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:: --merge-blktrace-only
105 Merge blktraces only, don't start any I/O.
107 .. option:: --output=filename
109 Write output to file `filename`.
111 .. option:: --output-format=format
113 Set the reporting `format` to `normal`, `terse`, `json`, or `json+`. Multiple
114 formats can be selected, separated by a comma. `terse` is a CSV based
115 format. `json+` is like `json`, except it adds a full dump of the latency
118 .. option:: --bandwidth-log
120 Generate aggregate bandwidth logs.
122 .. option:: --minimal
124 Print statistics in a terse, semicolon-delimited format.
126 .. option:: --append-terse
128 Print statistics in selected mode AND terse, semicolon-delimited format.
129 **Deprecated**, use :option:`--output-format` instead to select multiple
132 .. option:: --terse-version=version
134 Set terse `version` output format (default 3, or 2 or 4 or 5).
136 .. option:: --version
138 Print version information and exit.
142 Print a summary of the command line options and exit.
144 .. option:: --cpuclock-test
146 Perform test and validation of internal CPU clock.
148 .. option:: --crctest=[test]
150 Test the speed of the built-in checksumming functions. If no argument is
151 given, all of them are tested. Alternatively, a comma separated list can
152 be passed, in which case the given ones are tested.
154 .. option:: --cmdhelp=command
156 Print help information for `command`. May be ``all`` for all commands.
158 .. option:: --enghelp=[ioengine[,command]]
160 List all commands defined by `ioengine`, or print help for `command`
161 defined by `ioengine`. If no `ioengine` is given, list all
164 .. option:: --showcmd=jobfile
166 Convert `jobfile` to a set of command-line options.
168 .. option:: --readonly
170 Turn on safety read-only checks, preventing writes and trims. The
171 ``--readonly`` option is an extra safety guard to prevent users from
172 accidentally starting a write or trim workload when that is not desired.
173 Fio will only modify the device under test if
174 `rw=write/randwrite/rw/randrw/trim/randtrim/trimwrite` is given. This
175 safety net can be used as an extra precaution.
177 .. option:: --eta=when
179 Specifies when real-time ETA estimate should be printed. `when` may be
180 `always`, `never` or `auto`. `auto` is the default, it prints ETA
181 when requested if the output is a TTY. `always` disregards the output
182 type, and prints ETA when requested. `never` never prints ETA.
184 .. option:: --eta-interval=time
186 By default, fio requests client ETA status roughly every second. With
187 this option, the interval is configurable. Fio imposes a minimum
188 allowed time to avoid flooding the console, less than 250 msec is
191 .. option:: --eta-newline=time
193 Force a new line for every `time` period passed. When the unit is omitted,
194 the value is interpreted in seconds.
196 .. option:: --status-interval=time
198 Force a full status dump of cumulative (from job start) values at `time`
199 intervals. This option does *not* provide per-period measurements. So
200 values such as bandwidth are running averages. When the time unit is omitted,
201 `time` is interpreted in seconds. Note that using this option with
202 ``--output-format=json`` will yield output that technically isn't valid
203 json, since the output will be collated sets of valid json. It will need
204 to be split into valid sets of json after the run.
206 .. option:: --section=name
208 Only run specified section `name` in job file. Multiple sections can be specified.
209 The ``--section`` option allows one to combine related jobs into one file.
210 E.g. one job file could define light, moderate, and heavy sections. Tell
211 fio to run only the "heavy" section by giving ``--section=heavy``
212 command line option. One can also specify the "write" operations in one
213 section and "verify" operation in another section. The ``--section`` option
214 only applies to job sections. The reserved *global* section is always
217 .. option:: --alloc-size=kb
219 Set the internal smalloc pool size to `kb` in KiB. The
220 ``--alloc-size`` switch allows one to use a larger pool size for smalloc.
221 If running large jobs with randommap enabled, fio can run out of memory.
222 Smalloc is an internal allocator for shared structures from a fixed size
223 memory pool and can grow to 16 pools. The pool size defaults to 16MiB.
225 NOTE: While running :file:`.fio_smalloc.*` backing store files are visible
228 .. option:: --warnings-fatal
230 All fio parser warnings are fatal, causing fio to exit with an
233 .. option:: --max-jobs=nr
235 Set the maximum number of threads/processes to support to `nr`.
236 NOTE: On Linux, it may be necessary to increase the shared-memory
237 limit (:file:`/proc/sys/kernel/shmmax`) if fio runs into errors while
240 .. option:: --server=args
242 Start a backend server, with `args` specifying what to listen to.
243 See `Client/Server`_ section.
245 .. option:: --daemonize=pidfile
247 Background a fio server, writing the pid to the given `pidfile` file.
249 .. option:: --client=hostname
251 Instead of running the jobs locally, send and run them on the given `hostname`
252 or set of `hostname`\s. See `Client/Server`_ section.
254 .. option:: --remote-config=file
256 Tell fio server to load this local `file`.
258 .. option:: --idle-prof=option
260 Report CPU idleness. `option` is one of the following:
263 Run unit work calibration only and exit.
266 Show aggregate system idleness and unit work.
269 As **system** but also show per CPU idleness.
271 .. option:: --inflate-log=log
273 Inflate and output compressed `log`.
275 .. option:: --trigger-file=file
277 Execute trigger command when `file` exists.
279 .. option:: --trigger-timeout=time
281 Execute trigger at this `time`.
283 .. option:: --trigger=command
285 Set this `command` as local trigger.
287 .. option:: --trigger-remote=command
289 Set this `command` as remote trigger.
291 .. option:: --aux-path=path
293 Use the directory specified by `path` for generated state files instead
294 of the current working directory.
296 Any parameters following the options will be assumed to be job files, unless
297 they match a job file parameter. Multiple job files can be listed and each job
298 file will be regarded as a separate group. Fio will :option:`stonewall`
299 execution between each group.
305 As previously described, fio accepts one or more job files describing what it is
306 supposed to do. The job file format is the classic ini file, where the names
307 enclosed in [] brackets define the job name. You are free to use any ASCII name
308 you want, except *global* which has special meaning. Following the job name is
309 a sequence of zero or more parameters, one per line, that define the behavior of
310 the job. If the first character in a line is a ';' or a '#', the entire line is
311 discarded as a comment.
313 A *global* section sets defaults for the jobs described in that file. A job may
314 override a *global* section parameter, and a job file may even have several
315 *global* sections if so desired. A job is only affected by a *global* section
318 The :option:`--cmdhelp` option also lists all options. If used with a `command`
319 argument, :option:`--cmdhelp` will detail the given `command`.
321 See the `examples/` directory for inspiration on how to write job files. Note
322 the copyright and license requirements currently apply to `examples/` files.
324 So let's look at a really simple job file that defines two processes, each
325 randomly reading from a 128MiB file:
329 ; -- start job file --
340 As you can see, the job file sections themselves are empty as all the described
341 parameters are shared. As no :option:`filename` option is given, fio makes up a
342 `filename` for each of the jobs as it sees fit. On the command line, this job
343 would look as follows::
345 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
348 Let's look at an example that has a number of processes writing randomly to
353 ; -- start job file --
364 Here we have no *global* section, as we only have one job defined anyway. We
365 want to use async I/O here, with a depth of 4 for each file. We also increased
366 the buffer size used to 32KiB and define numjobs to 4 to fork 4 identical
367 jobs. The result is 4 processes each randomly writing to their own 64MiB
368 file. Instead of using the above job file, you could have given the parameters
369 on the command line. For this case, you would specify::
371 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
373 When fio is utilized as a basis of any reasonably large test suite, it might be
374 desirable to share a set of standardized settings across multiple job files.
375 Instead of copy/pasting such settings, any section may pull in an external
376 :file:`filename.fio` file with *include filename* directive, as in the following
379 ; -- start job file including.fio --
383 include glob-include.fio
390 include test-include.fio
391 ; -- end job file including.fio --
395 ; -- start job file glob-include.fio --
398 ; -- end job file glob-include.fio --
402 ; -- start job file test-include.fio --
405 ; -- end job file test-include.fio --
407 Settings pulled into a section apply to that section only (except *global*
408 section). Include directives may be nested in that any included file may contain
409 further include directive(s). Include files may not contain [] sections.
412 Environment variables
413 ~~~~~~~~~~~~~~~~~~~~~
415 Fio also supports environment variable expansion in job files. Any sub-string of
416 the form ``${VARNAME}`` as part of an option value (in other words, on the right
417 of the '='), will be expanded to the value of the environment variable called
418 `VARNAME`. If no such environment variable is defined, or `VARNAME` is the
419 empty string, the empty string will be substituted.
421 As an example, let's look at a sample fio invocation and job file::
423 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
427 ; -- start job file --
434 This will expand to the following equivalent job file at runtime:
438 ; -- start job file --
445 Fio ships with a few example job files, you can also look there for inspiration.
450 Additionally, fio has a set of reserved keywords that will be replaced
451 internally with the appropriate value. Those keywords are:
455 The architecture page size of the running system.
459 Megabytes of total memory in the system.
463 Number of online available CPUs.
465 These can be used on the command line or in the job file, and will be
466 automatically substituted with the current system values when the job is
467 run. Simple math is also supported on these keywords, so you can perform actions
472 and get that properly expanded to 8 times the size of memory in the machine.
478 This section describes in details each parameter associated with a job. Some
479 parameters take an option of a given type, such as an integer or a
480 string. Anywhere a numeric value is required, an arithmetic expression may be
481 used, provided it is surrounded by parentheses. Supported operators are:
490 For time values in expressions, units are microseconds by default. This is
491 different than for time values not in expressions (not enclosed in
492 parentheses). The following types are used:
499 String: A sequence of alphanumeric characters.
502 Integer with possible time suffix. Without a unit value is interpreted as
503 seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
504 hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and
505 'us' (or 'usec') for microseconds. For example, use 10m for 10 minutes.
510 Integer. A whole number value, which may contain an integer prefix
511 and an integer suffix:
513 [*integer prefix*] **number** [*integer suffix*]
515 The optional *integer prefix* specifies the number's base. The default
516 is decimal. *0x* specifies hexadecimal.
518 The optional *integer suffix* specifies the number's units, and includes an
519 optional unit prefix and an optional unit. For quantities of data, the
520 default unit is bytes. For quantities of time, the default unit is seconds
521 unless otherwise specified.
523 With :option:`kb_base`\=1000, fio follows international standards for unit
524 prefixes. To specify power-of-10 decimal values defined in the
525 International System of Units (SI):
527 * *K* -- means kilo (K) or 1000
528 * *M* -- means mega (M) or 1000**2
529 * *G* -- means giga (G) or 1000**3
530 * *T* -- means tera (T) or 1000**4
531 * *P* -- means peta (P) or 1000**5
533 To specify power-of-2 binary values defined in IEC 80000-13:
535 * *Ki* -- means kibi (Ki) or 1024
536 * *Mi* -- means mebi (Mi) or 1024**2
537 * *Gi* -- means gibi (Gi) or 1024**3
538 * *Ti* -- means tebi (Ti) or 1024**4
539 * *Pi* -- means pebi (Pi) or 1024**5
541 With :option:`kb_base`\=1024 (the default), the unit prefixes are opposite
542 from those specified in the SI and IEC 80000-13 standards to provide
543 compatibility with old scripts. For example, 4k means 4096.
545 For quantities of data, an optional unit of 'B' may be included
546 (e.g., 'kB' is the same as 'k').
548 The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
549 not milli). 'b' and 'B' both mean byte, not bit.
551 Examples with :option:`kb_base`\=1000:
553 * *4 KiB*: 4096, 4096b, 4096B, 4ki, 4kib, 4kiB, 4Ki, 4KiB
554 * *1 MiB*: 1048576, 1mi, 1024ki
555 * *1 MB*: 1000000, 1m, 1000k
556 * *1 TiB*: 1099511627776, 1ti, 1024gi, 1048576mi
557 * *1 TB*: 1000000000, 1t, 1000m, 1000000k
559 Examples with :option:`kb_base`\=1024 (default):
561 * *4 KiB*: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
562 * *1 MiB*: 1048576, 1m, 1024k
563 * *1 MB*: 1000000, 1mi, 1000ki
564 * *1 TiB*: 1099511627776, 1t, 1024g, 1048576m
565 * *1 TB*: 1000000000, 1ti, 1000mi, 1000000ki
567 To specify times (units are not case sensitive):
571 * *M* -- means minutes
572 * *s* -- or sec means seconds (default)
573 * *ms* -- or *msec* means milliseconds
574 * *us* -- or *usec* means microseconds
576 If the option accepts an upper and lower range, use a colon ':' or
577 minus '-' to separate such values. See :ref:`irange <irange>`.
578 If the lower value specified happens to be larger than the upper value
579 the two values are swapped.
584 Boolean. Usually parsed as an integer, however only defined for
585 true and false (1 and 0).
590 Integer range with suffix. Allows value range to be given, such as
591 1024-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
592 option allows two sets of ranges, they can be specified with a ',' or '/'
593 delimiter: 1k-4k/8k-32k. Also see :ref:`int <int>`.
596 A list of floating point numbers, separated by a ':' character.
598 With the above in mind, here follows the complete list of fio job parameters.
604 .. option:: kb_base=int
606 Select the interpretation of unit prefixes in input parameters.
609 Inputs comply with IEC 80000-13 and the International
610 System of Units (SI). Use:
612 - power-of-2 values with IEC prefixes (e.g., KiB)
613 - power-of-10 values with SI prefixes (e.g., kB)
616 Compatibility mode (default). To avoid breaking old scripts:
618 - power-of-2 values with SI prefixes
619 - power-of-10 values with IEC prefixes
621 See :option:`bs` for more details on input parameters.
623 Outputs always use correct prefixes. Most outputs include both
626 bw=2383.3kB/s (2327.4KiB/s)
628 If only one value is reported, then kb_base selects the one to use:
630 **1000** -- SI prefixes
632 **1024** -- IEC prefixes
634 .. option:: unit_base=int
636 Base unit for reporting. Allowed values are:
639 Use auto-detection (default).
651 ASCII name of the job. This may be used to override the name printed by fio
652 for this job. Otherwise the job name is used. On the command line this
653 parameter has the special purpose of also signaling the start of a new job.
655 .. option:: description=str
657 Text description of the job. Doesn't do anything except dump this text
658 description when this job is run. It's not parsed.
660 .. option:: loops=int
662 Run the specified number of iterations of this job. Used to repeat the same
663 workload a given number of times. Defaults to 1.
665 .. option:: numjobs=int
667 Create the specified number of clones of this job. Each clone of job
668 is spawned as an independent thread or process. May be used to setup a
669 larger number of threads/processes doing the same thing. Each thread is
670 reported separately; to see statistics for all clones as a whole, use
671 :option:`group_reporting` in conjunction with :option:`new_group`.
672 See :option:`--max-jobs`. Default: 1.
675 Time related parameters
676 ~~~~~~~~~~~~~~~~~~~~~~~
678 .. option:: runtime=time
680 Tell fio to terminate processing after the specified period of time. It
681 can be quite hard to determine for how long a specified job will run, so
682 this parameter is handy to cap the total runtime to a given time. When
683 the unit is omitted, the value is interpreted in seconds.
685 .. option:: time_based
687 If set, fio will run for the duration of the :option:`runtime` specified
688 even if the file(s) are completely read or written. It will simply loop over
689 the same workload as many times as the :option:`runtime` allows.
691 .. option:: startdelay=irange(time)
693 Delay the start of job for the specified amount of time. Can be a single
694 value or a range. When given as a range, each thread will choose a value
695 randomly from within the range. Value is in seconds if a unit is omitted.
697 .. option:: ramp_time=time
699 If set, fio will run the specified workload for this amount of time before
700 logging any performance numbers. Useful for letting performance settle
701 before logging results, thus minimizing the runtime required for stable
702 results. Note that the ``ramp_time`` is considered lead in time for a job,
703 thus it will increase the total runtime if a special timeout or
704 :option:`runtime` is specified. When the unit is omitted, the value is
707 .. option:: clocksource=str
709 Use the given clocksource as the base of timing. The supported options are:
712 :manpage:`gettimeofday(2)`
715 :manpage:`clock_gettime(2)`
718 Internal CPU clock source
720 cpu is the preferred clocksource if it is reliable, as it is very fast (and
721 fio is heavy on time calls). Fio will automatically use this clocksource if
722 it's supported and considered reliable on the system it is running on,
723 unless another clocksource is specifically set. For x86/x86-64 CPUs, this
724 means supporting TSC Invariant.
726 .. option:: gtod_reduce=bool
728 Enable all of the :manpage:`gettimeofday(2)` reducing options
729 (:option:`disable_clat`, :option:`disable_slat`, :option:`disable_bw_measurement`) plus
730 reduce precision of the timeout somewhat to really shrink the
731 :manpage:`gettimeofday(2)` call count. With this option enabled, we only do
732 about 0.4% of the :manpage:`gettimeofday(2)` calls we would have done if all
733 time keeping was enabled.
735 .. option:: gtod_cpu=int
737 Sometimes it's cheaper to dedicate a single thread of execution to just
738 getting the current time. Fio (and databases, for instance) are very
739 intensive on :manpage:`gettimeofday(2)` calls. With this option, you can set
740 one CPU aside for doing nothing but logging current time to a shared memory
741 location. Then the other threads/processes that run I/O workloads need only
742 copy that segment, instead of entering the kernel with a
743 :manpage:`gettimeofday(2)` call. The CPU set aside for doing these time
744 calls will be excluded from other uses. Fio will manually clear it from the
745 CPU mask of other jobs.
751 .. option:: directory=str
753 Prefix filenames with this directory. Used to place files in a different
754 location than :file:`./`. You can specify a number of directories by
755 separating the names with a ':' character. These directories will be
756 assigned equally distributed to job clones created by :option:`numjobs` as
757 long as they are using generated filenames. If specific `filename(s)` are
758 set fio will use the first listed directory, and thereby matching the
759 `filename` semantic (which generates a file for each clone if not
760 specified, but lets all clones use the same file if set).
762 See the :option:`filename` option for information on how to escape "``:``" and
763 "``\``" characters within the directory path itself.
765 Note: To control the directory fio will use for internal state files
766 use :option:`--aux-path`.
768 .. option:: filename=str
770 Fio normally makes up a `filename` based on the job name, thread number, and
771 file number (see :option:`filename_format`). If you want to share files
772 between threads in a job or several
773 jobs with fixed file paths, specify a `filename` for each of them to override
774 the default. If the ioengine is file based, you can specify a number of files
775 by separating the names with a ':' colon. So if you wanted a job to open
776 :file:`/dev/sda` and :file:`/dev/sdb` as the two working files, you would use
777 ``filename=/dev/sda:/dev/sdb``. This also means that whenever this option is
778 specified, :option:`nrfiles` is ignored. The size of regular files specified
779 by this option will be :option:`size` divided by number of files unless an
780 explicit size is specified by :option:`filesize`.
782 Each colon and backslash in the wanted path must be escaped with a ``\``
783 character. For instance, if the path is :file:`/dev/dsk/foo@3,0:c` then you
784 would use ``filename=/dev/dsk/foo@3,0\:c`` and if the path is
785 :file:`F:\\filename` then you would use ``filename=F\:\\filename``.
787 On Windows, disk devices are accessed as :file:`\\\\.\\PhysicalDrive0` for
788 the first device, :file:`\\\\.\\PhysicalDrive1` for the second etc.
789 Note: Windows and FreeBSD prevent write access to areas
790 of the disk containing in-use data (e.g. filesystems).
792 The filename "`-`" is a reserved name, meaning *stdin* or *stdout*. Which
793 of the two depends on the read/write direction set.
795 .. option:: filename_format=str
797 If sharing multiple files between jobs, it is usually necessary to have fio
798 generate the exact names that you want. By default, fio will name a file
799 based on the default file format specification of
800 :file:`jobname.jobnumber.filenumber`. With this option, that can be
801 customized. Fio will recognize and replace the following keywords in this
805 The name of the worker thread or process.
807 The incremental number of the worker thread or process.
809 The incremental number of the file for that worker thread or
812 To have dependent jobs share a set of files, this option can be set to have
813 fio generate filenames that are shared between the two. For instance, if
814 :file:`testfiles.$filenum` is specified, file number 4 for any job will be
815 named :file:`testfiles.4`. The default of :file:`$jobname.$jobnum.$filenum`
816 will be used if no other format specifier is given.
818 If you specify a path then the directories will be created up to the
819 main directory for the file. So for example if you specify
820 ``filename_format=a/b/c/$jobnum`` then the directories a/b/c will be
821 created before the file setup part of the job. If you specify
822 :option:`directory` then the path will be relative that directory,
823 otherwise it is treated as the absolute path.
825 .. option:: unique_filename=bool
827 To avoid collisions between networked clients, fio defaults to prefixing any
828 generated filenames (with a directory specified) with the source of the
829 client connecting. To disable this behavior, set this option to 0.
831 .. option:: opendir=str
833 Recursively open any files below directory `str`.
835 .. option:: lockfile=str
837 Fio defaults to not locking any files before it does I/O to them. If a file
838 or file descriptor is shared, fio can serialize I/O to that file to make the
839 end result consistent. This is usual for emulating real workloads that share
840 files. The lock modes are:
843 No locking. The default.
845 Only one thread or process may do I/O at a time, excluding all
848 Read-write locking on the file. Many readers may
849 access the file at the same time, but writes get exclusive access.
851 .. option:: nrfiles=int
853 Number of files to use for this job. Defaults to 1. The size of files
854 will be :option:`size` divided by this unless explicit size is specified by
855 :option:`filesize`. Files are created for each thread separately, and each
856 file will have a file number within its name by default, as explained in
857 :option:`filename` section.
860 .. option:: openfiles=int
862 Number of files to keep open at the same time. Defaults to the same as
863 :option:`nrfiles`, can be set smaller to limit the number simultaneous
866 .. option:: file_service_type=str
868 Defines how fio decides which file from a job to service next. The following
872 Choose a file at random.
875 Round robin over opened files. This is the default.
878 Finish one file before moving on to the next. Multiple files can
879 still be open depending on :option:`openfiles`.
882 Use a *Zipf* distribution to decide what file to access.
885 Use a *Pareto* distribution to decide what file to access.
888 Use a *Gaussian* (normal) distribution to decide what file to
894 For *random*, *roundrobin*, and *sequential*, a postfix can be appended to
895 tell fio how many I/Os to issue before switching to a new file. For example,
896 specifying ``file_service_type=random:8`` would cause fio to issue
897 8 I/Os before selecting a new file at random. For the non-uniform
898 distributions, a floating point postfix can be given to influence how the
899 distribution is skewed. See :option:`random_distribution` for a description
900 of how that would work.
902 .. option:: ioscheduler=str
904 Attempt to switch the device hosting the file to the specified I/O scheduler
907 .. option:: create_serialize=bool
909 If true, serialize the file creation for the jobs. This may be handy to
910 avoid interleaving of data files, which may greatly depend on the filesystem
911 used and even the number of processors in the system. Default: true.
913 .. option:: create_fsync=bool
915 :manpage:`fsync(2)` the data file after creation. This is the default.
917 .. option:: create_on_open=bool
919 If true, don't pre-create files but allow the job's open() to create a file
920 when it's time to do I/O. Default: false -- pre-create all necessary files
923 .. option:: create_only=bool
925 If true, fio will only run the setup phase of the job. If files need to be
926 laid out or updated on disk, only that will be done -- the actual job contents
927 are not executed. Default: false.
929 .. option:: allow_file_create=bool
931 If true, fio is permitted to create files as part of its workload. If this
932 option is false, then fio will error out if
933 the files it needs to use don't already exist. Default: true.
935 .. option:: allow_mounted_write=bool
937 If this isn't set, fio will abort jobs that are destructive (e.g. that write)
938 to what appears to be a mounted device or partition. This should help catch
939 creating inadvertently destructive tests, not realizing that the test will
940 destroy data on the mounted file system. Note that some platforms don't allow
941 writing against a mounted device regardless of this option. Default: false.
943 .. option:: pre_read=bool
945 If this is given, files will be pre-read into memory before starting the
946 given I/O operation. This will also clear the :option:`invalidate` flag,
947 since it is pointless to pre-read and then drop the cache. This will only
948 work for I/O engines that are seek-able, since they allow you to read the
949 same data multiple times. Thus it will not work on non-seekable I/O engines
950 (e.g. network, splice). Default: false.
952 .. option:: unlink=bool
954 Unlink the job files when done. Not the default, as repeated runs of that
955 job would then waste time recreating the file set again and again. Default:
958 .. option:: unlink_each_loop=bool
960 Unlink job files after each iteration or loop. Default: false.
962 .. option:: zonemode=str
967 The :option:`zonerange`, :option:`zonesize` and
968 :option:`zoneskip` parameters are ignored.
970 I/O happens in a single zone until
971 :option:`zonesize` bytes have been transferred.
972 After that number of bytes has been
973 transferred processing of the next zone
976 Zoned block device mode. I/O happens
977 sequentially in each zone, even if random I/O
978 has been selected. Random I/O happens across
979 all zones instead of being restricted to a
980 single zone. The :option:`zoneskip` parameter
981 is ignored. :option:`zonerange` and
982 :option:`zonesize` must be identical.
984 .. option:: zonerange=int
986 Size of a single zone. See also :option:`zonesize` and
989 .. option:: zonesize=int
991 For :option:`zonemode` =strided, this is the number of bytes to
992 transfer before skipping :option:`zoneskip` bytes. If this parameter
993 is smaller than :option:`zonerange` then only a fraction of each zone
994 with :option:`zonerange` bytes will be accessed. If this parameter is
995 larger than :option:`zonerange` then each zone will be accessed
996 multiple times before skipping to the next zone.
998 For :option:`zonemode` =zbd, this is the size of a single zone. The
999 :option:`zonerange` parameter is ignored in this mode.
1001 .. option:: zoneskip=int
1003 For :option:`zonemode` =strided, the number of bytes to skip after
1004 :option:`zonesize` bytes of data have been transferred. This parameter
1005 must be zero for :option:`zonemode` =zbd.
1007 .. option:: read_beyond_wp=bool
1009 This parameter applies to :option:`zonemode` =zbd only.
1011 Zoned block devices are block devices that consist of multiple zones.
1012 Each zone has a type, e.g. conventional or sequential. A conventional
1013 zone can be written at any offset that is a multiple of the block
1014 size. Sequential zones must be written sequentially. The position at
1015 which a write must occur is called the write pointer. A zoned block
1016 device can be either drive managed, host managed or host aware. For
1017 host managed devices the host must ensure that writes happen
1018 sequentially. Fio recognizes host managed devices and serializes
1019 writes to sequential zones for these devices.
1021 If a read occurs in a sequential zone beyond the write pointer then
1022 the zoned block device will complete the read without reading any data
1023 from the storage medium. Since such reads lead to unrealistically high
1024 bandwidth and IOPS numbers fio only reads beyond the write pointer if
1025 explicitly told to do so. Default: false.
1027 .. option:: max_open_zones=int
1029 When running a random write test across an entire drive many more
1030 zones will be open than in a typical application workload. Hence this
1031 command line option that allows to limit the number of open zones. The
1032 number of open zones is defined as the number of zones to which write
1033 commands are issued.
1035 .. option:: zone_reset_threshold=float
1037 A number between zero and one that indicates the ratio of logical
1038 blocks with data to the total number of logical blocks in the test
1039 above which zones should be reset periodically.
1041 .. option:: zone_reset_frequency=float
1043 A number between zero and one that indicates how often a zone reset
1044 should be issued if the zone reset threshold has been exceeded. A zone
1045 reset is submitted after each (1 / zone_reset_frequency) write
1046 requests. This and the previous parameter can be used to simulate
1047 garbage collection activity.
1053 .. option:: direct=bool
1055 If value is true, use non-buffered I/O. This is usually O_DIRECT. Note that
1056 OpenBSD and ZFS on Solaris don't support direct I/O. On Windows the synchronous
1057 ioengines don't support direct I/O. Default: false.
1059 .. option:: atomic=bool
1061 If value is true, attempt to use atomic direct I/O. Atomic writes are
1062 guaranteed to be stable once acknowledged by the operating system. Only
1063 Linux supports O_ATOMIC right now.
1065 .. option:: buffered=bool
1067 If value is true, use buffered I/O. This is the opposite of the
1068 :option:`direct` option. Defaults to true.
1070 .. option:: readwrite=str, rw=str
1072 Type of I/O pattern. Accepted values are:
1079 Sequential trims (Linux block devices and SCSI
1080 character devices only).
1086 Random trims (Linux block devices and SCSI
1087 character devices only).
1089 Sequential mixed reads and writes.
1091 Random mixed reads and writes.
1093 Sequential trim+write sequences. Blocks will be trimmed first,
1094 then the same blocks will be written to.
1096 Fio defaults to read if the option is not specified. For the mixed I/O
1097 types, the default is to split them 50/50. For certain types of I/O the
1098 result may still be skewed a bit, since the speed may be different.
1100 It is possible to specify the number of I/Os to do before getting a new
1101 offset by appending ``:<nr>`` to the end of the string given. For a
1102 random read, it would look like ``rw=randread:8`` for passing in an offset
1103 modifier with a value of 8. If the suffix is used with a sequential I/O
1104 pattern, then the *<nr>* value specified will be **added** to the generated
1105 offset for each I/O turning sequential I/O into sequential I/O with holes.
1106 For instance, using ``rw=write:4k`` will skip 4k for every write. Also see
1107 the :option:`rw_sequencer` option.
1109 .. option:: rw_sequencer=str
1111 If an offset modifier is given by appending a number to the ``rw=<str>``
1112 line, then this option controls how that number modifies the I/O offset
1113 being generated. Accepted values are:
1116 Generate sequential offset.
1118 Generate the same offset.
1120 ``sequential`` is only useful for random I/O, where fio would normally
1121 generate a new random offset for every I/O. If you append e.g. 8 to randread,
1122 you would get a new random offset for every 8 I/Os. The result would be a
1123 seek for only every 8 I/Os, instead of for every I/O. Use ``rw=randread:8``
1124 to specify that. As sequential I/O is already sequential, setting
1125 ``sequential`` for that would not result in any differences. ``identical``
1126 behaves in a similar fashion, except it sends the same offset 8 number of
1127 times before generating a new offset.
1129 .. option:: unified_rw_reporting=bool
1131 Fio normally reports statistics on a per data direction basis, meaning that
1132 reads, writes, and trims are accounted and reported separately. If this
1133 option is set fio sums the results and report them as "mixed" instead.
1135 .. option:: randrepeat=bool
1137 Seed the random number generator used for random I/O patterns in a
1138 predictable way so the pattern is repeatable across runs. Default: true.
1140 .. option:: allrandrepeat=bool
1142 Seed all random number generators in a predictable way so results are
1143 repeatable across runs. Default: false.
1145 .. option:: randseed=int
1147 Seed the random number generators based on this seed value, to be able to
1148 control what sequence of output is being generated. If not set, the random
1149 sequence depends on the :option:`randrepeat` setting.
1151 .. option:: fallocate=str
1153 Whether pre-allocation is performed when laying down files.
1154 Accepted values are:
1157 Do not pre-allocate space.
1160 Use a platform's native pre-allocation call but fall back to
1161 **none** behavior if it fails/is not implemented.
1164 Pre-allocate via :manpage:`posix_fallocate(3)`.
1167 Pre-allocate via :manpage:`fallocate(2)` with
1168 FALLOC_FL_KEEP_SIZE set.
1171 Backward-compatible alias for **none**.
1174 Backward-compatible alias for **posix**.
1176 May not be available on all supported platforms. **keep** is only available
1177 on Linux. If using ZFS on Solaris this cannot be set to **posix**
1178 because ZFS doesn't support pre-allocation. Default: **native** if any
1179 pre-allocation methods are available, **none** if not.
1181 .. option:: fadvise_hint=str
1183 Use :manpage:`posix_fadvise(2)` or :manpage:`posix_fadvise(2)` to
1184 advise the kernel on what I/O patterns are likely to be issued.
1185 Accepted values are:
1188 Backwards-compatible hint for "no hint".
1191 Backwards compatible hint for "advise with fio workload type". This
1192 uses **FADV_RANDOM** for a random workload, and **FADV_SEQUENTIAL**
1193 for a sequential workload.
1196 Advise using **FADV_SEQUENTIAL**.
1199 Advise using **FADV_RANDOM**.
1201 .. option:: write_hint=str
1203 Use :manpage:`fcntl(2)` to advise the kernel what life time to expect
1204 from a write. Only supported on Linux, as of version 4.13. Accepted
1208 No particular life time associated with this file.
1211 Data written to this file has a short life time.
1214 Data written to this file has a medium life time.
1217 Data written to this file has a long life time.
1220 Data written to this file has a very long life time.
1222 The values are all relative to each other, and no absolute meaning
1223 should be associated with them.
1225 .. option:: offset=int
1227 Start I/O at the provided offset in the file, given as either a fixed size in
1228 bytes or a percentage. If a percentage is given, the generated offset will be
1229 aligned to the minimum ``blocksize`` or to the value of ``offset_align`` if
1230 provided. Data before the given offset will not be touched. This
1231 effectively caps the file size at `real_size - offset`. Can be combined with
1232 :option:`size` to constrain the start and end range of the I/O workload.
1233 A percentage can be specified by a number between 1 and 100 followed by '%',
1234 for example, ``offset=20%`` to specify 20%.
1236 .. option:: offset_align=int
1238 If set to non-zero value, the byte offset generated by a percentage ``offset``
1239 is aligned upwards to this value. Defaults to 0 meaning that a percentage
1240 offset is aligned to the minimum block size.
1242 .. option:: offset_increment=int
1244 If this is provided, then the real offset becomes `offset + offset_increment
1245 * thread_number`, where the thread number is a counter that starts at 0 and
1246 is incremented for each sub-job (i.e. when :option:`numjobs` option is
1247 specified). This option is useful if there are several jobs which are
1248 intended to operate on a file in parallel disjoint segments, with even
1249 spacing between the starting points.
1251 .. option:: number_ios=int
1253 Fio will normally perform I/Os until it has exhausted the size of the region
1254 set by :option:`size`, or if it exhaust the allocated time (or hits an error
1255 condition). With this setting, the range/size can be set independently of
1256 the number of I/Os to perform. When fio reaches this number, it will exit
1257 normally and report status. Note that this does not extend the amount of I/O
1258 that will be done, it will only stop fio if this condition is met before
1259 other end-of-job criteria.
1261 .. option:: fsync=int
1263 If writing to a file, issue an :manpage:`fsync(2)` (or its equivalent) of
1264 the dirty data for every number of blocks given. For example, if you give 32
1265 as a parameter, fio will sync the file after every 32 writes issued. If fio is
1266 using non-buffered I/O, we may not sync the file. The exception is the sg
1267 I/O engine, which synchronizes the disk cache anyway. Defaults to 0, which
1268 means fio does not periodically issue and wait for a sync to complete. Also
1269 see :option:`end_fsync` and :option:`fsync_on_close`.
1271 .. option:: fdatasync=int
1273 Like :option:`fsync` but uses :manpage:`fdatasync(2)` to only sync data and
1274 not metadata blocks. In Windows, FreeBSD, and DragonFlyBSD there is no
1275 :manpage:`fdatasync(2)` so this falls back to using :manpage:`fsync(2)`.
1276 Defaults to 0, which means fio does not periodically issue and wait for a
1277 data-only sync to complete.
1279 .. option:: write_barrier=int
1281 Make every `N-th` write a barrier write.
1283 .. option:: sync_file_range=str:int
1285 Use :manpage:`sync_file_range(2)` for every `int` number of write
1286 operations. Fio will track range of writes that have happened since the last
1287 :manpage:`sync_file_range(2)` call. `str` can currently be one or more of:
1290 SYNC_FILE_RANGE_WAIT_BEFORE
1292 SYNC_FILE_RANGE_WRITE
1294 SYNC_FILE_RANGE_WAIT_AFTER
1296 So if you do ``sync_file_range=wait_before,write:8``, fio would use
1297 ``SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE`` for every 8
1298 writes. Also see the :manpage:`sync_file_range(2)` man page. This option is
1301 .. option:: overwrite=bool
1303 If true, writes to a file will always overwrite existing data. If the file
1304 doesn't already exist, it will be created before the write phase begins. If
1305 the file exists and is large enough for the specified write phase, nothing
1306 will be done. Default: false.
1308 .. option:: end_fsync=bool
1310 If true, :manpage:`fsync(2)` file contents when a write stage has completed.
1313 .. option:: fsync_on_close=bool
1315 If true, fio will :manpage:`fsync(2)` a dirty file on close. This differs
1316 from :option:`end_fsync` in that it will happen on every file close, not
1317 just at the end of the job. Default: false.
1319 .. option:: rwmixread=int
1321 Percentage of a mixed workload that should be reads. Default: 50.
1323 .. option:: rwmixwrite=int
1325 Percentage of a mixed workload that should be writes. If both
1326 :option:`rwmixread` and :option:`rwmixwrite` is given and the values do not
1327 add up to 100%, the latter of the two will be used to override the
1328 first. This may interfere with a given rate setting, if fio is asked to
1329 limit reads or writes to a certain rate. If that is the case, then the
1330 distribution may be skewed. Default: 50.
1332 .. option:: random_distribution=str:float[,str:float][,str:float]
1334 By default, fio will use a completely uniform random distribution when asked
1335 to perform random I/O. Sometimes it is useful to skew the distribution in
1336 specific ways, ensuring that some parts of the data is more hot than others.
1337 fio includes the following distribution models:
1340 Uniform random distribution
1349 Normal (Gaussian) distribution
1352 Zoned random distribution
1355 Zone absolute random distribution
1357 When using a **zipf** or **pareto** distribution, an input value is also
1358 needed to define the access pattern. For **zipf**, this is the `Zipf
1359 theta`. For **pareto**, it's the `Pareto power`. Fio includes a test
1360 program, :command:`fio-genzipf`, that can be used visualize what the given input
1361 values will yield in terms of hit rates. If you wanted to use **zipf** with
1362 a `theta` of 1.2, you would use ``random_distribution=zipf:1.2`` as the
1363 option. If a non-uniform model is used, fio will disable use of the random
1364 map. For the **normal** distribution, a normal (Gaussian) deviation is
1365 supplied as a value between 0 and 100.
1367 For a **zoned** distribution, fio supports specifying percentages of I/O
1368 access that should fall within what range of the file or device. For
1369 example, given a criteria of:
1371 * 60% of accesses should be to the first 10%
1372 * 30% of accesses should be to the next 20%
1373 * 8% of accesses should be to the next 30%
1374 * 2% of accesses should be to the next 40%
1376 we can define that through zoning of the random accesses. For the above
1377 example, the user would do::
1379 random_distribution=zoned:60/10:30/20:8/30:2/40
1381 A **zoned_abs** distribution works exactly like the **zoned**, except
1382 that it takes absolute sizes. For example, let's say you wanted to
1383 define access according to the following criteria:
1385 * 60% of accesses should be to the first 20G
1386 * 30% of accesses should be to the next 100G
1387 * 10% of accesses should be to the next 500G
1389 we can define an absolute zoning distribution with:
1391 random_distribution=zoned_abs=60/20G:30/100G:10/500g
1393 For both **zoned** and **zoned_abs**, fio supports defining up to
1396 Similarly to how :option:`bssplit` works for setting ranges and
1397 percentages of block sizes. Like :option:`bssplit`, it's possible to
1398 specify separate zones for reads, writes, and trims. If just one set
1399 is given, it'll apply to all of them. This goes for both **zoned**
1400 **zoned_abs** distributions.
1402 .. option:: percentage_random=int[,int][,int]
1404 For a random workload, set how big a percentage should be random. This
1405 defaults to 100%, in which case the workload is fully random. It can be set
1406 from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1407 sequential. Any setting in between will result in a random mix of sequential
1408 and random I/O, at the given percentages. Comma-separated values may be
1409 specified for reads, writes, and trims as described in :option:`blocksize`.
1411 .. option:: norandommap
1413 Normally fio will cover every block of the file when doing random I/O. If
1414 this option is given, fio will just get a new random offset without looking
1415 at past I/O history. This means that some blocks may not be read or written,
1416 and that some blocks may be read/written more than once. If this option is
1417 used with :option:`verify` and multiple blocksizes (via :option:`bsrange`),
1418 only intact blocks are verified, i.e., partially-overwritten blocks are
1419 ignored. With an async I/O engine and an I/O depth > 1, it is possible for
1420 the same block to be overwritten, which can cause verification errors. Either
1421 do not use norandommap in this case, or also use the lfsr random generator.
1423 .. option:: softrandommap=bool
1425 See :option:`norandommap`. If fio runs with the random block map enabled and
1426 it fails to allocate the map, if this option is set it will continue without
1427 a random block map. As coverage will not be as complete as with random maps,
1428 this option is disabled by default.
1430 .. option:: random_generator=str
1432 Fio supports the following engines for generating I/O offsets for random I/O:
1435 Strong 2^88 cycle random number generator.
1437 Linear feedback shift register generator.
1439 Strong 64-bit 2^258 cycle random number generator.
1441 **tausworthe** is a strong random number generator, but it requires tracking
1442 on the side if we want to ensure that blocks are only read or written
1443 once. **lfsr** guarantees that we never generate the same offset twice, and
1444 it's also less computationally expensive. It's not a true random generator,
1445 however, though for I/O purposes it's typically good enough. **lfsr** only
1446 works with single block sizes, not with workloads that use multiple block
1447 sizes. If used with such a workload, fio may read or write some blocks
1448 multiple times. The default value is **tausworthe**, unless the required
1449 space exceeds 2^32 blocks. If it does, then **tausworthe64** is
1450 selected automatically.
1456 .. option:: blocksize=int[,int][,int], bs=int[,int][,int]
1458 The block size in bytes used for I/O units. Default: 4096. A single value
1459 applies to reads, writes, and trims. Comma-separated values may be
1460 specified for reads, writes, and trims. A value not terminated in a comma
1461 applies to subsequent types.
1466 means 256k for reads, writes and trims.
1469 means 8k for reads, 32k for writes and trims.
1472 means 8k for reads, 32k for writes, and default for trims.
1475 means default for reads, 8k for writes and trims.
1478 means default for reads, 8k for writes, and default for trims.
1480 .. option:: blocksize_range=irange[,irange][,irange], bsrange=irange[,irange][,irange]
1482 A range of block sizes in bytes for I/O units. The issued I/O unit will
1483 always be a multiple of the minimum size, unless
1484 :option:`blocksize_unaligned` is set.
1486 Comma-separated ranges may be specified for reads, writes, and trims as
1487 described in :option:`blocksize`.
1489 Example: ``bsrange=1k-4k,2k-8k``.
1491 .. option:: bssplit=str[,str][,str]
1493 Sometimes you want even finer grained control of the block sizes
1494 issued, not just an even split between them. This option allows you to
1495 weight various block sizes, so that you are able to define a specific
1496 amount of block sizes issued. The format for this option is::
1498 bssplit=blocksize/percentage:blocksize/percentage
1500 for as many block sizes as needed. So if you want to define a workload
1501 that has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would
1504 bssplit=4k/10:64k/50:32k/40
1506 Ordering does not matter. If the percentage is left blank, fio will
1507 fill in the remaining values evenly. So a bssplit option like this one::
1509 bssplit=4k/50:1k/:32k/
1511 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always
1512 add up to 100, if bssplit is given a range that adds up to more, it
1515 Comma-separated values may be specified for reads, writes, and trims as
1516 described in :option:`blocksize`.
1518 If you want a workload that has 50% 2k reads and 50% 4k reads, while
1519 having 90% 4k writes and 10% 8k writes, you would specify::
1521 bssplit=2k/50:4k/50,4k/90:8k/10
1523 Fio supports defining up to 64 different weights for each data
1526 .. option:: blocksize_unaligned, bs_unaligned
1528 If set, fio will issue I/O units with any size within
1529 :option:`blocksize_range`, not just multiples of the minimum size. This
1530 typically won't work with direct I/O, as that normally requires sector
1533 .. option:: bs_is_seq_rand=bool
1535 If this option is set, fio will use the normal read,write blocksize settings
1536 as sequential,random blocksize settings instead. Any random read or write
1537 will use the WRITE blocksize settings, and any sequential read or write will
1538 use the READ blocksize settings.
1540 .. option:: blockalign=int[,int][,int], ba=int[,int][,int]
1542 Boundary to which fio will align random I/O units. Default:
1543 :option:`blocksize`. Minimum alignment is typically 512b for using direct
1544 I/O, though it usually depends on the hardware block size. This option is
1545 mutually exclusive with using a random map for files, so it will turn off
1546 that option. Comma-separated values may be specified for reads, writes, and
1547 trims as described in :option:`blocksize`.
1553 .. option:: zero_buffers
1555 Initialize buffers with all zeros. Default: fill buffers with random data.
1557 .. option:: refill_buffers
1559 If this option is given, fio will refill the I/O buffers on every
1560 submit. Only makes sense if :option:`zero_buffers` isn't specified,
1561 naturally. Defaults to being unset i.e., the buffer is only filled at
1562 init time and the data in it is reused when possible but if any of
1563 :option:`verify`, :option:`buffer_compress_percentage` or
1564 :option:`dedupe_percentage` are enabled then `refill_buffers` is also
1565 automatically enabled.
1567 .. option:: scramble_buffers=bool
1569 If :option:`refill_buffers` is too costly and the target is using data
1570 deduplication, then setting this option will slightly modify the I/O buffer
1571 contents to defeat normal de-dupe attempts. This is not enough to defeat
1572 more clever block compression attempts, but it will stop naive dedupe of
1573 blocks. Default: true.
1575 .. option:: buffer_compress_percentage=int
1577 If this is set, then fio will attempt to provide I/O buffer content
1578 (on WRITEs) that compresses to the specified level. Fio does this by
1579 providing a mix of random data followed by fixed pattern data. The
1580 fixed pattern is either zeros, or the pattern specified by
1581 :option:`buffer_pattern`. If the `buffer_pattern` option is used, it
1582 might skew the compression ratio slightly. Setting
1583 `buffer_compress_percentage` to a value other than 100 will also
1584 enable :option:`refill_buffers` in order to reduce the likelihood that
1585 adjacent blocks are so similar that they over compress when seen
1586 together. See :option:`buffer_compress_chunk` for how to set a finer or
1587 coarser granularity for the random/fixed data region. Defaults to unset
1588 i.e., buffer data will not adhere to any compression level.
1590 .. option:: buffer_compress_chunk=int
1592 This setting allows fio to manage how big the random/fixed data region
1593 is when using :option:`buffer_compress_percentage`. When
1594 `buffer_compress_chunk` is set to some non-zero value smaller than the
1595 block size, fio can repeat the random/fixed region throughout the I/O
1596 buffer at the specified interval (which particularly useful when
1597 bigger block sizes are used for a job). When set to 0, fio will use a
1598 chunk size that matches the block size resulting in a single
1599 random/fixed region within the I/O buffer. Defaults to 512. When the
1600 unit is omitted, the value is interpreted in bytes.
1602 .. option:: buffer_pattern=str
1604 If set, fio will fill the I/O buffers with this pattern or with the contents
1605 of a file. If not set, the contents of I/O buffers are defined by the other
1606 options related to buffer contents. The setting can be any pattern of bytes,
1607 and can be prefixed with 0x for hex values. It may also be a string, where
1608 the string must then be wrapped with ``""``. Or it may also be a filename,
1609 where the filename must be wrapped with ``''`` in which case the file is
1610 opened and read. Note that not all the file contents will be read if that
1611 would cause the buffers to overflow. So, for example::
1613 buffer_pattern='filename'
1617 buffer_pattern="abcd"
1625 buffer_pattern=0xdeadface
1627 Also you can combine everything together in any order::
1629 buffer_pattern=0xdeadface"abcd"-12'filename'
1631 .. option:: dedupe_percentage=int
1633 If set, fio will generate this percentage of identical buffers when
1634 writing. These buffers will be naturally dedupable. The contents of the
1635 buffers depend on what other buffer compression settings have been set. It's
1636 possible to have the individual buffers either fully compressible, or not at
1637 all -- this option only controls the distribution of unique buffers. Setting
1638 this option will also enable :option:`refill_buffers` to prevent every buffer
1641 .. option:: invalidate=bool
1643 Invalidate the buffer/page cache parts of the files to be used prior to
1644 starting I/O if the platform and file type support it. Defaults to true.
1645 This will be ignored if :option:`pre_read` is also specified for the
1648 .. option:: sync=bool
1650 Use synchronous I/O for buffered writes. For the majority of I/O engines,
1651 this means using O_SYNC. Default: false.
1653 .. option:: iomem=str, mem=str
1655 Fio can use various types of memory as the I/O unit buffer. The allowed
1659 Use memory from :manpage:`malloc(3)` as the buffers. Default memory
1663 Use shared memory as the buffers. Allocated through
1664 :manpage:`shmget(2)`.
1667 Same as shm, but use huge pages as backing.
1670 Use :manpage:`mmap(2)` to allocate buffers. May either be anonymous memory, or can
1671 be file backed if a filename is given after the option. The format
1672 is `mem=mmap:/path/to/file`.
1675 Use a memory mapped huge file as the buffer backing. Append filename
1676 after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file`.
1679 Same as mmap, but use a MMAP_SHARED mapping.
1682 Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1683 The :option:`ioengine` must be `rdma`.
1685 The area allocated is a function of the maximum allowed bs size for the job,
1686 multiplied by the I/O depth given. Note that for **shmhuge** and
1687 **mmaphuge** to work, the system must have free huge pages allocated. This
1688 can normally be checked and set by reading/writing
1689 :file:`/proc/sys/vm/nr_hugepages` on a Linux system. Fio assumes a huge page
1690 is 4MiB in size. So to calculate the number of huge pages you need for a
1691 given job file, add up the I/O depth of all jobs (normally one unless
1692 :option:`iodepth` is used) and multiply by the maximum bs set. Then divide
1693 that number by the huge page size. You can see the size of the huge pages in
1694 :file:`/proc/meminfo`. If no huge pages are allocated by having a non-zero
1695 number in `nr_hugepages`, using **mmaphuge** or **shmhuge** will fail. Also
1696 see :option:`hugepage-size`.
1698 **mmaphuge** also needs to have hugetlbfs mounted and the file location
1699 should point there. So if it's mounted in :file:`/huge`, you would use
1700 `mem=mmaphuge:/huge/somefile`.
1702 .. option:: iomem_align=int, mem_align=int
1704 This indicates the memory alignment of the I/O memory buffers. Note that
1705 the given alignment is applied to the first I/O unit buffer, if using
1706 :option:`iodepth` the alignment of the following buffers are given by the
1707 :option:`bs` used. In other words, if using a :option:`bs` that is a
1708 multiple of the page sized in the system, all buffers will be aligned to
1709 this value. If using a :option:`bs` that is not page aligned, the alignment
1710 of subsequent I/O memory buffers is the sum of the :option:`iomem_align` and
1713 .. option:: hugepage-size=int
1715 Defines the size of a huge page. Must at least be equal to the system
1716 setting, see :file:`/proc/meminfo`. Defaults to 4MiB. Should probably
1717 always be a multiple of megabytes, so using ``hugepage-size=Xm`` is the
1718 preferred way to set this to avoid setting a non-pow-2 bad value.
1720 .. option:: lockmem=int
1722 Pin the specified amount of memory with :manpage:`mlock(2)`. Can be used to
1723 simulate a smaller amount of memory. The amount specified is per worker.
1729 .. option:: size=int
1731 The total size of file I/O for each thread of this job. Fio will run until
1732 this many bytes has been transferred, unless runtime is limited by other options
1733 (such as :option:`runtime`, for instance, or increased/decreased by :option:`io_size`).
1734 Fio will divide this size between the available files determined by options
1735 such as :option:`nrfiles`, :option:`filename`, unless :option:`filesize` is
1736 specified by the job. If the result of division happens to be 0, the size is
1737 set to the physical size of the given files or devices if they exist.
1738 If this option is not specified, fio will use the full size of the given
1739 files or devices. If the files do not exist, size must be given. It is also
1740 possible to give size as a percentage between 1 and 100. If ``size=20%`` is
1741 given, fio will use 20% of the full size of the given files or devices.
1742 Can be combined with :option:`offset` to constrain the start and end range
1743 that I/O will be done within.
1745 .. option:: io_size=int, io_limit=int
1747 Normally fio operates within the region set by :option:`size`, which means
1748 that the :option:`size` option sets both the region and size of I/O to be
1749 performed. Sometimes that is not what you want. With this option, it is
1750 possible to define just the amount of I/O that fio should do. For instance,
1751 if :option:`size` is set to 20GiB and :option:`io_size` is set to 5GiB, fio
1752 will perform I/O within the first 20GiB but exit when 5GiB have been
1753 done. The opposite is also possible -- if :option:`size` is set to 20GiB,
1754 and :option:`io_size` is set to 40GiB, then fio will do 40GiB of I/O within
1755 the 0..20GiB region.
1757 .. option:: filesize=irange(int)
1759 Individual file sizes. May be a range, in which case fio will select sizes
1760 for files at random within the given range and limited to :option:`size` in
1761 total (if that is given). If not given, each created file is the same size.
1762 This option overrides :option:`size` in terms of file size, which means
1763 this value is used as a fixed size or possible range of each file.
1765 .. option:: file_append=bool
1767 Perform I/O after the end of the file. Normally fio will operate within the
1768 size of a file. If this option is set, then fio will append to the file
1769 instead. This has identical behavior to setting :option:`offset` to the size
1770 of a file. This option is ignored on non-regular files.
1772 .. option:: fill_device=bool, fill_fs=bool
1774 Sets size to something really large and waits for ENOSPC (no space left on
1775 device) as the terminating condition. Only makes sense with sequential
1776 write. For a read workload, the mount point will be filled first then I/O
1777 started on the result. This option doesn't make sense if operating on a raw
1778 device node, since the size of that is already known by the file system.
1779 Additionally, writing beyond end-of-device will not return ENOSPC there.
1785 .. option:: ioengine=str
1787 Defines how the job issues I/O to the file. The following types are defined:
1790 Basic :manpage:`read(2)` or :manpage:`write(2)`
1791 I/O. :manpage:`lseek(2)` is used to position the I/O location.
1792 See :option:`fsync` and :option:`fdatasync` for syncing write I/Os.
1795 Basic :manpage:`pread(2)` or :manpage:`pwrite(2)` I/O. Default on
1796 all supported operating systems except for Windows.
1799 Basic :manpage:`readv(2)` or :manpage:`writev(2)` I/O. Will emulate
1800 queuing by coalescing adjacent I/Os into a single submission.
1803 Basic :manpage:`preadv(2)` or :manpage:`pwritev(2)` I/O.
1806 Basic :manpage:`preadv2(2)` or :manpage:`pwritev2(2)` I/O.
1809 Linux native asynchronous I/O. Note that Linux may only support
1810 queued behavior with non-buffered I/O (set ``direct=1`` or
1812 This engine defines engine specific options.
1815 POSIX asynchronous I/O using :manpage:`aio_read(3)` and
1816 :manpage:`aio_write(3)`.
1819 Solaris native asynchronous I/O.
1822 Windows native asynchronous I/O. Default on Windows.
1825 File is memory mapped with :manpage:`mmap(2)` and data copied
1826 to/from using :manpage:`memcpy(3)`.
1829 :manpage:`splice(2)` is used to transfer the data and
1830 :manpage:`vmsplice(2)` to transfer data from user space to the
1834 SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1835 ioctl, or if the target is an sg character device we use
1836 :manpage:`read(2)` and :manpage:`write(2)` for asynchronous
1837 I/O. Requires :option:`filename` option to specify either block or
1838 character devices. This engine supports trim operations.
1839 The sg engine includes engine specific options.
1842 Doesn't transfer any data, just pretends to. This is mainly used to
1843 exercise fio itself and for debugging/testing purposes.
1846 Transfer over the network to given ``host:port``. Depending on the
1847 :option:`protocol` used, the :option:`hostname`, :option:`port`,
1848 :option:`listen` and :option:`filename` options are used to specify
1849 what sort of connection to make, while the :option:`protocol` option
1850 determines which protocol will be used. This engine defines engine
1854 Like **net**, but uses :manpage:`splice(2)` and
1855 :manpage:`vmsplice(2)` to map data and send/receive.
1856 This engine defines engine specific options.
1859 Doesn't transfer any data, but burns CPU cycles according to the
1860 :option:`cpuload` and :option:`cpuchunks` options. Setting
1861 :option:`cpuload`\=85 will cause that job to do nothing but burn 85%
1862 of the CPU. In case of SMP machines, use :option:`numjobs`\=<nr_of_cpu>
1863 to get desired CPU usage, as the cpuload only loads a
1864 single CPU at the desired rate. A job never finishes unless there is
1865 at least one non-cpuio job.
1868 The GUASI I/O engine is the Generic Userspace Asynchronous Syscall
1869 Interface approach to async I/O. See
1871 http://www.xmailserver.org/guasi-lib.html
1873 for more info on GUASI.
1876 The RDMA I/O engine supports both RDMA memory semantics
1877 (RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
1878 InfiniBand, RoCE and iWARP protocols. This engine defines engine
1882 I/O engine that does regular fallocate to simulate data transfer as
1886 does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1889 does fallocate(,mode = 0).
1892 does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1895 I/O engine that sends :manpage:`ftruncate(2)` operations in response
1896 to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
1897 size to the current block offset. :option:`blocksize` is ignored.
1900 I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1901 defragment activity in request to DDIR_WRITE event.
1904 I/O engine supporting direct access to Ceph Reliable Autonomic
1905 Distributed Object Store (RADOS) via librados. This ioengine
1906 defines engine specific options.
1909 I/O engine supporting direct access to Ceph Rados Block Devices
1910 (RBD) via librbd without the need to use the kernel rbd driver. This
1911 ioengine defines engine specific options.
1914 I/O engine supporting GET/PUT requests over HTTP(S) with libcurl to
1915 a WebDAV or S3 endpoint. This ioengine defines engine specific options.
1917 This engine only supports direct IO of iodepth=1; you need to scale this
1918 via numjobs. blocksize defines the size of the objects to be created.
1920 TRIM is translated to object deletion.
1923 Using GlusterFS libgfapi sync interface to direct access to
1924 GlusterFS volumes without having to go through FUSE. This ioengine
1925 defines engine specific options.
1928 Using GlusterFS libgfapi async interface to direct access to
1929 GlusterFS volumes without having to go through FUSE. This ioengine
1930 defines engine specific options.
1933 Read and write through Hadoop (HDFS). The :option:`filename` option
1934 is used to specify host,port of the hdfs name-node to connect. This
1935 engine interprets offsets a little differently. In HDFS, files once
1936 created cannot be modified so random writes are not possible. To
1937 imitate this the libhdfs engine expects a bunch of small files to be
1938 created over HDFS and will randomly pick a file from them
1939 based on the offset generated by fio backend (see the example
1940 job file to create such files, use ``rw=write`` option). Please
1941 note, it may be necessary to set environment variables to work
1942 with HDFS/libhdfs properly. Each job uses its own connection to
1946 Read, write and erase an MTD character device (e.g.,
1947 :file:`/dev/mtd0`). Discards are treated as erases. Depending on the
1948 underlying device type, the I/O may have to go in a certain pattern,
1949 e.g., on NAND, writing sequentially to erase blocks and discarding
1950 before overwriting. The `trimwrite` mode works well for this
1954 Read and write using filesystem DAX to a file on a filesystem
1955 mounted with DAX on a persistent memory device through the PMDK
1959 Read and write using device DAX to a persistent memory device (e.g.,
1960 /dev/dax0.0) through the PMDK libpmem library.
1963 Prefix to specify loading an external I/O engine object file. Append
1964 the engine filename, e.g. ``ioengine=external:/tmp/foo.o`` to load
1965 ioengine :file:`foo.o` in :file:`/tmp`. The path can be either
1966 absolute or relative. See :file:`engines/skeleton_external.c` for
1967 details of writing an external I/O engine.
1970 Simply create the files and do no I/O to them. You still need to
1971 set `filesize` so that all the accounting still occurs, but no
1972 actual I/O will be done other than creating the file.
1975 Read and write using mmap I/O to a file on a filesystem
1976 mounted with DAX on a persistent memory device through the PMDK
1980 Synchronous read and write using DDN's Infinite Memory Engine (IME).
1981 This engine is very basic and issues calls to IME whenever an IO is
1985 Synchronous read and write using DDN's Infinite Memory Engine (IME).
1986 This engine uses iovecs and will try to stack as much IOs as possible
1987 (if the IOs are "contiguous" and the IO depth is not exceeded)
1988 before issuing a call to IME.
1991 Asynchronous read and write using DDN's Infinite Memory Engine (IME).
1992 This engine will try to stack as much IOs as possible by creating
1993 requests for IME. FIO will then decide when to commit these requests.
1995 Read and write iscsi lun with libiscsi.
1997 I/O engine specific parameters
1998 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2000 In addition, there are some parameters which are only valid when a specific
2001 :option:`ioengine` is in use. These are used identically to normal parameters,
2002 with the caveat that when used on the command line, they must come after the
2003 :option:`ioengine` that defines them is selected.
2005 .. option:: userspace_reap : [libaio]
2007 Normally, with the libaio engine in use, fio will use the
2008 :manpage:`io_getevents(2)` system call to reap newly returned events. With
2009 this flag turned on, the AIO ring will be read directly from user-space to
2010 reap events. The reaping mode is only enabled when polling for a minimum of
2011 0 events (e.g. when :option:`iodepth_batch_complete` `=0`).
2013 .. option:: hipri : [pvsync2]
2015 Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
2018 .. option:: hipri_percentage : [pvsync2]
2020 When hipri is set this determines the probability of a pvsync2 I/O being high
2021 priority. The default is 100%.
2023 .. option:: cpuload=int : [cpuio]
2025 Attempt to use the specified percentage of CPU cycles. This is a mandatory
2026 option when using cpuio I/O engine.
2028 .. option:: cpuchunks=int : [cpuio]
2030 Split the load into cycles of the given time. In microseconds.
2032 .. option:: exit_on_io_done=bool : [cpuio]
2034 Detect when I/O threads are done, then exit.
2036 .. option:: namenode=str : [libhdfs]
2038 The hostname or IP address of a HDFS cluster namenode to contact.
2040 .. option:: port=int
2044 The listening port of the HFDS cluster namenode.
2048 The TCP or UDP port to bind to or connect to. If this is used with
2049 :option:`numjobs` to spawn multiple instances of the same job type, then
2050 this will be the starting port number since fio will use a range of
2055 The port to use for RDMA-CM communication. This should be the same value
2056 on the client and the server side.
2058 .. option:: hostname=str : [netsplice] [net] [rdma]
2060 The hostname or IP address to use for TCP, UDP or RDMA-CM based I/O. If the job
2061 is a TCP listener or UDP reader, the hostname is not used and must be omitted
2062 unless it is a valid UDP multicast address.
2064 .. option:: interface=str : [netsplice] [net]
2066 The IP address of the network interface used to send or receive UDP
2069 .. option:: ttl=int : [netsplice] [net]
2071 Time-to-live value for outgoing UDP multicast packets. Default: 1.
2073 .. option:: nodelay=bool : [netsplice] [net]
2075 Set TCP_NODELAY on TCP connections.
2077 .. option:: protocol=str, proto=str : [netsplice] [net]
2079 The network protocol to use. Accepted values are:
2082 Transmission control protocol.
2084 Transmission control protocol V6.
2086 User datagram protocol.
2088 User datagram protocol V6.
2092 When the protocol is TCP or UDP, the port must also be given, as well as the
2093 hostname if the job is a TCP listener or UDP reader. For unix sockets, the
2094 normal :option:`filename` option should be used and the port is invalid.
2096 .. option:: listen : [netsplice] [net]
2098 For TCP network connections, tell fio to listen for incoming connections
2099 rather than initiating an outgoing connection. The :option:`hostname` must
2100 be omitted if this option is used.
2102 .. option:: pingpong : [netsplice] [net]
2104 Normally a network writer will just continue writing data, and a network
2105 reader will just consume packages. If ``pingpong=1`` is set, a writer will
2106 send its normal payload to the reader, then wait for the reader to send the
2107 same payload back. This allows fio to measure network latencies. The
2108 submission and completion latencies then measure local time spent sending or
2109 receiving, and the completion latency measures how long it took for the
2110 other end to receive and send back. For UDP multicast traffic
2111 ``pingpong=1`` should only be set for a single reader when multiple readers
2112 are listening to the same address.
2114 .. option:: window_size : [netsplice] [net]
2116 Set the desired socket buffer size for the connection.
2118 .. option:: mss : [netsplice] [net]
2120 Set the TCP maximum segment size (TCP_MAXSEG).
2122 .. option:: donorname=str : [e4defrag]
2124 File will be used as a block donor (swap extents between files).
2126 .. option:: inplace=int : [e4defrag]
2128 Configure donor file blocks allocation strategy:
2131 Default. Preallocate donor's file on init.
2133 Allocate space immediately inside defragment event, and free right
2136 .. option:: clustername=str : [rbd,rados]
2138 Specifies the name of the Ceph cluster.
2140 .. option:: rbdname=str : [rbd]
2142 Specifies the name of the RBD.
2144 .. option:: pool=str : [rbd,rados]
2146 Specifies the name of the Ceph pool containing RBD or RADOS data.
2148 .. option:: clientname=str : [rbd,rados]
2150 Specifies the username (without the 'client.' prefix) used to access the
2151 Ceph cluster. If the *clustername* is specified, the *clientname* shall be
2152 the full *type.id* string. If no type. prefix is given, fio will add
2153 'client.' by default.
2155 .. option:: busy_poll=bool : [rbd,rados]
2157 Poll store instead of waiting for completion. Usually this provides better
2158 throughput at cost of higher(up to 100%) CPU utilization.
2160 .. option:: skip_bad=bool : [mtd]
2162 Skip operations against known bad blocks.
2164 .. option:: hdfsdirectory : [libhdfs]
2166 libhdfs will create chunk in this HDFS directory.
2168 .. option:: chunk_size : [libhdfs]
2170 The size of the chunk to use for each file.
2172 .. option:: verb=str : [rdma]
2174 The RDMA verb to use on this side of the RDMA ioengine connection. Valid
2175 values are write, read, send and recv. These correspond to the equivalent
2176 RDMA verbs (e.g. write = rdma_write etc.). Note that this only needs to be
2177 specified on the client side of the connection. See the examples folder.
2179 .. option:: bindname=str : [rdma]
2181 The name to use to bind the local RDMA-CM connection to a local RDMA device.
2182 This could be a hostname or an IPv4 or IPv6 address. On the server side this
2183 will be passed into the rdma_bind_addr() function and on the client site it
2184 will be used in the rdma_resolve_add() function. This can be useful when
2185 multiple paths exist between the client and the server or in certain loopback
2188 .. option:: readfua=bool : [sg]
2190 With readfua option set to 1, read operations include
2191 the force unit access (fua) flag. Default is 0.
2193 .. option:: writefua=bool : [sg]
2195 With writefua option set to 1, write operations include
2196 the force unit access (fua) flag. Default is 0.
2198 .. option:: sg_write_mode=str : [sg]
2200 Specify the type of write commands to issue. This option can take three values:
2203 This is the default where write opcodes are issued as usual.
2205 Issue WRITE AND VERIFY commands. The BYTCHK bit is set to 0. This
2206 directs the device to carry out a medium verification with no data
2207 comparison. The writefua option is ignored with this selection.
2209 Issue WRITE SAME commands. This transfers a single block to the device
2210 and writes this same block of data to a contiguous sequence of LBAs
2211 beginning at the specified offset. fio's block size parameter specifies
2212 the amount of data written with each command. However, the amount of data
2213 actually transferred to the device is equal to the device's block
2214 (sector) size. For a device with 512 byte sectors, blocksize=8k will
2215 write 16 sectors with each command. fio will still generate 8k of data
2216 for each command but only the first 512 bytes will be used and
2217 transferred to the device. The writefua option is ignored with this
2220 .. option:: http_host=str : [http]
2222 Hostname to connect to. For S3, this could be the bucket hostname.
2223 Default is **localhost**
2225 .. option:: http_user=str : [http]
2227 Username for HTTP authentication.
2229 .. option:: http_pass=str : [http]
2231 Password for HTTP authentication.
2233 .. option:: https=str : [http]
2235 Enable HTTPS instead of http. *on* enables HTTPS; *insecure*
2236 will enable HTTPS, but disable SSL peer verification (use with
2237 caution!). Default is **off**
2239 .. option:: http_mode=str : [http]
2241 Which HTTP access mode to use: *webdav*, *swift*, or *s3*.
2242 Default is **webdav**
2244 .. option:: http_s3_region=str : [http]
2246 The S3 region/zone string.
2247 Default is **us-east-1**
2249 .. option:: http_s3_key=str : [http]
2253 .. option:: http_s3_keyid=str : [http]
2255 The S3 key/access id.
2257 .. option:: http_swift_auth_token=str : [http]
2259 The Swift auth token. See the example configuration file on how
2262 .. option:: http_verbose=int : [http]
2264 Enable verbose requests from libcurl. Useful for debugging. 1
2265 turns on verbose logging from libcurl, 2 additionally enables
2266 HTTP IO tracing. Default is **0**
2271 .. option:: iodepth=int
2273 Number of I/O units to keep in flight against the file. Note that
2274 increasing *iodepth* beyond 1 will not affect synchronous ioengines (except
2275 for small degrees when :option:`verify_async` is in use). Even async
2276 engines may impose OS restrictions causing the desired depth not to be
2277 achieved. This may happen on Linux when using libaio and not setting
2278 :option:`direct`\=1, since buffered I/O is not async on that OS. Keep an
2279 eye on the I/O depth distribution in the fio output to verify that the
2280 achieved depth is as expected. Default: 1.
2282 .. option:: iodepth_batch_submit=int, iodepth_batch=int
2284 This defines how many pieces of I/O to submit at once. It defaults to 1
2285 which means that we submit each I/O as soon as it is available, but can be
2286 raised to submit bigger batches of I/O at the time. If it is set to 0 the
2287 :option:`iodepth` value will be used.
2289 .. option:: iodepth_batch_complete_min=int, iodepth_batch_complete=int
2291 This defines how many pieces of I/O to retrieve at once. It defaults to 1
2292 which means that we'll ask for a minimum of 1 I/O in the retrieval process
2293 from the kernel. The I/O retrieval will go on until we hit the limit set by
2294 :option:`iodepth_low`. If this variable is set to 0, then fio will always
2295 check for completed events before queuing more I/O. This helps reduce I/O
2296 latency, at the cost of more retrieval system calls.
2298 .. option:: iodepth_batch_complete_max=int
2300 This defines maximum pieces of I/O to retrieve at once. This variable should
2301 be used along with :option:`iodepth_batch_complete_min`\=int variable,
2302 specifying the range of min and max amount of I/O which should be
2303 retrieved. By default it is equal to the :option:`iodepth_batch_complete_min`
2308 iodepth_batch_complete_min=1
2309 iodepth_batch_complete_max=<iodepth>
2311 which means that we will retrieve at least 1 I/O and up to the whole
2312 submitted queue depth. If none of I/O has been completed yet, we will wait.
2316 iodepth_batch_complete_min=0
2317 iodepth_batch_complete_max=<iodepth>
2319 which means that we can retrieve up to the whole submitted queue depth, but
2320 if none of I/O has been completed yet, we will NOT wait and immediately exit
2321 the system call. In this example we simply do polling.
2323 .. option:: iodepth_low=int
2325 The low water mark indicating when to start filling the queue
2326 again. Defaults to the same as :option:`iodepth`, meaning that fio will
2327 attempt to keep the queue full at all times. If :option:`iodepth` is set to
2328 e.g. 16 and *iodepth_low* is set to 4, then after fio has filled the queue of
2329 16 requests, it will let the depth drain down to 4 before starting to fill
2332 .. option:: serialize_overlap=bool
2334 Serialize in-flight I/Os that might otherwise cause or suffer from data races.
2335 When two or more I/Os are submitted simultaneously, there is no guarantee that
2336 the I/Os will be processed or completed in the submitted order. Further, if
2337 two or more of those I/Os are writes, any overlapping region between them can
2338 become indeterminate/undefined on certain storage. These issues can cause
2339 verification to fail erratically when at least one of the racing I/Os is
2340 changing data and the overlapping region has a non-zero size. Setting
2341 ``serialize_overlap`` tells fio to avoid provoking this behavior by explicitly
2342 serializing in-flight I/Os that have a non-zero overlap. Note that setting
2343 this option can reduce both performance and the :option:`iodepth` achieved.
2345 This option only applies to I/Os issued for a single job except when it is
2346 enabled along with :option:`io_submit_mode`=offload. In offload mode, fio
2347 will check for overlap among all I/Os submitted by offload jobs with :option:`serialize_overlap`
2352 .. option:: io_submit_mode=str
2354 This option controls how fio submits the I/O to the I/O engine. The default
2355 is `inline`, which means that the fio job threads submit and reap I/O
2356 directly. If set to `offload`, the job threads will offload I/O submission
2357 to a dedicated pool of I/O threads. This requires some coordination and thus
2358 has a bit of extra overhead, especially for lower queue depth I/O where it
2359 can increase latencies. The benefit is that fio can manage submission rates
2360 independently of the device completion rates. This avoids skewed latency
2361 reporting if I/O gets backed up on the device side (the coordinated omission
2368 .. option:: thinktime=time
2370 Stall the job for the specified period of time after an I/O has completed before issuing the
2371 next. May be used to simulate processing being done by an application.
2372 When the unit is omitted, the value is interpreted in microseconds. See
2373 :option:`thinktime_blocks` and :option:`thinktime_spin`.
2375 .. option:: thinktime_spin=time
2377 Only valid if :option:`thinktime` is set - pretend to spend CPU time doing
2378 something with the data received, before falling back to sleeping for the
2379 rest of the period specified by :option:`thinktime`. When the unit is
2380 omitted, the value is interpreted in microseconds.
2382 .. option:: thinktime_blocks=int
2384 Only valid if :option:`thinktime` is set - control how many blocks to issue,
2385 before waiting :option:`thinktime` usecs. If not set, defaults to 1 which will make
2386 fio wait :option:`thinktime` usecs after every block. This effectively makes any
2387 queue depth setting redundant, since no more than 1 I/O will be queued
2388 before we have to complete it and do our :option:`thinktime`. In other words, this
2389 setting effectively caps the queue depth if the latter is larger.
2391 .. option:: rate=int[,int][,int]
2393 Cap the bandwidth used by this job. The number is in bytes/sec, the normal
2394 suffix rules apply. Comma-separated values may be specified for reads,
2395 writes, and trims as described in :option:`blocksize`.
2397 For example, using `rate=1m,500k` would limit reads to 1MiB/sec and writes to
2398 500KiB/sec. Capping only reads or writes can be done with `rate=,500k` or
2399 `rate=500k,` where the former will only limit writes (to 500KiB/sec) and the
2400 latter will only limit reads.
2402 .. option:: rate_min=int[,int][,int]
2404 Tell fio to do whatever it can to maintain at least this bandwidth. Failing
2405 to meet this requirement will cause the job to exit. Comma-separated values
2406 may be specified for reads, writes, and trims as described in
2407 :option:`blocksize`.
2409 .. option:: rate_iops=int[,int][,int]
2411 Cap the bandwidth to this number of IOPS. Basically the same as
2412 :option:`rate`, just specified independently of bandwidth. If the job is
2413 given a block size range instead of a fixed value, the smallest block size
2414 is used as the metric. Comma-separated values may be specified for reads,
2415 writes, and trims as described in :option:`blocksize`.
2417 .. option:: rate_iops_min=int[,int][,int]
2419 If fio doesn't meet this rate of I/O, it will cause the job to exit.
2420 Comma-separated values may be specified for reads, writes, and trims as
2421 described in :option:`blocksize`.
2423 .. option:: rate_process=str
2425 This option controls how fio manages rated I/O submissions. The default is
2426 `linear`, which submits I/O in a linear fashion with fixed delays between
2427 I/Os that gets adjusted based on I/O completion rates. If this is set to
2428 `poisson`, fio will submit I/O based on a more real world random request
2429 flow, known as the Poisson process
2430 (https://en.wikipedia.org/wiki/Poisson_point_process). The lambda will be
2431 10^6 / IOPS for the given workload.
2433 .. option:: rate_ignore_thinktime=bool
2435 By default, fio will attempt to catch up to the specified rate setting,
2436 if any kind of thinktime setting was used. If this option is set, then
2437 fio will ignore the thinktime and continue doing IO at the specified
2438 rate, instead of entering a catch-up mode after thinktime is done.
2444 .. option:: latency_target=time
2446 If set, fio will attempt to find the max performance point that the given
2447 workload will run at while maintaining a latency below this target. When
2448 the unit is omitted, the value is interpreted in microseconds. See
2449 :option:`latency_window` and :option:`latency_percentile`.
2451 .. option:: latency_window=time
2453 Used with :option:`latency_target` to specify the sample window that the job
2454 is run at varying queue depths to test the performance. When the unit is
2455 omitted, the value is interpreted in microseconds.
2457 .. option:: latency_percentile=float
2459 The percentage of I/Os that must fall within the criteria specified by
2460 :option:`latency_target` and :option:`latency_window`. If not set, this
2461 defaults to 100.0, meaning that all I/Os must be equal or below to the value
2462 set by :option:`latency_target`.
2464 .. option:: max_latency=time
2466 If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
2467 maximum latency. When the unit is omitted, the value is interpreted in
2470 .. option:: rate_cycle=int
2472 Average bandwidth for :option:`rate` and :option:`rate_min` over this number
2473 of milliseconds. Defaults to 1000.
2479 .. option:: write_iolog=str
2481 Write the issued I/O patterns to the specified file. See
2482 :option:`read_iolog`. Specify a separate file for each job, otherwise the
2483 iologs will be interspersed and the file may be corrupt.
2485 .. option:: read_iolog=str
2487 Open an iolog with the specified filename and replay the I/O patterns it
2488 contains. This can be used to store a workload and replay it sometime
2489 later. The iolog given may also be a blktrace binary file, which allows fio
2490 to replay a workload captured by :command:`blktrace`. See
2491 :manpage:`blktrace(8)` for how to capture such logging data. For blktrace
2492 replay, the file needs to be turned into a blkparse binary data file first
2493 (``blkparse <device> -o /dev/null -d file_for_fio.bin``).
2494 You can specify a number of files by separating the names with a ':'
2495 character. See the :option:`filename` option for information on how to
2496 escape ':' and '\' characters within the file names. These files will
2497 be sequentially assigned to job clones created by :option:`numjobs`.
2499 .. option:: read_iolog_chunked=bool
2501 Determines how iolog is read. If false(default) entire :option:`read_iolog`
2502 will be read at once. If selected true, input from iolog will be read
2503 gradually. Useful when iolog is very large, or it is generated.
2505 .. option:: merge_blktrace_file=str
2507 When specified, rather than replaying the logs passed to :option:`read_iolog`,
2508 the logs go through a merge phase which aggregates them into a single
2509 blktrace. The resulting file is then passed on as the :option:`read_iolog`
2510 parameter. The intention here is to make the order of events consistent.
2511 This limits the influence of the scheduler compared to replaying multiple
2512 blktraces via concurrent jobs.
2514 .. option:: merge_blktrace_scalars=float_list
2516 This is a percentage based option that is index paired with the list of
2517 files passed to :option:`read_iolog`. When merging is performed, scale
2518 the time of each event by the corresponding amount. For example,
2519 ``--merge_blktrace_scalars="50:100"`` runs the first trace in halftime
2520 and the second trace in realtime. This knob is separately tunable from
2521 :option:`replay_time_scale` which scales the trace during runtime and
2522 does not change the output of the merge unlike this option.
2524 .. option:: merge_blktrace_iters=float_list
2526 This is a whole number option that is index paired with the list of files
2527 passed to :option:`read_iolog`. When merging is performed, run each trace
2528 for the specified number of iterations. For example,
2529 ``--merge_blktrace_iters="2:1"`` runs the first trace for two iterations
2530 and the second trace for one iteration.
2532 .. option:: replay_no_stall=bool
2534 When replaying I/O with :option:`read_iolog` the default behavior is to
2535 attempt to respect the timestamps within the log and replay them with the
2536 appropriate delay between IOPS. By setting this variable fio will not
2537 respect the timestamps and attempt to replay them as fast as possible while
2538 still respecting ordering. The result is the same I/O pattern to a given
2539 device, but different timings.
2541 .. option:: replay_time_scale=int
2543 When replaying I/O with :option:`read_iolog`, fio will honor the
2544 original timing in the trace. With this option, it's possible to scale
2545 the time. It's a percentage option, if set to 50 it means run at 50%
2546 the original IO rate in the trace. If set to 200, run at twice the
2547 original IO rate. Defaults to 100.
2549 .. option:: replay_redirect=str
2551 While replaying I/O patterns using :option:`read_iolog` the default behavior
2552 is to replay the IOPS onto the major/minor device that each IOP was recorded
2553 from. This is sometimes undesirable because on a different machine those
2554 major/minor numbers can map to a different device. Changing hardware on the
2555 same system can also result in a different major/minor mapping.
2556 ``replay_redirect`` causes all I/Os to be replayed onto the single specified
2557 device regardless of the device it was recorded
2558 from. i.e. :option:`replay_redirect`\= :file:`/dev/sdc` would cause all I/O
2559 in the blktrace or iolog to be replayed onto :file:`/dev/sdc`. This means
2560 multiple devices will be replayed onto a single device, if the trace
2561 contains multiple devices. If you want multiple devices to be replayed
2562 concurrently to multiple redirected devices you must blkparse your trace
2563 into separate traces and replay them with independent fio invocations.
2564 Unfortunately this also breaks the strict time ordering between multiple
2567 .. option:: replay_align=int
2569 Force alignment of the byte offsets in a trace to this value. The value
2570 must be a power of 2.
2572 .. option:: replay_scale=int
2574 Scale byte offsets down by this factor when replaying traces. Should most
2575 likely use :option:`replay_align` as well.
2577 .. option:: replay_skip=str
2579 Sometimes it's useful to skip certain IO types in a replay trace.
2580 This could be, for instance, eliminating the writes in the trace.
2581 Or not replaying the trims/discards, if you are redirecting to
2582 a device that doesn't support them. This option takes a comma
2583 separated list of read, write, trim, sync.
2586 Threads, processes and job synchronization
2587 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2591 Fio defaults to creating jobs by using fork, however if this option is
2592 given, fio will create jobs by using POSIX Threads' function
2593 :manpage:`pthread_create(3)` to create threads instead.
2595 .. option:: wait_for=str
2597 If set, the current job won't be started until all workers of the specified
2598 waitee job are done.
2600 ``wait_for`` operates on the job name basis, so there are a few
2601 limitations. First, the waitee must be defined prior to the waiter job
2602 (meaning no forward references). Second, if a job is being referenced as a
2603 waitee, it must have a unique name (no duplicate waitees).
2605 .. option:: nice=int
2607 Run the job with the given nice value. See man :manpage:`nice(2)`.
2609 On Windows, values less than -15 set the process class to "High"; -1 through
2610 -15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2613 .. option:: prio=int
2615 Set the I/O priority value of this job. Linux limits us to a positive value
2616 between 0 and 7, with 0 being the highest. See man
2617 :manpage:`ionice(1)`. Refer to an appropriate manpage for other operating
2618 systems since meaning of priority may differ.
2620 .. option:: prioclass=int
2622 Set the I/O priority class. See man :manpage:`ionice(1)`.
2624 .. option:: cpus_allowed=str
2626 Controls the same options as :option:`cpumask`, but accepts a textual
2627 specification of the permitted CPUs instead and CPUs are indexed from 0. So
2628 to use CPUs 0 and 5 you would specify ``cpus_allowed=0,5``. This option also
2629 allows a range of CPUs to be specified -- say you wanted a binding to CPUs
2630 0, 5, and 8 to 15, you would set ``cpus_allowed=0,5,8-15``.
2632 On Windows, when ``cpus_allowed`` is unset only CPUs from fio's current
2633 processor group will be used and affinity settings are inherited from the
2634 system. An fio build configured to target Windows 7 makes options that set
2635 CPUs processor group aware and values will set both the processor group
2636 and a CPU from within that group. For example, on a system where processor
2637 group 0 has 40 CPUs and processor group 1 has 32 CPUs, ``cpus_allowed``
2638 values between 0 and 39 will bind CPUs from processor group 0 and
2639 ``cpus_allowed`` values between 40 and 71 will bind CPUs from processor
2640 group 1. When using ``cpus_allowed_policy=shared`` all CPUs specified by a
2641 single ``cpus_allowed`` option must be from the same processor group. For
2642 Windows fio builds not built for Windows 7, CPUs will only be selected from
2643 (and be relative to) whatever processor group fio happens to be running in
2644 and CPUs from other processor groups cannot be used.
2646 .. option:: cpus_allowed_policy=str
2648 Set the policy of how fio distributes the CPUs specified by
2649 :option:`cpus_allowed` or :option:`cpumask`. Two policies are supported:
2652 All jobs will share the CPU set specified.
2654 Each job will get a unique CPU from the CPU set.
2656 **shared** is the default behavior, if the option isn't specified. If
2657 **split** is specified, then fio will will assign one cpu per job. If not
2658 enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2661 .. option:: cpumask=int
2663 Set the CPU affinity of this job. The parameter given is a bit mask of
2664 allowed CPUs the job may run on. So if you want the allowed CPUs to be 1
2665 and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2666 :manpage:`sched_setaffinity(2)`. This may not work on all supported
2667 operating systems or kernel versions. This option doesn't work well for a
2668 higher CPU count than what you can store in an integer mask, so it can only
2669 control cpus 1-32. For boxes with larger CPU counts, use
2670 :option:`cpus_allowed`.
2672 .. option:: numa_cpu_nodes=str
2674 Set this job running on specified NUMA nodes' CPUs. The arguments allow
2675 comma delimited list of cpu numbers, A-B ranges, or `all`. Note, to enable
2676 NUMA options support, fio must be built on a system with libnuma-dev(el)
2679 .. option:: numa_mem_policy=str
2681 Set this job's memory policy and corresponding NUMA nodes. Format of the
2686 ``mode`` is one of the following memory policies: ``default``, ``prefer``,
2687 ``bind``, ``interleave`` or ``local``. For ``default`` and ``local`` memory
2688 policies, no node needs to be specified. For ``prefer``, only one node is
2689 allowed. For ``bind`` and ``interleave`` the ``nodelist`` may be as
2690 follows: a comma delimited list of numbers, A-B ranges, or `all`.
2692 .. option:: cgroup=str
2694 Add job to this control group. If it doesn't exist, it will be created. The
2695 system must have a mounted cgroup blkio mount point for this to work. If
2696 your system doesn't have it mounted, you can do so with::
2698 # mount -t cgroup -o blkio none /cgroup
2700 .. option:: cgroup_weight=int
2702 Set the weight of the cgroup to this value. See the documentation that comes
2703 with the kernel, allowed values are in the range of 100..1000.
2705 .. option:: cgroup_nodelete=bool
2707 Normally fio will delete the cgroups it has created after the job
2708 completion. To override this behavior and to leave cgroups around after the
2709 job completion, set ``cgroup_nodelete=1``. This can be useful if one wants
2710 to inspect various cgroup files after job completion. Default: false.
2712 .. option:: flow_id=int
2714 The ID of the flow. If not specified, it defaults to being a global
2715 flow. See :option:`flow`.
2717 .. option:: flow=int
2719 Weight in token-based flow control. If this value is used, then there is a
2720 'flow counter' which is used to regulate the proportion of activity between
2721 two or more jobs. Fio attempts to keep this flow counter near zero. The
2722 ``flow`` parameter stands for how much should be added or subtracted to the
2723 flow counter on each iteration of the main I/O loop. That is, if one job has
2724 ``flow=8`` and another job has ``flow=-1``, then there will be a roughly 1:8
2725 ratio in how much one runs vs the other.
2727 .. option:: flow_watermark=int
2729 The maximum value that the absolute value of the flow counter is allowed to
2730 reach before the job must wait for a lower value of the counter.
2732 .. option:: flow_sleep=int
2734 The period of time, in microseconds, to wait after the flow watermark has
2735 been exceeded before retrying operations.
2737 .. option:: stonewall, wait_for_previous
2739 Wait for preceding jobs in the job file to exit, before starting this
2740 one. Can be used to insert serialization points in the job file. A stone
2741 wall also implies starting a new reporting group, see
2742 :option:`group_reporting`.
2746 By default, fio will continue running all other jobs when one job finishes
2747 but sometimes this is not the desired action. Setting ``exitall`` will
2748 instead make fio terminate all other jobs when one job finishes.
2750 .. option:: exec_prerun=str
2752 Before running this job, issue the command specified through
2753 :manpage:`system(3)`. Output is redirected in a file called
2754 :file:`jobname.prerun.txt`.
2756 .. option:: exec_postrun=str
2758 After the job completes, issue the command specified though
2759 :manpage:`system(3)`. Output is redirected in a file called
2760 :file:`jobname.postrun.txt`.
2764 Instead of running as the invoking user, set the user ID to this value
2765 before the thread/process does any work.
2769 Set group ID, see :option:`uid`.
2775 .. option:: verify_only
2777 Do not perform specified workload, only verify data still matches previous
2778 invocation of this workload. This option allows one to check data multiple
2779 times at a later date without overwriting it. This option makes sense only
2780 for workloads that write data, and does not support workloads with the
2781 :option:`time_based` option set.
2783 .. option:: do_verify=bool
2785 Run the verify phase after a write phase. Only valid if :option:`verify` is
2788 .. option:: verify=str
2790 If writing to a file, fio can verify the file contents after each iteration
2791 of the job. Each verification method also implies verification of special
2792 header, which is written to the beginning of each block. This header also
2793 includes meta information, like offset of the block, block number, timestamp
2794 when block was written, etc. :option:`verify` can be combined with
2795 :option:`verify_pattern` option. The allowed values are:
2798 Use an md5 sum of the data area and store it in the header of
2802 Use an experimental crc64 sum of the data area and store it in the
2803 header of each block.
2806 Use a crc32c sum of the data area and store it in the header of
2807 each block. This will automatically use hardware acceleration
2808 (e.g. SSE4.2 on an x86 or CRC crypto extensions on ARM64) but will
2809 fall back to software crc32c if none is found. Generally the
2810 fastest checksum fio supports when hardware accelerated.
2816 Use a crc32 sum of the data area and store it in the header of each
2820 Use a crc16 sum of the data area and store it in the header of each
2824 Use a crc7 sum of the data area and store it in the header of each
2828 Use xxhash as the checksum function. Generally the fastest software
2829 checksum that fio supports.
2832 Use sha512 as the checksum function.
2835 Use sha256 as the checksum function.
2838 Use optimized sha1 as the checksum function.
2841 Use optimized sha3-224 as the checksum function.
2844 Use optimized sha3-256 as the checksum function.
2847 Use optimized sha3-384 as the checksum function.
2850 Use optimized sha3-512 as the checksum function.
2853 This option is deprecated, since now meta information is included in
2854 generic verification header and meta verification happens by
2855 default. For detailed information see the description of the
2856 :option:`verify` setting. This option is kept because of
2857 compatibility's sake with old configurations. Do not use it.
2860 Verify a strict pattern. Normally fio includes a header with some
2861 basic information and checksumming, but if this option is set, only
2862 the specific pattern set with :option:`verify_pattern` is verified.
2865 Only pretend to verify. Useful for testing internals with
2866 :option:`ioengine`\=null, not for much else.
2868 This option can be used for repeated burn-in tests of a system to make sure
2869 that the written data is also correctly read back. If the data direction
2870 given is a read or random read, fio will assume that it should verify a
2871 previously written file. If the data direction includes any form of write,
2872 the verify will be of the newly written data.
2874 To avoid false verification errors, do not use the norandommap option when
2875 verifying data with async I/O engines and I/O depths > 1. Or use the
2876 norandommap and the lfsr random generator together to avoid writing to the
2877 same offset with muliple outstanding I/Os.
2879 .. option:: verify_offset=int
2881 Swap the verification header with data somewhere else in the block before
2882 writing. It is swapped back before verifying.
2884 .. option:: verify_interval=int
2886 Write the verification header at a finer granularity than the
2887 :option:`blocksize`. It will be written for chunks the size of
2888 ``verify_interval``. :option:`blocksize` should divide this evenly.
2890 .. option:: verify_pattern=str
2892 If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2893 filling with totally random bytes, but sometimes it's interesting to fill
2894 with a known pattern for I/O verification purposes. Depending on the width
2895 of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time (it can
2896 be either a decimal or a hex number). The ``verify_pattern`` if larger than
2897 a 32-bit quantity has to be a hex number that starts with either "0x" or
2898 "0X". Use with :option:`verify`. Also, ``verify_pattern`` supports %o
2899 format, which means that for each block offset will be written and then
2900 verified back, e.g.::
2904 Or use combination of everything::
2906 verify_pattern=0xff%o"abcd"-12
2908 .. option:: verify_fatal=bool
2910 Normally fio will keep checking the entire contents before quitting on a
2911 block verification failure. If this option is set, fio will exit the job on
2912 the first observed failure. Default: false.
2914 .. option:: verify_dump=bool
2916 If set, dump the contents of both the original data block and the data block
2917 we read off disk to files. This allows later analysis to inspect just what
2918 kind of data corruption occurred. Off by default.
2920 .. option:: verify_async=int
2922 Fio will normally verify I/O inline from the submitting thread. This option
2923 takes an integer describing how many async offload threads to create for I/O
2924 verification instead, causing fio to offload the duty of verifying I/O
2925 contents to one or more separate threads. If using this offload option, even
2926 sync I/O engines can benefit from using an :option:`iodepth` setting higher
2927 than 1, as it allows them to have I/O in flight while verifies are running.
2928 Defaults to 0 async threads, i.e. verification is not asynchronous.
2930 .. option:: verify_async_cpus=str
2932 Tell fio to set the given CPU affinity on the async I/O verification
2933 threads. See :option:`cpus_allowed` for the format used.
2935 .. option:: verify_backlog=int
2937 Fio will normally verify the written contents of a job that utilizes verify
2938 once that job has completed. In other words, everything is written then
2939 everything is read back and verified. You may want to verify continually
2940 instead for a variety of reasons. Fio stores the meta data associated with
2941 an I/O block in memory, so for large verify workloads, quite a bit of memory
2942 would be used up holding this meta data. If this option is enabled, fio will
2943 write only N blocks before verifying these blocks.
2945 .. option:: verify_backlog_batch=int
2947 Control how many blocks fio will verify if :option:`verify_backlog` is
2948 set. If not set, will default to the value of :option:`verify_backlog`
2949 (meaning the entire queue is read back and verified). If
2950 ``verify_backlog_batch`` is less than :option:`verify_backlog` then not all
2951 blocks will be verified, if ``verify_backlog_batch`` is larger than
2952 :option:`verify_backlog`, some blocks will be verified more than once.
2954 .. option:: verify_state_save=bool
2956 When a job exits during the write phase of a verify workload, save its
2957 current state. This allows fio to replay up until that point, if the verify
2958 state is loaded for the verify read phase. The format of the filename is,
2961 <type>-<jobname>-<jobindex>-verify.state.
2963 <type> is "local" for a local run, "sock" for a client/server socket
2964 connection, and "ip" (192.168.0.1, for instance) for a networked
2965 client/server connection. Defaults to true.
2967 .. option:: verify_state_load=bool
2969 If a verify termination trigger was used, fio stores the current write state
2970 of each thread. This can be used at verification time so that fio knows how
2971 far it should verify. Without this information, fio will run a full
2972 verification pass, according to the settings in the job file used. Default
2975 .. option:: trim_percentage=int
2977 Number of verify blocks to discard/trim.
2979 .. option:: trim_verify_zero=bool
2981 Verify that trim/discarded blocks are returned as zeros.
2983 .. option:: trim_backlog=int
2985 Trim after this number of blocks are written.
2987 .. option:: trim_backlog_batch=int
2989 Trim this number of I/O blocks.
2991 .. option:: experimental_verify=bool
2993 Enable experimental verification.
2998 .. option:: steadystate=str:float, ss=str:float
3000 Define the criterion and limit for assessing steady state performance. The
3001 first parameter designates the criterion whereas the second parameter sets
3002 the threshold. When the criterion falls below the threshold for the
3003 specified duration, the job will stop. For example, `iops_slope:0.1%` will
3004 direct fio to terminate the job when the least squares regression slope
3005 falls below 0.1% of the mean IOPS. If :option:`group_reporting` is enabled
3006 this will apply to all jobs in the group. Below is the list of available
3007 steady state assessment criteria. All assessments are carried out using only
3008 data from the rolling collection window. Threshold limits can be expressed
3009 as a fixed value or as a percentage of the mean in the collection window.
3011 When using this feature, most jobs should include the :option:`time_based`
3012 and :option:`runtime` options or the :option:`loops` option so that fio does not
3013 stop running after it has covered the full size of the specified file(s) or device(s).
3016 Collect IOPS data. Stop the job if all individual IOPS measurements
3017 are within the specified limit of the mean IOPS (e.g., ``iops:2``
3018 means that all individual IOPS values must be within 2 of the mean,
3019 whereas ``iops:0.2%`` means that all individual IOPS values must be
3020 within 0.2% of the mean IOPS to terminate the job).
3023 Collect IOPS data and calculate the least squares regression
3024 slope. Stop the job if the slope falls below the specified limit.
3027 Collect bandwidth data. Stop the job if all individual bandwidth
3028 measurements are within the specified limit of the mean bandwidth.
3031 Collect bandwidth data and calculate the least squares regression
3032 slope. Stop the job if the slope falls below the specified limit.
3034 .. option:: steadystate_duration=time, ss_dur=time
3036 A rolling window of this duration will be used to judge whether steady state
3037 has been reached. Data will be collected once per second. The default is 0
3038 which disables steady state detection. When the unit is omitted, the
3039 value is interpreted in seconds.
3041 .. option:: steadystate_ramp_time=time, ss_ramp=time
3043 Allow the job to run for the specified duration before beginning data
3044 collection for checking the steady state job termination criterion. The
3045 default is 0. When the unit is omitted, the value is interpreted in seconds.
3048 Measurements and reporting
3049 ~~~~~~~~~~~~~~~~~~~~~~~~~~
3051 .. option:: per_job_logs=bool
3053 If set, this generates bw/clat/iops log with per file private filenames. If
3054 not set, jobs with identical names will share the log filename. Default:
3057 .. option:: group_reporting
3059 It may sometimes be interesting to display statistics for groups of jobs as
3060 a whole instead of for each individual job. This is especially true if
3061 :option:`numjobs` is used; looking at individual thread/process output
3062 quickly becomes unwieldy. To see the final report per-group instead of
3063 per-job, use :option:`group_reporting`. Jobs in a file will be part of the
3064 same reporting group, unless if separated by a :option:`stonewall`, or by
3065 using :option:`new_group`.
3067 .. option:: new_group
3069 Start a new reporting group. See: :option:`group_reporting`. If not given,
3070 all jobs in a file will be part of the same reporting group, unless
3071 separated by a :option:`stonewall`.
3073 .. option:: stats=bool
3075 By default, fio collects and shows final output results for all jobs
3076 that run. If this option is set to 0, then fio will ignore it in
3077 the final stat output.
3079 .. option:: write_bw_log=str
3081 If given, write a bandwidth log for this job. Can be used to store data of
3082 the bandwidth of the jobs in their lifetime.
3084 If no str argument is given, the default filename of
3085 :file:`jobname_type.x.log` is used. Even when the argument is given, fio
3086 will still append the type of log. So if one specifies::
3090 The actual log name will be :file:`foo_bw.x.log` where `x` is the index
3091 of the job (`1..N`, where `N` is the number of jobs). If
3092 :option:`per_job_logs` is false, then the filename will not include the
3095 The included :command:`fio_generate_plots` script uses :command:`gnuplot` to turn these
3096 text files into nice graphs. See `Log File Formats`_ for how data is
3097 structured within the file.
3099 .. option:: write_lat_log=str
3101 Same as :option:`write_bw_log`, except this option creates I/O
3102 submission (e.g., :file:`name_slat.x.log`), completion (e.g.,
3103 :file:`name_clat.x.log`), and total (e.g., :file:`name_lat.x.log`)
3104 latency files instead. See :option:`write_bw_log` for details about
3105 the filename format and `Log File Formats`_ for how data is structured
3108 .. option:: write_hist_log=str
3110 Same as :option:`write_bw_log` but writes an I/O completion latency
3111 histogram file (e.g., :file:`name_hist.x.log`) instead. Note that this
3112 file will be empty unless :option:`log_hist_msec` has also been set.
3113 See :option:`write_bw_log` for details about the filename format and
3114 `Log File Formats`_ for how data is structured within the file.
3116 .. option:: write_iops_log=str
3118 Same as :option:`write_bw_log`, but writes an IOPS file (e.g.
3119 :file:`name_iops.x.log`) instead. Because fio defaults to individual
3120 I/O logging, the value entry in the IOPS log will be 1 unless windowed
3121 logging (see :option:`log_avg_msec`) has been enabled. See
3122 :option:`write_bw_log` for details about the filename format and `Log
3123 File Formats`_ for how data is structured within the file.
3125 .. option:: log_avg_msec=int
3127 By default, fio will log an entry in the iops, latency, or bw log for every
3128 I/O that completes. When writing to the disk log, that can quickly grow to a
3129 very large size. Setting this option makes fio average the each log entry
3130 over the specified period of time, reducing the resolution of the log. See
3131 :option:`log_max_value` as well. Defaults to 0, logging all entries.
3132 Also see `Log File Formats`_.
3134 .. option:: log_hist_msec=int
3136 Same as :option:`log_avg_msec`, but logs entries for completion latency
3137 histograms. Computing latency percentiles from averages of intervals using
3138 :option:`log_avg_msec` is inaccurate. Setting this option makes fio log
3139 histogram entries over the specified period of time, reducing log sizes for
3140 high IOPS devices while retaining percentile accuracy. See
3141 :option:`log_hist_coarseness` and :option:`write_hist_log` as well.
3142 Defaults to 0, meaning histogram logging is disabled.
3144 .. option:: log_hist_coarseness=int
3146 Integer ranging from 0 to 6, defining the coarseness of the resolution of
3147 the histogram logs enabled with :option:`log_hist_msec`. For each increment
3148 in coarseness, fio outputs half as many bins. Defaults to 0, for which
3149 histogram logs contain 1216 latency bins. See :option:`write_hist_log`
3150 and `Log File Formats`_.
3152 .. option:: log_max_value=bool
3154 If :option:`log_avg_msec` is set, fio logs the average over that window. If
3155 you instead want to log the maximum value, set this option to 1. Defaults to
3156 0, meaning that averaged values are logged.
3158 .. option:: log_offset=bool
3160 If this is set, the iolog options will include the byte offset for the I/O
3161 entry as well as the other data values. Defaults to 0 meaning that
3162 offsets are not present in logs. Also see `Log File Formats`_.
3164 .. option:: log_compression=int
3166 If this is set, fio will compress the I/O logs as it goes, to keep the
3167 memory footprint lower. When a log reaches the specified size, that chunk is
3168 removed and compressed in the background. Given that I/O logs are fairly
3169 highly compressible, this yields a nice memory savings for longer runs. The
3170 downside is that the compression will consume some background CPU cycles, so
3171 it may impact the run. This, however, is also true if the logging ends up
3172 consuming most of the system memory. So pick your poison. The I/O logs are
3173 saved normally at the end of a run, by decompressing the chunks and storing
3174 them in the specified log file. This feature depends on the availability of
3177 .. option:: log_compression_cpus=str
3179 Define the set of CPUs that are allowed to handle online log compression for
3180 the I/O jobs. This can provide better isolation between performance
3181 sensitive jobs, and background compression work. See
3182 :option:`cpus_allowed` for the format used.
3184 .. option:: log_store_compressed=bool
3186 If set, fio will store the log files in a compressed format. They can be
3187 decompressed with fio, using the :option:`--inflate-log` command line
3188 parameter. The files will be stored with a :file:`.fz` suffix.
3190 .. option:: log_unix_epoch=bool
3192 If set, fio will log Unix timestamps to the log files produced by enabling
3193 write_type_log for each log type, instead of the default zero-based
3196 .. option:: block_error_percentiles=bool
3198 If set, record errors in trim block-sized units from writes and trims and
3199 output a histogram of how many trims it took to get to errors, and what kind
3200 of error was encountered.
3202 .. option:: bwavgtime=int
3204 Average the calculated bandwidth over the given time. Value is specified in
3205 milliseconds. If the job also does bandwidth logging through
3206 :option:`write_bw_log`, then the minimum of this option and
3207 :option:`log_avg_msec` will be used. Default: 500ms.
3209 .. option:: iopsavgtime=int
3211 Average the calculated IOPS over the given time. Value is specified in
3212 milliseconds. If the job also does IOPS logging through
3213 :option:`write_iops_log`, then the minimum of this option and
3214 :option:`log_avg_msec` will be used. Default: 500ms.
3216 .. option:: disk_util=bool
3218 Generate disk utilization statistics, if the platform supports it.
3221 .. option:: disable_lat=bool
3223 Disable measurements of total latency numbers. Useful only for cutting back
3224 the number of calls to :manpage:`gettimeofday(2)`, as that does impact
3225 performance at really high IOPS rates. Note that to really get rid of a
3226 large amount of these calls, this option must be used with
3227 :option:`disable_slat` and :option:`disable_bw_measurement` as well.
3229 .. option:: disable_clat=bool
3231 Disable measurements of completion latency numbers. See
3232 :option:`disable_lat`.
3234 .. option:: disable_slat=bool
3236 Disable measurements of submission latency numbers. See
3237 :option:`disable_lat`.
3239 .. option:: disable_bw_measurement=bool, disable_bw=bool
3241 Disable measurements of throughput/bandwidth numbers. See
3242 :option:`disable_lat`.
3244 .. option:: clat_percentiles=bool
3246 Enable the reporting of percentiles of completion latencies. This
3247 option is mutually exclusive with :option:`lat_percentiles`.
3249 .. option:: lat_percentiles=bool
3251 Enable the reporting of percentiles of I/O latencies. This is similar
3252 to :option:`clat_percentiles`, except that this includes the
3253 submission latency. This option is mutually exclusive with
3254 :option:`clat_percentiles`.
3256 .. option:: percentile_list=float_list
3258 Overwrite the default list of percentiles for completion latencies and
3259 the block error histogram. Each number is a floating number in the
3260 range (0,100], and the maximum length of the list is 20. Use ``:`` to
3261 separate the numbers, and list the numbers in ascending order. For
3262 example, ``--percentile_list=99.5:99.9`` will cause fio to report the
3263 values of completion latency below which 99.5% and 99.9% of the observed
3264 latencies fell, respectively.
3266 .. option:: significant_figures=int
3268 If using :option:`--output-format` of `normal`, set the significant
3269 figures to this value. Higher values will yield more precise IOPS and
3270 throughput units, while lower values will round. Requires a minimum
3271 value of 1 and a maximum value of 10. Defaults to 4.
3277 .. option:: exitall_on_error
3279 When one job finishes in error, terminate the rest. The default is to wait
3280 for each job to finish.
3282 .. option:: continue_on_error=str
3284 Normally fio will exit the job on the first observed failure. If this option
3285 is set, fio will continue the job when there is a 'non-fatal error' (EIO or
3286 EILSEQ) until the runtime is exceeded or the I/O size specified is
3287 completed. If this option is used, there are two more stats that are
3288 appended, the total error count and the first error. The error field given
3289 in the stats is the first error that was hit during the run.
3291 The allowed values are:
3294 Exit on any I/O or verify errors.
3297 Continue on read errors, exit on all others.
3300 Continue on write errors, exit on all others.
3303 Continue on any I/O error, exit on all others.
3306 Continue on verify errors, exit on all others.
3309 Continue on all errors.
3312 Backward-compatible alias for 'none'.
3315 Backward-compatible alias for 'all'.
3317 .. option:: ignore_error=str
3319 Sometimes you want to ignore some errors during test in that case you can
3320 specify error list for each error type, instead of only being able to
3321 ignore the default 'non-fatal error' using :option:`continue_on_error`.
3322 ``ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST`` errors for
3323 given error type is separated with ':'. Error may be symbol ('ENOSPC',
3324 'ENOMEM') or integer. Example::
3326 ignore_error=EAGAIN,ENOSPC:122
3328 This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
3329 WRITE. This option works by overriding :option:`continue_on_error` with
3330 the list of errors for each error type if any.
3332 .. option:: error_dump=bool
3334 If set dump every error even if it is non fatal, true by default. If
3335 disabled only fatal error will be dumped.
3337 Running predefined workloads
3338 ----------------------------
3340 Fio includes predefined profiles that mimic the I/O workloads generated by
3343 .. option:: profile=str
3345 The predefined workload to run. Current profiles are:
3348 Threaded I/O bench (tiotest/tiobench) like workload.
3351 Aerospike Certification Tool (ACT) like workload.
3353 To view a profile's additional options use :option:`--cmdhelp` after specifying
3354 the profile. For example::
3356 $ fio --profile=act --cmdhelp
3361 .. option:: device-names=str
3366 .. option:: load=int
3369 ACT load multiplier. Default: 1.
3371 .. option:: test-duration=time
3374 How long the entire test takes to run. When the unit is omitted, the value
3375 is given in seconds. Default: 24h.
3377 .. option:: threads-per-queue=int
3380 Number of read I/O threads per device. Default: 8.
3382 .. option:: read-req-num-512-blocks=int
3385 Number of 512B blocks to read at the time. Default: 3.
3387 .. option:: large-block-op-kbytes=int
3390 Size of large block ops in KiB (writes). Default: 131072.
3395 Set to run ACT prep phase.
3397 Tiobench profile options
3398 ~~~~~~~~~~~~~~~~~~~~~~~~
3400 .. option:: size=str
3405 .. option:: block=int
3408 Block size in bytes. Default: 4096.
3410 .. option:: numruns=int
3420 .. option:: threads=int
3425 Interpreting the output
3426 -----------------------
3429 Example output was based on the following:
3430 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --time_based \
3431 --rate=1256k --bs=14K --name=quick --runtime=1s --name=mixed \
3432 --runtime=2m --rw=rw
3434 Fio spits out a lot of output. While running, fio will display the status of the
3435 jobs created. An example of that would be::
3437 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]
3439 The characters inside the first set of square brackets denote the current status of
3440 each thread. The first character is the first job defined in the job file, and so
3441 forth. The possible values (in typical life cycle order) are:
3443 +------+-----+-----------------------------------------------------------+
3445 +======+=====+===========================================================+
3446 | P | | Thread setup, but not started. |
3447 +------+-----+-----------------------------------------------------------+
3448 | C | | Thread created. |
3449 +------+-----+-----------------------------------------------------------+
3450 | I | | Thread initialized, waiting or generating necessary data. |
3451 +------+-----+-----------------------------------------------------------+
3452 | | p | Thread running pre-reading file(s). |
3453 +------+-----+-----------------------------------------------------------+
3454 | | / | Thread is in ramp period. |
3455 +------+-----+-----------------------------------------------------------+
3456 | | R | Running, doing sequential reads. |
3457 +------+-----+-----------------------------------------------------------+
3458 | | r | Running, doing random reads. |
3459 +------+-----+-----------------------------------------------------------+
3460 | | W | Running, doing sequential writes. |
3461 +------+-----+-----------------------------------------------------------+
3462 | | w | Running, doing random writes. |
3463 +------+-----+-----------------------------------------------------------+
3464 | | M | Running, doing mixed sequential reads/writes. |
3465 +------+-----+-----------------------------------------------------------+
3466 | | m | Running, doing mixed random reads/writes. |
3467 +------+-----+-----------------------------------------------------------+
3468 | | D | Running, doing sequential trims. |
3469 +------+-----+-----------------------------------------------------------+
3470 | | d | Running, doing random trims. |
3471 +------+-----+-----------------------------------------------------------+
3472 | | F | Running, currently waiting for :manpage:`fsync(2)`. |
3473 +------+-----+-----------------------------------------------------------+
3474 | | V | Running, doing verification of written data. |
3475 +------+-----+-----------------------------------------------------------+
3476 | f | | Thread finishing. |
3477 +------+-----+-----------------------------------------------------------+
3478 | E | | Thread exited, not reaped by main thread yet. |
3479 +------+-----+-----------------------------------------------------------+
3480 | _ | | Thread reaped. |
3481 +------+-----+-----------------------------------------------------------+
3482 | X | | Thread reaped, exited with an error. |
3483 +------+-----+-----------------------------------------------------------+
3484 | K | | Thread reaped, exited due to signal. |
3485 +------+-----+-----------------------------------------------------------+
3488 Example output was based on the following:
3489 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --runtime=58m \
3490 --time_based --rate=2512k --bs=256K --numjobs=10 \
3491 --name=readers --rw=read --name=writers --rw=write
3493 Fio will condense the thread string as not to take up more space on the command
3494 line than needed. For instance, if you have 10 readers and 10 writers running,
3495 the output would look like this::
3497 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]
3499 Note that the status string is displayed in order, so it's possible to tell which of
3500 the jobs are currently doing what. In the example above this means that jobs 1--10
3501 are readers and 11--20 are writers.
3503 The other values are fairly self explanatory -- number of threads currently
3504 running and doing I/O, the number of currently open files (f=), the estimated
3505 completion percentage, the rate of I/O since last check (read speed listed first,
3506 then write speed and optionally trim speed) in terms of bandwidth and IOPS,
3507 and time to completion for the current running group. It's impossible to estimate
3508 runtime of the following groups (if any).
3511 Example output was based on the following:
3512 TZ=UTC fio --iodepth=16 --ioengine=posixaio --filename=/tmp/fiofile \
3513 --direct=1 --size=100M --time_based --runtime=50s --rate_iops=89 \
3514 --bs=7K --name=Client1 --rw=write
3516 When fio is done (or interrupted by :kbd:`Ctrl-C`), it will show the data for
3517 each thread, group of threads, and disks in that order. For each overall thread (or
3518 group) the output looks like::
3520 Client1: (groupid=0, jobs=1): err= 0: pid=16109: Sat Jun 24 12:07:54 2017
3521 write: IOPS=88, BW=623KiB/s (638kB/s)(30.4MiB/50032msec)
3522 slat (nsec): min=500, max=145500, avg=8318.00, stdev=4781.50
3523 clat (usec): min=170, max=78367, avg=4019.02, stdev=8293.31
3524 lat (usec): min=174, max=78375, avg=4027.34, stdev=8291.79
3525 clat percentiles (usec):
3526 | 1.00th=[ 302], 5.00th=[ 326], 10.00th=[ 343], 20.00th=[ 363],
3527 | 30.00th=[ 392], 40.00th=[ 404], 50.00th=[ 416], 60.00th=[ 445],
3528 | 70.00th=[ 816], 80.00th=[ 6718], 90.00th=[12911], 95.00th=[21627],
3529 | 99.00th=[43779], 99.50th=[51643], 99.90th=[68682], 99.95th=[72877],
3531 bw ( KiB/s): min= 532, max= 686, per=0.10%, avg=622.87, stdev=24.82, samples= 100
3532 iops : min= 76, max= 98, avg=88.98, stdev= 3.54, samples= 100
3533 lat (usec) : 250=0.04%, 500=64.11%, 750=4.81%, 1000=2.79%
3534 lat (msec) : 2=4.16%, 4=1.84%, 10=4.90%, 20=11.33%, 50=5.37%
3535 lat (msec) : 100=0.65%
3536 cpu : usr=0.27%, sys=0.18%, ctx=12072, majf=0, minf=21
3537 IO depths : 1=85.0%, 2=13.1%, 4=1.8%, 8=0.1%, 16=0.0%, 32=0.0%, >=64=0.0%
3538 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3539 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3540 issued rwt: total=0,4450,0, short=0,0,0, dropped=0,0,0
3541 latency : target=0, window=0, percentile=100.00%, depth=8
3543 The job name (or first job's name when using :option:`group_reporting`) is printed,
3544 along with the group id, count of jobs being aggregated, last error id seen (which
3545 is 0 when there are no errors), pid/tid of that thread and the time the job/group
3546 completed. Below are the I/O statistics for each data direction performed (showing
3547 writes in the example above). In the order listed, they denote:
3550 The string before the colon shows the I/O direction the statistics
3551 are for. **IOPS** is the average I/Os performed per second. **BW**
3552 is the average bandwidth rate shown as: value in power of 2 format
3553 (value in power of 10 format). The last two values show: (**total
3554 I/O performed** in power of 2 format / **runtime** of that thread).
3557 Submission latency (**min** being the minimum, **max** being the
3558 maximum, **avg** being the average, **stdev** being the standard
3559 deviation). This is the time it took to submit the I/O. For
3560 sync I/O this row is not displayed as the slat is really the
3561 completion latency (since queue/complete is one operation there).
3562 This value can be in nanoseconds, microseconds or milliseconds ---
3563 fio will choose the most appropriate base and print that (in the
3564 example above nanoseconds was the best scale). Note: in :option:`--minimal` mode
3565 latencies are always expressed in microseconds.
3568 Completion latency. Same names as slat, this denotes the time from
3569 submission to completion of the I/O pieces. For sync I/O, clat will
3570 usually be equal (or very close) to 0, as the time from submit to
3571 complete is basically just CPU time (I/O has already been done, see slat
3575 Total latency. Same names as slat and clat, this denotes the time from
3576 when fio created the I/O unit to completion of the I/O operation.
3579 Bandwidth statistics based on samples. Same names as the xlat stats,
3580 but also includes the number of samples taken (**samples**) and an
3581 approximate percentage of total aggregate bandwidth this thread
3582 received in its group (**per**). This last value is only really
3583 useful if the threads in this group are on the same disk, since they
3584 are then competing for disk access.
3587 IOPS statistics based on samples. Same names as bw.
3589 **lat (nsec/usec/msec)**
3590 The distribution of I/O completion latencies. This is the time from when
3591 I/O leaves fio and when it gets completed. Unlike the separate
3592 read/write/trim sections above, the data here and in the remaining
3593 sections apply to all I/Os for the reporting group. 250=0.04% means that
3594 0.04% of the I/Os completed in under 250us. 500=64.11% means that 64.11%
3595 of the I/Os required 250 to 499us for completion.
3598 CPU usage. User and system time, along with the number of context
3599 switches this thread went through, usage of system and user time, and
3600 finally the number of major and minor page faults. The CPU utilization
3601 numbers are averages for the jobs in that reporting group, while the
3602 context and fault counters are summed.
3605 The distribution of I/O depths over the job lifetime. The numbers are
3606 divided into powers of 2 and each entry covers depths from that value
3607 up to those that are lower than the next entry -- e.g., 16= covers
3608 depths from 16 to 31. Note that the range covered by a depth
3609 distribution entry can be different to the range covered by the
3610 equivalent submit/complete distribution entry.
3613 How many pieces of I/O were submitting in a single submit call. Each
3614 entry denotes that amount and below, until the previous entry -- e.g.,
3615 16=100% means that we submitted anywhere between 9 to 16 I/Os per submit
3616 call. Note that the range covered by a submit distribution entry can
3617 be different to the range covered by the equivalent depth distribution
3621 Like the above submit number, but for completions instead.
3624 The number of read/write/trim requests issued, and how many of them were
3628 These values are for :option:`latency_target` and related options. When
3629 these options are engaged, this section describes the I/O depth required
3630 to meet the specified latency target.
3633 Example output was based on the following:
3634 TZ=UTC fio --ioengine=null --iodepth=2 --size=100M --numjobs=2 \
3635 --rate_process=poisson --io_limit=32M --name=read --bs=128k \
3636 --rate=11M --name=write --rw=write --bs=2k --rate=700k
3638 After each client has been listed, the group statistics are printed. They
3639 will look like this::
3641 Run status group 0 (all jobs):
3642 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
3643 WRITE: bw=1231KiB/s (1261kB/s), 616KiB/s-621KiB/s (630kB/s-636kB/s), io=64.0MiB (67.1MB), run=52747-53223msec
3645 For each data direction it prints:
3648 Aggregate bandwidth of threads in this group followed by the
3649 minimum and maximum bandwidth of all the threads in this group.
3650 Values outside of brackets are power-of-2 format and those
3651 within are the equivalent value in a power-of-10 format.
3653 Aggregate I/O performed of all threads in this group. The
3654 format is the same as bw.
3656 The smallest and longest runtimes of the threads in this group.
3658 And finally, the disk statistics are printed. This is Linux specific. They will look like this::
3660 Disk stats (read/write):
3661 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3663 Each value is printed for both reads and writes, with reads first. The
3667 Number of I/Os performed by all groups.
3669 Number of merges performed by the I/O scheduler.
3671 Number of ticks we kept the disk busy.
3673 Total time spent in the disk queue.
3675 The disk utilization. A value of 100% means we kept the disk
3676 busy constantly, 50% would be a disk idling half of the time.
3678 It is also possible to get fio to dump the current output while it is running,
3679 without terminating the job. To do that, send fio the **USR1** signal. You can
3680 also get regularly timed dumps by using the :option:`--status-interval`
3681 parameter, or by creating a file in :file:`/tmp` named
3682 :file:`fio-dump-status`. If fio sees this file, it will unlink it and dump the
3683 current output status.
3689 For scripted usage where you typically want to generate tables or graphs of the
3690 results, fio can output the results in a semicolon separated format. The format
3691 is one long line of values, such as::
3693 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%
3694 A description of this job goes here.
3696 The job description (if provided) follows on a second line for terse v2.
3697 It appears on the same line for other terse versions.
3699 To enable terse output, use the :option:`--minimal` or
3700 :option:`--output-format`\=terse command line options. The
3701 first value is the version of the terse output format. If the output has to be
3702 changed for some reason, this number will be incremented by 1 to signify that
3705 Split up, the format is as follows (comments in brackets denote when a
3706 field was introduced or whether it's specific to some terse version):
3710 terse version, fio version [v3], jobname, groupid, error
3714 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3715 Submission latency: min, max, mean, stdev (usec)
3716 Completion latency: min, max, mean, stdev (usec)
3717 Completion latency percentiles: 20 fields (see below)
3718 Total latency: min, max, mean, stdev (usec)
3719 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3720 IOPS [v5]: min, max, mean, stdev, number of samples
3726 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3727 Submission latency: min, max, mean, stdev (usec)
3728 Completion latency: min, max, mean, stdev (usec)
3729 Completion latency percentiles: 20 fields (see below)
3730 Total latency: min, max, mean, stdev (usec)
3731 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3732 IOPS [v5]: min, max, mean, stdev, number of samples
3734 TRIM status [all but version 3]:
3736 Fields are similar to READ/WRITE status.
3740 user, system, context switches, major faults, minor faults
3744 <=1, 2, 4, 8, 16, 32, >=64
3746 I/O latencies microseconds::
3748 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3750 I/O latencies milliseconds::
3752 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3754 Disk utilization [v3]::
3756 disk name, read ios, write ios, read merges, write merges, read ticks, write ticks,
3757 time spent in queue, disk utilization percentage
3759 Additional Info (dependent on continue_on_error, default off)::
3761 total # errors, first error code
3763 Additional Info (dependent on description being set)::
3767 Completion latency percentiles can be a grouping of up to 20 sets, so for the
3768 terse output fio writes all of them. Each field will look like this::
3772 which is the Xth percentile, and the `usec` latency associated with it.
3774 For `Disk utilization`, all disks used by fio are shown. So for each disk there
3775 will be a disk utilization section.
3777 Below is a single line containing short names for each of the fields in the
3778 minimal output v3, separated by semicolons::
3780 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
3782 In client/server mode terse output differs from what appears when jobs are run
3783 locally. Disk utilization data is omitted from the standard terse output and
3784 for v3 and later appears on its own separate line at the end of each terse
3791 The `json` output format is intended to be both human readable and convenient
3792 for automated parsing. For the most part its sections mirror those of the
3793 `normal` output. The `runtime` value is reported in msec and the `bw` value is
3794 reported in 1024 bytes per second units.
3800 The `json+` output format is identical to the `json` output format except that it
3801 adds a full dump of the completion latency bins. Each `bins` object contains a
3802 set of (key, value) pairs where keys are latency durations and values count how
3803 many I/Os had completion latencies of the corresponding duration. For example,
3806 "bins" : { "87552" : 1, "89600" : 1, "94720" : 1, "96768" : 1, "97792" : 1, "99840" : 1, "100864" : 2, "103936" : 6, "104960" : 534, "105984" : 5995, "107008" : 7529, ... }
3808 This data indicates that one I/O required 87,552ns to complete, two I/Os required
3809 100,864ns to complete, and 7529 I/Os required 107,008ns to complete.
3811 Also included with fio is a Python script `fio_jsonplus_clat2csv` that takes
3812 json+ output and generates CSV-formatted latency data suitable for plotting.
3814 The latency durations actually represent the midpoints of latency intervals.
3815 For details refer to :file:`stat.h`.
3821 There are two trace file format that you can encounter. The older (v1) format is
3822 unsupported since version 1.20-rc3 (March 2008). It will still be described
3823 below in case that you get an old trace and want to understand it.
3825 In any case the trace is a simple text file with a single action per line.
3828 Trace file format v1
3829 ~~~~~~~~~~~~~~~~~~~~
3831 Each line represents a single I/O action in the following format::
3835 where `rw=0/1` for read/write, and the `offset` and `length` entries being in bytes.
3837 This format is not supported in fio versions >= 1.20-rc3.
3840 Trace file format v2
3841 ~~~~~~~~~~~~~~~~~~~~
3843 The second version of the trace file format was added in fio version 1.17. It
3844 allows to access more then one file per trace and has a bigger set of possible
3847 The first line of the trace file has to be::
3851 Following this can be lines in two different formats, which are described below.
3853 The file management format::
3857 The `filename` is given as an absolute path. The `action` can be one of these:
3860 Add the given `filename` to the trace.
3862 Open the file with the given `filename`. The `filename` has to have
3863 been added with the **add** action before.
3865 Close the file with the given `filename`. The file has to have been
3869 The file I/O action format::
3871 filename action offset length
3873 The `filename` is given as an absolute path, and has to have been added and
3874 opened before it can be used with this format. The `offset` and `length` are
3875 given in bytes. The `action` can be one of these:
3878 Wait for `offset` microseconds. Everything below 100 is discarded.
3879 The time is relative to the previous `wait` statement.
3881 Read `length` bytes beginning from `offset`.
3883 Write `length` bytes beginning from `offset`.
3885 :manpage:`fsync(2)` the file.
3887 :manpage:`fdatasync(2)` the file.
3889 Trim the given file from the given `offset` for `length` bytes.
3892 I/O Replay - Merging Traces
3893 ---------------------------
3895 Colocation is a common practice used to get the most out of a machine.
3896 Knowing which workloads play nicely with each other and which ones don't is
3897 a much harder task. While fio can replay workloads concurrently via multiple
3898 jobs, it leaves some variability up to the scheduler making results harder to
3899 reproduce. Merging is a way to make the order of events consistent.
3901 Merging is integrated into I/O replay and done when a
3902 :option:`merge_blktrace_file` is specified. The list of files passed to
3903 :option:`read_iolog` go through the merge process and output a single file
3904 stored to the specified file. The output file is passed on as if it were the
3905 only file passed to :option:`read_iolog`. An example would look like::
3907 $ fio --read_iolog="<file1>:<file2>" --merge_blktrace_file="<output_file>"
3909 Creating only the merged file can be done by passing the command line argument
3910 :option:`merge-blktrace-only`.
3912 Scaling traces can be done to see the relative impact of any particular trace
3913 being slowed down or sped up. :option:`merge_blktrace_scalars` takes in a colon
3914 separated list of percentage scalars. It is index paired with the files passed
3915 to :option:`read_iolog`.
3917 With scaling, it may be desirable to match the running time of all traces.
3918 This can be done with :option:`merge_blktrace_iters`. It is index paired with
3919 :option:`read_iolog` just like :option:`merge_blktrace_scalars`.
3921 In an example, given two traces, A and B, each 60s long. If we want to see
3922 the impact of trace A issuing IOs twice as fast and repeat trace A over the
3923 runtime of trace B, the following can be done::
3925 $ fio --read_iolog="<trace_a>:"<trace_b>" --merge_blktrace_file"<output_file>" --merge_blktrace_scalars="50:100" --merge_blktrace_iters="2:1"
3927 This runs trace A at 2x the speed twice for approximately the same runtime as
3928 a single run of trace B.
3931 CPU idleness profiling
3932 ----------------------
3934 In some cases, we want to understand CPU overhead in a test. For example, we
3935 test patches for the specific goodness of whether they reduce CPU usage.
3936 Fio implements a balloon approach to create a thread per CPU that runs at idle
3937 priority, meaning that it only runs when nobody else needs the cpu.
3938 By measuring the amount of work completed by the thread, idleness of each CPU
3939 can be derived accordingly.
3941 An unit work is defined as touching a full page of unsigned characters. Mean and
3942 standard deviation of time to complete an unit work is reported in "unit work"
3943 section. Options can be chosen to report detailed percpu idleness or overall
3944 system idleness by aggregating percpu stats.
3947 Verification and triggers
3948 -------------------------
3950 Fio is usually run in one of two ways, when data verification is done. The first
3951 is a normal write job of some sort with verify enabled. When the write phase has
3952 completed, fio switches to reads and verifies everything it wrote. The second
3953 model is running just the write phase, and then later on running the same job
3954 (but with reads instead of writes) to repeat the same I/O patterns and verify
3955 the contents. Both of these methods depend on the write phase being completed,
3956 as fio otherwise has no idea how much data was written.
3958 With verification triggers, fio supports dumping the current write state to
3959 local files. Then a subsequent read verify workload can load this state and know
3960 exactly where to stop. This is useful for testing cases where power is cut to a
3961 server in a managed fashion, for instance.
3963 A verification trigger consists of two things:
3965 1) Storing the write state of each job.
3966 2) Executing a trigger command.
3968 The write state is relatively small, on the order of hundreds of bytes to single
3969 kilobytes. It contains information on the number of completions done, the last X
3972 A trigger is invoked either through creation ('touch') of a specified file in
3973 the system, or through a timeout setting. If fio is run with
3974 :option:`--trigger-file`\= :file:`/tmp/trigger-file`, then it will continually
3975 check for the existence of :file:`/tmp/trigger-file`. When it sees this file, it
3976 will fire off the trigger (thus saving state, and executing the trigger
3979 For client/server runs, there's both a local and remote trigger. If fio is
3980 running as a server backend, it will send the job states back to the client for
3981 safe storage, then execute the remote trigger, if specified. If a local trigger
3982 is specified, the server will still send back the write state, but the client
3983 will then execute the trigger.
3985 Verification trigger example
3986 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3988 Let's say we want to run a powercut test on the remote Linux machine 'server'.
3989 Our write workload is in :file:`write-test.fio`. We want to cut power to 'server' at
3990 some point during the run, and we'll run this test from the safety or our local
3991 machine, 'localbox'. On the server, we'll start the fio backend normally::
3993 server# fio --server
3995 and on the client, we'll fire off the workload::
3997 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
3999 We set :file:`/tmp/my-trigger` as the trigger file, and we tell fio to execute::
4001 echo b > /proc/sysrq-trigger
4003 on the server once it has received the trigger and sent us the write state. This
4004 will work, but it's not **really** cutting power to the server, it's merely
4005 abruptly rebooting it. If we have a remote way of cutting power to the server
4006 through IPMI or similar, we could do that through a local trigger command
4007 instead. Let's assume we have a script that does IPMI reboot of a given hostname,
4008 ipmi-reboot. On localbox, we could then have run fio with a local trigger
4011 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
4013 For this case, fio would wait for the server to send us the write state, then
4014 execute ``ipmi-reboot server`` when that happened.
4016 Loading verify state
4017 ~~~~~~~~~~~~~~~~~~~~
4019 To load stored write state, a read verification job file must contain the
4020 :option:`verify_state_load` option. If that is set, fio will load the previously
4021 stored state. For a local fio run this is done by loading the files directly,
4022 and on a client/server run, the server backend will ask the client to send the
4023 files over and load them from there.
4029 Fio supports a variety of log file formats, for logging latencies, bandwidth,
4030 and IOPS. The logs share a common format, which looks like this:
4032 *time* (`msec`), *value*, *data direction*, *block size* (`bytes`),
4035 *Time* for the log entry is always in milliseconds. The *value* logged depends
4036 on the type of log, it will be one of the following:
4039 Value is latency in nsecs
4045 *Data direction* is one of the following:
4054 The entry's *block size* is always in bytes. The *offset* is the position in bytes
4055 from the start of the file for that particular I/O. The logging of the offset can be
4056 toggled with :option:`log_offset`.
4058 Fio defaults to logging every individual I/O but when windowed logging is set
4059 through :option:`log_avg_msec`, either the average (by default) or the maximum
4060 (:option:`log_max_value` is set) *value* seen over the specified period of time
4061 is recorded. Each *data direction* seen within the window period will aggregate
4062 its values in a separate row. Further, when using windowed logging the *block
4063 size* and *offset* entries will always contain 0.
4069 Normally fio is invoked as a stand-alone application on the machine where the
4070 I/O workload should be generated. However, the backend and frontend of fio can
4071 be run separately i.e., the fio server can generate an I/O workload on the "Device
4072 Under Test" while being controlled by a client on another machine.
4074 Start the server on the machine which has access to the storage DUT::
4078 where `args` defines what fio listens to. The arguments are of the form
4079 ``type,hostname`` or ``IP,port``. *type* is either ``ip`` (or ip4) for TCP/IP
4080 v4, ``ip6`` for TCP/IP v6, or ``sock`` for a local unix domain socket.
4081 *hostname* is either a hostname or IP address, and *port* is the port to listen
4082 to (only valid for TCP/IP, not a local socket). Some examples:
4086 Start a fio server, listening on all interfaces on the default port (8765).
4088 2) ``fio --server=ip:hostname,4444``
4090 Start a fio server, listening on IP belonging to hostname and on port 4444.
4092 3) ``fio --server=ip6:::1,4444``
4094 Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
4096 4) ``fio --server=,4444``
4098 Start a fio server, listening on all interfaces on port 4444.
4100 5) ``fio --server=1.2.3.4``
4102 Start a fio server, listening on IP 1.2.3.4 on the default port.
4104 6) ``fio --server=sock:/tmp/fio.sock``
4106 Start a fio server, listening on the local socket :file:`/tmp/fio.sock`.
4108 Once a server is running, a "client" can connect to the fio server with::
4110 fio <local-args> --client=<server> <remote-args> <job file(s)>
4112 where `local-args` are arguments for the client where it is running, `server`
4113 is the connect string, and `remote-args` and `job file(s)` are sent to the
4114 server. The `server` string follows the same format as it does on the server
4115 side, to allow IP/hostname/socket and port strings.
4117 Fio can connect to multiple servers this way::
4119 fio --client=<server1> <job file(s)> --client=<server2> <job file(s)>
4121 If the job file is located on the fio server, then you can tell the server to
4122 load a local file as well. This is done by using :option:`--remote-config` ::
4124 fio --client=server --remote-config /path/to/file.fio
4126 Then fio will open this local (to the server) job file instead of being passed
4127 one from the client.
4129 If you have many servers (example: 100 VMs/containers), you can input a pathname
4130 of a file containing host IPs/names as the parameter value for the
4131 :option:`--client` option. For example, here is an example :file:`host.list`
4132 file containing 2 hostnames::
4134 host1.your.dns.domain
4135 host2.your.dns.domain
4137 The fio command would then be::
4139 fio --client=host.list <job file(s)>
4141 In this mode, you cannot input server-specific parameters or job files -- all
4142 servers receive the same job file.
4144 In order to let ``fio --client`` runs use a shared filesystem from multiple
4145 hosts, ``fio --client`` now prepends the IP address of the server to the
4146 filename. For example, if fio is using the directory :file:`/mnt/nfs/fio` and is
4147 writing filename :file:`fileio.tmp`, with a :option:`--client` `hostfile`
4148 containing two hostnames ``h1`` and ``h2`` with IP addresses 192.168.10.120 and
4149 192.168.10.121, then fio will create two files::
4151 /mnt/nfs/fio/192.168.10.120.fileio.tmp
4152 /mnt/nfs/fio/192.168.10.121.fileio.tmp
4154 Terse output in client/server mode will differ slightly from what is produced
4155 when fio is run in stand-alone mode. See the terse output section for details.