 number in `nr_hugepages`, using **mmaphuge** or **shmhuge** will fail. Also
1692 see :option:`hugepage-size`.
1694 **mmaphuge** also needs to have hugetlbfs mounted and the file location
1695 should point there. So if it's mounted in :file:`/huge`, you would use
1696 `mem=mmaphuge:/huge/somefile`.
1698 .. option:: iomem_align=int, mem_align=int
1700 This indicates the memory alignment of the I/O memory buffers. Note that
1701 the given alignment is applied to the first I/O unit buffer, if using
1702 :option:`iodepth` the alignment of the following buffers are given by the
1703 :option:`bs` used. In other words, if using a :option:`bs` that is a
1704 multiple of the page sized in the system, all buffers will be aligned to
1705 this value. If using a :option:`bs` that is not page aligned, the alignment
1706 of subsequent I/O memory buffers is the sum of the :option:`iomem_align` and
1709 .. option:: hugepage-size=int
1711 Defines the size of a huge page. Must at least be equal to the system
1712 setting, see :file:`/proc/meminfo`. Defaults to 4MiB. Should probably
1713 always be a multiple of megabytes, so using ``hugepage-size=Xm`` is the
1714 preferred way to set this to avoid setting a non-pow-2 bad value.
1716 .. option:: lockmem=int
1718 Pin the specified amount of memory with :manpage:`mlock(2)`. Can be used to
1719 simulate a smaller amount of memory. The amount specified is per worker.
1725 .. option:: size=int
1727 The total size of file I/O for each thread of this job. Fio will run until
1728 this many bytes has been transferred, unless runtime is limited by other options
1729 (such as :option:`runtime`, for instance, or increased/decreased by :option:`io_size`).
1730 Fio will divide this size between the available files determined by options
1731 such as :option:`nrfiles`, :option:`filename`, unless :option:`filesize` is
1732 specified by the job. If the result of division happens to be 0, the size is
1733 set to the physical size of the given files or devices if they exist.
1734 If this option is not specified, fio will use the full size of the given
1735 files or devices. If the files do not exist, size must be given. It is also
1736 possible to give size as a percentage between 1 and 100. If ``size=20%`` is
1737 given, fio will use 20% of the full size of the given files or devices.
1738 Can be combined with :option:`offset` to constrain the start and end range
1739 that I/O will be done within.
1741 .. option:: io_size=int, io_limit=int
1743 Normally fio operates within the region set by :option:`size`, which means
1744 that the :option:`size` option sets both the region and size of I/O to be
1745 performed. Sometimes that is not what you want. With this option, it is
1746 possible to define just the amount of I/O that fio should do. For instance,
1747 if :option:`size` is set to 20GiB and :option:`io_size` is set to 5GiB, fio
1748 will perform I/O within the first 20GiB but exit when 5GiB have been
1749 done. The opposite is also possible -- if :option:`size` is set to 20GiB,
1750 and :option:`io_size` is set to 40GiB, then fio will do 40GiB of I/O within
1751 the 0..20GiB region.
1753 .. option:: filesize=irange(int)
1755 Individual file sizes. May be a range, in which case fio will select sizes
1756 for files at random within the given range and limited to :option:`size` in
1757 total (if that is given). If not given, each created file is the same size.
1758 This option overrides :option:`size` in terms of file size, which means
1759 this value is used as a fixed size or possible range of each file.
1761 .. option:: file_append=bool
1763 Perform I/O after the end of the file. Normally fio will operate within the
1764 size of a file. If this option is set, then fio will append to the file
1765 instead. This has identical behavior to setting :option:`offset` to the size
1766 of a file. This option is ignored on non-regular files.
1768 .. option:: fill_device=bool, fill_fs=bool
1770 Sets size to something really large and waits for ENOSPC (no space left on
1771 device) as the terminating condition. Only makes sense with sequential
1772 write. For a read workload, the mount point will be filled first then I/O
1773 started on the result. This option doesn't make sense if operating on a raw
1774 device node, since the size of that is already known by the file system.
1775 Additionally, writing beyond end-of-device will not return ENOSPC there.
1781 .. option:: ioengine=str
1783 Defines how the job issues I/O to the file. The following types are defined:
1786 Basic :manpage:`read(2)` or :manpage:`write(2)`
1787 I/O. :manpage:`lseek(2)` is used to position the I/O location.
1788 See :option:`fsync` and :option:`fdatasync` for syncing write I/Os.
1791 Basic :manpage:`pread(2)` or :manpage:`pwrite(2)` I/O. Default on
1792 all supported operating systems except for Windows.
1795 Basic :manpage:`readv(2)` or :manpage:`writev(2)` I/O. Will emulate
1796 queuing by coalescing adjacent I/Os into a single submission.
1799 Basic :manpage:`preadv(2)` or :manpage:`pwritev(2)` I/O.
1802 Basic :manpage:`preadv2(2)` or :manpage:`pwritev2(2)` I/O.
1805 Linux native asynchronous I/O. Note that Linux may only support
1806 queued behavior with non-buffered I/O (set ``direct=1`` or
1808 This engine defines engine specific options.
1811 POSIX asynchronous I/O using :manpage:`aio_read(3)` and
1812 :manpage:`aio_write(3)`.
1815 Solaris native asynchronous I/O.
1818 Windows native asynchronous I/O. Default on Windows.
1821 File is memory mapped with :manpage:`mmap(2)` and data copied
1822 to/from using :manpage:`memcpy(3)`.
1825 :manpage:`splice(2)` is used to transfer the data and
1826 :manpage:`vmsplice(2)` to transfer data from user space to the
1830 SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1831 ioctl, or if the target is an sg character device we use
1832 :manpage:`read(2)` and :manpage:`write(2)` for asynchronous
1833 I/O. Requires :option:`filename` option to specify either block or
1834 character devices. This engine supports trim operations.
1835 The sg engine includes engine specific options.
1838 Doesn't transfer any data, just pretends to. This is mainly used to
1839 exercise fio itself and for debugging/testing purposes.
1842 Transfer over the network to given ``host:port``. Depending on the
1843 :option:`protocol` used, the :option:`hostname`, :option:`port`,
1844 :option:`listen` and :option:`filename` options are used to specify
1845 what sort of connection to make, while the :option:`protocol` option
1846 determines which protocol will be used. This engine defines engine
1850 Like **net**, but uses :manpage:`splice(2)` and
1851 :manpage:`vmsplice(2)` to map data and send/receive.
1852 This engine defines engine specific options.
1855 Doesn't transfer any data, but burns CPU cycles according to the
1856 :option:`cpuload` and :option:`cpuchunks` options. Setting
1857 :option:`cpuload`\=85 will cause that job to do nothing but burn 85%
1858 of the CPU. In case of SMP machines, use :option:`numjobs`\=<nr_of_cpu>
1859 to get desired CPU usage, as the cpuload only loads a
1860 single CPU at the desired rate. A job never finishes unless there is
1861 at least one non-cpuio job.
1864 The GUASI I/O engine is the Generic Userspace Asynchronous Syscall
1865 Interface approach to async I/O. See
1867 http://www.xmailserver.org/guasi-lib.html
1869 for more info on GUASI.
1872 The RDMA I/O engine supports both RDMA memory semantics
1873 (RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
1874 InfiniBand, RoCE and iWARP protocols. This engine defines engine
1878 I/O engine that does regular fallocate to simulate data transfer as
1882 does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1885 does fallocate(,mode = 0).
1888 does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1891 I/O engine that sends :manpage:`ftruncate(2)` operations in response
1892 to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
1893 size to the current block offset. :option:`blocksize` is ignored.
1896 I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1897 defragment activity in request to DDIR_WRITE event.
1900 I/O engine supporting direct access to Ceph Reliable Autonomic
1901 Distributed Object Store (RADOS) via librados. This ioengine
1902 defines engine specific options.
1905 I/O engine supporting direct access to Ceph Rados Block Devices
1906 (RBD) via librbd without the need to use the kernel rbd driver. This
1907 ioengine defines engine specific options.
1910 I/O engine supporting GET/PUT requests over HTTP(S) with libcurl to
1911 a WebDAV or S3 endpoint. This ioengine defines engine specific options.
1913 This engine only supports direct IO of iodepth=1; you need to scale this
1914 via numjobs. blocksize defines the size of the objects to be created.
1916 TRIM is translated to object deletion.
1919 Using GlusterFS libgfapi sync interface to direct access to
1920 GlusterFS volumes without having to go through FUSE. This ioengine
1921 defines engine specific options.
1924 Using GlusterFS libgfapi async interface to direct access to
1925 GlusterFS volumes without having to go through FUSE. This ioengine
1926 defines engine specific options.
1929 Read and write through Hadoop (HDFS). The :option:`filename` option
1930 is used to specify host,port of the hdfs name-node to connect. This
1931 engine interprets offsets a little differently. In HDFS, files once
1932 created cannot be modified so random writes are not possible. To
1933 imitate this the libhdfs engine expects a bunch of small files to be
1934 created over HDFS and will randomly pick a file from them
1935 based on the offset generated by fio backend (see the example
1936 job file to create such files, use ``rw=write`` option). Please
1937 note, it may be necessary to set environment variables to work
1938 with HDFS/libhdfs properly. Each job uses its own connection to
1942 Read, write and erase an MTD character device (e.g.,
1943 :file:`/dev/mtd0`). Discards are treated as erases. Depending on the
1944 underlying device type, the I/O may have to go in a certain pattern,
1945 e.g., on NAND, writing sequentially to erase blocks and discarding
1946 before overwriting. The `trimwrite` mode works well for this
1950 Read and write using filesystem DAX to a file on a filesystem
1951 mounted with DAX on a persistent memory device through the PMDK
1955 Read and write using device DAX to a persistent memory device (e.g.,
1956 /dev/dax0.0) through the PMDK libpmem library.
1959 Prefix to specify loading an external I/O engine object file. Append
1960 the engine filename, e.g. ``ioengine=external:/tmp/foo.o`` to load
1961 ioengine :file:`foo.o` in :file:`/tmp`. The path can be either
1962 absolute or relative. See :file:`engines/skeleton_external.c` for
1963 details of writing an external I/O engine.
1966 Simply create the files and do no I/O to them. You still need to
1967 set `filesize` so that all the accounting still occurs, but no
1968 actual I/O will be done other than creating the file.
1971 Read and write using mmap I/O to a file on a filesystem
1972 mounted with DAX on a persistent memory device through the PMDK
1976 Synchronous read and write using DDN's Infinite Memory Engine (IME).
1977 This engine is very basic and issues calls to IME whenever an IO is
1981 Synchronous read and write using DDN's Infinite Memory Engine (IME).
1982 This engine uses iovecs and will try to stack as much IOs as possible
1983 (if the IOs are "contiguous" and the IO depth is not exceeded)
1984 before issuing a call to IME.
1987 Asynchronous read and write using DDN's Infinite Memory Engine (IME).
1988 This engine will try to stack as much IOs as possible by creating
1989 requests for IME. FIO will then decide when to commit these requests.
1991 I/O engine specific parameters
1992 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1994 In addition, there are some parameters which are only valid when a specific
1995 :option:`ioengine` is in use. These are used identically to normal parameters,
1996 with the caveat that when used on the command line, they must come after the
1997 :option:`ioengine` that defines them is selected.
1999 .. option:: userspace_reap : [libaio]
2001 Normally, with the libaio engine in use, fio will use the
2002 :manpage:`io_getevents(2)` system call to reap newly returned events. With
2003 this flag turned on, the AIO ring will be read directly from user-space to
2004 reap events. The reaping mode is only enabled when polling for a minimum of
2005 0 events (e.g. when :option:`iodepth_batch_complete` `=0`).
2007 .. option:: hipri : [pvsync2]
2009 Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
2012 .. option:: hipri_percentage : [pvsync2]
2014 When hipri is set this determines the probability of a pvsync2 I/O being high
2015 priority. The default is 100%.
2017 .. option:: cpuload=int : [cpuio]
2019 Attempt to use the specified percentage of CPU cycles. This is a mandatory
2020 option when using cpuio I/O engine.
2022 .. option:: cpuchunks=int : [cpuio]
2024 Split the load into cycles of the given time. In microseconds.
2026 .. option:: exit_on_io_done=bool : [cpuio]
2028 Detect when I/O threads are done, then exit.
2030 .. option:: namenode=str : [libhdfs]
2032 The hostname or IP address of a HDFS cluster namenode to contact.
2034 .. option:: port=int
2038 The listening port of the HFDS cluster namenode.
2042 The TCP or UDP port to bind to or connect to. If this is used with
2043 :option:`numjobs` to spawn multiple instances of the same job type, then
2044 this will be the starting port number since fio will use a range of
2049 The port to use for RDMA-CM communication. This should be the same value
2050 on the client and the server side.
2052 .. option:: hostname=str : [netsplice] [net] [rdma]
2054 The hostname or IP address to use for TCP, UDP or RDMA-CM based I/O. If the job
2055 is a TCP listener or UDP reader, the hostname is not used and must be omitted
2056 unless it is a valid UDP multicast address.
2058 .. option:: interface=str : [netsplice] [net]
2060 The IP address of the network interface used to send or receive UDP
2063 .. option:: ttl=int : [netsplice] [net]
2065 Time-to-live value for outgoing UDP multicast packets. Default: 1.
2067 .. option:: nodelay=bool : [netsplice] [net]
2069 Set TCP_NODELAY on TCP connections.
2071 .. option:: protocol=str, proto=str : [netsplice] [net]
2073 The network protocol to use. Accepted values are:
2076 Transmission control protocol.
2078 Transmission control protocol V6.
2080 User datagram protocol.
2082 User datagram protocol V6.
2086 When the protocol is TCP or UDP, the port must also be given, as well as the
2087 hostname if the job is a TCP listener or UDP reader. For unix sockets, the
2088 normal :option:`filename` option should be used and the port is invalid.
2090 .. option:: listen : [netsplice] [net]
2092 For TCP network connections, tell fio to listen for incoming connections
2093 rather than initiating an outgoing connection. The :option:`hostname` must
2094 be omitted if this option is used.
2096 .. option:: pingpong : [netsplice] [net]
2098 Normally a network writer will just continue writing data, and a network
2099 reader will just consume packages. If ``pingpong=1`` is set, a writer will
2100 send its normal payload to the reader, then wait for the reader to send the
2101 same payload back. This allows fio to measure network latencies. The
2102 submission and completion latencies then measure local time spent sending or
2103 receiving, and the completion latency measures how long it took for the
2104 other end to receive and send back. For UDP multicast traffic
2105 ``pingpong=1`` should only be set for a single reader when multiple readers
2106 are listening to the same address.
2108 .. option:: window_size : [netsplice] [net]
2110 Set the desired socket buffer size for the connection.
2112 .. option:: mss : [netsplice] [net]
2114 Set the TCP maximum segment size (TCP_MAXSEG).
2116 .. option:: donorname=str : [e4defrag]
2118 File will be used as a block donor (swap extents between files).
2120 .. option:: inplace=int : [e4defrag]
2122 Configure donor file blocks allocation strategy:
2125 Default. Preallocate donor's file on init.
2127 Allocate space immediately inside defragment event, and free right
2130 .. option:: clustername=str : [rbd,rados]
2132 Specifies the name of the Ceph cluster.
2134 .. option:: rbdname=str : [rbd]
2136 Specifies the name of the RBD.
2138 .. option:: pool=str : [rbd,rados]
2140 Specifies the name of the Ceph pool containing RBD or RADOS data.
2142 .. option:: clientname=str : [rbd,rados]
2144 Specifies the username (without the 'client.' prefix) used to access the
2145 Ceph cluster. If the *clustername* is specified, the *clientname* shall be
2146 the full *type.id* string. If no type. prefix is given, fio will add
2147 'client.' by default.
2149 .. option:: busy_poll=bool : [rbd,rados]
2151 Poll store instead of waiting for completion. Usually this provides better
2152 throughput at cost of higher(up to 100%) CPU utilization.
2154 .. option:: skip_bad=bool : [mtd]
2156 Skip operations against known bad blocks.
2158 .. option:: hdfsdirectory : [libhdfs]
2160 libhdfs will create chunk in this HDFS directory.
2162 .. option:: chunk_size : [libhdfs]
2164 The size of the chunk to use for each file.
2166 .. option:: verb=str : [rdma]
2168 The RDMA verb to use on this side of the RDMA ioengine connection. Valid
2169 values are write, read, send and recv. These correspond to the equivalent
2170 RDMA verbs (e.g. write = rdma_write etc.). Note that this only needs to be
2171 specified on the client side of the connection. See the examples folder.
2173 .. option:: bindname=str : [rdma]
2175 The name to use to bind the local RDMA-CM connection to a local RDMA device.
2176 This could be a hostname or an IPv4 or IPv6 address. On the server side this
2177 will be passed into the rdma_bind_addr() function and on the client site it
2178 will be used in the rdma_resolve_add() function. This can be useful when
2179 multiple paths exist between the client and the server or in certain loopback
2182 .. option:: readfua=bool : [sg]
2184 With readfua option set to 1, read operations include
2185 the force unit access (fua) flag. Default is 0.
2187 .. option:: writefua=bool : [sg]
2189 With writefua option set to 1, write operations include
2190 the force unit access (fua) flag. Default is 0.
2192 .. option:: sg_write_mode=str : [sg]
2194 Specify the type of write commands to issue. This option can take three values:
2197 This is the default where write opcodes are issued as usual.
2199 Issue WRITE AND VERIFY commands. The BYTCHK bit is set to 0. This
2200 directs the device to carry out a medium verification with no data
2201 comparison. The writefua option is ignored with this selection.
2203 Issue WRITE SAME commands. This transfers a single block to the device
2204 and writes this same block of data to a contiguous sequence of LBAs
2205 beginning at the specified offset. fio's block size parameter specifies
2206 the amount of data written with each command. However, the amount of data
2207 actually transferred to the device is equal to the device's block
2208 (sector) size. For a device with 512 byte sectors, blocksize=8k will
2209 write 16 sectors with each command. fio will still generate 8k of data
2210 for each command but only the first 512 bytes will be used and
2211 transferred to the device. The writefua option is ignored with this
2214 .. option:: http_host=str : [http]
2216 Hostname to connect to. For S3, this could be the bucket hostname.
2217 Default is **localhost**
2219 .. option:: http_user=str : [http]
2221 Username for HTTP authentication.
2223 .. option:: http_pass=str : [http]
2225 Password for HTTP authentication.
2227 .. option:: https=str : [http]
2229 Enable HTTPS instead of http. *on* enables HTTPS; *insecure*
2230 will enable HTTPS, but disable SSL peer verification (use with
2231 caution!). Default is **off**
2233 .. option:: http_mode=str : [http]
2235 Which HTTP access mode to use: *webdav*, *swift*, or *s3*.
2236 Default is **webdav**
2238 .. option:: http_s3_region=str : [http]
2240 The S3 region/zone string.
2241 Default is **us-east-1**
2243 .. option:: http_s3_key=str : [http]
2247 .. option:: http_s3_keyid=str : [http]
2249 The S3 key/access id.
2251 .. option:: http_swift_auth_token=str : [http]
2253 The Swift auth token. See the example configuration file on how
2256 .. option:: http_verbose=int : [http]
2258 Enable verbose requests from libcurl. Useful for debugging. 1
2259 turns on verbose logging from libcurl, 2 additionally enables
2260 HTTP IO tracing. Default is **0**
2265 .. option:: iodepth=int
2267 Number of I/O units to keep in flight against the file. Note that
2268 increasing *iodepth* beyond 1 will not affect synchronous ioengines (except
2269 for small degrees when :option:`verify_async` is in use). Even async
2270 engines may impose OS restrictions causing the desired depth not to be
2271 achieved. This may happen on Linux when using libaio and not setting
2272 :option:`direct`\=1, since buffered I/O is not async on that OS. Keep an
2273 eye on the I/O depth distribution in the fio output to verify that the
2274 achieved depth is as expected. Default: 1.
2276 .. option:: iodepth_batch_submit=int, iodepth_batch=int
2278 This defines how many pieces of I/O to submit at once. It defaults to 1
2279 which means that we submit each I/O as soon as it is available, but can be
2280 raised to submit bigger batches of I/O at the time. If it is set to 0 the
2281 :option:`iodepth` value will be used.
2283 .. option:: iodepth_batch_complete_min=int, iodepth_batch_complete=int
2285 This defines how many pieces of I/O to retrieve at once. It defaults to 1
2286 which means that we'll ask for a minimum of 1 I/O in the retrieval process
2287 from the kernel. The I/O retrieval will go on until we hit the limit set by
2288 :option:`iodepth_low`. If this variable is set to 0, then fio will always
2289 check for completed events before queuing more I/O. This helps reduce I/O
2290 latency, at the cost of more retrieval system calls.
2292 .. option:: iodepth_batch_complete_max=int
2294 This defines maximum pieces of I/O to retrieve at once. This variable should
2295 be used along with :option:`iodepth_batch_complete_min`\=int variable,
2296 specifying the range of min and max amount of I/O which should be
2297 retrieved. By default it is equal to the :option:`iodepth_batch_complete_min`
2302 iodepth_batch_complete_min=1
2303 iodepth_batch_complete_max=<iodepth>
2305 which means that we will retrieve at least 1 I/O and up to the whole
2306 submitted queue depth. If none of I/O has been completed yet, we will wait.
2310 iodepth_batch_complete_min=0
2311 iodepth_batch_complete_max=<iodepth>
2313 which means that we can retrieve up to the whole submitted queue depth, but
2314 if none of I/O has been completed yet, we will NOT wait and immediately exit
2315 the system call. In this example we simply do polling.
2317 .. option:: iodepth_low=int
2319 The low water mark indicating when to start filling the queue
2320 again. Defaults to the same as :option:`iodepth`, meaning that fio will
2321 attempt to keep the queue full at all times. If :option:`iodepth` is set to
2322 e.g. 16 and *iodepth_low* is set to 4, then after fio has filled the queue of
2323 16 requests, it will let the depth drain down to 4 before starting to fill
2326 .. option:: serialize_overlap=bool
2328 Serialize in-flight I/Os that might otherwise cause or suffer from data races.
2329 When two or more I/Os are submitted simultaneously, there is no guarantee that
2330 the I/Os will be processed or completed in the submitted order. Further, if
2331 two or more of those I/Os are writes, any overlapping region between them can
2332 become indeterminate/undefined on certain storage. These issues can cause
2333 verification to fail erratically when at least one of the racing I/Os is
2334 changing data and the overlapping region has a non-zero size. Setting
2335 ``serialize_overlap`` tells fio to avoid provoking this behavior by explicitly
2336 serializing in-flight I/Os that have a non-zero overlap. Note that setting
2337 this option can reduce both performance and the :option:`iodepth` achieved.
2338 Additionally this option does not work when :option:`io_submit_mode` is set to
2339 offload. Default: false.
2341 .. option:: io_submit_mode=str
2343 This option controls how fio submits the I/O to the I/O engine. The default
2344 is `inline`, which means that the fio job threads submit and reap I/O
2345 directly. If set to `offload`, the job threads will offload I/O submission
2346 to a dedicated pool of I/O threads. This requires some coordination and thus
2347 has a bit of extra overhead, especially for lower queue depth I/O where it
2348 can increase latencies. The benefit is that fio can manage submission rates
2349 independently of the device completion rates. This avoids skewed latency
2350 reporting if I/O gets backed up on the device side (the coordinated omission
2357 .. option:: thinktime=time
2359 Stall the job for the specified period of time after an I/O has completed before issuing the
2360 next. May be used to simulate processing being done by an application.
2361 When the unit is omitted, the value is interpreted in microseconds. See
2362 :option:`thinktime_blocks` and :option:`thinktime_spin`.
2364 .. option:: thinktime_spin=time
2366 Only valid if :option:`thinktime` is set - pretend to spend CPU time doing
2367 something with the data received, before falling back to sleeping for the
2368 rest of the period specified by :option:`thinktime`. When the unit is
2369 omitted, the value is interpreted in microseconds.
2371 .. option:: thinktime_blocks=int
2373 Only valid if :option:`thinktime` is set - control how many blocks to issue,
2374 before waiting :option:`thinktime` usecs. If not set, defaults to 1 which will make
2375 fio wait :option:`thinktime` usecs after every block. This effectively makes any
2376 queue depth setting redundant, since no more than 1 I/O will be queued
2377 before we have to complete it and do our :option:`thinktime`. In other words, this
2378 setting effectively caps the queue depth if the latter is larger.
2380 .. option:: rate=int[,int][,int]
2382 Cap the bandwidth used by this job. The number is in bytes/sec, the normal
2383 suffix rules apply. Comma-separated values may be specified for reads,
2384 writes, and trims as described in :option:`blocksize`.
2386 For example, using `rate=1m,500k` would limit reads to 1MiB/sec and writes to
2387 500KiB/sec. Capping only reads or writes can be done with `rate=,500k` or
2388 `rate=500k,` where the former will only limit writes (to 500KiB/sec) and the
2389 latter will only limit reads.
2391 .. option:: rate_min=int[,int][,int]
2393 Tell fio to do whatever it can to maintain at least this bandwidth. Failing
2394 to meet this requirement will cause the job to exit. Comma-separated values
2395 may be specified for reads, writes, and trims as described in
2396 :option:`blocksize`.
2398 .. option:: rate_iops=int[,int][,int]
2400 Cap the bandwidth to this number of IOPS. Basically the same as
2401 :option:`rate`, just specified independently of bandwidth. If the job is
2402 given a block size range instead of a fixed value, the smallest block size
2403 is used as the metric. Comma-separated values may be specified for reads,
2404 writes, and trims as described in :option:`blocksize`.
2406 .. option:: rate_iops_min=int[,int][,int]
2408 If fio doesn't meet this rate of I/O, it will cause the job to exit.
2409 Comma-separated values may be specified for reads, writes, and trims as
2410 described in :option:`blocksize`.
2412 .. option:: rate_process=str
2414 This option controls how fio manages rated I/O submissions. The default is
2415 `linear`, which submits I/O in a linear fashion with fixed delays between
2416 I/Os that gets adjusted based on I/O completion rates. If this is set to
2417 `poisson`, fio will submit I/O based on a more real world random request
2418 flow, known as the Poisson process
2419 (https://en.wikipedia.org/wiki/Poisson_point_process). The lambda will be
2420 10^6 / IOPS for the given workload.
2422 .. option:: rate_ignore_thinktime=bool
2424 By default, fio will attempt to catch up to the specified rate setting,
2425 if any kind of thinktime setting was used. If this option is set, then
2426 fio will ignore the thinktime and continue doing IO at the specified
2427 rate, instead of entering a catch-up mode after thinktime is done.
2433 .. option:: latency_target=time
2435 If set, fio will attempt to find the max performance point that the given
2436 workload will run at while maintaining a latency below this target. When
2437 the unit is omitted, the value is interpreted in microseconds. See
2438 :option:`latency_window` and :option:`latency_percentile`.
2440 .. option:: latency_window=time
2442 Used with :option:`latency_target` to specify the sample window that the job
2443 is run at varying queue depths to test the performance. When the unit is
2444 omitted, the value is interpreted in microseconds.
2446 .. option:: latency_percentile=float
2448 The percentage of I/Os that must fall within the criteria specified by
2449 :option:`latency_target` and :option:`latency_window`. If not set, this
2450 defaults to 100.0, meaning that all I/Os must be equal or below to the value
2451 set by :option:`latency_target`.
2453 .. option:: max_latency=time
2455 If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
2456 maximum latency. When the unit is omitted, the value is interpreted in
2459 .. option:: rate_cycle=int
2461 Average bandwidth for :option:`rate` and :option:`rate_min` over this number
2462 of milliseconds. Defaults to 1000.
2468 .. option:: write_iolog=str
2470 Write the issued I/O patterns to the specified file. See
2471 :option:`read_iolog`. Specify a separate file for each job, otherwise the
2472 iologs will be interspersed and the file may be corrupt.
2474 .. option:: read_iolog=str
2476 Open an iolog with the specified filename and replay the I/O patterns it
2477 contains. This can be used to store a workload and replay it sometime
2478 later. The iolog given may also be a blktrace binary file, which allows fio
2479 to replay a workload captured by :command:`blktrace`. See
2480 :manpage:`blktrace(8)` for how to capture such logging data. For blktrace
2481 replay, the file needs to be turned into a blkparse binary data file first
2482 (``blkparse <device> -o /dev/null -d file_for_fio.bin``).
2483 You can specify a number of files by separating the names with a ':'
2484 character. See the :option:`filename` option for information on how to
2485 escape ':' and '\' characters within the file names. These files will
2486 be sequentially assigned to job clones created by :option:`numjobs`.
2488 .. option:: read_iolog_chunked=bool
2490 Determines how iolog is read. If false(default) entire :option:`read_iolog`
2491 will be read at once. If selected true, input from iolog will be read
2492 gradually. Useful when iolog is very large, or it is generated.
2494 .. option:: replay_no_stall=bool
2496 When replaying I/O with :option:`read_iolog` the default behavior is to
2497 attempt to respect the timestamps within the log and replay them with the
2498 appropriate delay between IOPS. By setting this variable fio will not
2499 respect the timestamps and attempt to replay them as fast as possible while
2500 still respecting ordering. The result is the same I/O pattern to a given
2501 device, but different timings.
2503 .. option:: replay_time_scale=int
2505 When replaying I/O with :option:`read_iolog`, fio will honor the
2506 original timing in the trace. With this option, it's possible to scale
2507 the time. It's a percentage option, if set to 50 it means run at 50%
2508 the original IO rate in the trace. If set to 200, run at twice the
2509 original IO rate. Defaults to 100.
2511 .. option:: replay_redirect=str
2513 While replaying I/O patterns using :option:`read_iolog` the default behavior
2514 is to replay the IOPS onto the major/minor device that each IOP was recorded
2515 from. This is sometimes undesirable because on a different machine those
2516 major/minor numbers can map to a different device. Changing hardware on the
2517 same system can also result in a different major/minor mapping.
2518 ``replay_redirect`` causes all I/Os to be replayed onto the single specified
2519 device regardless of the device it was recorded
2520 from. i.e. :option:`replay_redirect`\= :file:`/dev/sdc` would cause all I/O
2521 in the blktrace or iolog to be replayed onto :file:`/dev/sdc`. This means
2522 multiple devices will be replayed onto a single device, if the trace
2523 contains multiple devices. If you want multiple devices to be replayed
2524 concurrently to multiple redirected devices you must blkparse your trace
2525 into separate traces and replay them with independent fio invocations.
2526 Unfortunately this also breaks the strict time ordering between multiple
2529 .. option:: replay_align=int
2531 Force alignment of I/O offsets and lengths in a trace to this power of 2
2534 .. option:: replay_scale=int
2536 Scale sector offsets down by this factor when replaying traces.
2538 .. option:: replay_skip=str
2540 Sometimes it's useful to skip certain IO types in a replay trace.
2541 This could be, for instance, eliminating the writes in the trace.
2542 Or not replaying the trims/discards, if you are redirecting to
2543 a device that doesn't support them. This option takes a comma
2544 separated list of read, write, trim, sync.
2547 Threads, processes and job synchronization
2548 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2552 Fio defaults to creating jobs by using fork, however if this option is
2553 given, fio will create jobs by using POSIX Threads' function
2554 :manpage:`pthread_create(3)` to create threads instead.
2556 .. option:: wait_for=str
2558 If set, the current job won't be started until all workers of the specified
2559 waitee job are done.
2561 ``wait_for`` operates on the job name basis, so there are a few
2562 limitations. First, the waitee must be defined prior to the waiter job
2563 (meaning no forward references). Second, if a job is being referenced as a
2564 waitee, it must have a unique name (no duplicate waitees).
2566 .. option:: nice=int
2568 Run the job with the given nice value. See man :manpage:`nice(2)`.
2570 On Windows, values less than -15 set the process class to "High"; -1 through
2571 -15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2574 .. option:: prio=int
2576 Set the I/O priority value of this job. Linux limits us to a positive value
2577 between 0 and 7, with 0 being the highest. See man
2578 :manpage:`ionice(1)`. Refer to an appropriate manpage for other operating
2579 systems since meaning of priority may differ.
2581 .. option:: prioclass=int
2583 Set the I/O priority class. See man :manpage:`ionice(1)`.
2585 .. option:: cpus_allowed=str
2587 Controls the same options as :option:`cpumask`, but accepts a textual
2588 specification of the permitted CPUs instead and CPUs are indexed from 0. So
2589 to use CPUs 0 and 5 you would specify ``cpus_allowed=0,5``. This option also
2590 allows a range of CPUs to be specified -- say you wanted a binding to CPUs
2591 0, 5, and 8 to 15, you would set ``cpus_allowed=0,5,8-15``.
2593 On Windows, when ``cpus_allowed`` is unset only CPUs from fio's current
2594 processor group will be used and affinity settings are inherited from the
2595 system. An fio build configured to target Windows 7 makes options that set
2596 CPUs processor group aware and values will set both the processor group
2597 and a CPU from within that group. For example, on a system where processor
2598 group 0 has 40 CPUs and processor group 1 has 32 CPUs, ``cpus_allowed``
2599 values between 0 and 39 will bind CPUs from processor group 0 and
2600 ``cpus_allowed`` values between 40 and 71 will bind CPUs from processor
2601 group 1. When using ``cpus_allowed_policy=shared`` all CPUs specified by a
2602 single ``cpus_allowed`` option must be from the same processor group. For
2603 Windows fio builds not built for Windows 7, CPUs will only be selected from
2604 (and be relative to) whatever processor group fio happens to be running in
2605 and CPUs from other processor groups cannot be used.
2607 .. option:: cpus_allowed_policy=str
2609 Set the policy of how fio distributes the CPUs specified by
2610 :option:`cpus_allowed` or :option:`cpumask`. Two policies are supported:
2613 All jobs will share the CPU set specified.
2615 Each job will get a unique CPU from the CPU set.
2617 **shared** is the default behavior, if the option isn't specified. If
2618 **split** is specified, then fio will will assign one cpu per job. If not
2619 enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2622 .. option:: cpumask=int
2624 Set the CPU affinity of this job. The parameter given is a bit mask of
2625 allowed CPUs the job may run on. So if you want the allowed CPUs to be 1
2626 and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2627 :manpage:`sched_setaffinity(2)`. This may not work on all supported
2628 operating systems or kernel versions. This option doesn't work well for a
2629 higher CPU count than what you can store in an integer mask, so it can only
2630 control cpus 1-32. For boxes with larger CPU counts, use
2631 :option:`cpus_allowed`.
2633 .. option:: numa_cpu_nodes=str
2635 Set this job running on specified NUMA nodes' CPUs. The arguments allow
2636 comma delimited list of cpu numbers, A-B ranges, or `all`. Note, to enable
2637 NUMA options support, fio must be built on a system with libnuma-dev(el)
2640 .. option:: numa_mem_policy=str
2642 Set this job's memory policy and corresponding NUMA nodes. Format of the
2647 ``mode`` is one of the following memory policies: ``default``, ``prefer``,
2648 ``bind``, ``interleave`` or ``local``. For ``default`` and ``local`` memory
2649 policies, no node needs to be specified. For ``prefer``, only one node is
2650 allowed. For ``bind`` and ``interleave`` the ``nodelist`` may be as
2651 follows: a comma delimited list of numbers, A-B ranges, or `all`.
2653 .. option:: cgroup=str
2655 Add job to this control group. If it doesn't exist, it will be created. The
2656 system must have a mounted cgroup blkio mount point for this to work. If
2657 your system doesn't have it mounted, you can do so with::
2659 # mount -t cgroup -o blkio none /cgroup
2661 .. option:: cgroup_weight=int
2663 Set the weight of the cgroup to this value. See the documentation that comes
2664 with the kernel, allowed values are in the range of 100..1000.
2666 .. option:: cgroup_nodelete=bool
2668 Normally fio will delete the cgroups it has created after the job
2669 completion. To override this behavior and to leave cgroups around after the
2670 job completion, set ``cgroup_nodelete=1``. This can be useful if one wants
2671 to inspect various cgroup files after job completion. Default: false.
2673 .. option:: flow_id=int
2675 The ID of the flow. If not specified, it defaults to being a global
2676 flow. See :option:`flow`.
2678 .. option:: flow=int
2680 Weight in token-based flow control. If this value is used, then there is a
2681 'flow counter' which is used to regulate the proportion of activity between
2682 two or more jobs. Fio attempts to keep this flow counter near zero. The
2683 ``flow`` parameter stands for how much should be added or subtracted to the
2684 flow counter on each iteration of the main I/O loop. That is, if one job has
2685 ``flow=8`` and another job has ``flow=-1``, then there will be a roughly 1:8
2686 ratio in how much one runs vs the other.
2688 .. option:: flow_watermark=int
2690 The maximum value that the absolute value of the flow counter is allowed to
2691 reach before the job must wait for a lower value of the counter.
2693 .. option:: flow_sleep=int
2695 The period of time, in microseconds, to wait after the flow watermark has
2696 been exceeded before retrying operations.
2698 .. option:: stonewall, wait_for_previous
2700 Wait for preceding jobs in the job file to exit, before starting this
2701 one. Can be used to insert serialization points in the job file. A stone
2702 wall also implies starting a new reporting group, see
2703 :option:`group_reporting`.
2707 By default, fio will continue running all other jobs when one job finishes
2708 but sometimes this is not the desired action. Setting ``exitall`` will
2709 instead make fio terminate all other jobs when one job finishes.
2711 .. option:: exec_prerun=str
2713 Before running this job, issue the command specified through
2714 :manpage:`system(3)`. Output is redirected in a file called
2715 :file:`jobname.prerun.txt`.
2717 .. option:: exec_postrun=str
2719 After the job completes, issue the command specified though
2720 :manpage:`system(3)`. Output is redirected in a file called
2721 :file:`jobname.postrun.txt`.
2725 Instead of running as the invoking user, set the user ID to this value
2726 before the thread/process does any work.
2730 Set group ID, see :option:`uid`.
2736 .. option:: verify_only
2738 Do not perform specified workload, only verify data still matches previous
2739 invocation of this workload. This option allows one to check data multiple
2740 times at a later date without overwriting it. This option makes sense only
2741 for workloads that write data, and does not support workloads with the
2742 :option:`time_based` option set.
2744 .. option:: do_verify=bool
2746 Run the verify phase after a write phase. Only valid if :option:`verify` is
2749 .. option:: verify=str
2751 If writing to a file, fio can verify the file contents after each iteration
2752 of the job. Each verification method also implies verification of special
2753 header, which is written to the beginning of each block. This header also
2754 includes meta information, like offset of the block, block number, timestamp
2755 when block was written, etc. :option:`verify` can be combined with
2756 :option:`verify_pattern` option. The allowed values are:
2759 Use an md5 sum of the data area and store it in the header of
2763 Use an experimental crc64 sum of the data area and store it in the
2764 header of each block.
2767 Use a crc32c sum of the data area and store it in the header of
2768 each block. This will automatically use hardware acceleration
2769 (e.g. SSE4.2 on an x86 or CRC crypto extensions on ARM64) but will
2770 fall back to software crc32c if none is found. Generally the
2771 fastest checksum fio supports when hardware accelerated.
2777 Use a crc32 sum of the data area and store it in the header of each
2781 Use a crc16 sum of the data area and store it in the header of each
2785 Use a crc7 sum of the data area and store it in the header of each
2789 Use xxhash as the checksum function. Generally the fastest software
2790 checksum that fio supports.
2793 Use sha512 as the checksum function.
2796 Use sha256 as the checksum function.
2799 Use optimized sha1 as the checksum function.
2802 Use optimized sha3-224 as the checksum function.
2805 Use optimized sha3-256 as the checksum function.
2808 Use optimized sha3-384 as the checksum function.
2811 Use optimized sha3-512 as the checksum function.
2814 This option is deprecated, since now meta information is included in
2815 generic verification header and meta verification happens by
2816 default. For detailed information see the description of the
2817 :option:`verify` setting. This option is kept because of
2818 compatibility's sake with old configurations. Do not use it.
2821 Verify a strict pattern. Normally fio includes a header with some
2822 basic information and checksumming, but if this option is set, only
2823 the specific pattern set with :option:`verify_pattern` is verified.
2826 Only pretend to verify. Useful for testing internals with
2827 :option:`ioengine`\=null, not for much else.
2829 This option can be used for repeated burn-in tests of a system to make sure
2830 that the written data is also correctly read back. If the data direction
2831 given is a read or random read, fio will assume that it should verify a
2832 previously written file. If the data direction includes any form of write,
2833 the verify will be of the newly written data.
2835 To avoid false verification errors, do not use the norandommap option when
2836 verifying data with async I/O engines and I/O depths > 1. Or use the
2837 norandommap and the lfsr random generator together to avoid writing to the
2838 same offset with muliple outstanding I/Os.
2840 .. option:: verify_offset=int
2842 Swap the verification header with data somewhere else in the block before
2843 writing. It is swapped back before verifying.
2845 .. option:: verify_interval=int
2847 Write the verification header at a finer granularity than the
2848 :option:`blocksize`. It will be written for chunks the size of
2849 ``verify_interval``. :option:`blocksize` should divide this evenly.
2851 .. option:: verify_pattern=str
2853 If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2854 filling with totally random bytes, but sometimes it's interesting to fill
2855 with a known pattern for I/O verification purposes. Depending on the width
2856 of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time (it can
2857 be either a decimal or a hex number). The ``verify_pattern`` if larger than
2858 a 32-bit quantity has to be a hex number that starts with either "0x" or
2859 "0X". Use with :option:`verify`. Also, ``verify_pattern`` supports %o
2860 format, which means that for each block offset will be written and then
2861 verified back, e.g.::
2865 Or use combination of everything::
2867 verify_pattern=0xff%o"abcd"-12
2869 .. option:: verify_fatal=bool
2871 Normally fio will keep checking the entire contents before quitting on a
2872 block verification failure. If this option is set, fio will exit the job on
2873 the first observed failure. Default: false.
2875 .. option:: verify_dump=bool
2877 If set, dump the contents of both the original data block and the data block
2878 we read off disk to files. This allows later analysis to inspect just what
2879 kind of data corruption occurred. Off by default.
2881 .. option:: verify_async=int
2883 Fio will normally verify I/O inline from the submitting thread. This option
2884 takes an integer describing how many async offload threads to create for I/O
2885 verification instead, causing fio to offload the duty of verifying I/O
2886 contents to one or more separate threads. If using this offload option, even
2887 sync I/O engines can benefit from using an :option:`iodepth` setting higher
2888 than 1, as it allows them to have I/O in flight while verifies are running.
2889 Defaults to 0 async threads, i.e. verification is not asynchronous.
2891 .. option:: verify_async_cpus=str
2893 Tell fio to set the given CPU affinity on the async I/O verification
2894 threads. See :option:`cpus_allowed` for the format used.
2896 .. option:: verify_backlog=int
2898 Fio will normally verify the written contents of a job that utilizes verify
2899 once that job has completed. In other words, everything is written then
2900 everything is read back and verified. You may want to verify continually
2901 instead for a variety of reasons. Fio stores the meta data associated with
2902 an I/O block in memory, so for large verify workloads, quite a bit of memory
2903 would be used up holding this meta data. If this option is enabled, fio will
2904 write only N blocks before verifying these blocks.
2906 .. option:: verify_backlog_batch=int
2908 Control how many blocks fio will verify if :option:`verify_backlog` is
2909 set. If not set, will default to the value of :option:`verify_backlog`
2910 (meaning the entire queue is read back and verified). If
2911 ``verify_backlog_batch`` is less than :option:`verify_backlog` then not all
2912 blocks will be verified, if ``verify_backlog_batch`` is larger than
2913 :option:`verify_backlog`, some blocks will be verified more than once.
2915 .. option:: verify_state_save=bool
2917 When a job exits during the write phase of a verify workload, save its
2918 current state. This allows fio to replay up until that point, if the verify
2919 state is loaded for the verify read phase. The format of the filename is,
2922 <type>-<jobname>-<jobindex>-verify.state.
2924 <type> is "local" for a local run, "sock" for a client/server socket
2925 connection, and "ip" (192.168.0.1, for instance) for a networked
2926 client/server connection. Defaults to true.
2928 .. option:: verify_state_load=bool
2930 If a verify termination trigger was used, fio stores the current write state
2931 of each thread. This can be used at verification time so that fio knows how
2932 far it should verify. Without this information, fio will run a full
2933 verification pass, according to the settings in the job file used. Default
2936 .. option:: trim_percentage=int
2938 Number of verify blocks to discard/trim.
2940 .. option:: trim_verify_zero=bool
2942 Verify that trim/discarded blocks are returned as zeros.
2944 .. option:: trim_backlog=int
2946 Trim after this number of blocks are written.
2948 .. option:: trim_backlog_batch=int
2950 Trim this number of I/O blocks.
2952 .. option:: experimental_verify=bool
2954 Enable experimental verification.
2959 .. option:: steadystate=str:float, ss=str:float
2961 Define the criterion and limit for assessing steady state performance. The
2962 first parameter designates the criterion whereas the second parameter sets
2963 the threshold. When the criterion falls below the threshold for the
2964 specified duration, the job will stop. For example, `iops_slope:0.1%` will
2965 direct fio to terminate the job when the least squares regression slope
2966 falls below 0.1% of the mean IOPS. If :option:`group_reporting` is enabled
2967 this will apply to all jobs in the group. Below is the list of available
2968 steady state assessment criteria. All assessments are carried out using only
2969 data from the rolling collection window. Threshold limits can be expressed
2970 as a fixed value or as a percentage of the mean in the collection window.
2973 Collect IOPS data. Stop the job if all individual IOPS measurements
2974 are within the specified limit of the mean IOPS (e.g., ``iops:2``
2975 means that all individual IOPS values must be within 2 of the mean,
2976 whereas ``iops:0.2%`` means that all individual IOPS values must be
2977 within 0.2% of the mean IOPS to terminate the job).
2980 Collect IOPS data and calculate the least squares regression
2981 slope. Stop the job if the slope falls below the specified limit.
2984 Collect bandwidth data. Stop the job if all individual bandwidth
2985 measurements are within the specified limit of the mean bandwidth.
2988 Collect bandwidth data and calculate the least squares regression
2989 slope. Stop the job if the slope falls below the specified limit.
2991 .. option:: steadystate_duration=time, ss_dur=time
2993 A rolling window of this duration will be used to judge whether steady state
2994 has been reached. Data will be collected once per second. The default is 0
2995 which disables steady state detection. When the unit is omitted, the
2996 value is interpreted in seconds.
2998 .. option:: steadystate_ramp_time=time, ss_ramp=time
3000 Allow the job to run for the specified duration before beginning data
3001 collection for checking the steady state job termination criterion. The
3002 default is 0. When the unit is omitted, the value is interpreted in seconds.
3005 Measurements and reporting
3006 ~~~~~~~~~~~~~~~~~~~~~~~~~~
3008 .. option:: per_job_logs=bool
3010 If set, this generates bw/clat/iops log with per file private filenames. If
3011 not set, jobs with identical names will share the log filename. Default:
3014 .. option:: group_reporting
3016 It may sometimes be interesting to display statistics for groups of jobs as
3017 a whole instead of for each individual job. This is especially true if
3018 :option:`numjobs` is used; looking at individual thread/process output
3019 quickly becomes unwieldy. To see the final report per-group instead of
3020 per-job, use :option:`group_reporting`. Jobs in a file will be part of the
3021 same reporting group, unless if separated by a :option:`stonewall`, or by
3022 using :option:`new_group`.
3024 .. option:: new_group
3026 Start a new reporting group. See: :option:`group_reporting`. If not given,
3027 all jobs in a file will be part of the same reporting group, unless
3028 separated by a :option:`stonewall`.
3030 .. option:: stats=bool
3032 By default, fio collects and shows final output results for all jobs
3033 that run. If this option is set to 0, then fio will ignore it in
3034 the final stat output.
3036 .. option:: write_bw_log=str
3038 If given, write a bandwidth log for this job. Can be used to store data of
3039 the bandwidth of the jobs in their lifetime.
3041 If no str argument is given, the default filename of
3042 :file:`jobname_type.x.log` is used. Even when the argument is given, fio
3043 will still append the type of log. So if one specifies::
3047 The actual log name will be :file:`foo_bw.x.log` where `x` is the index
3048 of the job (`1..N`, where `N` is the number of jobs). If
3049 :option:`per_job_logs` is false, then the filename will not include the
3052 The included :command:`fio_generate_plots` script uses :command:`gnuplot` to turn these
3053 text files into nice graphs. See `Log File Formats`_ for how data is
3054 structured within the file.
3056 .. option:: write_lat_log=str
3058 Same as :option:`write_bw_log`, except this option creates I/O
3059 submission (e.g., :file:`name_slat.x.log`), completion (e.g.,
3060 :file:`name_clat.x.log`), and total (e.g., :file:`name_lat.x.log`)
3061 latency files instead. See :option:`write_bw_log` for details about
3062 the filename format and `Log File Formats`_ for how data is structured
3065 .. option:: write_hist_log=str
3067 Same as :option:`write_bw_log` but writes an I/O completion latency
3068 histogram file (e.g., :file:`name_hist.x.log`) instead. Note that this
3069 file will be empty unless :option:`log_hist_msec` has also been set.
3070 See :option:`write_bw_log` for details about the filename format and
3071 `Log File Formats`_ for how data is structured within the file.
3073 .. option:: write_iops_log=str
3075 Same as :option:`write_bw_log`, but writes an IOPS file (e.g.
3076 :file:`name_iops.x.log`) instead. Because fio defaults to individual
3077 I/O logging, the value entry in the IOPS log will be 1 unless windowed
3078 logging (see :option:`log_avg_msec`) has been enabled. See
3079 :option:`write_bw_log` for details about the filename format and `Log
3080 File Formats`_ for how data is structured within the file.
3082 .. option:: log_avg_msec=int
3084 By default, fio will log an entry in the iops, latency, or bw log for every
3085 I/O that completes. When writing to the disk log, that can quickly grow to a
3086 very large size. Setting this option makes fio average the each log entry
3087 over the specified period of time, reducing the resolution of the log. See
3088 :option:`log_max_value` as well. Defaults to 0, logging all entries.
3089 Also see `Log File Formats`_.
3091 .. option:: log_hist_msec=int
3093 Same as :option:`log_avg_msec`, but logs entries for completion latency
3094 histograms. Computing latency percentiles from averages of intervals using
3095 :option:`log_avg_msec` is inaccurate. Setting this option makes fio log
3096 histogram entries over the specified period of time, reducing log sizes for
3097 high IOPS devices while retaining percentile accuracy. See
3098 :option:`log_hist_coarseness` and :option:`write_hist_log` as well.
3099 Defaults to 0, meaning histogram logging is disabled.
3101 .. option:: log_hist_coarseness=int
3103 Integer ranging from 0 to 6, defining the coarseness of the resolution of
3104 the histogram logs enabled with :option:`log_hist_msec`. For each increment
3105 in coarseness, fio outputs half as many bins. Defaults to 0, for which
3106 histogram logs contain 1216 latency bins. See :option:`write_hist_log`
3107 and `Log File Formats`_.
3109 .. option:: log_max_value=bool
3111 If :option:`log_avg_msec` is set, fio logs the average over that window. If
3112 you instead want to log the maximum value, set this option to 1. Defaults to
3113 0, meaning that averaged values are logged.
3115 .. option:: log_offset=bool
3117 If this is set, the iolog options will include the byte offset for the I/O
3118 entry as well as the other data values. Defaults to 0 meaning that
3119 offsets are not present in logs. Also see `Log File Formats`_.
3121 .. option:: log_compression=int
3123 If this is set, fio will compress the I/O logs as it goes, to keep the
3124 memory footprint lower. When a log reaches the specified size, that chunk is
3125 removed and compressed in the background. Given that I/O logs are fairly
3126 highly compressible, this yields a nice memory savings for longer runs. The
3127 downside is that the compression will consume some background CPU cycles, so
3128 it may impact the run. This, however, is also true if the logging ends up
3129 consuming most of the system memory. So pick your poison. The I/O logs are
3130 saved normally at the end of a run, by decompressing the chunks and storing
3131 them in the specified log file. This feature depends on the availability of
3134 .. option:: log_compression_cpus=str
3136 Define the set of CPUs that are allowed to handle online log compression for
3137 the I/O jobs. This can provide better isolation between performance
3138 sensitive jobs, and background compression work. See
3139 :option:`cpus_allowed` for the format used.
3141 .. option:: log_store_compressed=bool
3143 If set, fio will store the log files in a compressed format. They can be
3144 decompressed with fio, using the :option:`--inflate-log` command line
3145 parameter. The files will be stored with a :file:`.fz` suffix.
3147 .. option:: log_unix_epoch=bool
3149 If set, fio will log Unix timestamps to the log files produced by enabling
3150 write_type_log for each log type, instead of the default zero-based
3153 .. option:: block_error_percentiles=bool
3155 If set, record errors in trim block-sized units from writes and trims and
3156 output a histogram of how many trims it took to get to errors, and what kind
3157 of error was encountered.
3159 .. option:: bwavgtime=int
3161 Average the calculated bandwidth over the given time. Value is specified in
3162 milliseconds. If the job also does bandwidth logging through
3163 :option:`write_bw_log`, then the minimum of this option and
3164 :option:`log_avg_msec` will be used. Default: 500ms.
3166 .. option:: iopsavgtime=int
3168 Average the calculated IOPS over the given time. Value is specified in
3169 milliseconds. If the job also does IOPS logging through
3170 :option:`write_iops_log`, then the minimum of this option and
3171 :option:`log_avg_msec` will be used. Default: 500ms.
3173 .. option:: disk_util=bool
3175 Generate disk utilization statistics, if the platform supports it.
3178 .. option:: disable_lat=bool
3180 Disable measurements of total latency numbers. Useful only for cutting back
3181 the number of calls to :manpage:`gettimeofday(2)`, as that does impact
3182 performance at really high IOPS rates. Note that to really get rid of a
3183 large amount of these calls, this option must be used with
3184 :option:`disable_slat` and :option:`disable_bw_measurement` as well.
3186 .. option:: disable_clat=bool
3188 Disable measurements of completion latency numbers. See
3189 :option:`disable_lat`.
3191 .. option:: disable_slat=bool
3193 Disable measurements of submission latency numbers. See
3194 :option:`disable_lat`.
3196 .. option:: disable_bw_measurement=bool, disable_bw=bool
3198 Disable measurements of throughput/bandwidth numbers. See
3199 :option:`disable_lat`.
3201 .. option:: clat_percentiles=bool
3203 Enable the reporting of percentiles of completion latencies. This
3204 option is mutually exclusive with :option:`lat_percentiles`.
3206 .. option:: lat_percentiles=bool
3208 Enable the reporting of percentiles of I/O latencies. This is similar
3209 to :option:`clat_percentiles`, except that this includes the
3210 submission latency. This option is mutually exclusive with
3211 :option:`clat_percentiles`.
3213 .. option:: percentile_list=float_list
3215 Overwrite the default list of percentiles for completion latencies and
3216 the block error histogram. Each number is a floating number in the
3217 range (0,100], and the maximum length of the list is 20. Use ``:`` to
3218 separate the numbers, and list the numbers in ascending order. For
3219 example, ``--percentile_list=99.5:99.9`` will cause fio to report the
3220 values of completion latency below which 99.5% and 99.9% of the observed
3221 latencies fell, respectively.
3223 .. option:: significant_figures=int
3225 If using :option:`--output-format` of `normal`, set the significant
3226 figures to this value. Higher values will yield more precise IOPS and
3227 throughput units, while lower values will round. Requires a minimum
3228 value of 1 and a maximum value of 10. Defaults to 4.
3234 .. option:: exitall_on_error
3236 When one job finishes in error, terminate the rest. The default is to wait
3237 for each job to finish.
3239 .. option:: continue_on_error=str
3241 Normally fio will exit the job on the first observed failure. If this option
3242 is set, fio will continue the job when there is a 'non-fatal error' (EIO or
3243 EILSEQ) until the runtime is exceeded or the I/O size specified is
3244 completed. If this option is used, there are two more stats that are
3245 appended, the total error count and the first error. The error field given
3246 in the stats is the first error that was hit during the run.
3248 The allowed values are:
3251 Exit on any I/O or verify errors.
3254 Continue on read errors, exit on all others.
3257 Continue on write errors, exit on all others.
3260 Continue on any I/O error, exit on all others.
3263 Continue on verify errors, exit on all others.
3266 Continue on all errors.
3269 Backward-compatible alias for 'none'.
3272 Backward-compatible alias for 'all'.
3274 .. option:: ignore_error=str
3276 Sometimes you want to ignore some errors during test in that case you can
3277 specify error list for each error type, instead of only being able to
3278 ignore the default 'non-fatal error' using :option:`continue_on_error`.
3279 ``ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST`` errors for
3280 given error type is separated with ':'. Error may be symbol ('ENOSPC',
3281 'ENOMEM') or integer. Example::
3283 ignore_error=EAGAIN,ENOSPC:122
3285 This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
3286 WRITE. This option works by overriding :option:`continue_on_error` with
3287 the list of errors for each error type if any.
3289 .. option:: error_dump=bool
3291 If set dump every error even if it is non fatal, true by default. If
3292 disabled only fatal error will be dumped.
3294 Running predefined workloads
3295 ----------------------------
3297 Fio includes predefined profiles that mimic the I/O workloads generated by
3300 .. option:: profile=str
3302 The predefined workload to run. Current profiles are:
3305 Threaded I/O bench (tiotest/tiobench) like workload.
3308 Aerospike Certification Tool (ACT) like workload.
3310 To view a profile's additional options use :option:`--cmdhelp` after specifying
3311 the profile. For example::
3313 $ fio --profile=act --cmdhelp
3318 .. option:: device-names=str
3323 .. option:: load=int
3326 ACT load multiplier. Default: 1.
3328 .. option:: test-duration=time
3331 How long the entire test takes to run. When the unit is omitted, the value
3332 is given in seconds. Default: 24h.
3334 .. option:: threads-per-queue=int
3337 Number of read I/O threads per device. Default: 8.
3339 .. option:: read-req-num-512-blocks=int
3342 Number of 512B blocks to read at the time. Default: 3.
3344 .. option:: large-block-op-kbytes=int
3347 Size of large block ops in KiB (writes). Default: 131072.
3352 Set to run ACT prep phase.
3354 Tiobench profile options
3355 ~~~~~~~~~~~~~~~~~~~~~~~~
3357 .. option:: size=str
3362 .. option:: block=int
3365 Block size in bytes. Default: 4096.
3367 .. option:: numruns=int
3377 .. option:: threads=int
3382 Interpreting the output
3383 -----------------------
3386 Example output was based on the following:
3387 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --time_based \
3388 --rate=1256k --bs=14K --name=quick --runtime=1s --name=mixed \
3389 --runtime=2m --rw=rw
3391 Fio spits out a lot of output. While running, fio will display the status of the
3392 jobs created. An example of that would be::
3394 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]
3396 The characters inside the first set of square brackets denote the current status of
3397 each thread. The first character is the first job defined in the job file, and so
3398 forth. The possible values (in typical life cycle order) are:
3400 +------+-----+-----------------------------------------------------------+
3402 +======+=====+===========================================================+
3403 | P | | Thread setup, but not started. |
3404 +------+-----+-----------------------------------------------------------+
3405 | C | | Thread created. |
3406 +------+-----+-----------------------------------------------------------+
3407 | I | | Thread initialized, waiting or generating necessary data. |
3408 +------+-----+-----------------------------------------------------------+
3409 | | p | Thread running pre-reading file(s). |
3410 +------+-----+-----------------------------------------------------------+
3411 | | / | Thread is in ramp period. |
3412 +------+-----+-----------------------------------------------------------+
3413 | | R | Running, doing sequential reads. |
3414 +------+-----+-----------------------------------------------------------+
3415 | | r | Running, doing random reads. |
3416 +------+-----+-----------------------------------------------------------+
3417 | | W | Running, doing sequential writes. |
3418 +------+-----+-----------------------------------------------------------+
3419 | | w | Running, doing random writes. |
3420 +------+-----+-----------------------------------------------------------+
3421 | | M | Running, doing mixed sequential reads/writes. |
3422 +------+-----+-----------------------------------------------------------+
3423 | | m | Running, doing mixed random reads/writes. |
3424 +------+-----+-----------------------------------------------------------+
3425 | | D | Running, doing sequential trims. |
3426 +------+-----+-----------------------------------------------------------+
3427 | | d | Running, doing random trims. |
3428 +------+-----+-----------------------------------------------------------+
3429 | | F | Running, currently waiting for :manpage:`fsync(2)`. |
3430 +------+-----+-----------------------------------------------------------+
3431 | | V | Running, doing verification of written data. |
3432 +------+-----+-----------------------------------------------------------+
3433 | f | | Thread finishing. |
3434 +------+-----+-----------------------------------------------------------+
3435 | E | | Thread exited, not reaped by main thread yet. |
3436 +------+-----+-----------------------------------------------------------+
3437 | _ | | Thread reaped. |
3438 +------+-----+-----------------------------------------------------------+
3439 | X | | Thread reaped, exited with an error. |
3440 +------+-----+-----------------------------------------------------------+
3441 | K | | Thread reaped, exited due to signal. |
3442 +------+-----+-----------------------------------------------------------+
3445 Example output was based on the following:
3446 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --runtime=58m \
3447 --time_based --rate=2512k --bs=256K --numjobs=10 \
3448 --name=readers --rw=read --name=writers --rw=write
3450 Fio will condense the thread string as not to take up more space on the command
3451 line than needed. For instance, if you have 10 readers and 10 writers running,
3452 the output would look like this::
3454 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]
3456 Note that the status string is displayed in order, so it's possible to tell which of
3457 the jobs are currently doing what. In the example above this means that jobs 1--10
3458 are readers and 11--20 are writers.
3460 The other values are fairly self explanatory -- number of threads currently
3461 running and doing I/O, the number of currently open files (f=), the estimated
3462 completion percentage, the rate of I/O since last check (read speed listed first,
3463 then write speed and optionally trim speed) in terms of bandwidth and IOPS,
3464 and time to completion for the current running group. It's impossible to estimate
3465 runtime of the following groups (if any).
3468 Example output was based on the following:
3469 TZ=UTC fio --iodepth=16 --ioengine=posixaio --filename=/tmp/fiofile \
3470 --direct=1 --size=100M --time_based --runtime=50s --rate_iops=89 \
3471 --bs=7K --name=Client1 --rw=write
3473 When fio is done (or interrupted by :kbd:`Ctrl-C`), it will show the data for
3474 each thread, group of threads, and disks in that order. For each overall thread (or
3475 group) the output looks like::
3477 Client1: (groupid=0, jobs=1): err= 0: pid=16109: Sat Jun 24 12:07:54 2017
3478 write: IOPS=88, BW=623KiB/s (638kB/s)(30.4MiB/50032msec)
3479 slat (nsec): min=500, max=145500, avg=8318.00, stdev=4781.50
3480 clat (usec): min=170, max=78367, avg=4019.02, stdev=8293.31
3481 lat (usec): min=174, max=78375, avg=4027.34, stdev=8291.79
3482 clat percentiles (usec):
3483 | 1.00th=[ 302], 5.00th=[ 326], 10.00th=[ 343], 20.00th=[ 363],
3484 | 30.00th=[ 392], 40.00th=[ 404], 50.00th=[ 416], 60.00th=[ 445],
3485 | 70.00th=[ 816], 80.00th=[ 6718], 90.00th=[12911], 95.00th=[21627],
3486 | 99.00th=[43779], 99.50th=[51643], 99.90th=[68682], 99.95th=[72877],
3488 bw ( KiB/s): min= 532, max= 686, per=0.10%, avg=622.87, stdev=24.82, samples= 100
3489 iops : min= 76, max= 98, avg=88.98, stdev= 3.54, samples= 100
3490 lat (usec) : 250=0.04%, 500=64.11%, 750=4.81%, 1000=2.79%
3491 lat (msec) : 2=4.16%, 4=1.84%, 10=4.90%, 20=11.33%, 50=5.37%
3492 lat (msec) : 100=0.65%
3493 cpu : usr=0.27%, sys=0.18%, ctx=12072, majf=0, minf=21
3494 IO depths : 1=85.0%, 2=13.1%, 4=1.8%, 8=0.1%, 16=0.0%, 32=0.0%, >=64=0.0%
3495 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3496 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3497 issued rwt: total=0,4450,0, short=0,0,0, dropped=0,0,0
3498 latency : target=0, window=0, percentile=100.00%, depth=8
3500 The job name (or first job's name when using :option:`group_reporting`) is printed,
3501 along with the group id, count of jobs being aggregated, last error id seen (which
3502 is 0 when there are no errors), pid/tid of that thread and the time the job/group
3503 completed. Below are the I/O statistics for each data direction performed (showing
3504 writes in the example above). In the order listed, they denote:
3507 The string before the colon shows the I/O direction the statistics
3508 are for. **IOPS** is the average I/Os performed per second. **BW**
3509 is the average bandwidth rate shown as: value in power of 2 format
3510 (value in power of 10 format). The last two values show: (**total
3511 I/O performed** in power of 2 format / **runtime** of that thread).
3514 Submission latency (**min** being the minimum, **max** being the
3515 maximum, **avg** being the average, **stdev** being the standard
3516 deviation). This is the time it took to submit the I/O. For
3517 sync I/O this row is not displayed as the slat is really the
3518 completion latency (since queue/complete is one operation there).
3519 This value can be in nanoseconds, microseconds or milliseconds ---
3520 fio will choose the most appropriate base and print that (in the
3521 example above nanoseconds was the best scale). Note: in :option:`--minimal` mode
3522 latencies are always expressed in microseconds.
3525 Completion latency. Same names as slat, this denotes the time from
3526 submission to completion of the I/O pieces. For sync I/O, clat will
3527 usually be equal (or very close) to 0, as the time from submit to
3528 complete is basically just CPU time (I/O has already been done, see slat
3532 Total latency. Same names as slat and clat, this denotes the time from
3533 when fio created the I/O unit to completion of the I/O operation.
3536 Bandwidth statistics based on samples. Same names as the xlat stats,
3537 but also includes the number of samples taken (**samples**) and an
3538 approximate percentage of total aggregate bandwidth this thread
3539 received in its group (**per**). This last value is only really
3540 useful if the threads in this group are on the same disk, since they
3541 are then competing for disk access.
3544 IOPS statistics based on samples. Same names as bw.
3546 **lat (nsec/usec/msec)**
3547 The distribution of I/O completion latencies. This is the time from when
3548 I/O leaves fio and when it gets completed. Unlike the separate
3549 read/write/trim sections above, the data here and in the remaining
3550 sections apply to all I/Os for the reporting group. 250=0.04% means that
3551 0.04% of the I/Os completed in under 250us. 500=64.11% means that 64.11%
3552 of the I/Os required 250 to 499us for completion.
3555 CPU usage. User and system time, along with the number of context
3556 switches this thread went through, usage of system and user time, and
3557 finally the number of major and minor page faults. The CPU utilization
3558 numbers are averages for the jobs in that reporting group, while the
3559 context and fault counters are summed.
3562 The distribution of I/O depths over the job lifetime. The numbers are
3563 divided into powers of 2 and each entry covers depths from that value
3564 up to those that are lower than the next entry -- e.g., 16= covers
3565 depths from 16 to 31. Note that the range covered by a depth
3566 distribution entry can be different to the range covered by the
3567 equivalent submit/complete distribution entry.
3570 How many pieces of I/O were submitting in a single submit call. Each
3571 entry denotes that amount and below, until the previous entry -- e.g.,
3572 16=100% means that we submitted anywhere between 9 to 16 I/Os per submit
3573 call. Note that the range covered by a submit distribution entry can
3574 be different to the range covered by the equivalent depth distribution
3578 Like the above submit number, but for completions instead.
3581 The number of read/write/trim requests issued, and how many of them were
3585 These values are for :option:`latency_target` and related options. When
3586 these options are engaged, this section describes the I/O depth required
3587 to meet the specified latency target.
3590 Example output was based on the following:
3591 TZ=UTC fio --ioengine=null --iodepth=2 --size=100M --numjobs=2 \
3592 --rate_process=poisson --io_limit=32M --name=read --bs=128k \
3593 --rate=11M --name=write --rw=write --bs=2k --rate=700k
3595 After each client has been listed, the group statistics are printed. They
3596 will look like this::
3598 Run status group 0 (all jobs):
3599 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
3600 WRITE: bw=1231KiB/s (1261kB/s), 616KiB/s-621KiB/s (630kB/s-636kB/s), io=64.0MiB (67.1MB), run=52747-53223msec
3602 For each data direction it prints:
3605 Aggregate bandwidth of threads in this group followed by the
3606 minimum and maximum bandwidth of all the threads in this group.
3607 Values outside of brackets are power-of-2 format and those
3608 within are the equivalent value in a power-of-10 format.
3610 Aggregate I/O performed of all threads in this group. The
3611 format is the same as bw.
3613 The smallest and longest runtimes of the threads in this group.
3615 And finally, the disk statistics are printed. This is Linux specific. They will look like this::
3617 Disk stats (read/write):
3618 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3620 Each value is printed for both reads and writes, with reads first. The
3624 Number of I/Os performed by all groups.
3626 Number of merges performed by the I/O scheduler.
3628 Number of ticks we kept the disk busy.
3630 Total time spent in the disk queue.
3632 The disk utilization. A value of 100% means we kept the disk
3633 busy constantly, 50% would be a disk idling half of the time.
3635 It is also possible to get fio to dump the current output while it is running,
3636 without terminating the job. To do that, send fio the **USR1** signal. You can
3637 also get regularly timed dumps by using the :option:`--status-interval`
3638 parameter, or by creating a file in :file:`/tmp` named
3639 :file:`fio-dump-status`. If fio sees this file, it will unlink it and dump the
3640 current output status.
3646 For scripted usage where you typically want to generate tables or graphs of the
3647 results, fio can output the results in a semicolon separated format. The format
3648 is one long line of values, such as::
3650 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%
3651 A description of this job goes here.
3653 The job description (if provided) follows on a second line.
3655 To enable terse output, use the :option:`--minimal` or
3656 :option:`--output-format`\=terse command line options. The
3657 first value is the version of the terse output format. If the output has to be
3658 changed for some reason, this number will be incremented by 1 to signify that
3661 Split up, the format is as follows (comments in brackets denote when a
3662 field was introduced or whether it's specific to some terse version):
3666 terse version, fio version [v3], jobname, groupid, error
3670 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3671 Submission latency: min, max, mean, stdev (usec)
3672 Completion latency: min, max, mean, stdev (usec)
3673 Completion latency percentiles: 20 fields (see below)
3674 Total latency: min, max, mean, stdev (usec)
3675 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3676 IOPS [v5]: min, max, mean, stdev, number of samples
3682 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3683 Submission latency: min, max, mean, stdev (usec)
3684 Completion latency: min, max, mean, stdev (usec)
3685 Completion latency percentiles: 20 fields (see below)
3686 Total latency: min, max, mean, stdev (usec)
3687 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3688 IOPS [v5]: min, max, mean, stdev, number of samples
3690 TRIM status [all but version 3]:
3692 Fields are similar to READ/WRITE status.
3696 user, system, context switches, major faults, minor faults
3700 <=1, 2, 4, 8, 16, 32, >=64
3702 I/O latencies microseconds::
3704 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3706 I/O latencies milliseconds::
3708 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3710 Disk utilization [v3]::
3712 disk name, read ios, write ios, read merges, write merges, read ticks, write ticks,
3713 time spent in queue, disk utilization percentage
3715 Additional Info (dependent on continue_on_error, default off)::
3717 total # errors, first error code
3719 Additional Info (dependent on description being set)::
3723 Completion latency percentiles can be a grouping of up to 20 sets, so for the
3724 terse output fio writes all of them. Each field will look like this::
3728 which is the Xth percentile, and the `usec` latency associated with it.
3730 For `Disk utilization`, all disks used by fio are shown. So for each disk there
3731 will be a disk utilization section.
3733 Below is a single line containing short names for each of the fields in the
3734 minimal output v3, separated by semicolons::
3736 terse_version_3;fio_version;jobname;groupid;error;read_kb;read_bandwidth;read_iops;read_runtime_ms;read_slat_min;read_slat_max;read_slat_mean;read_slat_dev;read_clat_min;read_clat_max;read_clat_mean;read_clat_dev;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;read_lat_max;read_lat_mean;read_lat_dev;read_bw_min;read_bw_max;read_bw_agg_pct;read_bw_mean;read_bw_dev;write_kb;write_bandwidth;write_iops;write_runtime_ms;write_slat_min;write_slat_max;write_slat_mean;write_slat_dev;write_clat_min;write_clat_max;write_clat_mean;write_clat_dev;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;write_lat_max;write_lat_mean;write_lat_dev;write_bw_min;write_bw_max;write_bw_agg_pct;write_bw_mean;write_bw_dev;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
3742 The `json` output format is intended to be both human readable and convenient
3743 for automated parsing. For the most part its sections mirror those of the
3744 `normal` output. The `runtime` value is reported in msec and the `bw` value is
3745 reported in 1024 bytes per second units.
3751 The `json+` output format is identical to the `json` output format except that it
3752 adds a full dump of the completion latency bins. Each `bins` object contains a
3753 set of (key, value) pairs where keys are latency durations and values count how
3754 many I/Os had completion latencies of the corresponding duration. For example,
3757 "bins" : { "87552" : 1, "89600" : 1, "94720" : 1, "96768" : 1, "97792" : 1, "99840" : 1, "100864" : 2, "103936" : 6, "104960" : 534, "105984" : 5995, "107008" : 7529, ... }
3759 This data indicates that one I/O required 87,552ns to complete, two I/Os required
3760 100,864ns to complete, and 7529 I/Os required 107,008ns to complete.
3762 Also included with fio is a Python script `fio_jsonplus_clat2csv` that takes
3763 json+ output and generates CSV-formatted latency data suitable for plotting.
3765 The latency durations actually represent the midpoints of latency intervals.
3766 For details refer to :file:`stat.h`.
3772 There are two trace file format that you can encounter. The older (v1) format is
3773 unsupported since version 1.20-rc3 (March 2008). It will still be described
3774 below in case that you get an old trace and want to understand it.
3776 In any case the trace is a simple text file with a single action per line.
3779 Trace file format v1
3780 ~~~~~~~~~~~~~~~~~~~~
3782 Each line represents a single I/O action in the following format::
3786 where `rw=0/1` for read/write, and the `offset` and `length` entries being in bytes.
3788 This format is not supported in fio versions >= 1.20-rc3.
3791 Trace file format v2
3792 ~~~~~~~~~~~~~~~~~~~~
3794 The second version of the trace file format was added in fio version 1.17. It
3795 allows to access more then one file per trace and has a bigger set of possible
3798 The first line of the trace file has to be::
3802 Following this can be lines in two different formats, which are described below.
3804 The file management format::
3808 The `filename` is given as an absolute path. The `action` can be one of these:
3811 Add the given `filename` to the trace.
3813 Open the file with the given `filename`. The `filename` has to have
3814 been added with the **add** action before.
3816 Close the file with the given `filename`. The file has to have been
3820 The file I/O action format::
3822 filename action offset length
3824 The `filename` is given as an absolute path, and has to have been added and
3825 opened before it can be used with this format. The `offset` and `length` are
3826 given in bytes. The `action` can be one of these:
3829 Wait for `offset` microseconds. Everything below 100 is discarded.
3830 The time is relative to the previous `wait` statement.
3832 Read `length` bytes beginning from `offset`.
3834 Write `length` bytes beginning from `offset`.
3836 :manpage:`fsync(2)` the file.
3838 :manpage:`fdatasync(2)` the file.
3840 Trim the given file from the given `offset` for `length` bytes.
3842 CPU idleness profiling
3843 ----------------------
3845 In some cases, we want to understand CPU overhead in a test. For example, we
3846 test patches for the specific goodness of whether they reduce CPU usage.
3847 Fio implements a balloon approach to create a thread per CPU that runs at idle
3848 priority, meaning that it only runs when nobody else needs the cpu.
3849 By measuring the amount of work completed by the thread, idleness of each CPU
3850 can be derived accordingly.
3852 An unit work is defined as touching a full page of unsigned characters. Mean and
3853 standard deviation of time to complete an unit work is reported in "unit work"
3854 section. Options can be chosen to report detailed percpu idleness or overall
3855 system idleness by aggregating percpu stats.
3858 Verification and triggers
3859 -------------------------
3861 Fio is usually run in one of two ways, when data verification is done. The first
3862 is a normal write job of some sort with verify enabled. When the write phase has
3863 completed, fio switches to reads and verifies everything it wrote. The second
3864 model is running just the write phase, and then later on running the same job
3865 (but with reads instead of writes) to repeat the same I/O patterns and verify
3866 the contents. Both of these methods depend on the write phase being completed,
3867 as fio otherwise has no idea how much data was written.
3869 With verification triggers, fio supports dumping the current write state to
3870 local files. Then a subsequent read verify workload can load this state and know
3871 exactly where to stop. This is useful for testing cases where power is cut to a
3872 server in a managed fashion, for instance.
3874 A verification trigger consists of two things:
3876 1) Storing the write state of each job.
3877 2) Executing a trigger command.
3879 The write state is relatively small, on the order of hundreds of bytes to single
3880 kilobytes. It contains information on the number of completions done, the last X
3883 A trigger is invoked either through creation ('touch') of a specified file in
3884 the system, or through a timeout setting. If fio is run with
3885 :option:`--trigger-file`\= :file:`/tmp/trigger-file`, then it will continually
3886 check for the existence of :file:`/tmp/trigger-file`. When it sees this file, it
3887 will fire off the trigger (thus saving state, and executing the trigger
3890 For client/server runs, there's both a local and remote trigger. If fio is
3891 running as a server backend, it will send the job states back to the client for
3892 safe storage, then execute the remote trigger, if specified. If a local trigger
3893 is specified, the server will still send back the write state, but the client
3894 will then execute the trigger.
3896 Verification trigger example
3897 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3899 Let's say we want to run a powercut test on the remote Linux machine 'server'.
3900 Our write workload is in :file:`write-test.fio`. We want to cut power to 'server' at
3901 some point during the run, and we'll run this test from the safety or our local
3902 machine, 'localbox'. On the server, we'll start the fio backend normally::
3904 server# fio --server
3906 and on the client, we'll fire off the workload::
3908 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
3910 We set :file:`/tmp/my-trigger` as the trigger file, and we tell fio to execute::
3912 echo b > /proc/sysrq-trigger
3914 on the server once it has received the trigger and sent us the write state. This
3915 will work, but it's not **really** cutting power to the server, it's merely
3916 abruptly rebooting it. If we have a remote way of cutting power to the server
3917 through IPMI or similar, we could do that through a local trigger command
3918 instead. Let's assume we have a script that does IPMI reboot of a given hostname,
3919 ipmi-reboot. On localbox, we could then have run fio with a local trigger
3922 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
3924 For this case, fio would wait for the server to send us the write state, then
3925 execute ``ipmi-reboot server`` when that happened.
3927 Loading verify state
3928 ~~~~~~~~~~~~~~~~~~~~
3930 To load stored write state, a read verification job file must contain the
3931 :option:`verify_state_load` option. If that is set, fio will load the previously
3932 stored state. For a local fio run this is done by loading the files directly,
3933 and on a client/server run, the server backend will ask the client to send the
3934 files over and load them from there.
3940 Fio supports a variety of log file formats, for logging latencies, bandwidth,
3941 and IOPS. The logs share a common format, which looks like this:
3943 *time* (`msec`), *value*, *data direction*, *block size* (`bytes`),
3946 *Time* for the log entry is always in milliseconds. The *value* logged depends
3947 on the type of log, it will be one of the following:
3950 Value is latency in nsecs
3956 *Data direction* is one of the following:
3965 The entry's *block size* is always in bytes. The *offset* is the position in bytes
3966 from the start of the file for that particular I/O. The logging of the offset can be
3967 toggled with :option:`log_offset`.
3969 Fio defaults to logging every individual I/O but when windowed logging is set
3970 through :option:`log_avg_msec`, either the average (by default) or the maximum
3971 (:option:`log_max_value` is set) *value* seen over the specified period of time
3972 is recorded. Each *data direction* seen within the window period will aggregate
3973 its values in a separate row. Further, when using windowed logging the *block
3974 size* and *offset* entries will always contain 0.
3979 Normally fio is invoked as a stand-alone application on the machine where the
3980 I/O workload should be generated. However, the backend and frontend of fio can
3981 be run separately i.e., the fio server can generate an I/O workload on the "Device
3982 Under Test" while being controlled by a client on another machine.
3984 Start the server on the machine which has access to the storage DUT::
3988 where `args` defines what fio listens to. The arguments are of the form
3989 ``type,hostname`` or ``IP,port``. *type* is either ``ip`` (or ip4) for TCP/IP
3990 v4, ``ip6`` for TCP/IP v6, or ``sock`` for a local unix domain socket.
3991 *hostname* is either a hostname or IP address, and *port* is the port to listen
3992 to (only valid for TCP/IP, not a local socket). Some examples:
3996 Start a fio server, listening on all interfaces on the default port (8765).
3998 2) ``fio --server=ip:hostname,4444``
4000 Start a fio server, listening on IP belonging to hostname and on port 4444.
4002 3) ``fio --server=ip6:::1,4444``
4004 Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
4006 4) ``fio --server=,4444``
4008 Start a fio server, listening on all interfaces on port 4444.
4010 5) ``fio --server=1.2.3.4``
4012 Start a fio server, listening on IP 1.2.3.4 on the default port.
4014 6) ``fio --server=sock:/tmp/fio.sock``
4016 Start a fio server, listening on the local socket :file:`/tmp/fio.sock`.
4018 Once a server is running, a "client" can connect to the fio server with::
4020 fio <local-args> --client=<server> <remote-args> <job file(s)>
4022 where `local-args` are arguments for the client where it is running, `server`
4023 is the connect string, and `remote-args` and `job file(s)` are sent to the
4024 server. The `server` string follows the same format as it does on the server
4025 side, to allow IP/hostname/socket and port strings.
4027 Fio can connect to multiple servers this way::
4029 fio --client=<server1> <job file(s)> --client=<server2> <job file(s)>
4031 If the job file is located on the fio server, then you can tell the server to
4032 load a local file as well. This is done by using :option:`--remote-config` ::
4034 fio --client=server --remote-config /path/to/file.fio
4036 Then fio will open this local (to the server) job file instead of being passed
4037 one from the client.
4039 If you have many servers (example: 100 VMs/containers), you can input a pathname
4040 of a file containing host IPs/names as the parameter value for the
4041 :option:`--client` option. For example, here is an example :file:`host.list`
4042 file containing 2 hostnames::
4044 host1.your.dns.domain
4045 host2.your.dns.domain
4047 The fio command would then be::
4049 fio --client=host.list <job file(s)>
4051 In this mode, you cannot input server-specific parameters or job files -- all
4052 servers receive the same job file.
4054 In order to let ``fio --client`` runs use a shared filesystem from multiple
4055 hosts, ``fio --client`` now prepends the IP address of the server to the
4056 filename. For example, if fio is using the directory :file:`/mnt/nfs/fio` and is
4057 writing filename :file:`fileio.tmp`, with a :option:`--client` `hostfile`
4058 containing two hostnames ``h1`` and ``h2`` with IP addresses 192.168.10.120 and
4059 192.168.10.121, then fio will create two files::
4061 /mnt/nfs/fio/192.168.10.120.fileio.tmp
4062 /mnt/nfs/fio/192.168.10.121.fileio.tmp