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.
199 .. option:: --section=name
201 Only run specified section `name` in job file. Multiple sections can be specified.
202 The ``--section`` option allows one to combine related jobs into one file.
203 E.g. one job file could define light, moderate, and heavy sections. Tell
204 fio to run only the "heavy" section by giving ``--section=heavy``
205 command line option. One can also specify the "write" operations in one
206 section and "verify" operation in another section. The ``--section`` option
207 only applies to job sections. The reserved *global* section is always
210 .. option:: --alloc-size=kb
212 Set the internal smalloc pool size to `kb` in KiB. The
213 ``--alloc-size`` switch allows one to use a larger pool size for smalloc.
214 If running large jobs with randommap enabled, fio can run out of memory.
215 Smalloc is an internal allocator for shared structures from a fixed size
216 memory pool and can grow to 16 pools. The pool size defaults to 16MiB.
218 NOTE: While running :file:`.fio_smalloc.*` backing store files are visible
221 .. option:: --warnings-fatal
223 All fio parser warnings are fatal, causing fio to exit with an
226 .. option:: --max-jobs=nr
228 Set the maximum number of threads/processes to support to `nr`.
229 NOTE: On Linux, it may be necessary to increase the shared-memory
230 limit (:file:`/proc/sys/kernel/shmmax`) if fio runs into errors while
233 .. option:: --server=args
235 Start a backend server, with `args` specifying what to listen to.
236 See `Client/Server`_ section.
238 .. option:: --daemonize=pidfile
240 Background a fio server, writing the pid to the given `pidfile` file.
242 .. option:: --client=hostname
244 Instead of running the jobs locally, send and run them on the given `hostname`
245 or set of `hostname`\s. See `Client/Server`_ section.
247 .. option:: --remote-config=file
249 Tell fio server to load this local `file`.
251 .. option:: --idle-prof=option
253 Report CPU idleness. `option` is one of the following:
256 Run unit work calibration only and exit.
259 Show aggregate system idleness and unit work.
262 As **system** but also show per CPU idleness.
264 .. option:: --inflate-log=log
266 Inflate and output compressed `log`.
268 .. option:: --trigger-file=file
270 Execute trigger command when `file` exists.
272 .. option:: --trigger-timeout=time
274 Execute trigger at this `time`.
276 .. option:: --trigger=command
278 Set this `command` as local trigger.
280 .. option:: --trigger-remote=command
282 Set this `command` as remote trigger.
284 .. option:: --aux-path=path
286 Use the directory specified by `path` for generated state files instead
287 of the current working directory.
289 Any parameters following the options will be assumed to be job files, unless
290 they match a job file parameter. Multiple job files can be listed and each job
291 file will be regarded as a separate group. Fio will :option:`stonewall`
292 execution between each group.
298 As previously described, fio accepts one or more job files describing what it is
299 supposed to do. The job file format is the classic ini file, where the names
300 enclosed in [] brackets define the job name. You are free to use any ASCII name
301 you want, except *global* which has special meaning. Following the job name is
302 a sequence of zero or more parameters, one per line, that define the behavior of
303 the job. If the first character in a line is a ';' or a '#', the entire line is
304 discarded as a comment.
306 A *global* section sets defaults for the jobs described in that file. A job may
307 override a *global* section parameter, and a job file may even have several
308 *global* sections if so desired. A job is only affected by a *global* section
311 The :option:`--cmdhelp` option also lists all options. If used with a `command`
312 argument, :option:`--cmdhelp` will detail the given `command`.
314 See the `examples/` directory for inspiration on how to write job files. Note
315 the copyright and license requirements currently apply to `examples/` files.
317 So let's look at a really simple job file that defines two processes, each
318 randomly reading from a 128MiB file:
322 ; -- start job file --
333 As you can see, the job file sections themselves are empty as all the described
334 parameters are shared. As no :option:`filename` option is given, fio makes up a
335 `filename` for each of the jobs as it sees fit. On the command line, this job
336 would look as follows::
338 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
341 Let's look at an example that has a number of processes writing randomly to
346 ; -- start job file --
357 Here we have no *global* section, as we only have one job defined anyway. We
358 want to use async I/O here, with a depth of 4 for each file. We also increased
359 the buffer size used to 32KiB and define numjobs to 4 to fork 4 identical
360 jobs. The result is 4 processes each randomly writing to their own 64MiB
361 file. Instead of using the above job file, you could have given the parameters
362 on the command line. For this case, you would specify::
364 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
366 When fio is utilized as a basis of any reasonably large test suite, it might be
367 desirable to share a set of standardized settings across multiple job files.
368 Instead of copy/pasting such settings, any section may pull in an external
369 :file:`filename.fio` file with *include filename* directive, as in the following
372 ; -- start job file including.fio --
376 include glob-include.fio
383 include test-include.fio
384 ; -- end job file including.fio --
388 ; -- start job file glob-include.fio --
391 ; -- end job file glob-include.fio --
395 ; -- start job file test-include.fio --
398 ; -- end job file test-include.fio --
400 Settings pulled into a section apply to that section only (except *global*
401 section). Include directives may be nested in that any included file may contain
402 further include directive(s). Include files may not contain [] sections.
405 Environment variables
406 ~~~~~~~~~~~~~~~~~~~~~
408 Fio also supports environment variable expansion in job files. Any sub-string of
409 the form ``${VARNAME}`` as part of an option value (in other words, on the right
410 of the '='), will be expanded to the value of the environment variable called
411 `VARNAME`. If no such environment variable is defined, or `VARNAME` is the
412 empty string, the empty string will be substituted.
414 As an example, let's look at a sample fio invocation and job file::
416 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
420 ; -- start job file --
427 This will expand to the following equivalent job file at runtime:
431 ; -- start job file --
438 Fio ships with a few example job files, you can also look there for inspiration.
443 Additionally, fio has a set of reserved keywords that will be replaced
444 internally with the appropriate value. Those keywords are:
448 The architecture page size of the running system.
452 Megabytes of total memory in the system.
456 Number of online available CPUs.
458 These can be used on the command line or in the job file, and will be
459 automatically substituted with the current system values when the job is
460 run. Simple math is also supported on these keywords, so you can perform actions
465 and get that properly expanded to 8 times the size of memory in the machine.
471 This section describes in details each parameter associated with a job. Some
472 parameters take an option of a given type, such as an integer or a
473 string. Anywhere a numeric value is required, an arithmetic expression may be
474 used, provided it is surrounded by parentheses. Supported operators are:
483 For time values in expressions, units are microseconds by default. This is
484 different than for time values not in expressions (not enclosed in
485 parentheses). The following types are used:
492 String: A sequence of alphanumeric characters.
495 Integer with possible time suffix. Without a unit value is interpreted as
496 seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
497 hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and
498 'us' (or 'usec') for microseconds. For example, use 10m for 10 minutes.
503 Integer. A whole number value, which may contain an integer prefix
504 and an integer suffix:
506 [*integer prefix*] **number** [*integer suffix*]
508 The optional *integer prefix* specifies the number's base. The default
509 is decimal. *0x* specifies hexadecimal.
511 The optional *integer suffix* specifies the number's units, and includes an
512 optional unit prefix and an optional unit. For quantities of data, the
513 default unit is bytes. For quantities of time, the default unit is seconds
514 unless otherwise specified.
516 With :option:`kb_base`\=1000, fio follows international standards for unit
517 prefixes. To specify power-of-10 decimal values defined in the
518 International System of Units (SI):
520 * *K* -- means kilo (K) or 1000
521 * *M* -- means mega (M) or 1000**2
522 * *G* -- means giga (G) or 1000**3
523 * *T* -- means tera (T) or 1000**4
524 * *P* -- means peta (P) or 1000**5
526 To specify power-of-2 binary values defined in IEC 80000-13:
528 * *Ki* -- means kibi (Ki) or 1024
529 * *Mi* -- means mebi (Mi) or 1024**2
530 * *Gi* -- means gibi (Gi) or 1024**3
531 * *Ti* -- means tebi (Ti) or 1024**4
532 * *Pi* -- means pebi (Pi) or 1024**5
534 With :option:`kb_base`\=1024 (the default), the unit prefixes are opposite
535 from those specified in the SI and IEC 80000-13 standards to provide
536 compatibility with old scripts. For example, 4k means 4096.
538 For quantities of data, an optional unit of 'B' may be included
539 (e.g., 'kB' is the same as 'k').
541 The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
542 not milli). 'b' and 'B' both mean byte, not bit.
544 Examples with :option:`kb_base`\=1000:
546 * *4 KiB*: 4096, 4096b, 4096B, 4ki, 4kib, 4kiB, 4Ki, 4KiB
547 * *1 MiB*: 1048576, 1mi, 1024ki
548 * *1 MB*: 1000000, 1m, 1000k
549 * *1 TiB*: 1099511627776, 1ti, 1024gi, 1048576mi
550 * *1 TB*: 1000000000, 1t, 1000m, 1000000k
552 Examples with :option:`kb_base`\=1024 (default):
554 * *4 KiB*: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
555 * *1 MiB*: 1048576, 1m, 1024k
556 * *1 MB*: 1000000, 1mi, 1000ki
557 * *1 TiB*: 1099511627776, 1t, 1024g, 1048576m
558 * *1 TB*: 1000000000, 1ti, 1000mi, 1000000ki
560 To specify times (units are not case sensitive):
564 * *M* -- means minutes
565 * *s* -- or sec means seconds (default)
566 * *ms* -- or *msec* means milliseconds
567 * *us* -- or *usec* means microseconds
569 If the option accepts an upper and lower range, use a colon ':' or
570 minus '-' to separate such values. See :ref:`irange <irange>`.
571 If the lower value specified happens to be larger than the upper value
572 the two values are swapped.
577 Boolean. Usually parsed as an integer, however only defined for
578 true and false (1 and 0).
583 Integer range with suffix. Allows value range to be given, such as
584 1024-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
585 option allows two sets of ranges, they can be specified with a ',' or '/'
586 delimiter: 1k-4k/8k-32k. Also see :ref:`int <int>`.
589 A list of floating point numbers, separated by a ':' character.
591 With the above in mind, here follows the complete list of fio job parameters.
597 .. option:: kb_base=int
599 Select the interpretation of unit prefixes in input parameters.
602 Inputs comply with IEC 80000-13 and the International
603 System of Units (SI). Use:
605 - power-of-2 values with IEC prefixes (e.g., KiB)
606 - power-of-10 values with SI prefixes (e.g., kB)
609 Compatibility mode (default). To avoid breaking old scripts:
611 - power-of-2 values with SI prefixes
612 - power-of-10 values with IEC prefixes
614 See :option:`bs` for more details on input parameters.
616 Outputs always use correct prefixes. Most outputs include both
619 bw=2383.3kB/s (2327.4KiB/s)
621 If only one value is reported, then kb_base selects the one to use:
623 **1000** -- SI prefixes
625 **1024** -- IEC prefixes
627 .. option:: unit_base=int
629 Base unit for reporting. Allowed values are:
632 Use auto-detection (default).
644 ASCII name of the job. This may be used to override the name printed by fio
645 for this job. Otherwise the job name is used. On the command line this
646 parameter has the special purpose of also signaling the start of a new job.
648 .. option:: description=str
650 Text description of the job. Doesn't do anything except dump this text
651 description when this job is run. It's not parsed.
653 .. option:: loops=int
655 Run the specified number of iterations of this job. Used to repeat the same
656 workload a given number of times. Defaults to 1.
658 .. option:: numjobs=int
660 Create the specified number of clones of this job. Each clone of job
661 is spawned as an independent thread or process. May be used to setup a
662 larger number of threads/processes doing the same thing. Each thread is
663 reported separately; to see statistics for all clones as a whole, use
664 :option:`group_reporting` in conjunction with :option:`new_group`.
665 See :option:`--max-jobs`. Default: 1.
668 Time related parameters
669 ~~~~~~~~~~~~~~~~~~~~~~~
671 .. option:: runtime=time
673 Tell fio to terminate processing after the specified period of time. It
674 can be quite hard to determine for how long a specified job will run, so
675 this parameter is handy to cap the total runtime to a given time. When
676 the unit is omitted, the value is interpreted in seconds.
678 .. option:: time_based
680 If set, fio will run for the duration of the :option:`runtime` specified
681 even if the file(s) are completely read or written. It will simply loop over
682 the same workload as many times as the :option:`runtime` allows.
684 .. option:: startdelay=irange(time)
686 Delay the start of job for the specified amount of time. Can be a single
687 value or a range. When given as a range, each thread will choose a value
688 randomly from within the range. Value is in seconds if a unit is omitted.
690 .. option:: ramp_time=time
692 If set, fio will run the specified workload for this amount of time before
693 logging any performance numbers. Useful for letting performance settle
694 before logging results, thus minimizing the runtime required for stable
695 results. Note that the ``ramp_time`` is considered lead in time for a job,
696 thus it will increase the total runtime if a special timeout or
697 :option:`runtime` is specified. When the unit is omitted, the value is
700 .. option:: clocksource=str
702 Use the given clocksource as the base of timing. The supported options are:
705 :manpage:`gettimeofday(2)`
708 :manpage:`clock_gettime(2)`
711 Internal CPU clock source
713 cpu is the preferred clocksource if it is reliable, as it is very fast (and
714 fio is heavy on time calls). Fio will automatically use this clocksource if
715 it's supported and considered reliable on the system it is running on,
716 unless another clocksource is specifically set. For x86/x86-64 CPUs, this
717 means supporting TSC Invariant.
719 .. option:: gtod_reduce=bool
721 Enable all of the :manpage:`gettimeofday(2)` reducing options
722 (:option:`disable_clat`, :option:`disable_slat`, :option:`disable_bw_measurement`) plus
723 reduce precision of the timeout somewhat to really shrink the
724 :manpage:`gettimeofday(2)` call count. With this option enabled, we only do
725 about 0.4% of the :manpage:`gettimeofday(2)` calls we would have done if all
726 time keeping was enabled.
728 .. option:: gtod_cpu=int
730 Sometimes it's cheaper to dedicate a single thread of execution to just
731 getting the current time. Fio (and databases, for instance) are very
732 intensive on :manpage:`gettimeofday(2)` calls. With this option, you can set
733 one CPU aside for doing nothing but logging current time to a shared memory
734 location. Then the other threads/processes that run I/O workloads need only
735 copy that segment, instead of entering the kernel with a
736 :manpage:`gettimeofday(2)` call. The CPU set aside for doing these time
737 calls will be excluded from other uses. Fio will manually clear it from the
738 CPU mask of other jobs.
744 .. option:: directory=str
746 Prefix filenames with this directory. Used to place files in a different
747 location than :file:`./`. You can specify a number of directories by
748 separating the names with a ':' character. These directories will be
749 assigned equally distributed to job clones created by :option:`numjobs` as
750 long as they are using generated filenames. If specific `filename(s)` are
751 set fio will use the first listed directory, and thereby matching the
752 `filename` semantic (which generates a file for each clone if not
753 specified, but lets all clones use the same file if set).
755 See the :option:`filename` option for information on how to escape "``:``" and
756 "``\``" characters within the directory path itself.
758 Note: To control the directory fio will use for internal state files
759 use :option:`--aux-path`.
761 .. option:: filename=str
763 Fio normally makes up a `filename` based on the job name, thread number, and
764 file number (see :option:`filename_format`). If you want to share files
765 between threads in a job or several
766 jobs with fixed file paths, specify a `filename` for each of them to override
767 the default. If the ioengine is file based, you can specify a number of files
768 by separating the names with a ':' colon. So if you wanted a job to open
769 :file:`/dev/sda` and :file:`/dev/sdb` as the two working files, you would use
770 ``filename=/dev/sda:/dev/sdb``. This also means that whenever this option is
771 specified, :option:`nrfiles` is ignored. The size of regular files specified
772 by this option will be :option:`size` divided by number of files unless an
773 explicit size is specified by :option:`filesize`.
775 Each colon and backslash in the wanted path must be escaped with a ``\``
776 character. For instance, if the path is :file:`/dev/dsk/foo@3,0:c` then you
777 would use ``filename=/dev/dsk/foo@3,0\:c`` and if the path is
778 :file:`F:\\filename` then you would use ``filename=F\:\\filename``.
780 On Windows, disk devices are accessed as :file:`\\\\.\\PhysicalDrive0` for
781 the first device, :file:`\\\\.\\PhysicalDrive1` for the second etc.
782 Note: Windows and FreeBSD prevent write access to areas
783 of the disk containing in-use data (e.g. filesystems).
785 The filename "`-`" is a reserved name, meaning *stdin* or *stdout*. Which
786 of the two depends on the read/write direction set.
788 .. option:: filename_format=str
790 If sharing multiple files between jobs, it is usually necessary to have fio
791 generate the exact names that you want. By default, fio will name a file
792 based on the default file format specification of
793 :file:`jobname.jobnumber.filenumber`. With this option, that can be
794 customized. Fio will recognize and replace the following keywords in this
798 The name of the worker thread or process.
800 The incremental number of the worker thread or process.
802 The incremental number of the file for that worker thread or
805 To have dependent jobs share a set of files, this option can be set to have
806 fio generate filenames that are shared between the two. For instance, if
807 :file:`testfiles.$filenum` is specified, file number 4 for any job will be
808 named :file:`testfiles.4`. The default of :file:`$jobname.$jobnum.$filenum`
809 will be used if no other format specifier is given.
811 If you specify a path then the directories will be created up to the
812 main directory for the file. So for example if you specify
813 ``filename_format=a/b/c/$jobnum`` then the directories a/b/c will be
814 created before the file setup part of the job. If you specify
815 :option:`directory` then the path will be relative that directory,
816 otherwise it is treated as the absolute path.
818 .. option:: unique_filename=bool
820 To avoid collisions between networked clients, fio defaults to prefixing any
821 generated filenames (with a directory specified) with the source of the
822 client connecting. To disable this behavior, set this option to 0.
824 .. option:: opendir=str
826 Recursively open any files below directory `str`.
828 .. option:: lockfile=str
830 Fio defaults to not locking any files before it does I/O to them. If a file
831 or file descriptor is shared, fio can serialize I/O to that file to make the
832 end result consistent. This is usual for emulating real workloads that share
833 files. The lock modes are:
836 No locking. The default.
838 Only one thread or process may do I/O at a time, excluding all
841 Read-write locking on the file. Many readers may
842 access the file at the same time, but writes get exclusive access.
844 .. option:: nrfiles=int
846 Number of files to use for this job. Defaults to 1. The size of files
847 will be :option:`size` divided by this unless explicit size is specified by
848 :option:`filesize`. Files are created for each thread separately, and each
849 file will have a file number within its name by default, as explained in
850 :option:`filename` section.
853 .. option:: openfiles=int
855 Number of files to keep open at the same time. Defaults to the same as
856 :option:`nrfiles`, can be set smaller to limit the number simultaneous
859 .. option:: file_service_type=str
861 Defines how fio decides which file from a job to service next. The following
865 Choose a file at random.
868 Round robin over opened files. This is the default.
871 Finish one file before moving on to the next. Multiple files can
872 still be open depending on :option:`openfiles`.
875 Use a *Zipf* distribution to decide what file to access.
878 Use a *Pareto* distribution to decide what file to access.
881 Use a *Gaussian* (normal) distribution to decide what file to
887 For *random*, *roundrobin*, and *sequential*, a postfix can be appended to
888 tell fio how many I/Os to issue before switching to a new file. For example,
889 specifying ``file_service_type=random:8`` would cause fio to issue
890 8 I/Os before selecting a new file at random. For the non-uniform
891 distributions, a floating point postfix can be given to influence how the
892 distribution is skewed. See :option:`random_distribution` for a description
893 of how that would work.
895 .. option:: ioscheduler=str
897 Attempt to switch the device hosting the file to the specified I/O scheduler
900 .. option:: create_serialize=bool
902 If true, serialize the file creation for the jobs. This may be handy to
903 avoid interleaving of data files, which may greatly depend on the filesystem
904 used and even the number of processors in the system. Default: true.
906 .. option:: create_fsync=bool
908 :manpage:`fsync(2)` the data file after creation. This is the default.
910 .. option:: create_on_open=bool
912 If true, don't pre-create files but allow the job's open() to create a file
913 when it's time to do I/O. Default: false -- pre-create all necessary files
916 .. option:: create_only=bool
918 If true, fio will only run the setup phase of the job. If files need to be
919 laid out or updated on disk, only that will be done -- the actual job contents
920 are not executed. Default: false.
922 .. option:: allow_file_create=bool
924 If true, fio is permitted to create files as part of its workload. If this
925 option is false, then fio will error out if
926 the files it needs to use don't already exist. Default: true.
928 .. option:: allow_mounted_write=bool
930 If this isn't set, fio will abort jobs that are destructive (e.g. that write)
931 to what appears to be a mounted device or partition. This should help catch
932 creating inadvertently destructive tests, not realizing that the test will
933 destroy data on the mounted file system. Note that some platforms don't allow
934 writing against a mounted device regardless of this option. Default: false.
936 .. option:: pre_read=bool
938 If this is given, files will be pre-read into memory before starting the
939 given I/O operation. This will also clear the :option:`invalidate` flag,
940 since it is pointless to pre-read and then drop the cache. This will only
941 work for I/O engines that are seek-able, since they allow you to read the
942 same data multiple times. Thus it will not work on non-seekable I/O engines
943 (e.g. network, splice). Default: false.
945 .. option:: unlink=bool
947 Unlink the job files when done. Not the default, as repeated runs of that
948 job would then waste time recreating the file set again and again. Default:
951 .. option:: unlink_each_loop=bool
953 Unlink job files after each iteration or loop. Default: false.
955 .. option:: zonerange=int
957 Size of a single zone in which I/O occurs. See also :option:`zonesize`
958 and :option:`zoneskip`.
960 .. option:: zonesize=int
962 Number of bytes to transfer before skipping :option:`zoneskip`
963 bytes. If this parameter is smaller than :option:`zonerange` then only
964 a fraction of each zone with :option:`zonerange` bytes will be
965 accessed. If this parameter is larger than :option:`zonerange` then
966 each zone will be accessed multiple times before skipping
968 .. option:: zoneskip=int
970 Skip the specified number of bytes when :option:`zonesize` data have
971 been transferred. The three zone options can be used to do strided I/O
978 .. option:: direct=bool
980 If value is true, use non-buffered I/O. This is usually O_DIRECT. Note that
981 OpenBSD and ZFS on Solaris don't support direct I/O. On Windows the synchronous
982 ioengines don't support direct I/O. Default: false.
984 .. option:: atomic=bool
986 If value is true, attempt to use atomic direct I/O. Atomic writes are
987 guaranteed to be stable once acknowledged by the operating system. Only
988 Linux supports O_ATOMIC right now.
990 .. option:: buffered=bool
992 If value is true, use buffered I/O. This is the opposite of the
993 :option:`direct` option. Defaults to true.
995 .. option:: readwrite=str, rw=str
997 Type of I/O pattern. Accepted values are:
1004 Sequential trims (Linux block devices and SCSI
1005 character devices only).
1011 Random trims (Linux block devices and SCSI
1012 character devices only).
1014 Sequential mixed reads and writes.
1016 Random mixed reads and writes.
1018 Sequential trim+write sequences. Blocks will be trimmed first,
1019 then the same blocks will be written to.
1021 Fio defaults to read if the option is not specified. For the mixed I/O
1022 types, the default is to split them 50/50. For certain types of I/O the
1023 result may still be skewed a bit, since the speed may be different.
1025 It is possible to specify the number of I/Os to do before getting a new
1026 offset by appending ``:<nr>`` to the end of the string given. For a
1027 random read, it would look like ``rw=randread:8`` for passing in an offset
1028 modifier with a value of 8. If the suffix is used with a sequential I/O
1029 pattern, then the *<nr>* value specified will be **added** to the generated
1030 offset for each I/O turning sequential I/O into sequential I/O with holes.
1031 For instance, using ``rw=write:4k`` will skip 4k for every write. Also see
1032 the :option:`rw_sequencer` option.
1034 .. option:: rw_sequencer=str
1036 If an offset modifier is given by appending a number to the ``rw=<str>``
1037 line, then this option controls how that number modifies the I/O offset
1038 being generated. Accepted values are:
1041 Generate sequential offset.
1043 Generate the same offset.
1045 ``sequential`` is only useful for random I/O, where fio would normally
1046 generate a new random offset for every I/O. If you append e.g. 8 to randread,
1047 you would get a new random offset for every 8 I/Os. The result would be a
1048 seek for only every 8 I/Os, instead of for every I/O. Use ``rw=randread:8``
1049 to specify that. As sequential I/O is already sequential, setting
1050 ``sequential`` for that would not result in any differences. ``identical``
1051 behaves in a similar fashion, except it sends the same offset 8 number of
1052 times before generating a new offset.
1054 .. option:: unified_rw_reporting=bool
1056 Fio normally reports statistics on a per data direction basis, meaning that
1057 reads, writes, and trims are accounted and reported separately. If this
1058 option is set fio sums the results and report them as "mixed" instead.
1060 .. option:: randrepeat=bool
1062 Seed the random number generator used for random I/O patterns in a
1063 predictable way so the pattern is repeatable across runs. Default: true.
1065 .. option:: allrandrepeat=bool
1067 Seed all random number generators in a predictable way so results are
1068 repeatable across runs. Default: false.
1070 .. option:: randseed=int
1072 Seed the random number generators based on this seed value, to be able to
1073 control what sequence of output is being generated. If not set, the random
1074 sequence depends on the :option:`randrepeat` setting.
1076 .. option:: fallocate=str
1078 Whether pre-allocation is performed when laying down files.
1079 Accepted values are:
1082 Do not pre-allocate space.
1085 Use a platform's native pre-allocation call but fall back to
1086 **none** behavior if it fails/is not implemented.
1089 Pre-allocate via :manpage:`posix_fallocate(3)`.
1092 Pre-allocate via :manpage:`fallocate(2)` with
1093 FALLOC_FL_KEEP_SIZE set.
1096 Backward-compatible alias for **none**.
1099 Backward-compatible alias for **posix**.
1101 May not be available on all supported platforms. **keep** is only available
1102 on Linux. If using ZFS on Solaris this cannot be set to **posix**
1103 because ZFS doesn't support pre-allocation. Default: **native** if any
1104 pre-allocation methods are available, **none** if not.
1106 .. option:: fadvise_hint=str
1108 Use :manpage:`posix_fadvise(2)` or :manpage:`posix_fadvise(2)` to
1109 advise the kernel on what I/O patterns are likely to be issued.
1110 Accepted values are:
1113 Backwards-compatible hint for "no hint".
1116 Backwards compatible hint for "advise with fio workload type". This
1117 uses **FADV_RANDOM** for a random workload, and **FADV_SEQUENTIAL**
1118 for a sequential workload.
1121 Advise using **FADV_SEQUENTIAL**.
1124 Advise using **FADV_RANDOM**.
1126 .. option:: write_hint=str
1128 Use :manpage:`fcntl(2)` to advise the kernel what life time to expect
1129 from a write. Only supported on Linux, as of version 4.13. Accepted
1133 No particular life time associated with this file.
1136 Data written to this file has a short life time.
1139 Data written to this file has a medium life time.
1142 Data written to this file has a long life time.
1145 Data written to this file has a very long life time.
1147 The values are all relative to each other, and no absolute meaning
1148 should be associated with them.
1150 .. option:: offset=int
1152 Start I/O at the provided offset in the file, given as either a fixed size in
1153 bytes or a percentage. If a percentage is given, the generated offset will be
1154 aligned to the minimum ``blocksize`` or to the value of ``offset_align`` if
1155 provided. Data before the given offset will not be touched. This
1156 effectively caps the file size at `real_size - offset`. Can be combined with
1157 :option:`size` to constrain the start and end range of the I/O workload.
1158 A percentage can be specified by a number between 1 and 100 followed by '%',
1159 for example, ``offset=20%`` to specify 20%.
1161 .. option:: offset_align=int
1163 If set to non-zero value, the byte offset generated by a percentage ``offset``
1164 is aligned upwards to this value. Defaults to 0 meaning that a percentage
1165 offset is aligned to the minimum block size.
1167 .. option:: offset_increment=int
1169 If this is provided, then the real offset becomes `offset + offset_increment
1170 * thread_number`, where the thread number is a counter that starts at 0 and
1171 is incremented for each sub-job (i.e. when :option:`numjobs` option is
1172 specified). This option is useful if there are several jobs which are
1173 intended to operate on a file in parallel disjoint segments, with even
1174 spacing between the starting points.
1176 .. option:: number_ios=int
1178 Fio will normally perform I/Os until it has exhausted the size of the region
1179 set by :option:`size`, or if it exhaust the allocated time (or hits an error
1180 condition). With this setting, the range/size can be set independently of
1181 the number of I/Os to perform. When fio reaches this number, it will exit
1182 normally and report status. Note that this does not extend the amount of I/O
1183 that will be done, it will only stop fio if this condition is met before
1184 other end-of-job criteria.
1186 .. option:: fsync=int
1188 If writing to a file, issue an :manpage:`fsync(2)` (or its equivalent) of
1189 the dirty data for every number of blocks given. For example, if you give 32
1190 as a parameter, fio will sync the file after every 32 writes issued. If fio is
1191 using non-buffered I/O, we may not sync the file. The exception is the sg
1192 I/O engine, which synchronizes the disk cache anyway. Defaults to 0, which
1193 means fio does not periodically issue and wait for a sync to complete. Also
1194 see :option:`end_fsync` and :option:`fsync_on_close`.
1196 .. option:: fdatasync=int
1198 Like :option:`fsync` but uses :manpage:`fdatasync(2)` to only sync data and
1199 not metadata blocks. In Windows, FreeBSD, and DragonFlyBSD there is no
1200 :manpage:`fdatasync(2)` so this falls back to using :manpage:`fsync(2)`.
1201 Defaults to 0, which means fio does not periodically issue and wait for a
1202 data-only sync to complete.
1204 .. option:: write_barrier=int
1206 Make every `N-th` write a barrier write.
1208 .. option:: sync_file_range=str:int
1210 Use :manpage:`sync_file_range(2)` for every `int` number of write
1211 operations. Fio will track range of writes that have happened since the last
1212 :manpage:`sync_file_range(2)` call. `str` can currently be one or more of:
1215 SYNC_FILE_RANGE_WAIT_BEFORE
1217 SYNC_FILE_RANGE_WRITE
1219 SYNC_FILE_RANGE_WAIT_AFTER
1221 So if you do ``sync_file_range=wait_before,write:8``, fio would use
1222 ``SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE`` for every 8
1223 writes. Also see the :manpage:`sync_file_range(2)` man page. This option is
1226 .. option:: overwrite=bool
1228 If true, writes to a file will always overwrite existing data. If the file
1229 doesn't already exist, it will be created before the write phase begins. If
1230 the file exists and is large enough for the specified write phase, nothing
1231 will be done. Default: false.
1233 .. option:: end_fsync=bool
1235 If true, :manpage:`fsync(2)` file contents when a write stage has completed.
1238 .. option:: fsync_on_close=bool
1240 If true, fio will :manpage:`fsync(2)` a dirty file on close. This differs
1241 from :option:`end_fsync` in that it will happen on every file close, not
1242 just at the end of the job. Default: false.
1244 .. option:: rwmixread=int
1246 Percentage of a mixed workload that should be reads. Default: 50.
1248 .. option:: rwmixwrite=int
1250 Percentage of a mixed workload that should be writes. If both
1251 :option:`rwmixread` and :option:`rwmixwrite` is given and the values do not
1252 add up to 100%, the latter of the two will be used to override the
1253 first. This may interfere with a given rate setting, if fio is asked to
1254 limit reads or writes to a certain rate. If that is the case, then the
1255 distribution may be skewed. Default: 50.
1257 .. option:: random_distribution=str:float[,str:float][,str:float]
1259 By default, fio will use a completely uniform random distribution when asked
1260 to perform random I/O. Sometimes it is useful to skew the distribution in
1261 specific ways, ensuring that some parts of the data is more hot than others.
1262 fio includes the following distribution models:
1265 Uniform random distribution
1274 Normal (Gaussian) distribution
1277 Zoned random distribution
1280 Zone absolute random distribution
1282 When using a **zipf** or **pareto** distribution, an input value is also
1283 needed to define the access pattern. For **zipf**, this is the `Zipf
1284 theta`. For **pareto**, it's the `Pareto power`. Fio includes a test
1285 program, :command:`fio-genzipf`, that can be used visualize what the given input
1286 values will yield in terms of hit rates. If you wanted to use **zipf** with
1287 a `theta` of 1.2, you would use ``random_distribution=zipf:1.2`` as the
1288 option. If a non-uniform model is used, fio will disable use of the random
1289 map. For the **normal** distribution, a normal (Gaussian) deviation is
1290 supplied as a value between 0 and 100.
1292 For a **zoned** distribution, fio supports specifying percentages of I/O
1293 access that should fall within what range of the file or device. For
1294 example, given a criteria of:
1296 * 60% of accesses should be to the first 10%
1297 * 30% of accesses should be to the next 20%
1298 * 8% of accesses should be to the next 30%
1299 * 2% of accesses should be to the next 40%
1301 we can define that through zoning of the random accesses. For the above
1302 example, the user would do::
1304 random_distribution=zoned:60/10:30/20:8/30:2/40
1306 A **zoned_abs** distribution works exactly like the **zoned**, except
1307 that it takes absolute sizes. For example, let's say you wanted to
1308 define access according to the following criteria:
1310 * 60% of accesses should be to the first 20G
1311 * 30% of accesses should be to the next 100G
1312 * 10% of accesses should be to the next 500G
1314 we can define an absolute zoning distribution with:
1316 random_distribution=zoned_abs=60/20G:30/100G:10/500g
1318 For both **zoned** and **zoned_abs**, fio supports defining up to
1321 Similarly to how :option:`bssplit` works for setting ranges and
1322 percentages of block sizes. Like :option:`bssplit`, it's possible to
1323 specify separate zones for reads, writes, and trims. If just one set
1324 is given, it'll apply to all of them. This goes for both **zoned**
1325 **zoned_abs** distributions.
1327 .. option:: percentage_random=int[,int][,int]
1329 For a random workload, set how big a percentage should be random. This
1330 defaults to 100%, in which case the workload is fully random. It can be set
1331 from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1332 sequential. Any setting in between will result in a random mix of sequential
1333 and random I/O, at the given percentages. Comma-separated values may be
1334 specified for reads, writes, and trims as described in :option:`blocksize`.
1336 .. option:: norandommap
1338 Normally fio will cover every block of the file when doing random I/O. If
1339 this option is given, fio will just get a new random offset without looking
1340 at past I/O history. This means that some blocks may not be read or written,
1341 and that some blocks may be read/written more than once. If this option is
1342 used with :option:`verify` and multiple blocksizes (via :option:`bsrange`),
1343 only intact blocks are verified, i.e., partially-overwritten blocks are
1344 ignored. With an async I/O engine and an I/O depth > 1, it is possible for
1345 the same block to be overwritten, which can cause verification errors. Either
1346 do not use norandommap in this case, or also use the lfsr random generator.
1348 .. option:: softrandommap=bool
1350 See :option:`norandommap`. If fio runs with the random block map enabled and
1351 it fails to allocate the map, if this option is set it will continue without
1352 a random block map. As coverage will not be as complete as with random maps,
1353 this option is disabled by default.
1355 .. option:: random_generator=str
1357 Fio supports the following engines for generating I/O offsets for random I/O:
1360 Strong 2^88 cycle random number generator.
1362 Linear feedback shift register generator.
1364 Strong 64-bit 2^258 cycle random number generator.
1366 **tausworthe** is a strong random number generator, but it requires tracking
1367 on the side if we want to ensure that blocks are only read or written
1368 once. **lfsr** guarantees that we never generate the same offset twice, and
1369 it's also less computationally expensive. It's not a true random generator,
1370 however, though for I/O purposes it's typically good enough. **lfsr** only
1371 works with single block sizes, not with workloads that use multiple block
1372 sizes. If used with such a workload, fio may read or write some blocks
1373 multiple times. The default value is **tausworthe**, unless the required
1374 space exceeds 2^32 blocks. If it does, then **tausworthe64** is
1375 selected automatically.
1381 .. option:: blocksize=int[,int][,int], bs=int[,int][,int]
1383 The block size in bytes used for I/O units. Default: 4096. A single value
1384 applies to reads, writes, and trims. Comma-separated values may be
1385 specified for reads, writes, and trims. A value not terminated in a comma
1386 applies to subsequent types.
1391 means 256k for reads, writes and trims.
1394 means 8k for reads, 32k for writes and trims.
1397 means 8k for reads, 32k for writes, and default for trims.
1400 means default for reads, 8k for writes and trims.
1403 means default for reads, 8k for writes, and default for trims.
1405 .. option:: blocksize_range=irange[,irange][,irange], bsrange=irange[,irange][,irange]
1407 A range of block sizes in bytes for I/O units. The issued I/O unit will
1408 always be a multiple of the minimum size, unless
1409 :option:`blocksize_unaligned` is set.
1411 Comma-separated ranges may be specified for reads, writes, and trims as
1412 described in :option:`blocksize`.
1414 Example: ``bsrange=1k-4k,2k-8k``.
1416 .. option:: bssplit=str[,str][,str]
1418 Sometimes you want even finer grained control of the block sizes
1419 issued, not just an even split between them. This option allows you to
1420 weight various block sizes, so that you are able to define a specific
1421 amount of block sizes issued. The format for this option is::
1423 bssplit=blocksize/percentage:blocksize/percentage
1425 for as many block sizes as needed. So if you want to define a workload
1426 that has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would
1429 bssplit=4k/10:64k/50:32k/40
1431 Ordering does not matter. If the percentage is left blank, fio will
1432 fill in the remaining values evenly. So a bssplit option like this one::
1434 bssplit=4k/50:1k/:32k/
1436 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always
1437 add up to 100, if bssplit is given a range that adds up to more, it
1440 Comma-separated values may be specified for reads, writes, and trims as
1441 described in :option:`blocksize`.
1443 If you want a workload that has 50% 2k reads and 50% 4k reads, while
1444 having 90% 4k writes and 10% 8k writes, you would specify::
1446 bssplit=2k/50:4k/50,4k/90:8k/10
1448 Fio supports defining up to 64 different weights for each data
1451 .. option:: blocksize_unaligned, bs_unaligned
1453 If set, fio will issue I/O units with any size within
1454 :option:`blocksize_range`, not just multiples of the minimum size. This
1455 typically won't work with direct I/O, as that normally requires sector
1458 .. option:: bs_is_seq_rand=bool
1460 If this option is set, fio will use the normal read,write blocksize settings
1461 as sequential,random blocksize settings instead. Any random read or write
1462 will use the WRITE blocksize settings, and any sequential read or write will
1463 use the READ blocksize settings.
1465 .. option:: blockalign=int[,int][,int], ba=int[,int][,int]
1467 Boundary to which fio will align random I/O units. Default:
1468 :option:`blocksize`. Minimum alignment is typically 512b for using direct
1469 I/O, though it usually depends on the hardware block size. This option is
1470 mutually exclusive with using a random map for files, so it will turn off
1471 that option. Comma-separated values may be specified for reads, writes, and
1472 trims as described in :option:`blocksize`.
1478 .. option:: zero_buffers
1480 Initialize buffers with all zeros. Default: fill buffers with random data.
1482 .. option:: refill_buffers
1484 If this option is given, fio will refill the I/O buffers on every
1485 submit. Only makes sense if :option:`zero_buffers` isn't specified,
1486 naturally. Defaults to being unset i.e., the buffer is only filled at
1487 init time and the data in it is reused when possible but if any of
1488 :option:`verify`, :option:`buffer_compress_percentage` or
1489 :option:`dedupe_percentage` are enabled then `refill_buffers` is also
1490 automatically enabled.
1492 .. option:: scramble_buffers=bool
1494 If :option:`refill_buffers` is too costly and the target is using data
1495 deduplication, then setting this option will slightly modify the I/O buffer
1496 contents to defeat normal de-dupe attempts. This is not enough to defeat
1497 more clever block compression attempts, but it will stop naive dedupe of
1498 blocks. Default: true.
1500 .. option:: buffer_compress_percentage=int
1502 If this is set, then fio will attempt to provide I/O buffer content
1503 (on WRITEs) that compresses to the specified level. Fio does this by
1504 providing a mix of random data followed by fixed pattern data. The
1505 fixed pattern is either zeros, or the pattern specified by
1506 :option:`buffer_pattern`. If the `buffer_pattern` option is used, it
1507 might skew the compression ratio slightly. Setting
1508 `buffer_compress_percentage` to a value other than 100 will also
1509 enable :option:`refill_buffers` in order to reduce the likelihood that
1510 adjacent blocks are so similar that they over compress when seen
1511 together. See :option:`buffer_compress_chunk` for how to set a finer or
1512 coarser granularity for the random/fixed data region. Defaults to unset
1513 i.e., buffer data will not adhere to any compression level.
1515 .. option:: buffer_compress_chunk=int
1517 This setting allows fio to manage how big the random/fixed data region
1518 is when using :option:`buffer_compress_percentage`. When
1519 `buffer_compress_chunk` is set to some non-zero value smaller than the
1520 block size, fio can repeat the random/fixed region throughout the I/O
1521 buffer at the specified interval (which particularly useful when
1522 bigger block sizes are used for a job). When set to 0, fio will use a
1523 chunk size that matches the block size resulting in a single
1524 random/fixed region within the I/O buffer. Defaults to 512. When the
1525 unit is omitted, the value is interpreted in bytes.
1527 .. option:: buffer_pattern=str
1529 If set, fio will fill the I/O buffers with this pattern or with the contents
1530 of a file. If not set, the contents of I/O buffers are defined by the other
1531 options related to buffer contents. The setting can be any pattern of bytes,
1532 and can be prefixed with 0x for hex values. It may also be a string, where
1533 the string must then be wrapped with ``""``. Or it may also be a filename,
1534 where the filename must be wrapped with ``''`` in which case the file is
1535 opened and read. Note that not all the file contents will be read if that
1536 would cause the buffers to overflow. So, for example::
1538 buffer_pattern='filename'
1542 buffer_pattern="abcd"
1550 buffer_pattern=0xdeadface
1552 Also you can combine everything together in any order::
1554 buffer_pattern=0xdeadface"abcd"-12'filename'
1556 .. option:: dedupe_percentage=int
1558 If set, fio will generate this percentage of identical buffers when
1559 writing. These buffers will be naturally dedupable. The contents of the
1560 buffers depend on what other buffer compression settings have been set. It's
1561 possible to have the individual buffers either fully compressible, or not at
1562 all -- this option only controls the distribution of unique buffers. Setting
1563 this option will also enable :option:`refill_buffers` to prevent every buffer
1566 .. option:: invalidate=bool
1568 Invalidate the buffer/page cache parts of the files to be used prior to
1569 starting I/O if the platform and file type support it. Defaults to true.
1570 This will be ignored if :option:`pre_read` is also specified for the
1573 .. option:: sync=bool
1575 Use synchronous I/O for buffered writes. For the majority of I/O engines,
1576 this means using O_SYNC. Default: false.
1578 .. option:: iomem=str, mem=str
1580 Fio can use various types of memory as the I/O unit buffer. The allowed
1584 Use memory from :manpage:`malloc(3)` as the buffers. Default memory
1588 Use shared memory as the buffers. Allocated through
1589 :manpage:`shmget(2)`.
1592 Same as shm, but use huge pages as backing.
1595 Use :manpage:`mmap(2)` to allocate buffers. May either be anonymous memory, or can
1596 be file backed if a filename is given after the option. The format
1597 is `mem=mmap:/path/to/file`.
1600 Use a memory mapped huge file as the buffer backing. Append filename
1601 after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file`.
1604 Same as mmap, but use a MMAP_SHARED mapping.
1607 Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1608 The :option:`ioengine` must be `rdma`.
1610 The area allocated is a function of the maximum allowed bs size for the job,
1611 multiplied by the I/O depth given. Note that for **shmhuge** and
1612 **mmaphuge** to work, the system must have free huge pages allocated. This
1613 can normally be checked and set by reading/writing
1614 :file:`/proc/sys/vm/nr_hugepages` on a Linux system. Fio assumes a huge page
1615 is 4MiB in size. So to calculate the number of huge pages you need for a
1616 given job file, add up the I/O depth of all jobs (normally one unless
1617 :option:`iodepth` is used) and multiply by the maximum bs set. Then divide
1618 that number by the huge page size. You can see the size of the huge pages in
1619 :file:`/proc/meminfo`. If no huge pages are allocated by having a non-zero
1620 number in `nr_hugepages`, using **mmaphuge** or **shmhuge** will fail. Also
1621 see :option:`hugepage-size`.
1623 **mmaphuge** also needs to have hugetlbfs mounted and the file location
1624 should point there. So if it's mounted in :file:`/huge`, you would use
1625 `mem=mmaphuge:/huge/somefile`.
1627 .. option:: iomem_align=int, mem_align=int
1629 This indicates the memory alignment of the I/O memory buffers. Note that
1630 the given alignment is applied to the first I/O unit buffer, if using
1631 :option:`iodepth` the alignment of the following buffers are given by the
1632 :option:`bs` used. In other words, if using a :option:`bs` that is a
1633 multiple of the page sized in the system, all buffers will be aligned to
1634 this value. If using a :option:`bs` that is not page aligned, the alignment
1635 of subsequent I/O memory buffers is the sum of the :option:`iomem_align` and
1638 .. option:: hugepage-size=int
1640 Defines the size of a huge page. Must at least be equal to the system
1641 setting, see :file:`/proc/meminfo`. Defaults to 4MiB. Should probably
1642 always be a multiple of megabytes, so using ``hugepage-size=Xm`` is the
1643 preferred way to set this to avoid setting a non-pow-2 bad value.
1645 .. option:: lockmem=int
1647 Pin the specified amount of memory with :manpage:`mlock(2)`. Can be used to
1648 simulate a smaller amount of memory. The amount specified is per worker.
1654 .. option:: size=int
1656 The total size of file I/O for each thread of this job. Fio will run until
1657 this many bytes has been transferred, unless runtime is limited by other options
1658 (such as :option:`runtime`, for instance, or increased/decreased by :option:`io_size`).
1659 Fio will divide this size between the available files determined by options
1660 such as :option:`nrfiles`, :option:`filename`, unless :option:`filesize` is
1661 specified by the job. If the result of division happens to be 0, the size is
1662 set to the physical size of the given files or devices if they exist.
1663 If this option is not specified, fio will use the full size of the given
1664 files or devices. If the files do not exist, size must be given. It is also
1665 possible to give size as a percentage between 1 and 100. If ``size=20%`` is
1666 given, fio will use 20% of the full size of the given files or devices.
1667 Can be combined with :option:`offset` to constrain the start and end range
1668 that I/O will be done within.
1670 .. option:: io_size=int, io_limit=int
1672 Normally fio operates within the region set by :option:`size`, which means
1673 that the :option:`size` option sets both the region and size of I/O to be
1674 performed. Sometimes that is not what you want. With this option, it is
1675 possible to define just the amount of I/O that fio should do. For instance,
1676 if :option:`size` is set to 20GiB and :option:`io_size` is set to 5GiB, fio
1677 will perform I/O within the first 20GiB but exit when 5GiB have been
1678 done. The opposite is also possible -- if :option:`size` is set to 20GiB,
1679 and :option:`io_size` is set to 40GiB, then fio will do 40GiB of I/O within
1680 the 0..20GiB region.
1682 .. option:: filesize=irange(int)
1684 Individual file sizes. May be a range, in which case fio will select sizes
1685 for files at random within the given range and limited to :option:`size` in
1686 total (if that is given). If not given, each created file is the same size.
1687 This option overrides :option:`size` in terms of file size, which means
1688 this value is used as a fixed size or possible range of each file.
1690 .. option:: file_append=bool
1692 Perform I/O after the end of the file. Normally fio will operate within the
1693 size of a file. If this option is set, then fio will append to the file
1694 instead. This has identical behavior to setting :option:`offset` to the size
1695 of a file. This option is ignored on non-regular files.
1697 .. option:: fill_device=bool, fill_fs=bool
1699 Sets size to something really large and waits for ENOSPC (no space left on
1700 device) as the terminating condition. Only makes sense with sequential
1701 write. For a read workload, the mount point will be filled first then I/O
1702 started on the result. This option doesn't make sense if operating on a raw
1703 device node, since the size of that is already known by the file system.
1704 Additionally, writing beyond end-of-device will not return ENOSPC there.
1710 .. option:: ioengine=str
1712 Defines how the job issues I/O to the file. The following types are defined:
1715 Basic :manpage:`read(2)` or :manpage:`write(2)`
1716 I/O. :manpage:`lseek(2)` is used to position the I/O location.
1717 See :option:`fsync` and :option:`fdatasync` for syncing write I/Os.
1720 Basic :manpage:`pread(2)` or :manpage:`pwrite(2)` I/O. Default on
1721 all supported operating systems except for Windows.
1724 Basic :manpage:`readv(2)` or :manpage:`writev(2)` I/O. Will emulate
1725 queuing by coalescing adjacent I/Os into a single submission.
1728 Basic :manpage:`preadv(2)` or :manpage:`pwritev(2)` I/O.
1731 Basic :manpage:`preadv2(2)` or :manpage:`pwritev2(2)` I/O.
1734 Linux native asynchronous I/O. Note that Linux may only support
1735 queued behavior with non-buffered I/O (set ``direct=1`` or
1737 This engine defines engine specific options.
1740 POSIX asynchronous I/O using :manpage:`aio_read(3)` and
1741 :manpage:`aio_write(3)`.
1744 Solaris native asynchronous I/O.
1747 Windows native asynchronous I/O. Default on Windows.
1750 File is memory mapped with :manpage:`mmap(2)` and data copied
1751 to/from using :manpage:`memcpy(3)`.
1754 :manpage:`splice(2)` is used to transfer the data and
1755 :manpage:`vmsplice(2)` to transfer data from user space to the
1759 SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1760 ioctl, or if the target is an sg character device we use
1761 :manpage:`read(2)` and :manpage:`write(2)` for asynchronous
1762 I/O. Requires :option:`filename` option to specify either block or
1763 character devices. This engine supports trim operations.
1764 The sg engine includes engine specific options.
1767 Doesn't transfer any data, just pretends to. This is mainly used to
1768 exercise fio itself and for debugging/testing purposes.
1771 Transfer over the network to given ``host:port``. Depending on the
1772 :option:`protocol` used, the :option:`hostname`, :option:`port`,
1773 :option:`listen` and :option:`filename` options are used to specify
1774 what sort of connection to make, while the :option:`protocol` option
1775 determines which protocol will be used. This engine defines engine
1779 Like **net**, but uses :manpage:`splice(2)` and
1780 :manpage:`vmsplice(2)` to map data and send/receive.
1781 This engine defines engine specific options.
1784 Doesn't transfer any data, but burns CPU cycles according to the
1785 :option:`cpuload` and :option:`cpuchunks` options. Setting
1786 :option:`cpuload`\=85 will cause that job to do nothing but burn 85%
1787 of the CPU. In case of SMP machines, use :option:`numjobs`\=<nr_of_cpu>
1788 to get desired CPU usage, as the cpuload only loads a
1789 single CPU at the desired rate. A job never finishes unless there is
1790 at least one non-cpuio job.
1793 The GUASI I/O engine is the Generic Userspace Asynchronous Syscall
1794 Interface approach to async I/O. See
1796 http://www.xmailserver.org/guasi-lib.html
1798 for more info on GUASI.
1801 The RDMA I/O engine supports both RDMA memory semantics
1802 (RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
1803 InfiniBand, RoCE and iWARP protocols. This engine defines engine
1807 I/O engine that does regular fallocate to simulate data transfer as
1811 does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1814 does fallocate(,mode = 0).
1817 does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1820 I/O engine that sends :manpage:`ftruncate(2)` operations in response
1821 to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
1822 size to the current block offset. :option:`blocksize` is ignored.
1825 I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1826 defragment activity in request to DDIR_WRITE event.
1829 I/O engine supporting direct access to Ceph Reliable Autonomic
1830 Distributed Object Store (RADOS) via librados. This ioengine
1831 defines engine specific options.
1834 I/O engine supporting direct access to Ceph Rados Block Devices
1835 (RBD) via librbd without the need to use the kernel rbd driver. This
1836 ioengine defines engine specific options.
1839 Using GlusterFS libgfapi sync interface to direct access to
1840 GlusterFS volumes without having to go through FUSE. This ioengine
1841 defines engine specific options.
1844 Using GlusterFS libgfapi async interface to direct access to
1845 GlusterFS volumes without having to go through FUSE. This ioengine
1846 defines engine specific options.
1849 Read and write through Hadoop (HDFS). The :option:`filename` option
1850 is used to specify host,port of the hdfs name-node to connect. This
1851 engine interprets offsets a little differently. In HDFS, files once
1852 created cannot be modified so random writes are not possible. To
1853 imitate this the libhdfs engine expects a bunch of small files to be
1854 created over HDFS and will randomly pick a file from them
1855 based on the offset generated by fio backend (see the example
1856 job file to create such files, use ``rw=write`` option). Please
1857 note, it may be necessary to set environment variables to work
1858 with HDFS/libhdfs properly. Each job uses its own connection to
1862 Read, write and erase an MTD character device (e.g.,
1863 :file:`/dev/mtd0`). Discards are treated as erases. Depending on the
1864 underlying device type, the I/O may have to go in a certain pattern,
1865 e.g., on NAND, writing sequentially to erase blocks and discarding
1866 before overwriting. The `trimwrite` mode works well for this
1870 Read and write using filesystem DAX to a file on a filesystem
1871 mounted with DAX on a persistent memory device through the PMDK
1875 Read and write using device DAX to a persistent memory device (e.g.,
1876 /dev/dax0.0) through the PMDK libpmem library.
1879 Prefix to specify loading an external I/O engine object file. Append
1880 the engine filename, e.g. ``ioengine=external:/tmp/foo.o`` to load
1881 ioengine :file:`foo.o` in :file:`/tmp`. The path can be either
1882 absolute or relative. See :file:`engines/skeleton_external.c` for
1883 details of writing an external I/O engine.
1886 Simply create the files and do no I/O to them. You still need to
1887 set `filesize` so that all the accounting still occurs, but no
1888 actual I/O will be done other than creating the file.
1891 Read and write using mmap I/O to a file on a filesystem
1892 mounted with DAX on a persistent memory device through the PMDK
1895 I/O engine specific parameters
1896 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1898 In addition, there are some parameters which are only valid when a specific
1899 :option:`ioengine` is in use. These are used identically to normal parameters,
1900 with the caveat that when used on the command line, they must come after the
1901 :option:`ioengine` that defines them is selected.
1903 .. option:: userspace_reap : [libaio]
1905 Normally, with the libaio engine in use, fio will use the
1906 :manpage:`io_getevents(2)` system call to reap newly returned events. With
1907 this flag turned on, the AIO ring will be read directly from user-space to
1908 reap events. The reaping mode is only enabled when polling for a minimum of
1909 0 events (e.g. when :option:`iodepth_batch_complete` `=0`).
1911 .. option:: hipri : [pvsync2]
1913 Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
1916 .. option:: hipri_percentage : [pvsync2]
1918 When hipri is set this determines the probability of a pvsync2 I/O being high
1919 priority. The default is 100%.
1921 .. option:: cpuload=int : [cpuio]
1923 Attempt to use the specified percentage of CPU cycles. This is a mandatory
1924 option when using cpuio I/O engine.
1926 .. option:: cpuchunks=int : [cpuio]
1928 Split the load into cycles of the given time. In microseconds.
1930 .. option:: exit_on_io_done=bool : [cpuio]
1932 Detect when I/O threads are done, then exit.
1934 .. option:: namenode=str : [libhdfs]
1936 The hostname or IP address of a HDFS cluster namenode to contact.
1938 .. option:: port=int
1942 The listening port of the HFDS cluster namenode.
1946 The TCP or UDP port to bind to or connect to. If this is used with
1947 :option:`numjobs` to spawn multiple instances of the same job type, then
1948 this will be the starting port number since fio will use a range of
1953 The port to use for RDMA-CM communication. This should be the same value
1954 on the client and the server side.
1956 .. option:: hostname=str : [netsplice] [net] [rdma]
1958 The hostname or IP address to use for TCP, UDP or RDMA-CM based I/O. If the job
1959 is a TCP listener or UDP reader, the hostname is not used and must be omitted
1960 unless it is a valid UDP multicast address.
1962 .. option:: interface=str : [netsplice] [net]
1964 The IP address of the network interface used to send or receive UDP
1967 .. option:: ttl=int : [netsplice] [net]
1969 Time-to-live value for outgoing UDP multicast packets. Default: 1.
1971 .. option:: nodelay=bool : [netsplice] [net]
1973 Set TCP_NODELAY on TCP connections.
1975 .. option:: protocol=str, proto=str : [netsplice] [net]
1977 The network protocol to use. Accepted values are:
1980 Transmission control protocol.
1982 Transmission control protocol V6.
1984 User datagram protocol.
1986 User datagram protocol V6.
1990 When the protocol is TCP or UDP, the port must also be given, as well as the
1991 hostname if the job is a TCP listener or UDP reader. For unix sockets, the
1992 normal :option:`filename` option should be used and the port is invalid.
1994 .. option:: listen : [netsplice] [net]
1996 For TCP network connections, tell fio to listen for incoming connections
1997 rather than initiating an outgoing connection. The :option:`hostname` must
1998 be omitted if this option is used.
2000 .. option:: pingpong : [netsplice] [net]
2002 Normally a network writer will just continue writing data, and a network
2003 reader will just consume packages. If ``pingpong=1`` is set, a writer will
2004 send its normal payload to the reader, then wait for the reader to send the
2005 same payload back. This allows fio to measure network latencies. The
2006 submission and completion latencies then measure local time spent sending or
2007 receiving, and the completion latency measures how long it took for the
2008 other end to receive and send back. For UDP multicast traffic
2009 ``pingpong=1`` should only be set for a single reader when multiple readers
2010 are listening to the same address.
2012 .. option:: window_size : [netsplice] [net]
2014 Set the desired socket buffer size for the connection.
2016 .. option:: mss : [netsplice] [net]
2018 Set the TCP maximum segment size (TCP_MAXSEG).
2020 .. option:: donorname=str : [e4defrag]
2022 File will be used as a block donor (swap extents between files).
2024 .. option:: inplace=int : [e4defrag]
2026 Configure donor file blocks allocation strategy:
2029 Default. Preallocate donor's file on init.
2031 Allocate space immediately inside defragment event, and free right
2034 .. option:: clustername=str : [rbd,rados]
2036 Specifies the name of the Ceph cluster.
2038 .. option:: rbdname=str : [rbd]
2040 Specifies the name of the RBD.
2042 .. option:: pool=str : [rbd,rados]
2044 Specifies the name of the Ceph pool containing RBD or RADOS data.
2046 .. option:: clientname=str : [rbd,rados]
2048 Specifies the username (without the 'client.' prefix) used to access the
2049 Ceph cluster. If the *clustername* is specified, the *clientname* shall be
2050 the full *type.id* string. If no type. prefix is given, fio will add
2051 'client.' by default.
2053 .. option:: busy_poll=bool : [rbd,rados]
2055 Poll store instead of waiting for completion. Usually this provides better
2056 throughput at cost of higher(up to 100%) CPU utilization.
2058 .. option:: skip_bad=bool : [mtd]
2060 Skip operations against known bad blocks.
2062 .. option:: hdfsdirectory : [libhdfs]
2064 libhdfs will create chunk in this HDFS directory.
2066 .. option:: chunk_size : [libhdfs]
2068 The size of the chunk to use for each file.
2070 .. option:: verb=str : [rdma]
2072 The RDMA verb to use on this side of the RDMA ioengine connection. Valid
2073 values are write, read, send and recv. These correspond to the equivalent
2074 RDMA verbs (e.g. write = rdma_write etc.). Note that this only needs to be
2075 specified on the client side of the connection. See the examples folder.
2077 .. option:: bindname=str : [rdma]
2079 The name to use to bind the local RDMA-CM connection to a local RDMA device.
2080 This could be a hostname or an IPv4 or IPv6 address. On the server side this
2081 will be passed into the rdma_bind_addr() function and on the client site it
2082 will be used in the rdma_resolve_add() function. This can be useful when
2083 multiple paths exist between the client and the server or in certain loopback
2086 .. option:: readfua=bool : [sg]
2088 With readfua option set to 1, read operations include
2089 the force unit access (fua) flag. Default is 0.
2091 .. option:: writefua=bool : [sg]
2093 With writefua option set to 1, write operations include
2094 the force unit access (fua) flag. Default is 0.
2096 .. option:: sg_write_mode=str : [sg]
2098 Specify the type of write commands to issue. This option can take three values:
2101 This is the default where write opcodes are issued as usual.
2103 Issue WRITE AND VERIFY commands. The BYTCHK bit is set to 0. This
2104 directs the device to carry out a medium verification with no data
2105 comparison. The writefua option is ignored with this selection.
2107 Issue WRITE SAME commands. This transfers a single block to the device
2108 and writes this same block of data to a contiguous sequence of LBAs
2109 beginning at the specified offset. fio's block size parameter specifies
2110 the amount of data written with each command. However, the amount of data
2111 actually transferred to the device is equal to the device's block
2112 (sector) size. For a device with 512 byte sectors, blocksize=8k will
2113 write 16 sectors with each command. fio will still generate 8k of data
2114 for each command but only the first 512 bytes will be used and
2115 transferred to the device. The writefua option is ignored with this
2121 .. option:: iodepth=int
2123 Number of I/O units to keep in flight against the file. Note that
2124 increasing *iodepth* beyond 1 will not affect synchronous ioengines (except
2125 for small degrees when :option:`verify_async` is in use). Even async
2126 engines may impose OS restrictions causing the desired depth not to be
2127 achieved. This may happen on Linux when using libaio and not setting
2128 :option:`direct`\=1, since buffered I/O is not async on that OS. Keep an
2129 eye on the I/O depth distribution in the fio output to verify that the
2130 achieved depth is as expected. Default: 1.
2132 .. option:: iodepth_batch_submit=int, iodepth_batch=int
2134 This defines how many pieces of I/O to submit at once. It defaults to 1
2135 which means that we submit each I/O as soon as it is available, but can be
2136 raised to submit bigger batches of I/O at the time. If it is set to 0 the
2137 :option:`iodepth` value will be used.
2139 .. option:: iodepth_batch_complete_min=int, iodepth_batch_complete=int
2141 This defines how many pieces of I/O to retrieve at once. It defaults to 1
2142 which means that we'll ask for a minimum of 1 I/O in the retrieval process
2143 from the kernel. The I/O retrieval will go on until we hit the limit set by
2144 :option:`iodepth_low`. If this variable is set to 0, then fio will always
2145 check for completed events before queuing more I/O. This helps reduce I/O
2146 latency, at the cost of more retrieval system calls.
2148 .. option:: iodepth_batch_complete_max=int
2150 This defines maximum pieces of I/O to retrieve at once. This variable should
2151 be used along with :option:`iodepth_batch_complete_min`\=int variable,
2152 specifying the range of min and max amount of I/O which should be
2153 retrieved. By default it is equal to the :option:`iodepth_batch_complete_min`
2158 iodepth_batch_complete_min=1
2159 iodepth_batch_complete_max=<iodepth>
2161 which means that we will retrieve at least 1 I/O and up to the whole
2162 submitted queue depth. If none of I/O has been completed yet, we will wait.
2166 iodepth_batch_complete_min=0
2167 iodepth_batch_complete_max=<iodepth>
2169 which means that we can retrieve up to the whole submitted queue depth, but
2170 if none of I/O has been completed yet, we will NOT wait and immediately exit
2171 the system call. In this example we simply do polling.
2173 .. option:: iodepth_low=int
2175 The low water mark indicating when to start filling the queue
2176 again. Defaults to the same as :option:`iodepth`, meaning that fio will
2177 attempt to keep the queue full at all times. If :option:`iodepth` is set to
2178 e.g. 16 and *iodepth_low* is set to 4, then after fio has filled the queue of
2179 16 requests, it will let the depth drain down to 4 before starting to fill
2182 .. option:: serialize_overlap=bool
2184 Serialize in-flight I/Os that might otherwise cause or suffer from data races.
2185 When two or more I/Os are submitted simultaneously, there is no guarantee that
2186 the I/Os will be processed or completed in the submitted order. Further, if
2187 two or more of those I/Os are writes, any overlapping region between them can
2188 become indeterminate/undefined on certain storage. These issues can cause
2189 verification to fail erratically when at least one of the racing I/Os is
2190 changing data and the overlapping region has a non-zero size. Setting
2191 ``serialize_overlap`` tells fio to avoid provoking this behavior by explicitly
2192 serializing in-flight I/Os that have a non-zero overlap. Note that setting
2193 this option can reduce both performance and the :option:`iodepth` achieved.
2194 Additionally this option does not work when :option:`io_submit_mode` is set to
2195 offload. Default: false.
2197 .. option:: io_submit_mode=str
2199 This option controls how fio submits the I/O to the I/O engine. The default
2200 is `inline`, which means that the fio job threads submit and reap I/O
2201 directly. If set to `offload`, the job threads will offload I/O submission
2202 to a dedicated pool of I/O threads. This requires some coordination and thus
2203 has a bit of extra overhead, especially for lower queue depth I/O where it
2204 can increase latencies. The benefit is that fio can manage submission rates
2205 independently of the device completion rates. This avoids skewed latency
2206 reporting if I/O gets backed up on the device side (the coordinated omission
2213 .. option:: thinktime=time
2215 Stall the job for the specified period of time after an I/O has completed before issuing the
2216 next. May be used to simulate processing being done by an application.
2217 When the unit is omitted, the value is interpreted in microseconds. See
2218 :option:`thinktime_blocks` and :option:`thinktime_spin`.
2220 .. option:: thinktime_spin=time
2222 Only valid if :option:`thinktime` is set - pretend to spend CPU time doing
2223 something with the data received, before falling back to sleeping for the
2224 rest of the period specified by :option:`thinktime`. When the unit is
2225 omitted, the value is interpreted in microseconds.
2227 .. option:: thinktime_blocks=int
2229 Only valid if :option:`thinktime` is set - control how many blocks to issue,
2230 before waiting :option:`thinktime` usecs. If not set, defaults to 1 which will make
2231 fio wait :option:`thinktime` usecs after every block. This effectively makes any
2232 queue depth setting redundant, since no more than 1 I/O will be queued
2233 before we have to complete it and do our :option:`thinktime`. In other words, this
2234 setting effectively caps the queue depth if the latter is larger.
2236 .. option:: rate=int[,int][,int]
2238 Cap the bandwidth used by this job. The number is in bytes/sec, the normal
2239 suffix rules apply. Comma-separated values may be specified for reads,
2240 writes, and trims as described in :option:`blocksize`.
2242 For example, using `rate=1m,500k` would limit reads to 1MiB/sec and writes to
2243 500KiB/sec. Capping only reads or writes can be done with `rate=,500k` or
2244 `rate=500k,` where the former will only limit writes (to 500KiB/sec) and the
2245 latter will only limit reads.
2247 .. option:: rate_min=int[,int][,int]
2249 Tell fio to do whatever it can to maintain at least this bandwidth. Failing
2250 to meet this requirement will cause the job to exit. Comma-separated values
2251 may be specified for reads, writes, and trims as described in
2252 :option:`blocksize`.
2254 .. option:: rate_iops=int[,int][,int]
2256 Cap the bandwidth to this number of IOPS. Basically the same as
2257 :option:`rate`, just specified independently of bandwidth. If the job is
2258 given a block size range instead of a fixed value, the smallest block size
2259 is used as the metric. Comma-separated values may be specified for reads,
2260 writes, and trims as described in :option:`blocksize`.
2262 .. option:: rate_iops_min=int[,int][,int]
2264 If fio doesn't meet this rate of I/O, it will cause the job to exit.
2265 Comma-separated values may be specified for reads, writes, and trims as
2266 described in :option:`blocksize`.
2268 .. option:: rate_process=str
2270 This option controls how fio manages rated I/O submissions. The default is
2271 `linear`, which submits I/O in a linear fashion with fixed delays between
2272 I/Os that gets adjusted based on I/O completion rates. If this is set to
2273 `poisson`, fio will submit I/O based on a more real world random request
2274 flow, known as the Poisson process
2275 (https://en.wikipedia.org/wiki/Poisson_point_process). The lambda will be
2276 10^6 / IOPS for the given workload.
2278 .. option:: rate_ignore_thinktime=bool
2280 By default, fio will attempt to catch up to the specified rate setting,
2281 if any kind of thinktime setting was used. If this option is set, then
2282 fio will ignore the thinktime and continue doing IO at the specified
2283 rate, instead of entering a catch-up mode after thinktime is done.
2289 .. option:: latency_target=time
2291 If set, fio will attempt to find the max performance point that the given
2292 workload will run at while maintaining a latency below this target. When
2293 the unit is omitted, the value is interpreted in microseconds. See
2294 :option:`latency_window` and :option:`latency_percentile`.
2296 .. option:: latency_window=time
2298 Used with :option:`latency_target` to specify the sample window that the job
2299 is run at varying queue depths to test the performance. When the unit is
2300 omitted, the value is interpreted in microseconds.
2302 .. option:: latency_percentile=float
2304 The percentage of I/Os that must fall within the criteria specified by
2305 :option:`latency_target` and :option:`latency_window`. If not set, this
2306 defaults to 100.0, meaning that all I/Os must be equal or below to the value
2307 set by :option:`latency_target`.
2309 .. option:: max_latency=time
2311 If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
2312 maximum latency. When the unit is omitted, the value is interpreted in
2315 .. option:: rate_cycle=int
2317 Average bandwidth for :option:`rate` and :option:`rate_min` over this number
2318 of milliseconds. Defaults to 1000.
2324 .. option:: write_iolog=str
2326 Write the issued I/O patterns to the specified file. See
2327 :option:`read_iolog`. Specify a separate file for each job, otherwise the
2328 iologs will be interspersed and the file may be corrupt.
2330 .. option:: read_iolog=str
2332 Open an iolog with the specified filename and replay the I/O patterns it
2333 contains. This can be used to store a workload and replay it sometime
2334 later. The iolog given may also be a blktrace binary file, which allows fio
2335 to replay a workload captured by :command:`blktrace`. See
2336 :manpage:`blktrace(8)` for how to capture such logging data. For blktrace
2337 replay, the file needs to be turned into a blkparse binary data file first
2338 (``blkparse <device> -o /dev/null -d file_for_fio.bin``).
2340 .. option:: read_iolog_chunked=bool
2342 Determines how iolog is read. If false(default) entire :option:`read_iolog`
2343 will be read at once. If selected true, input from iolog will be read
2344 gradually. Useful when iolog is very large, or it is generated.
2346 .. option:: replay_no_stall=bool
2348 When replaying I/O with :option:`read_iolog` the default behavior is to
2349 attempt to respect the timestamps within the log and replay them with the
2350 appropriate delay between IOPS. By setting this variable fio will not
2351 respect the timestamps and attempt to replay them as fast as possible while
2352 still respecting ordering. The result is the same I/O pattern to a given
2353 device, but different timings.
2355 .. option:: replay_time_scale=int
2357 When replaying I/O with :option:`read_iolog`, fio will honor the
2358 original timing in the trace. With this option, it's possible to scale
2359 the time. It's a percentage option, if set to 50 it means run at 50%
2360 the original IO rate in the trace. If set to 200, run at twice the
2361 original IO rate. Defaults to 100.
2363 .. option:: replay_redirect=str
2365 While replaying I/O patterns using :option:`read_iolog` the default behavior
2366 is to replay the IOPS onto the major/minor device that each IOP was recorded
2367 from. This is sometimes undesirable because on a different machine those
2368 major/minor numbers can map to a different device. Changing hardware on the
2369 same system can also result in a different major/minor mapping.
2370 ``replay_redirect`` causes all I/Os to be replayed onto the single specified
2371 device regardless of the device it was recorded
2372 from. i.e. :option:`replay_redirect`\= :file:`/dev/sdc` would cause all I/O
2373 in the blktrace or iolog to be replayed onto :file:`/dev/sdc`. This means
2374 multiple devices will be replayed onto a single device, if the trace
2375 contains multiple devices. If you want multiple devices to be replayed
2376 concurrently to multiple redirected devices you must blkparse your trace
2377 into separate traces and replay them with independent fio invocations.
2378 Unfortunately this also breaks the strict time ordering between multiple
2381 .. option:: replay_align=int
2383 Force alignment of I/O offsets and lengths in a trace to this power of 2
2386 .. option:: replay_scale=int
2388 Scale sector offsets down by this factor when replaying traces.
2390 .. option:: replay_skip=str
2392 Sometimes it's useful to skip certain IO types in a replay trace.
2393 This could be, for instance, eliminating the writes in the trace.
2394 Or not replaying the trims/discards, if you are redirecting to
2395 a device that doesn't support them. This option takes a comma
2396 separated list of read, write, trim, sync.
2399 Threads, processes and job synchronization
2400 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2404 Fio defaults to creating jobs by using fork, however if this option is
2405 given, fio will create jobs by using POSIX Threads' function
2406 :manpage:`pthread_create(3)` to create threads instead.
2408 .. option:: wait_for=str
2410 If set, the current job won't be started until all workers of the specified
2411 waitee job are done.
2413 ``wait_for`` operates on the job name basis, so there are a few
2414 limitations. First, the waitee must be defined prior to the waiter job
2415 (meaning no forward references). Second, if a job is being referenced as a
2416 waitee, it must have a unique name (no duplicate waitees).
2418 .. option:: nice=int
2420 Run the job with the given nice value. See man :manpage:`nice(2)`.
2422 On Windows, values less than -15 set the process class to "High"; -1 through
2423 -15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2426 .. option:: prio=int
2428 Set the I/O priority value of this job. Linux limits us to a positive value
2429 between 0 and 7, with 0 being the highest. See man
2430 :manpage:`ionice(1)`. Refer to an appropriate manpage for other operating
2431 systems since meaning of priority may differ.
2433 .. option:: prioclass=int
2435 Set the I/O priority class. See man :manpage:`ionice(1)`.
2437 .. option:: cpus_allowed=str
2439 Controls the same options as :option:`cpumask`, but accepts a textual
2440 specification of the permitted CPUs instead and CPUs are indexed from 0. So
2441 to use CPUs 0 and 5 you would specify ``cpus_allowed=0,5``. This option also
2442 allows a range of CPUs to be specified -- say you wanted a binding to CPUs
2443 0, 5, and 8 to 15, you would set ``cpus_allowed=0,5,8-15``.
2445 On Windows, when ``cpus_allowed`` is unset only CPUs from fio's current
2446 processor group will be used and affinity settings are inherited from the
2447 system. An fio build configured to target Windows 7 makes options that set
2448 CPUs processor group aware and values will set both the processor group
2449 and a CPU from within that group. For example, on a system where processor
2450 group 0 has 40 CPUs and processor group 1 has 32 CPUs, ``cpus_allowed``
2451 values between 0 and 39 will bind CPUs from processor group 0 and
2452 ``cpus_allowed`` values between 40 and 71 will bind CPUs from processor
2453 group 1. When using ``cpus_allowed_policy=shared`` all CPUs specified by a
2454 single ``cpus_allowed`` option must be from the same processor group. For
2455 Windows fio builds not built for Windows 7, CPUs will only be selected from
2456 (and be relative to) whatever processor group fio happens to be running in
2457 and CPUs from other processor groups cannot be used.
2459 .. option:: cpus_allowed_policy=str
2461 Set the policy of how fio distributes the CPUs specified by
2462 :option:`cpus_allowed` or :option:`cpumask`. Two policies are supported:
2465 All jobs will share the CPU set specified.
2467 Each job will get a unique CPU from the CPU set.
2469 **shared** is the default behavior, if the option isn't specified. If
2470 **split** is specified, then fio will will assign one cpu per job. If not
2471 enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2474 .. option:: cpumask=int
2476 Set the CPU affinity of this job. The parameter given is a bit mask of
2477 allowed CPUs the job may run on. So if you want the allowed CPUs to be 1
2478 and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2479 :manpage:`sched_setaffinity(2)`. This may not work on all supported
2480 operating systems or kernel versions. This option doesn't work well for a
2481 higher CPU count than what you can store in an integer mask, so it can only
2482 control cpus 1-32. For boxes with larger CPU counts, use
2483 :option:`cpus_allowed`.
2485 .. option:: numa_cpu_nodes=str
2487 Set this job running on specified NUMA nodes' CPUs. The arguments allow
2488 comma delimited list of cpu numbers, A-B ranges, or `all`. Note, to enable
2489 NUMA options support, fio must be built on a system with libnuma-dev(el)
2492 .. option:: numa_mem_policy=str
2494 Set this job's memory policy and corresponding NUMA nodes. Format of the
2499 ``mode`` is one of the following memory policies: ``default``, ``prefer``,
2500 ``bind``, ``interleave`` or ``local``. For ``default`` and ``local`` memory
2501 policies, no node needs to be specified. For ``prefer``, only one node is
2502 allowed. For ``bind`` and ``interleave`` the ``nodelist`` may be as
2503 follows: a comma delimited list of numbers, A-B ranges, or `all`.
2505 .. option:: cgroup=str
2507 Add job to this control group. If it doesn't exist, it will be created. The
2508 system must have a mounted cgroup blkio mount point for this to work. If
2509 your system doesn't have it mounted, you can do so with::
2511 # mount -t cgroup -o blkio none /cgroup
2513 .. option:: cgroup_weight=int
2515 Set the weight of the cgroup to this value. See the documentation that comes
2516 with the kernel, allowed values are in the range of 100..1000.
2518 .. option:: cgroup_nodelete=bool
2520 Normally fio will delete the cgroups it has created after the job
2521 completion. To override this behavior and to leave cgroups around after the
2522 job completion, set ``cgroup_nodelete=1``. This can be useful if one wants
2523 to inspect various cgroup files after job completion. Default: false.
2525 .. option:: flow_id=int
2527 The ID of the flow. If not specified, it defaults to being a global
2528 flow. See :option:`flow`.
2530 .. option:: flow=int
2532 Weight in token-based flow control. If this value is used, then there is a
2533 'flow counter' which is used to regulate the proportion of activity between
2534 two or more jobs. Fio attempts to keep this flow counter near zero. The
2535 ``flow`` parameter stands for how much should be added or subtracted to the
2536 flow counter on each iteration of the main I/O loop. That is, if one job has
2537 ``flow=8`` and another job has ``flow=-1``, then there will be a roughly 1:8
2538 ratio in how much one runs vs the other.
2540 .. option:: flow_watermark=int
2542 The maximum value that the absolute value of the flow counter is allowed to
2543 reach before the job must wait for a lower value of the counter.
2545 .. option:: flow_sleep=int
2547 The period of time, in microseconds, to wait after the flow watermark has
2548 been exceeded before retrying operations.
2550 .. option:: stonewall, wait_for_previous
2552 Wait for preceding jobs in the job file to exit, before starting this
2553 one. Can be used to insert serialization points in the job file. A stone
2554 wall also implies starting a new reporting group, see
2555 :option:`group_reporting`.
2559 By default, fio will continue running all other jobs when one job finishes
2560 but sometimes this is not the desired action. Setting ``exitall`` will
2561 instead make fio terminate all other jobs when one job finishes.
2563 .. option:: exec_prerun=str
2565 Before running this job, issue the command specified through
2566 :manpage:`system(3)`. Output is redirected in a file called
2567 :file:`jobname.prerun.txt`.
2569 .. option:: exec_postrun=str
2571 After the job completes, issue the command specified though
2572 :manpage:`system(3)`. Output is redirected in a file called
2573 :file:`jobname.postrun.txt`.
2577 Instead of running as the invoking user, set the user ID to this value
2578 before the thread/process does any work.
2582 Set group ID, see :option:`uid`.
2588 .. option:: verify_only
2590 Do not perform specified workload, only verify data still matches previous
2591 invocation of this workload. This option allows one to check data multiple
2592 times at a later date without overwriting it. This option makes sense only
2593 for workloads that write data, and does not support workloads with the
2594 :option:`time_based` option set.
2596 .. option:: do_verify=bool
2598 Run the verify phase after a write phase. Only valid if :option:`verify` is
2601 .. option:: verify=str
2603 If writing to a file, fio can verify the file contents after each iteration
2604 of the job. Each verification method also implies verification of special
2605 header, which is written to the beginning of each block. This header also
2606 includes meta information, like offset of the block, block number, timestamp
2607 when block was written, etc. :option:`verify` can be combined with
2608 :option:`verify_pattern` option. The allowed values are:
2611 Use an md5 sum of the data area and store it in the header of
2615 Use an experimental crc64 sum of the data area and store it in the
2616 header of each block.
2619 Use a crc32c sum of the data area and store it in the header of
2620 each block. This will automatically use hardware acceleration
2621 (e.g. SSE4.2 on an x86 or CRC crypto extensions on ARM64) but will
2622 fall back to software crc32c if none is found. Generally the
2623 fastest checksum fio supports when hardware accelerated.
2629 Use a crc32 sum of the data area and store it in the header of each
2633 Use a crc16 sum of the data area and store it in the header of each
2637 Use a crc7 sum of the data area and store it in the header of each
2641 Use xxhash as the checksum function. Generally the fastest software
2642 checksum that fio supports.
2645 Use sha512 as the checksum function.
2648 Use sha256 as the checksum function.
2651 Use optimized sha1 as the checksum function.
2654 Use optimized sha3-224 as the checksum function.
2657 Use optimized sha3-256 as the checksum function.
2660 Use optimized sha3-384 as the checksum function.
2663 Use optimized sha3-512 as the checksum function.
2666 This option is deprecated, since now meta information is included in
2667 generic verification header and meta verification happens by
2668 default. For detailed information see the description of the
2669 :option:`verify` setting. This option is kept because of
2670 compatibility's sake with old configurations. Do not use it.
2673 Verify a strict pattern. Normally fio includes a header with some
2674 basic information and checksumming, but if this option is set, only
2675 the specific pattern set with :option:`verify_pattern` is verified.
2678 Only pretend to verify. Useful for testing internals with
2679 :option:`ioengine`\=null, not for much else.
2681 This option can be used for repeated burn-in tests of a system to make sure
2682 that the written data is also correctly read back. If the data direction
2683 given is a read or random read, fio will assume that it should verify a
2684 previously written file. If the data direction includes any form of write,
2685 the verify will be of the newly written data.
2687 To avoid false verification errors, do not use the norandommap option when
2688 verifying data with async I/O engines and I/O depths > 1. Or use the
2689 norandommap and the lfsr random generator together to avoid writing to the
2690 same offset with muliple outstanding I/Os.
2692 .. option:: verify_offset=int
2694 Swap the verification header with data somewhere else in the block before
2695 writing. It is swapped back before verifying.
2697 .. option:: verify_interval=int
2699 Write the verification header at a finer granularity than the
2700 :option:`blocksize`. It will be written for chunks the size of
2701 ``verify_interval``. :option:`blocksize` should divide this evenly.
2703 .. option:: verify_pattern=str
2705 If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2706 filling with totally random bytes, but sometimes it's interesting to fill
2707 with a known pattern for I/O verification purposes. Depending on the width
2708 of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time (it can
2709 be either a decimal or a hex number). The ``verify_pattern`` if larger than
2710 a 32-bit quantity has to be a hex number that starts with either "0x" or
2711 "0X". Use with :option:`verify`. Also, ``verify_pattern`` supports %o
2712 format, which means that for each block offset will be written and then
2713 verified back, e.g.::
2717 Or use combination of everything::
2719 verify_pattern=0xff%o"abcd"-12
2721 .. option:: verify_fatal=bool
2723 Normally fio will keep checking the entire contents before quitting on a
2724 block verification failure. If this option is set, fio will exit the job on
2725 the first observed failure. Default: false.
2727 .. option:: verify_dump=bool
2729 If set, dump the contents of both the original data block and the data block
2730 we read off disk to files. This allows later analysis to inspect just what
2731 kind of data corruption occurred. Off by default.
2733 .. option:: verify_async=int
2735 Fio will normally verify I/O inline from the submitting thread. This option
2736 takes an integer describing how many async offload threads to create for I/O
2737 verification instead, causing fio to offload the duty of verifying I/O
2738 contents to one or more separate threads. If using this offload option, even
2739 sync I/O engines can benefit from using an :option:`iodepth` setting higher
2740 than 1, as it allows them to have I/O in flight while verifies are running.
2741 Defaults to 0 async threads, i.e. verification is not asynchronous.
2743 .. option:: verify_async_cpus=str
2745 Tell fio to set the given CPU affinity on the async I/O verification
2746 threads. See :option:`cpus_allowed` for the format used.
2748 .. option:: verify_backlog=int
2750 Fio will normally verify the written contents of a job that utilizes verify
2751 once that job has completed. In other words, everything is written then
2752 everything is read back and verified. You may want to verify continually
2753 instead for a variety of reasons. Fio stores the meta data associated with
2754 an I/O block in memory, so for large verify workloads, quite a bit of memory
2755 would be used up holding this meta data. If this option is enabled, fio will
2756 write only N blocks before verifying these blocks.
2758 .. option:: verify_backlog_batch=int
2760 Control how many blocks fio will verify if :option:`verify_backlog` is
2761 set. If not set, will default to the value of :option:`verify_backlog`
2762 (meaning the entire queue is read back and verified). If
2763 ``verify_backlog_batch`` is less than :option:`verify_backlog` then not all
2764 blocks will be verified, if ``verify_backlog_batch`` is larger than
2765 :option:`verify_backlog`, some blocks will be verified more than once.
2767 .. option:: verify_state_save=bool
2769 When a job exits during the write phase of a verify workload, save its
2770 current state. This allows fio to replay up until that point, if the verify
2771 state is loaded for the verify read phase. The format of the filename is,
2774 <type>-<jobname>-<jobindex>-verify.state.
2776 <type> is "local" for a local run, "sock" for a client/server socket
2777 connection, and "ip" (192.168.0.1, for instance) for a networked
2778 client/server connection. Defaults to true.
2780 .. option:: verify_state_load=bool
2782 If a verify termination trigger was used, fio stores the current write state
2783 of each thread. This can be used at verification time so that fio knows how
2784 far it should verify. Without this information, fio will run a full
2785 verification pass, according to the settings in the job file used. Default
2788 .. option:: trim_percentage=int
2790 Number of verify blocks to discard/trim.
2792 .. option:: trim_verify_zero=bool
2794 Verify that trim/discarded blocks are returned as zeros.
2796 .. option:: trim_backlog=int
2798 Trim after this number of blocks are written.
2800 .. option:: trim_backlog_batch=int
2802 Trim this number of I/O blocks.
2804 .. option:: experimental_verify=bool
2806 Enable experimental verification.
2811 .. option:: steadystate=str:float, ss=str:float
2813 Define the criterion and limit for assessing steady state performance. The
2814 first parameter designates the criterion whereas the second parameter sets
2815 the threshold. When the criterion falls below the threshold for the
2816 specified duration, the job will stop. For example, `iops_slope:0.1%` will
2817 direct fio to terminate the job when the least squares regression slope
2818 falls below 0.1% of the mean IOPS. If :option:`group_reporting` is enabled
2819 this will apply to all jobs in the group. Below is the list of available
2820 steady state assessment criteria. All assessments are carried out using only
2821 data from the rolling collection window. Threshold limits can be expressed
2822 as a fixed value or as a percentage of the mean in the collection window.
2825 Collect IOPS data. Stop the job if all individual IOPS measurements
2826 are within the specified limit of the mean IOPS (e.g., ``iops:2``
2827 means that all individual IOPS values must be within 2 of the mean,
2828 whereas ``iops:0.2%`` means that all individual IOPS values must be
2829 within 0.2% of the mean IOPS to terminate the job).
2832 Collect IOPS data and calculate the least squares regression
2833 slope. Stop the job if the slope falls below the specified limit.
2836 Collect bandwidth data. Stop the job if all individual bandwidth
2837 measurements are within the specified limit of the mean bandwidth.
2840 Collect bandwidth data and calculate the least squares regression
2841 slope. Stop the job if the slope falls below the specified limit.
2843 .. option:: steadystate_duration=time, ss_dur=time
2845 A rolling window of this duration will be used to judge whether steady state
2846 has been reached. Data will be collected once per second. The default is 0
2847 which disables steady state detection. When the unit is omitted, the
2848 value is interpreted in seconds.
2850 .. option:: steadystate_ramp_time=time, ss_ramp=time
2852 Allow the job to run for the specified duration before beginning data
2853 collection for checking the steady state job termination criterion. The
2854 default is 0. When the unit is omitted, the value is interpreted in seconds.
2857 Measurements and reporting
2858 ~~~~~~~~~~~~~~~~~~~~~~~~~~
2860 .. option:: per_job_logs=bool
2862 If set, this generates bw/clat/iops log with per file private filenames. If
2863 not set, jobs with identical names will share the log filename. Default:
2866 .. option:: group_reporting
2868 It may sometimes be interesting to display statistics for groups of jobs as
2869 a whole instead of for each individual job. This is especially true if
2870 :option:`numjobs` is used; looking at individual thread/process output
2871 quickly becomes unwieldy. To see the final report per-group instead of
2872 per-job, use :option:`group_reporting`. Jobs in a file will be part of the
2873 same reporting group, unless if separated by a :option:`stonewall`, or by
2874 using :option:`new_group`.
2876 .. option:: new_group
2878 Start a new reporting group. See: :option:`group_reporting`. If not given,
2879 all jobs in a file will be part of the same reporting group, unless
2880 separated by a :option:`stonewall`.
2882 .. option:: stats=bool
2884 By default, fio collects and shows final output results for all jobs
2885 that run. If this option is set to 0, then fio will ignore it in
2886 the final stat output.
2888 .. option:: write_bw_log=str
2890 If given, write a bandwidth log for this job. Can be used to store data of
2891 the bandwidth of the jobs in their lifetime.
2893 If no str argument is given, the default filename of
2894 :file:`jobname_type.x.log` is used. Even when the argument is given, fio
2895 will still append the type of log. So if one specifies::
2899 The actual log name will be :file:`foo_bw.x.log` where `x` is the index
2900 of the job (`1..N`, where `N` is the number of jobs). If
2901 :option:`per_job_logs` is false, then the filename will not include the
2904 The included :command:`fio_generate_plots` script uses :command:`gnuplot` to turn these
2905 text files into nice graphs. See `Log File Formats`_ for how data is
2906 structured within the file.
2908 .. option:: write_lat_log=str
2910 Same as :option:`write_bw_log`, except this option creates I/O
2911 submission (e.g., :file:`name_slat.x.log`), completion (e.g.,
2912 :file:`name_clat.x.log`), and total (e.g., :file:`name_lat.x.log`)
2913 latency files instead. See :option:`write_bw_log` for details about
2914 the filename format and `Log File Formats`_ for how data is structured
2917 .. option:: write_hist_log=str
2919 Same as :option:`write_bw_log` but writes an I/O completion latency
2920 histogram file (e.g., :file:`name_hist.x.log`) instead. Note that this
2921 file will be empty unless :option:`log_hist_msec` has also been set.
2922 See :option:`write_bw_log` for details about the filename format and
2923 `Log File Formats`_ for how data is structured within the file.
2925 .. option:: write_iops_log=str
2927 Same as :option:`write_bw_log`, but writes an IOPS file (e.g.
2928 :file:`name_iops.x.log`) instead. Because fio defaults to individual
2929 I/O logging, the value entry in the IOPS log will be 1 unless windowed
2930 logging (see :option:`log_avg_msec`) has been enabled. See
2931 :option:`write_bw_log` for details about the filename format and `Log
2932 File Formats`_ for how data is structured within the file.
2934 .. option:: log_avg_msec=int
2936 By default, fio will log an entry in the iops, latency, or bw log for every
2937 I/O that completes. When writing to the disk log, that can quickly grow to a
2938 very large size. Setting this option makes fio average the each log entry
2939 over the specified period of time, reducing the resolution of the log. See
2940 :option:`log_max_value` as well. Defaults to 0, logging all entries.
2941 Also see `Log File Formats`_.
2943 .. option:: log_hist_msec=int
2945 Same as :option:`log_avg_msec`, but logs entries for completion latency
2946 histograms. Computing latency percentiles from averages of intervals using
2947 :option:`log_avg_msec` is inaccurate. Setting this option makes fio log
2948 histogram entries over the specified period of time, reducing log sizes for
2949 high IOPS devices while retaining percentile accuracy. See
2950 :option:`log_hist_coarseness` and :option:`write_hist_log` as well.
2951 Defaults to 0, meaning histogram logging is disabled.
2953 .. option:: log_hist_coarseness=int
2955 Integer ranging from 0 to 6, defining the coarseness of the resolution of
2956 the histogram logs enabled with :option:`log_hist_msec`. For each increment
2957 in coarseness, fio outputs half as many bins. Defaults to 0, for which
2958 histogram logs contain 1216 latency bins. See :option:`write_hist_log`
2959 and `Log File Formats`_.
2961 .. option:: log_max_value=bool
2963 If :option:`log_avg_msec` is set, fio logs the average over that window. If
2964 you instead want to log the maximum value, set this option to 1. Defaults to
2965 0, meaning that averaged values are logged.
2967 .. option:: log_offset=bool
2969 If this is set, the iolog options will include the byte offset for the I/O
2970 entry as well as the other data values. Defaults to 0 meaning that
2971 offsets are not present in logs. Also see `Log File Formats`_.
2973 .. option:: log_compression=int
2975 If this is set, fio will compress the I/O logs as it goes, to keep the
2976 memory footprint lower. When a log reaches the specified size, that chunk is
2977 removed and compressed in the background. Given that I/O logs are fairly
2978 highly compressible, this yields a nice memory savings for longer runs. The
2979 downside is that the compression will consume some background CPU cycles, so
2980 it may impact the run. This, however, is also true if the logging ends up
2981 consuming most of the system memory. So pick your poison. The I/O logs are
2982 saved normally at the end of a run, by decompressing the chunks and storing
2983 them in the specified log file. This feature depends on the availability of
2986 .. option:: log_compression_cpus=str
2988 Define the set of CPUs that are allowed to handle online log compression for
2989 the I/O jobs. This can provide better isolation between performance
2990 sensitive jobs, and background compression work. See
2991 :option:`cpus_allowed` for the format used.
2993 .. option:: log_store_compressed=bool
2995 If set, fio will store the log files in a compressed format. They can be
2996 decompressed with fio, using the :option:`--inflate-log` command line
2997 parameter. The files will be stored with a :file:`.fz` suffix.
2999 .. option:: log_unix_epoch=bool
3001 If set, fio will log Unix timestamps to the log files produced by enabling
3002 write_type_log for each log type, instead of the default zero-based
3005 .. option:: block_error_percentiles=bool
3007 If set, record errors in trim block-sized units from writes and trims and
3008 output a histogram of how many trims it took to get to errors, and what kind
3009 of error was encountered.
3011 .. option:: bwavgtime=int
3013 Average the calculated bandwidth over the given time. Value is specified in
3014 milliseconds. If the job also does bandwidth logging through
3015 :option:`write_bw_log`, then the minimum of this option and
3016 :option:`log_avg_msec` will be used. Default: 500ms.
3018 .. option:: iopsavgtime=int
3020 Average the calculated IOPS over the given time. Value is specified in
3021 milliseconds. If the job also does IOPS logging through
3022 :option:`write_iops_log`, then the minimum of this option and
3023 :option:`log_avg_msec` will be used. Default: 500ms.
3025 .. option:: disk_util=bool
3027 Generate disk utilization statistics, if the platform supports it.
3030 .. option:: disable_lat=bool
3032 Disable measurements of total latency numbers. Useful only for cutting back
3033 the number of calls to :manpage:`gettimeofday(2)`, as that does impact
3034 performance at really high IOPS rates. Note that to really get rid of a
3035 large amount of these calls, this option must be used with
3036 :option:`disable_slat` and :option:`disable_bw_measurement` as well.
3038 .. option:: disable_clat=bool
3040 Disable measurements of completion latency numbers. See
3041 :option:`disable_lat`.
3043 .. option:: disable_slat=bool
3045 Disable measurements of submission latency numbers. See
3046 :option:`disable_lat`.
3048 .. option:: disable_bw_measurement=bool, disable_bw=bool
3050 Disable measurements of throughput/bandwidth numbers. See
3051 :option:`disable_lat`.
3053 .. option:: clat_percentiles=bool
3055 Enable the reporting of percentiles of completion latencies. This
3056 option is mutually exclusive with :option:`lat_percentiles`.
3058 .. option:: lat_percentiles=bool
3060 Enable the reporting of percentiles of I/O latencies. This is similar
3061 to :option:`clat_percentiles`, except that this includes the
3062 submission latency. This option is mutually exclusive with
3063 :option:`clat_percentiles`.
3065 .. option:: percentile_list=float_list
3067 Overwrite the default list of percentiles for completion latencies and
3068 the block error histogram. Each number is a floating number in the
3069 range (0,100], and the maximum length of the list is 20. Use ``:`` to
3070 separate the numbers, and list the numbers in ascending order. For
3071 example, ``--percentile_list=99.5:99.9`` will cause fio to report the
3072 values of completion latency below which 99.5% and 99.9% of the observed
3073 latencies fell, respectively.
3075 .. option:: significant_figures=int
3077 If using :option:`--output-format` of `normal`, set the significant
3078 figures to this value. Higher values will yield more precise IOPS and
3079 throughput units, while lower values will round. Requires a minimum
3080 value of 1 and a maximum value of 10. Defaults to 4.
3086 .. option:: exitall_on_error
3088 When one job finishes in error, terminate the rest. The default is to wait
3089 for each job to finish.
3091 .. option:: continue_on_error=str
3093 Normally fio will exit the job on the first observed failure. If this option
3094 is set, fio will continue the job when there is a 'non-fatal error' (EIO or
3095 EILSEQ) until the runtime is exceeded or the I/O size specified is
3096 completed. If this option is used, there are two more stats that are
3097 appended, the total error count and the first error. The error field given
3098 in the stats is the first error that was hit during the run.
3100 The allowed values are:
3103 Exit on any I/O or verify errors.
3106 Continue on read errors, exit on all others.
3109 Continue on write errors, exit on all others.
3112 Continue on any I/O error, exit on all others.
3115 Continue on verify errors, exit on all others.
3118 Continue on all errors.
3121 Backward-compatible alias for 'none'.
3124 Backward-compatible alias for 'all'.
3126 .. option:: ignore_error=str
3128 Sometimes you want to ignore some errors during test in that case you can
3129 specify error list for each error type, instead of only being able to
3130 ignore the default 'non-fatal error' using :option:`continue_on_error`.
3131 ``ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST`` errors for
3132 given error type is separated with ':'. Error may be symbol ('ENOSPC',
3133 'ENOMEM') or integer. Example::
3135 ignore_error=EAGAIN,ENOSPC:122
3137 This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
3138 WRITE. This option works by overriding :option:`continue_on_error` with
3139 the list of errors for each error type if any.
3141 .. option:: error_dump=bool
3143 If set dump every error even if it is non fatal, true by default. If
3144 disabled only fatal error will be dumped.
3146 Running predefined workloads
3147 ----------------------------
3149 Fio includes predefined profiles that mimic the I/O workloads generated by
3152 .. option:: profile=str
3154 The predefined workload to run. Current profiles are:
3157 Threaded I/O bench (tiotest/tiobench) like workload.
3160 Aerospike Certification Tool (ACT) like workload.
3162 To view a profile's additional options use :option:`--cmdhelp` after specifying
3163 the profile. For example::
3165 $ fio --profile=act --cmdhelp
3170 .. option:: device-names=str
3175 .. option:: load=int
3178 ACT load multiplier. Default: 1.
3180 .. option:: test-duration=time
3183 How long the entire test takes to run. When the unit is omitted, the value
3184 is given in seconds. Default: 24h.
3186 .. option:: threads-per-queue=int
3189 Number of read I/O threads per device. Default: 8.
3191 .. option:: read-req-num-512-blocks=int
3194 Number of 512B blocks to read at the time. Default: 3.
3196 .. option:: large-block-op-kbytes=int
3199 Size of large block ops in KiB (writes). Default: 131072.
3204 Set to run ACT prep phase.
3206 Tiobench profile options
3207 ~~~~~~~~~~~~~~~~~~~~~~~~
3209 .. option:: size=str
3214 .. option:: block=int
3217 Block size in bytes. Default: 4096.
3219 .. option:: numruns=int
3229 .. option:: threads=int
3234 Interpreting the output
3235 -----------------------
3238 Example output was based on the following:
3239 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --time_based \
3240 --rate=1256k --bs=14K --name=quick --runtime=1s --name=mixed \
3241 --runtime=2m --rw=rw
3243 Fio spits out a lot of output. While running, fio will display the status of the
3244 jobs created. An example of that would be::
3246 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]
3248 The characters inside the first set of square brackets denote the current status of
3249 each thread. The first character is the first job defined in the job file, and so
3250 forth. The possible values (in typical life cycle order) are:
3252 +------+-----+-----------------------------------------------------------+
3254 +======+=====+===========================================================+
3255 | P | | Thread setup, but not started. |
3256 +------+-----+-----------------------------------------------------------+
3257 | C | | Thread created. |
3258 +------+-----+-----------------------------------------------------------+
3259 | I | | Thread initialized, waiting or generating necessary data. |
3260 +------+-----+-----------------------------------------------------------+
3261 | | p | Thread running pre-reading file(s). |
3262 +------+-----+-----------------------------------------------------------+
3263 | | / | Thread is in ramp period. |
3264 +------+-----+-----------------------------------------------------------+
3265 | | R | Running, doing sequential reads. |
3266 +------+-----+-----------------------------------------------------------+
3267 | | r | Running, doing random reads. |
3268 +------+-----+-----------------------------------------------------------+
3269 | | W | Running, doing sequential writes. |
3270 +------+-----+-----------------------------------------------------------+
3271 | | w | Running, doing random writes. |
3272 +------+-----+-----------------------------------------------------------+
3273 | | M | Running, doing mixed sequential reads/writes. |
3274 +------+-----+-----------------------------------------------------------+
3275 | | m | Running, doing mixed random reads/writes. |
3276 +------+-----+-----------------------------------------------------------+
3277 | | D | Running, doing sequential trims. |
3278 +------+-----+-----------------------------------------------------------+
3279 | | d | Running, doing random trims. |
3280 +------+-----+-----------------------------------------------------------+
3281 | | F | Running, currently waiting for :manpage:`fsync(2)`. |
3282 +------+-----+-----------------------------------------------------------+
3283 | | V | Running, doing verification of written data. |
3284 +------+-----+-----------------------------------------------------------+
3285 | f | | Thread finishing. |
3286 +------+-----+-----------------------------------------------------------+
3287 | E | | Thread exited, not reaped by main thread yet. |
3288 +------+-----+-----------------------------------------------------------+
3289 | _ | | Thread reaped. |
3290 +------+-----+-----------------------------------------------------------+
3291 | X | | Thread reaped, exited with an error. |
3292 +------+-----+-----------------------------------------------------------+
3293 | K | | Thread reaped, exited due to signal. |
3294 +------+-----+-----------------------------------------------------------+
3297 Example output was based on the following:
3298 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --runtime=58m \
3299 --time_based --rate=2512k --bs=256K --numjobs=10 \
3300 --name=readers --rw=read --name=writers --rw=write
3302 Fio will condense the thread string as not to take up more space on the command
3303 line than needed. For instance, if you have 10 readers and 10 writers running,
3304 the output would look like this::
3306 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]
3308 Note that the status string is displayed in order, so it's possible to tell which of
3309 the jobs are currently doing what. In the example above this means that jobs 1--10
3310 are readers and 11--20 are writers.
3312 The other values are fairly self explanatory -- number of threads currently
3313 running and doing I/O, the number of currently open files (f=), the estimated
3314 completion percentage, the rate of I/O since last check (read speed listed first,
3315 then write speed and optionally trim speed) in terms of bandwidth and IOPS,
3316 and time to completion for the current running group. It's impossible to estimate
3317 runtime of the following groups (if any).
3320 Example output was based on the following:
3321 TZ=UTC fio --iodepth=16 --ioengine=posixaio --filename=/tmp/fiofile \
3322 --direct=1 --size=100M --time_based --runtime=50s --rate_iops=89 \
3323 --bs=7K --name=Client1 --rw=write
3325 When fio is done (or interrupted by :kbd:`Ctrl-C`), it will show the data for
3326 each thread, group of threads, and disks in that order. For each overall thread (or
3327 group) the output looks like::
3329 Client1: (groupid=0, jobs=1): err= 0: pid=16109: Sat Jun 24 12:07:54 2017
3330 write: IOPS=88, BW=623KiB/s (638kB/s)(30.4MiB/50032msec)
3331 slat (nsec): min=500, max=145500, avg=8318.00, stdev=4781.50
3332 clat (usec): min=170, max=78367, avg=4019.02, stdev=8293.31
3333 lat (usec): min=174, max=78375, avg=4027.34, stdev=8291.79
3334 clat percentiles (usec):
3335 | 1.00th=[ 302], 5.00th=[ 326], 10.00th=[ 343], 20.00th=[ 363],
3336 | 30.00th=[ 392], 40.00th=[ 404], 50.00th=[ 416], 60.00th=[ 445],
3337 | 70.00th=[ 816], 80.00th=[ 6718], 90.00th=[12911], 95.00th=[21627],
3338 | 99.00th=[43779], 99.50th=[51643], 99.90th=[68682], 99.95th=[72877],
3340 bw ( KiB/s): min= 532, max= 686, per=0.10%, avg=622.87, stdev=24.82, samples= 100
3341 iops : min= 76, max= 98, avg=88.98, stdev= 3.54, samples= 100
3342 lat (usec) : 250=0.04%, 500=64.11%, 750=4.81%, 1000=2.79%
3343 lat (msec) : 2=4.16%, 4=1.84%, 10=4.90%, 20=11.33%, 50=5.37%
3344 lat (msec) : 100=0.65%
3345 cpu : usr=0.27%, sys=0.18%, ctx=12072, majf=0, minf=21
3346 IO depths : 1=85.0%, 2=13.1%, 4=1.8%, 8=0.1%, 16=0.0%, 32=0.0%, >=64=0.0%
3347 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3348 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3349 issued rwt: total=0,4450,0, short=0,0,0, dropped=0,0,0
3350 latency : target=0, window=0, percentile=100.00%, depth=8
3352 The job name (or first job's name when using :option:`group_reporting`) is printed,
3353 along with the group id, count of jobs being aggregated, last error id seen (which
3354 is 0 when there are no errors), pid/tid of that thread and the time the job/group
3355 completed. Below are the I/O statistics for each data direction performed (showing
3356 writes in the example above). In the order listed, they denote:
3359 The string before the colon shows the I/O direction the statistics
3360 are for. **IOPS** is the average I/Os performed per second. **BW**
3361 is the average bandwidth rate shown as: value in power of 2 format
3362 (value in power of 10 format). The last two values show: (**total
3363 I/O performed** in power of 2 format / **runtime** of that thread).
3366 Submission latency (**min** being the minimum, **max** being the
3367 maximum, **avg** being the average, **stdev** being the standard
3368 deviation). This is the time it took to submit the I/O. For
3369 sync I/O this row is not displayed as the slat is really the
3370 completion latency (since queue/complete is one operation there).
3371 This value can be in nanoseconds, microseconds or milliseconds ---
3372 fio will choose the most appropriate base and print that (in the
3373 example above nanoseconds was the best scale). Note: in :option:`--minimal` mode
3374 latencies are always expressed in microseconds.
3377 Completion latency. Same names as slat, this denotes the time from
3378 submission to completion of the I/O pieces. For sync I/O, clat will
3379 usually be equal (or very close) to 0, as the time from submit to
3380 complete is basically just CPU time (I/O has already been done, see slat
3384 Total latency. Same names as slat and clat, this denotes the time from
3385 when fio created the I/O unit to completion of the I/O operation.
3388 Bandwidth statistics based on samples. Same names as the xlat stats,
3389 but also includes the number of samples taken (**samples**) and an
3390 approximate percentage of total aggregate bandwidth this thread
3391 received in its group (**per**). This last value is only really
3392 useful if the threads in this group are on the same disk, since they
3393 are then competing for disk access.
3396 IOPS statistics based on samples. Same names as bw.
3398 **lat (nsec/usec/msec)**
3399 The distribution of I/O completion latencies. This is the time from when
3400 I/O leaves fio and when it gets completed. Unlike the separate
3401 read/write/trim sections above, the data here and in the remaining
3402 sections apply to all I/Os for the reporting group. 250=0.04% means that
3403 0.04% of the I/Os completed in under 250us. 500=64.11% means that 64.11%
3404 of the I/Os required 250 to 499us for completion.
3407 CPU usage. User and system time, along with the number of context
3408 switches this thread went through, usage of system and user time, and
3409 finally the number of major and minor page faults. The CPU utilization
3410 numbers are averages for the jobs in that reporting group, while the
3411 context and fault counters are summed.
3414 The distribution of I/O depths over the job lifetime. The numbers are
3415 divided into powers of 2 and each entry covers depths from that value
3416 up to those that are lower than the next entry -- e.g., 16= covers
3417 depths from 16 to 31. Note that the range covered by a depth
3418 distribution entry can be different to the range covered by the
3419 equivalent submit/complete distribution entry.
3422 How many pieces of I/O were submitting in a single submit call. Each
3423 entry denotes that amount and below, until the previous entry -- e.g.,
3424 16=100% means that we submitted anywhere between 9 to 16 I/Os per submit
3425 call. Note that the range covered by a submit distribution entry can
3426 be different to the range covered by the equivalent depth distribution
3430 Like the above submit number, but for completions instead.
3433 The number of read/write/trim requests issued, and how many of them were
3437 These values are for :option:`latency_target` and related options. When
3438 these options are engaged, this section describes the I/O depth required
3439 to meet the specified latency target.
3442 Example output was based on the following:
3443 TZ=UTC fio --ioengine=null --iodepth=2 --size=100M --numjobs=2 \
3444 --rate_process=poisson --io_limit=32M --name=read --bs=128k \
3445 --rate=11M --name=write --rw=write --bs=2k --rate=700k
3447 After each client has been listed, the group statistics are printed. They
3448 will look like this::
3450 Run status group 0 (all jobs):
3451 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
3452 WRITE: bw=1231KiB/s (1261kB/s), 616KiB/s-621KiB/s (630kB/s-636kB/s), io=64.0MiB (67.1MB), run=52747-53223msec
3454 For each data direction it prints:
3457 Aggregate bandwidth of threads in this group followed by the
3458 minimum and maximum bandwidth of all the threads in this group.
3459 Values outside of brackets are power-of-2 format and those
3460 within are the equivalent value in a power-of-10 format.
3462 Aggregate I/O performed of all threads in this group. The
3463 format is the same as bw.
3465 The smallest and longest runtimes of the threads in this group.
3467 And finally, the disk statistics are printed. This is Linux specific. They will look like this::
3469 Disk stats (read/write):
3470 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3472 Each value is printed for both reads and writes, with reads first. The
3476 Number of I/Os performed by all groups.
3478 Number of merges performed by the I/O scheduler.
3480 Number of ticks we kept the disk busy.
3482 Total time spent in the disk queue.
3484 The disk utilization. A value of 100% means we kept the disk
3485 busy constantly, 50% would be a disk idling half of the time.
3487 It is also possible to get fio to dump the current output while it is running,
3488 without terminating the job. To do that, send fio the **USR1** signal. You can
3489 also get regularly timed dumps by using the :option:`--status-interval`
3490 parameter, or by creating a file in :file:`/tmp` named
3491 :file:`fio-dump-status`. If fio sees this file, it will unlink it and dump the
3492 current output status.
3498 For scripted usage where you typically want to generate tables or graphs of the
3499 results, fio can output the results in a semicolon separated format. The format
3500 is one long line of values, such as::
3502 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%
3503 A description of this job goes here.
3505 The job description (if provided) follows on a second line.
3507 To enable terse output, use the :option:`--minimal` or
3508 :option:`--output-format`\=terse command line options. The
3509 first value is the version of the terse output format. If the output has to be
3510 changed for some reason, this number will be incremented by 1 to signify that
3513 Split up, the format is as follows (comments in brackets denote when a
3514 field was introduced or whether it's specific to some terse version):
3518 terse version, fio version [v3], jobname, groupid, error
3522 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3523 Submission latency: min, max, mean, stdev (usec)
3524 Completion latency: min, max, mean, stdev (usec)
3525 Completion latency percentiles: 20 fields (see below)
3526 Total latency: min, max, mean, stdev (usec)
3527 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3528 IOPS [v5]: min, max, mean, stdev, number of samples
3534 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3535 Submission latency: min, max, mean, stdev (usec)
3536 Completion latency: min, max, mean, stdev (usec)
3537 Completion latency percentiles: 20 fields (see below)
3538 Total latency: min, max, mean, stdev (usec)
3539 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3540 IOPS [v5]: min, max, mean, stdev, number of samples
3542 TRIM status [all but version 3]:
3544 Fields are similar to READ/WRITE status.
3548 user, system, context switches, major faults, minor faults
3552 <=1, 2, 4, 8, 16, 32, >=64
3554 I/O latencies microseconds::
3556 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3558 I/O latencies milliseconds::
3560 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3562 Disk utilization [v3]::
3564 disk name, read ios, write ios, read merges, write merges, read ticks, write ticks,
3565 time spent in queue, disk utilization percentage
3567 Additional Info (dependent on continue_on_error, default off)::
3569 total # errors, first error code
3571 Additional Info (dependent on description being set)::
3575 Completion latency percentiles can be a grouping of up to 20 sets, so for the
3576 terse output fio writes all of them. Each field will look like this::
3580 which is the Xth percentile, and the `usec` latency associated with it.
3582 For `Disk utilization`, all disks used by fio are shown. So for each disk there
3583 will be a disk utilization section.
3585 Below is a single line containing short names for each of the fields in the
3586 minimal output v3, separated by semicolons::
3588 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
3594 The `json` output format is intended to be both human readable and convenient
3595 for automated parsing. For the most part its sections mirror those of the
3596 `normal` output. The `runtime` value is reported in msec and the `bw` value is
3597 reported in 1024 bytes per second units.
3603 The `json+` output format is identical to the `json` output format except that it
3604 adds a full dump of the completion latency bins. Each `bins` object contains a
3605 set of (key, value) pairs where keys are latency durations and values count how
3606 many I/Os had completion latencies of the corresponding duration. For example,
3609 "bins" : { "87552" : 1, "89600" : 1, "94720" : 1, "96768" : 1, "97792" : 1, "99840" : 1, "100864" : 2, "103936" : 6, "104960" : 534, "105984" : 5995, "107008" : 7529, ... }
3611 This data indicates that one I/O required 87,552ns to complete, two I/Os required
3612 100,864ns to complete, and 7529 I/Os required 107,008ns to complete.
3614 Also included with fio is a Python script `fio_jsonplus_clat2csv` that takes
3615 json+ output and generates CSV-formatted latency data suitable for plotting.
3617 The latency durations actually represent the midpoints of latency intervals.
3618 For details refer to :file:`stat.h`.
3624 There are two trace file format that you can encounter. The older (v1) format is
3625 unsupported since version 1.20-rc3 (March 2008). It will still be described
3626 below in case that you get an old trace and want to understand it.
3628 In any case the trace is a simple text file with a single action per line.
3631 Trace file format v1
3632 ~~~~~~~~~~~~~~~~~~~~
3634 Each line represents a single I/O action in the following format::
3638 where `rw=0/1` for read/write, and the `offset` and `length` entries being in bytes.
3640 This format is not supported in fio versions >= 1.20-rc3.
3643 Trace file format v2
3644 ~~~~~~~~~~~~~~~~~~~~
3646 The second version of the trace file format was added in fio version 1.17. It
3647 allows to access more then one file per trace and has a bigger set of possible
3650 The first line of the trace file has to be::
3654 Following this can be lines in two different formats, which are described below.
3656 The file management format::
3660 The `filename` is given as an absolute path. The `action` can be one of these:
3663 Add the given `filename` to the trace.
3665 Open the file with the given `filename`. The `filename` has to have
3666 been added with the **add** action before.
3668 Close the file with the given `filename`. The file has to have been
3672 The file I/O action format::
3674 filename action offset length
3676 The `filename` is given as an absolute path, and has to have been added and
3677 opened before it can be used with this format. The `offset` and `length` are
3678 given in bytes. The `action` can be one of these:
3681 Wait for `offset` microseconds. Everything below 100 is discarded.
3682 The time is relative to the previous `wait` statement.
3684 Read `length` bytes beginning from `offset`.
3686 Write `length` bytes beginning from `offset`.
3688 :manpage:`fsync(2)` the file.
3690 :manpage:`fdatasync(2)` the file.
3692 Trim the given file from the given `offset` for `length` bytes.
3694 CPU idleness profiling
3695 ----------------------
3697 In some cases, we want to understand CPU overhead in a test. For example, we
3698 test patches for the specific goodness of whether they reduce CPU usage.
3699 Fio implements a balloon approach to create a thread per CPU that runs at idle
3700 priority, meaning that it only runs when nobody else needs the cpu.
3701 By measuring the amount of work completed by the thread, idleness of each CPU
3702 can be derived accordingly.
3704 An unit work is defined as touching a full page of unsigned characters. Mean and
3705 standard deviation of time to complete an unit work is reported in "unit work"
3706 section. Options can be chosen to report detailed percpu idleness or overall
3707 system idleness by aggregating percpu stats.
3710 Verification and triggers
3711 -------------------------
3713 Fio is usually run in one of two ways, when data verification is done. The first
3714 is a normal write job of some sort with verify enabled. When the write phase has
3715 completed, fio switches to reads and verifies everything it wrote. The second
3716 model is running just the write phase, and then later on running the same job
3717 (but with reads instead of writes) to repeat the same I/O patterns and verify
3718 the contents. Both of these methods depend on the write phase being completed,
3719 as fio otherwise has no idea how much data was written.
3721 With verification triggers, fio supports dumping the current write state to
3722 local files. Then a subsequent read verify workload can load this state and know
3723 exactly where to stop. This is useful for testing cases where power is cut to a
3724 server in a managed fashion, for instance.
3726 A verification trigger consists of two things:
3728 1) Storing the write state of each job.
3729 2) Executing a trigger command.
3731 The write state is relatively small, on the order of hundreds of bytes to single
3732 kilobytes. It contains information on the number of completions done, the last X
3735 A trigger is invoked either through creation ('touch') of a specified file in
3736 the system, or through a timeout setting. If fio is run with
3737 :option:`--trigger-file`\= :file:`/tmp/trigger-file`, then it will continually
3738 check for the existence of :file:`/tmp/trigger-file`. When it sees this file, it
3739 will fire off the trigger (thus saving state, and executing the trigger
3742 For client/server runs, there's both a local and remote trigger. If fio is
3743 running as a server backend, it will send the job states back to the client for
3744 safe storage, then execute the remote trigger, if specified. If a local trigger
3745 is specified, the server will still send back the write state, but the client
3746 will then execute the trigger.
3748 Verification trigger example
3749 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3751 Let's say we want to run a powercut test on the remote Linux machine 'server'.
3752 Our write workload is in :file:`write-test.fio`. We want to cut power to 'server' at
3753 some point during the run, and we'll run this test from the safety or our local
3754 machine, 'localbox'. On the server, we'll start the fio backend normally::
3756 server# fio --server
3758 and on the client, we'll fire off the workload::
3760 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
3762 We set :file:`/tmp/my-trigger` as the trigger file, and we tell fio to execute::
3764 echo b > /proc/sysrq-trigger
3766 on the server once it has received the trigger and sent us the write state. This
3767 will work, but it's not **really** cutting power to the server, it's merely
3768 abruptly rebooting it. If we have a remote way of cutting power to the server
3769 through IPMI or similar, we could do that through a local trigger command
3770 instead. Let's assume we have a script that does IPMI reboot of a given hostname,
3771 ipmi-reboot. On localbox, we could then have run fio with a local trigger
3774 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
3776 For this case, fio would wait for the server to send us the write state, then
3777 execute ``ipmi-reboot server`` when that happened.
3779 Loading verify state
3780 ~~~~~~~~~~~~~~~~~~~~
3782 To load stored write state, a read verification job file must contain the
3783 :option:`verify_state_load` option. If that is set, fio will load the previously
3784 stored state. For a local fio run this is done by loading the files directly,
3785 and on a client/server run, the server backend will ask the client to send the
3786 files over and load them from there.
3792 Fio supports a variety of log file formats, for logging latencies, bandwidth,
3793 and IOPS. The logs share a common format, which looks like this:
3795 *time* (`msec`), *value*, *data direction*, *block size* (`bytes`),
3798 *Time* for the log entry is always in milliseconds. The *value* logged depends
3799 on the type of log, it will be one of the following:
3802 Value is latency in nsecs
3808 *Data direction* is one of the following:
3817 The entry's *block size* is always in bytes. The *offset* is the position in bytes
3818 from the start of the file for that particular I/O. The logging of the offset can be
3819 toggled with :option:`log_offset`.
3821 Fio defaults to logging every individual I/O but when windowed logging is set
3822 through :option:`log_avg_msec`, either the average (by default) or the maximum
3823 (:option:`log_max_value` is set) *value* seen over the specified period of time
3824 is recorded. Each *data direction* seen within the window period will aggregate
3825 its values in a separate row. Further, when using windowed logging the *block
3826 size* and *offset* entries will always contain 0.
3831 Normally fio is invoked as a stand-alone application on the machine where the
3832 I/O workload should be generated. However, the backend and frontend of fio can
3833 be run separately i.e., the fio server can generate an I/O workload on the "Device
3834 Under Test" while being controlled by a client on another machine.
3836 Start the server on the machine which has access to the storage DUT::
3840 where `args` defines what fio listens to. The arguments are of the form
3841 ``type,hostname`` or ``IP,port``. *type* is either ``ip`` (or ip4) for TCP/IP
3842 v4, ``ip6`` for TCP/IP v6, or ``sock`` for a local unix domain socket.
3843 *hostname* is either a hostname or IP address, and *port* is the port to listen
3844 to (only valid for TCP/IP, not a local socket). Some examples:
3848 Start a fio server, listening on all interfaces on the default port (8765).
3850 2) ``fio --server=ip:hostname,4444``
3852 Start a fio server, listening on IP belonging to hostname and on port 4444.
3854 3) ``fio --server=ip6:::1,4444``
3856 Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
3858 4) ``fio --server=,4444``
3860 Start a fio server, listening on all interfaces on port 4444.
3862 5) ``fio --server=1.2.3.4``
3864 Start a fio server, listening on IP 1.2.3.4 on the default port.
3866 6) ``fio --server=sock:/tmp/fio.sock``
3868 Start a fio server, listening on the local socket :file:`/tmp/fio.sock`.
3870 Once a server is running, a "client" can connect to the fio server with::
3872 fio <local-args> --client=<server> <remote-args> <job file(s)>
3874 where `local-args` are arguments for the client where it is running, `server`
3875 is the connect string, and `remote-args` and `job file(s)` are sent to the
3876 server. The `server` string follows the same format as it does on the server
3877 side, to allow IP/hostname/socket and port strings.
3879 Fio can connect to multiple servers this way::
3881 fio --client=<server1> <job file(s)> --client=<server2> <job file(s)>
3883 If the job file is located on the fio server, then you can tell the server to
3884 load a local file as well. This is done by using :option:`--remote-config` ::
3886 fio --client=server --remote-config /path/to/file.fio
3888 Then fio will open this local (to the server) job file instead of being passed
3889 one from the client.
3891 If you have many servers (example: 100 VMs/containers), you can input a pathname
3892 of a file containing host IPs/names as the parameter value for the
3893 :option:`--client` option. For example, here is an example :file:`host.list`
3894 file containing 2 hostnames::
3896 host1.your.dns.domain
3897 host2.your.dns.domain
3899 The fio command would then be::
3901 fio --client=host.list <job file(s)>
3903 In this mode, you cannot input server-specific parameters or job files -- all
3904 servers receive the same job file.
3906 In order to let ``fio --client`` runs use a shared filesystem from multiple
3907 hosts, ``fio --client`` now prepends the IP address of the server to the
3908 filename. For example, if fio is using the directory :file:`/mnt/nfs/fio` and is
3909 writing filename :file:`fileio.tmp`, with a :option:`--client` `hostfile`
3910 containing two hostnames ``h1`` and ``h2`` with IP addresses 192.168.10.120 and
3911 192.168.10.121, then fio will create two files::
3913 /mnt/nfs/fio/192.168.10.120.fileio.tmp
3914 /mnt/nfs/fio/192.168.10.121.fileio.tmp