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 of various fio actions. May be ``all`` for all types
58 or individual types separated by a comma (e.g. ``--debug=file,mem`` will
59 enable file and memory debugging). Currently, additional logging is
63 Dump info related to processes.
65 Dump info related to file actions.
67 Dump info related to I/O queuing.
69 Dump info related to memory allocations.
71 Dump info related to blktrace setup.
73 Dump info related to I/O verification.
75 Enable all debug options.
77 Dump info related to random offset generation.
79 Dump info related to option matching and parsing.
81 Dump info related to disk utilization updates.
83 Dump info only related to job number x.
85 Dump info only related to mutex up/down ops.
87 Dump info related to profile extensions.
89 Dump info related to internal time keeping.
91 Dump info related to networking connections.
93 Dump info related to I/O rate switching.
95 Dump info related to log compress/decompress.
97 Show available debug options.
99 .. option:: --parse-only
101 Parse options only, don\'t start any I/O.
103 .. option:: --output=filename
105 Write output to file `filename`.
107 .. option:: --output-format=type
109 Set the reporting format to `normal`, `terse`, `json`, or `json+`. Multiple
110 formats can be selected, separated by a comma. `terse` is a CSV based
111 format. `json+` is like `json`, except it adds a full dump of the latency
114 .. option:: --runtime
115 Limit run time to runtime seconds.
117 .. option:: --bandwidth-log
119 Generate aggregate bandwidth logs.
121 .. option:: --minimal
123 Print statistics in a terse, semicolon-delimited format.
125 .. option:: --append-terse
127 Print statistics in selected mode AND terse, semicolon-delimited format.
128 **Deprecated**, use :option:`--output-format` instead to select multiple
131 .. option:: --terse-version=type
133 Set terse version output format (default 3, or 2 or 4 or 5).
135 .. option:: --version
137 Print version information and exit.
141 Print a summary of the command line options and exit.
143 .. option:: --cpuclock-test
145 Perform test and validation of internal CPU clock.
147 .. option:: --crctest=[test]
149 Test the speed of the built-in checksumming functions. If no argument is
150 given, all of them are tested. Alternatively, a comma separated list can
151 be passed, in which case the given ones are tested.
153 .. option:: --cmdhelp=command
155 Print help information for `command`. May be ``all`` for all commands.
157 .. option:: --enghelp=[ioengine[,command]]
159 List all commands defined by :option:`ioengine`, or print help for `command`
160 defined by :option:`ioengine`. If no :option:`ioengine` is given, list all
163 .. option:: --showcmd=jobfile
165 Convert `jobfile` to a set of command-line options.
167 .. option:: --readonly
169 Turn on safety read-only checks, preventing writes. The ``--readonly``
170 option is an extra safety guard to prevent users from accidentally starting
171 a write workload when that is not desired. Fio will only write if
172 `rw=write/randwrite/rw/randrw` is given. This extra safety net can be used
173 as an extra precaution as ``--readonly`` will also enable a write check in
174 the I/O engine core to prevent writes due to unknown user space bug(s).
176 .. option:: --eta=when
178 Specifies when real-time ETA estimate should be printed. `when` may be
179 `always`, `never` or `auto`.
181 .. option:: --eta-newline=time
183 Force a new line for every `time` period passed. When the unit is omitted,
184 the value is interpreted in seconds.
186 .. option:: --status-interval=time
188 Force full status dump every `time` period passed. When the unit is
189 omitted, the value is interpreted in seconds.
191 .. option:: --section=name
193 Only run specified section `name` in job file. Multiple sections can be specified.
194 The ``--section`` option allows one to combine related jobs into one file.
195 E.g. one job file could define light, moderate, and heavy sections. Tell
196 fio to run only the "heavy" section by giving ``--section=heavy``
197 command line option. One can also specify the "write" operations in one
198 section and "verify" operation in another section. The ``--section`` option
199 only applies to job sections. The reserved *global* section is always
202 .. option:: --alloc-size=kb
204 Set the internal smalloc pool size to `kb` in KiB. The
205 ``--alloc-size`` switch allows one to use a larger pool size for smalloc.
206 If running large jobs with randommap enabled, fio can run out of memory.
207 Smalloc is an internal allocator for shared structures from a fixed size
208 memory pool and can grow to 16 pools. The pool size defaults to 16MiB.
210 NOTE: While running :file:`.fio_smalloc.*` backing store files are visible
213 .. option:: --warnings-fatal
215 All fio parser warnings are fatal, causing fio to exit with an
218 .. option:: --max-jobs=nr
220 Set the maximum number of threads/processes to support.
222 .. option:: --server=args
224 Start a backend server, with `args` specifying what to listen to.
225 See `Client/Server`_ section.
227 .. option:: --daemonize=pidfile
229 Background a fio server, writing the pid to the given `pidfile` file.
231 .. option:: --client=hostname
233 Instead of running the jobs locally, send and run them on the given host or
234 set of hosts. See `Client/Server`_ section.
236 .. option:: --remote-config=file
238 Tell fio server to load this local file.
240 .. option:: --idle-prof=option
242 Report CPU idleness. `option` is one of the following:
245 Run unit work calibration only and exit.
248 Show aggregate system idleness and unit work.
251 As **system** but also show per CPU idleness.
253 .. option:: --inflate-log=log
255 Inflate and output compressed log.
257 .. option:: --trigger-file=file
259 Execute trigger cmd when file exists.
261 .. option:: --trigger-timeout=t
263 Execute trigger at this time.
265 .. option:: --trigger=cmd
267 Set this command as local trigger.
269 .. option:: --trigger-remote=cmd
271 Set this command as remote trigger.
273 .. option:: --aux-path=path
275 Use this path for fio state generated files.
277 Any parameters following the options will be assumed to be job files, unless
278 they match a job file parameter. Multiple job files can be listed and each job
279 file will be regarded as a separate group. Fio will :option:`stonewall`
280 execution between each group.
286 As previously described, fio accepts one or more job files describing what it is
287 supposed to do. The job file format is the classic ini file, where the names
288 enclosed in [] brackets define the job name. You are free to use any ASCII name
289 you want, except *global* which has special meaning. Following the job name is
290 a sequence of zero or more parameters, one per line, that define the behavior of
291 the job. If the first character in a line is a ';' or a '#', the entire line is
292 discarded as a comment.
294 A *global* section sets defaults for the jobs described in that file. A job may
295 override a *global* section parameter, and a job file may even have several
296 *global* sections if so desired. A job is only affected by a *global* section
299 The :option:`--cmdhelp` option also lists all options. If used with an `option`
300 argument, :option:`--cmdhelp` will detail the given `option`.
302 See the `examples/` directory for inspiration on how to write job files. Note
303 the copyright and license requirements currently apply to `examples/` files.
305 So let's look at a really simple job file that defines two processes, each
306 randomly reading from a 128MiB file:
310 ; -- start job file --
321 As you can see, the job file sections themselves are empty as all the described
322 parameters are shared. As no :option:`filename` option is given, fio makes up a
323 `filename` for each of the jobs as it sees fit. On the command line, this job
324 would look as follows::
326 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
329 Let's look at an example that has a number of processes writing randomly to
334 ; -- start job file --
345 Here we have no *global* section, as we only have one job defined anyway. We
346 want to use async I/O here, with a depth of 4 for each file. We also increased
347 the buffer size used to 32KiB and define numjobs to 4 to fork 4 identical
348 jobs. The result is 4 processes each randomly writing to their own 64MiB
349 file. Instead of using the above job file, you could have given the parameters
350 on the command line. For this case, you would specify::
352 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
354 When fio is utilized as a basis of any reasonably large test suite, it might be
355 desirable to share a set of standardized settings across multiple job files.
356 Instead of copy/pasting such settings, any section may pull in an external
357 :file:`filename.fio` file with *include filename* directive, as in the following
360 ; -- start job file including.fio --
364 include glob-include.fio
371 include test-include.fio
372 ; -- end job file including.fio --
376 ; -- start job file glob-include.fio --
379 ; -- end job file glob-include.fio --
383 ; -- start job file test-include.fio --
386 ; -- end job file test-include.fio --
388 Settings pulled into a section apply to that section only (except *global*
389 section). Include directives may be nested in that any included file may contain
390 further include directive(s). Include files may not contain [] sections.
393 Environment variables
394 ~~~~~~~~~~~~~~~~~~~~~
396 Fio also supports environment variable expansion in job files. Any sub-string of
397 the form ``${VARNAME}`` as part of an option value (in other words, on the right
398 of the '='), will be expanded to the value of the environment variable called
399 `VARNAME`. If no such environment variable is defined, or `VARNAME` is the
400 empty string, the empty string will be substituted.
402 As an example, let's look at a sample fio invocation and job file::
404 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
408 ; -- start job file --
415 This will expand to the following equivalent job file at runtime:
419 ; -- start job file --
426 Fio ships with a few example job files, you can also look there for inspiration.
431 Additionally, fio has a set of reserved keywords that will be replaced
432 internally with the appropriate value. Those keywords are:
436 The architecture page size of the running system.
440 Megabytes of total memory in the system.
444 Number of online available CPUs.
446 These can be used on the command line or in the job file, and will be
447 automatically substituted with the current system values when the job is
448 run. Simple math is also supported on these keywords, so you can perform actions
453 and get that properly expanded to 8 times the size of memory in the machine.
459 This section describes in details each parameter associated with a job. Some
460 parameters take an option of a given type, such as an integer or a
461 string. Anywhere a numeric value is required, an arithmetic expression may be
462 used, provided it is surrounded by parentheses. Supported operators are:
471 For time values in expressions, units are microseconds by default. This is
472 different than for time values not in expressions (not enclosed in
473 parentheses). The following types are used:
480 String: A sequence of alphanumeric characters.
483 Integer with possible time suffix. Without a unit value is interpreted as
484 seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
485 hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and
486 'us' (or 'usec') for microseconds. For example, use 10m for 10 minutes.
491 Integer. A whole number value, which may contain an integer prefix
492 and an integer suffix:
494 [*integer prefix*] **number** [*integer suffix*]
496 The optional *integer prefix* specifies the number's base. The default
497 is decimal. *0x* specifies hexadecimal.
499 The optional *integer suffix* specifies the number's units, and includes an
500 optional unit prefix and an optional unit. For quantities of data, the
501 default unit is bytes. For quantities of time, the default unit is seconds
502 unless otherwise specified.
504 With :option:`kb_base`\=1000, fio follows international standards for unit
505 prefixes. To specify power-of-10 decimal values defined in the
506 International System of Units (SI):
508 * *Ki* -- means kilo (K) or 1000
509 * *Mi* -- means mega (M) or 1000**2
510 * *Gi* -- means giga (G) or 1000**3
511 * *Ti* -- means tera (T) or 1000**4
512 * *Pi* -- means peta (P) or 1000**5
514 To specify power-of-2 binary values defined in IEC 80000-13:
516 * *k* -- means kibi (Ki) or 1024
517 * *M* -- means mebi (Mi) or 1024**2
518 * *G* -- means gibi (Gi) or 1024**3
519 * *T* -- means tebi (Ti) or 1024**4
520 * *P* -- means pebi (Pi) or 1024**5
522 With :option:`kb_base`\=1024 (the default), the unit prefixes are opposite
523 from those specified in the SI and IEC 80000-13 standards to provide
524 compatibility with old scripts. For example, 4k means 4096.
526 For quantities of data, an optional unit of 'B' may be included
527 (e.g., 'kB' is the same as 'k').
529 The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
530 not milli). 'b' and 'B' both mean byte, not bit.
532 Examples with :option:`kb_base`\=1000:
534 * *4 KiB*: 4096, 4096b, 4096B, 4ki, 4kib, 4kiB, 4Ki, 4KiB
535 * *1 MiB*: 1048576, 1mi, 1024ki
536 * *1 MB*: 1000000, 1m, 1000k
537 * *1 TiB*: 1099511627776, 1ti, 1024gi, 1048576mi
538 * *1 TB*: 1000000000, 1t, 1000m, 1000000k
540 Examples with :option:`kb_base`\=1024 (default):
542 * *4 KiB*: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
543 * *1 MiB*: 1048576, 1m, 1024k
544 * *1 MB*: 1000000, 1mi, 1000ki
545 * *1 TiB*: 1099511627776, 1t, 1024g, 1048576m
546 * *1 TB*: 1000000000, 1ti, 1000mi, 1000000ki
548 To specify times (units are not case sensitive):
552 * *M* -- means minutes
553 * *s* -- or sec means seconds (default)
554 * *ms* -- or *msec* means milliseconds
555 * *us* -- or *usec* means microseconds
557 If the option accepts an upper and lower range, use a colon ':' or
558 minus '-' to separate such values. See :ref:`irange <irange>`.
559 If the lower value specified happens to be larger than the upper value
560 the two values are swapped.
565 Boolean. Usually parsed as an integer, however only defined for
566 true and false (1 and 0).
571 Integer range with suffix. Allows value range to be given, such as
572 1024-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
573 option allows two sets of ranges, they can be specified with a ',' or '/'
574 delimiter: 1k-4k/8k-32k. Also see :ref:`int <int>`.
577 A list of floating point numbers, separated by a ':' character.
583 .. option:: kb_base=int
585 Select the interpretation of unit prefixes in input parameters.
588 Inputs comply with IEC 80000-13 and the International
589 System of Units (SI). Use:
591 - power-of-2 values with IEC prefixes (e.g., KiB)
592 - power-of-10 values with SI prefixes (e.g., kB)
595 Compatibility mode (default). To avoid breaking old scripts:
597 - power-of-2 values with SI prefixes
598 - power-of-10 values with IEC prefixes
600 See :option:`bs` for more details on input parameters.
602 Outputs always use correct prefixes. Most outputs include both
605 bw=2383.3kB/s (2327.4KiB/s)
607 If only one value is reported, then kb_base selects the one to use:
609 **1000** -- SI prefixes
611 **1024** -- IEC prefixes
613 .. option:: unit_base=int
615 Base unit for reporting. Allowed values are:
618 Use auto-detection (default).
625 With the above in mind, here follows the complete list of fio job parameters.
633 ASCII name of the job. This may be used to override the name printed by fio
634 for this job. Otherwise the job name is used. On the command line this
635 parameter has the special purpose of also signaling the start of a new job.
637 .. option:: description=str
639 Text description of the job. Doesn't do anything except dump this text
640 description when this job is run. It's not parsed.
642 .. option:: loops=int
644 Run the specified number of iterations of this job. Used to repeat the same
645 workload a given number of times. Defaults to 1.
647 .. option:: numjobs=int
649 Create the specified number of clones of this job. Each clone of job
650 is spawned as an independent thread or process. May be used to setup a
651 larger number of threads/processes doing the same thing. Each thread is
652 reported separately; to see statistics for all clones as a whole, use
653 :option:`group_reporting` in conjunction with :option:`new_group`.
654 See :option:`--max-jobs`. Default: 1.
657 Time related parameters
658 ~~~~~~~~~~~~~~~~~~~~~~~
660 .. option:: runtime=time
662 Tell fio to terminate processing after the specified period of time. It
663 can be quite hard to determine for how long a specified job will run, so
664 this parameter is handy to cap the total runtime to a given time. When
665 the unit is omitted, the value is intepreted in seconds.
667 .. option:: time_based
669 If set, fio will run for the duration of the :option:`runtime` specified
670 even if the file(s) are completely read or written. It will simply loop over
671 the same workload as many times as the :option:`runtime` allows.
673 .. option:: startdelay=irange(time)
675 Delay the start of job for the specified amount of time. Can be a single
676 value or a range. When given as a range, each thread will choose a value
677 randomly from within the range. Value is in seconds if a unit is omitted.
679 .. option:: ramp_time=time
681 If set, fio will run the specified workload for this amount of time before
682 logging any performance numbers. Useful for letting performance settle
683 before logging results, thus minimizing the runtime required for stable
684 results. Note that the ``ramp_time`` is considered lead in time for a job,
685 thus it will increase the total runtime if a special timeout or
686 :option:`runtime` is specified. When the unit is omitted, the value is
689 .. option:: clocksource=str
691 Use the given clocksource as the base of timing. The supported options are:
694 :manpage:`gettimeofday(2)`
697 :manpage:`clock_gettime(2)`
700 Internal CPU clock source
702 cpu is the preferred clocksource if it is reliable, as it is very fast (and
703 fio is heavy on time calls). Fio will automatically use this clocksource if
704 it's supported and considered reliable on the system it is running on,
705 unless another clocksource is specifically set. For x86/x86-64 CPUs, this
706 means supporting TSC Invariant.
708 .. option:: gtod_reduce=bool
710 Enable all of the :manpage:`gettimeofday(2)` reducing options
711 (:option:`disable_clat`, :option:`disable_slat`, :option:`disable_bw_measurement`) plus
712 reduce precision of the timeout somewhat to really shrink the
713 :manpage:`gettimeofday(2)` call count. With this option enabled, we only do
714 about 0.4% of the :manpage:`gettimeofday(2)` calls we would have done if all
715 time keeping was enabled.
717 .. option:: gtod_cpu=int
719 Sometimes it's cheaper to dedicate a single thread of execution to just
720 getting the current time. Fio (and databases, for instance) are very
721 intensive on :manpage:`gettimeofday(2)` calls. With this option, you can set
722 one CPU aside for doing nothing but logging current time to a shared memory
723 location. Then the other threads/processes that run I/O workloads need only
724 copy that segment, instead of entering the kernel with a
725 :manpage:`gettimeofday(2)` call. The CPU set aside for doing these time
726 calls will be excluded from other uses. Fio will manually clear it from the
727 CPU mask of other jobs.
733 .. option:: directory=str
735 Prefix filenames with this directory. Used to place files in a different
736 location than :file:`./`. You can specify a number of directories by
737 separating the names with a ':' character. These directories will be
738 assigned equally distributed to job clones created by :option:`numjobs` as
739 long as they are using generated filenames. If specific `filename(s)` are
740 set fio will use the first listed directory, and thereby matching the
741 `filename` semantic which generates a file each clone if not specified, but
742 let all clones use the same if set.
744 See the :option:`filename` option for information on how to escape "``:``" and
745 "``\``" characters within the directory path itself.
747 .. option:: filename=str
749 Fio normally makes up a `filename` based on the job name, thread number, and
750 file number (see :option:`filename_format`). If you want to share files
751 between threads in a job or several
752 jobs with fixed file paths, specify a `filename` for each of them to override
753 the default. If the ioengine is file based, you can specify a number of files
754 by separating the names with a ':' colon. So if you wanted a job to open
755 :file:`/dev/sda` and :file:`/dev/sdb` as the two working files, you would use
756 ``filename=/dev/sda:/dev/sdb``. This also means that whenever this option is
757 specified, :option:`nrfiles` is ignored. The size of regular files specified
758 by this option will be :option:`size` divided by number of files unless an
759 explicit size is specified by :option:`filesize`.
761 Each colon and backslash in the wanted path must be escaped with a ``\``
762 character. For instance, if the path is :file:`/dev/dsk/foo@3,0:c` then you
763 would use ``filename=/dev/dsk/foo@3,0\:c`` and if the path is
764 :file:`F:\\filename` then you would use ``filename=F\:\\filename``.
766 On Windows, disk devices are accessed as :file:`\\\\.\\PhysicalDrive0` for
767 the first device, :file:`\\\\.\\PhysicalDrive1` for the second etc.
768 Note: Windows and FreeBSD prevent write access to areas
769 of the disk containing in-use data (e.g. filesystems).
771 The filename "`-`" is a reserved name, meaning *stdin* or *stdout*. Which
772 of the two depends on the read/write direction set.
774 .. option:: filename_format=str
776 If sharing multiple files between jobs, it is usually necessary to have fio
777 generate the exact names that you want. By default, fio will name a file
778 based on the default file format specification of
779 :file:`jobname.jobnumber.filenumber`. With this option, that can be
780 customized. Fio will recognize and replace the following keywords in this
784 The name of the worker thread or process.
786 The incremental number of the worker thread or process.
788 The incremental number of the file for that worker thread or
791 To have dependent jobs share a set of files, this option can be set to have
792 fio generate filenames that are shared between the two. For instance, if
793 :file:`testfiles.$filenum` is specified, file number 4 for any job will be
794 named :file:`testfiles.4`. The default of :file:`$jobname.$jobnum.$filenum`
795 will be used if no other format specifier is given.
797 .. option:: unique_filename=bool
799 To avoid collisions between networked clients, fio defaults to prefixing any
800 generated filenames (with a directory specified) with the source of the
801 client connecting. To disable this behavior, set this option to 0.
803 .. option:: opendir=str
805 Recursively open any files below directory `str`.
807 .. option:: lockfile=str
809 Fio defaults to not locking any files before it does I/O to them. If a file
810 or file descriptor is shared, fio can serialize I/O to that file to make the
811 end result consistent. This is usual for emulating real workloads that share
812 files. The lock modes are:
815 No locking. The default.
817 Only one thread or process may do I/O at a time, excluding all
820 Read-write locking on the file. Many readers may
821 access the file at the same time, but writes get exclusive access.
823 .. option:: nrfiles=int
825 Number of files to use for this job. Defaults to 1. The size of files
826 will be :option:`size` divided by this unless explicit size is specified by
827 :option:`filesize`. Files are created for each thread separately, and each
828 file will have a file number within its name by default, as explained in
829 :option:`filename` section.
832 .. option:: openfiles=int
834 Number of files to keep open at the same time. Defaults to the same as
835 :option:`nrfiles`, can be set smaller to limit the number simultaneous
838 .. option:: file_service_type=str
840 Defines how fio decides which file from a job to service next. The following
844 Choose a file at random.
847 Round robin over opened files. This is the default.
850 Finish one file before moving on to the next. Multiple files can
851 still be open depending on 'openfiles'.
854 Use a *Zipf* distribution to decide what file to access.
857 Use a *Pareto* distribution to decide what file to access.
860 Use a *Gaussian* (normal) distribution to decide what file to
866 For *random*, *roundrobin*, and *sequential*, a postfix can be appended to
867 tell fio how many I/Os to issue before switching to a new file. For example,
868 specifying ``file_service_type=random:8`` would cause fio to issue
869 8 I/Os before selecting a new file at random. For the non-uniform
870 distributions, a floating point postfix can be given to influence how the
871 distribution is skewed. See :option:`random_distribution` for a description
872 of how that would work.
874 .. option:: ioscheduler=str
876 Attempt to switch the device hosting the file to the specified I/O scheduler
879 .. option:: create_serialize=bool
881 If true, serialize the file creation for the jobs. This may be handy to
882 avoid interleaving of data files, which may greatly depend on the filesystem
883 used and even the number of processors in the system. Default: true.
885 .. option:: create_fsync=bool
887 :manpage:`fsync(2)` the data file after creation. This is the default.
889 .. option:: create_on_open=bool
891 If true, don't pre-create files but allow the job's open() to create a file
892 when it's time to do I/O. Default: false -- pre-create all necessary files
895 .. option:: create_only=bool
897 If true, fio will only run the setup phase of the job. If files need to be
898 laid out or updated on disk, only that will be done -- the actual job contents
899 are not executed. Default: false.
901 .. option:: allow_file_create=bool
903 If true, fio is permitted to create files as part of its workload. If this
904 option is false, then fio will error out if
905 the files it needs to use don't already exist. Default: true.
907 .. option:: allow_mounted_write=bool
909 If this isn't set, fio will abort jobs that are destructive (e.g. that write)
910 to what appears to be a mounted device or partition. This should help catch
911 creating inadvertently destructive tests, not realizing that the test will
912 destroy data on the mounted file system. Note that some platforms don't allow
913 writing against a mounted device regardless of this option. Default: false.
915 .. option:: pre_read=bool
917 If this is given, files will be pre-read into memory before starting the
918 given I/O operation. This will also clear the :option:`invalidate` flag,
919 since it is pointless to pre-read and then drop the cache. This will only
920 work for I/O engines that are seek-able, since they allow you to read the
921 same data multiple times. Thus it will not work on non-seekable I/O engines
922 (e.g. network, splice). Default: false.
924 .. option:: unlink=bool
926 Unlink the job files when done. Not the default, as repeated runs of that
927 job would then waste time recreating the file set again and again. Default:
930 .. option:: unlink_each_loop=bool
932 Unlink job files after each iteration or loop. Default: false.
934 .. option:: zonesize=int
936 Divide a file into zones of the specified size. See :option:`zoneskip`.
938 .. option:: zonerange=int
940 Give size of an I/O zone. See :option:`zoneskip`.
942 .. option:: zoneskip=int
944 Skip the specified number of bytes when :option:`zonesize` data has been
945 read. The two zone options can be used to only do I/O on zones of a file.
951 .. option:: direct=bool
953 If value is true, use non-buffered I/O. This is usually O_DIRECT. Note that
954 ZFS on Solaris doesn't support direct I/O. On Windows the synchronous
955 ioengines don't support direct I/O. Default: false.
957 .. option:: atomic=bool
959 If value is true, attempt to use atomic direct I/O. Atomic writes are
960 guaranteed to be stable once acknowledged by the operating system. Only
961 Linux supports O_ATOMIC right now.
963 .. option:: buffered=bool
965 If value is true, use buffered I/O. This is the opposite of the
966 :option:`direct` option. Defaults to true.
968 .. option:: readwrite=str, rw=str
970 Type of I/O pattern. Accepted values are:
977 Sequential trims (Linux block devices only).
983 Random trims (Linux block devices only).
985 Sequential mixed reads and writes.
987 Random mixed reads and writes.
989 Sequential trim+write sequences. Blocks will be trimmed first,
990 then the same blocks will be written to.
992 Fio defaults to read if the option is not specified. For the mixed I/O
993 types, the default is to split them 50/50. For certain types of I/O the
994 result may still be skewed a bit, since the speed may be different.
996 It is possible to specify the number of I/Os to do before getting a new
997 offset by appending ``:<nr>`` to the end of the string given. For a
998 random read, it would look like ``rw=randread:8`` for passing in an offset
999 modifier with a value of 8. If the suffix is used with a sequential I/O
1000 pattern, then the *<nr>* value specified will be **added** to the generated
1001 offset for each I/O turning sequential I/O into sequential I/O with holes.
1002 For instance, using ``rw=write:4k`` will skip 4k for every write. Also see
1003 the :option:`rw_sequencer` option.
1005 .. option:: rw_sequencer=str
1007 If an offset modifier is given by appending a number to the ``rw=<str>``
1008 line, then this option controls how that number modifies the I/O offset
1009 being generated. Accepted values are:
1012 Generate sequential offset.
1014 Generate the same offset.
1016 ``sequential`` is only useful for random I/O, where fio would normally
1017 generate a new random offset for every I/O. If you append e.g. 8 to randread,
1018 you would get a new random offset for every 8 I/O's. The result would be a
1019 seek for only every 8 I/O's, instead of for every I/O. Use ``rw=randread:8``
1020 to specify that. As sequential I/O is already sequential, setting
1021 ``sequential`` for that would not result in any differences. ``identical``
1022 behaves in a similar fashion, except it sends the same offset 8 number of
1023 times before generating a new offset.
1025 .. option:: unified_rw_reporting=bool
1027 Fio normally reports statistics on a per data direction basis, meaning that
1028 reads, writes, and trims are accounted and reported separately. If this
1029 option is set fio sums the results and report them as "mixed" instead.
1031 .. option:: randrepeat=bool
1033 Seed the random number generator used for random I/O patterns in a
1034 predictable way so the pattern is repeatable across runs. Default: true.
1036 .. option:: allrandrepeat=bool
1038 Seed all random number generators in a predictable way so results are
1039 repeatable across runs. Default: false.
1041 .. option:: randseed=int
1043 Seed the random number generators based on this seed value, to be able to
1044 control what sequence of output is being generated. If not set, the random
1045 sequence depends on the :option:`randrepeat` setting.
1047 .. option:: fallocate=str
1049 Whether pre-allocation is performed when laying down files.
1050 Accepted values are:
1053 Do not pre-allocate space.
1056 Use a platform's native pre-allocation call but fall back to
1057 **none** behavior if it fails/is not implemented.
1060 Pre-allocate via :manpage:`posix_fallocate(3)`.
1063 Pre-allocate via :manpage:`fallocate(2)` with
1064 FALLOC_FL_KEEP_SIZE set.
1067 Backward-compatible alias for **none**.
1070 Backward-compatible alias for **posix**.
1072 May not be available on all supported platforms. **keep** is only available
1073 on Linux. If using ZFS on Solaris this cannot be set to **posix**
1074 because ZFS doesn't support pre-allocation. Default: **native** if any
1075 pre-allocation methods are available, **none** if not.
1077 .. option:: fadvise_hint=str
1079 Use :manpage:`posix_fadvise(2)` to advise the kernel on what I/O patterns
1080 are likely to be issued. Accepted values are:
1083 Backwards-compatible hint for "no hint".
1086 Backwards compatible hint for "advise with fio workload type". This
1087 uses **FADV_RANDOM** for a random workload, and **FADV_SEQUENTIAL**
1088 for a sequential workload.
1091 Advise using **FADV_SEQUENTIAL**.
1094 Advise using **FADV_RANDOM**.
1096 .. option:: fadvise_stream=int
1098 Use :manpage:`posix_fadvise(2)` to advise the kernel what stream ID the
1099 writes issued belong to. Only supported on Linux. Note, this option may
1100 change going forward.
1102 .. option:: offset=int
1104 Start I/O at the provided offset in the file, given as either a fixed size in
1105 bytes or a percentage. If a percentage is given, the next ``blockalign``-ed
1106 offset will be used. Data before the given offset will not be touched. This
1107 effectively caps the file size at `real_size - offset`. Can be combined with
1108 :option:`size` to constrain the start and end range of the I/O workload.
1109 A percentage can be specified by a number between 1 and 100 followed by '%',
1110 for example, ``offset=20%`` to specify 20%.
1112 .. option:: offset_increment=int
1114 If this is provided, then the real offset becomes `offset + offset_increment
1115 * thread_number`, where the thread number is a counter that starts at 0 and
1116 is incremented for each sub-job (i.e. when :option:`numjobs` option is
1117 specified). This option is useful if there are several jobs which are
1118 intended to operate on a file in parallel disjoint segments, with even
1119 spacing between the starting points.
1121 .. option:: number_ios=int
1123 Fio will normally perform I/Os until it has exhausted the size of the region
1124 set by :option:`size`, or if it exhaust the allocated time (or hits an error
1125 condition). With this setting, the range/size can be set independently of
1126 the number of I/Os to perform. When fio reaches this number, it will exit
1127 normally and report status. Note that this does not extend the amount of I/O
1128 that will be done, it will only stop fio if this condition is met before
1129 other end-of-job criteria.
1131 .. option:: fsync=int
1133 If writing to a file, issue an :manpage:`fsync(2)` (or its equivalent) of
1134 the dirty data for every number of blocks given. For example, if you give 32
1135 as a parameter, fio will sync the file after every 32 writes issued. If fio is
1136 using non-buffered I/O, we may not sync the file. The exception is the sg
1137 I/O engine, which synchronizes the disk cache anyway. Defaults to 0, which
1138 means fio does not periodically issue and wait for a sync to complete. Also
1139 see :option:`end_fsync` and :option:`fsync_on_close`.
1141 .. option:: fdatasync=int
1143 Like :option:`fsync` but uses :manpage:`fdatasync(2)` to only sync data and
1144 not metadata blocks. In Windows, FreeBSD, and DragonFlyBSD there is no
1145 :manpage:`fdatasync(2)` so this falls back to using :manpage:`fsync(2)`.
1146 Defaults to 0, which means fio does not periodically issue and wait for a
1147 data-only sync to complete.
1149 .. option:: write_barrier=int
1151 Make every `N-th` write a barrier write.
1153 .. option:: sync_file_range=str:val
1155 Use :manpage:`sync_file_range(2)` for every `val` number of write
1156 operations. Fio will track range of writes that have happened since the last
1157 :manpage:`sync_file_range(2)` call. `str` can currently be one or more of:
1160 SYNC_FILE_RANGE_WAIT_BEFORE
1162 SYNC_FILE_RANGE_WRITE
1164 SYNC_FILE_RANGE_WAIT_AFTER
1166 So if you do ``sync_file_range=wait_before,write:8``, fio would use
1167 ``SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE`` for every 8
1168 writes. Also see the :manpage:`sync_file_range(2)` man page. This option is
1171 .. option:: overwrite=bool
1173 If true, writes to a file will always overwrite existing data. If the file
1174 doesn't already exist, it will be created before the write phase begins. If
1175 the file exists and is large enough for the specified write phase, nothing
1176 will be done. Default: false.
1178 .. option:: end_fsync=bool
1180 If true, :manpage:`fsync(2)` file contents when a write stage has completed.
1183 .. option:: fsync_on_close=bool
1185 If true, fio will :manpage:`fsync(2)` a dirty file on close. This differs
1186 from :option:`end_fsync` in that it will happen on every file close, not
1187 just at the end of the job. Default: false.
1189 .. option:: rwmixread=int
1191 Percentage of a mixed workload that should be reads. Default: 50.
1193 .. option:: rwmixwrite=int
1195 Percentage of a mixed workload that should be writes. If both
1196 :option:`rwmixread` and :option:`rwmixwrite` is given and the values do not
1197 add up to 100%, the latter of the two will be used to override the
1198 first. This may interfere with a given rate setting, if fio is asked to
1199 limit reads or writes to a certain rate. If that is the case, then the
1200 distribution may be skewed. Default: 50.
1202 .. option:: random_distribution=str:float[,str:float][,str:float]
1204 By default, fio will use a completely uniform random distribution when asked
1205 to perform random I/O. Sometimes it is useful to skew the distribution in
1206 specific ways, ensuring that some parts of the data is more hot than others.
1207 fio includes the following distribution models:
1210 Uniform random distribution
1219 Normal (Gaussian) distribution
1222 Zoned random distribution
1224 When using a **zipf** or **pareto** distribution, an input value is also
1225 needed to define the access pattern. For **zipf**, this is the `zipf
1226 theta`. For **pareto**, it's the `Pareto power`. Fio includes a test
1227 program, :command:`genzipf`, that can be used visualize what the given input
1228 values will yield in terms of hit rates. If you wanted to use **zipf** with
1229 a `theta` of 1.2, you would use ``random_distribution=zipf:1.2`` as the
1230 option. If a non-uniform model is used, fio will disable use of the random
1231 map. For the **normal** distribution, a normal (Gaussian) deviation is
1232 supplied as a value between 0 and 100.
1234 For a **zoned** distribution, fio supports specifying percentages of I/O
1235 access that should fall within what range of the file or device. For
1236 example, given a criteria of:
1238 * 60% of accesses should be to the first 10%
1239 * 30% of accesses should be to the next 20%
1240 * 8% of accesses should be to to the next 30%
1241 * 2% of accesses should be to the next 40%
1243 we can define that through zoning of the random accesses. For the above
1244 example, the user would do::
1246 random_distribution=zoned:60/10:30/20:8/30:2/40
1248 similarly to how :option:`bssplit` works for setting ranges and percentages
1249 of block sizes. Like :option:`bssplit`, it's possible to specify separate
1250 zones for reads, writes, and trims. If just one set is given, it'll apply to
1253 .. option:: percentage_random=int[,int][,int]
1255 For a random workload, set how big a percentage should be random. This
1256 defaults to 100%, in which case the workload is fully random. It can be set
1257 from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1258 sequential. Any setting in between will result in a random mix of sequential
1259 and random I/O, at the given percentages. Comma-separated values may be
1260 specified for reads, writes, and trims as described in :option:`blocksize`.
1262 .. option:: norandommap
1264 Normally fio will cover every block of the file when doing random I/O. If
1265 this option is given, fio will just get a new random offset without looking
1266 at past I/O history. This means that some blocks may not be read or written,
1267 and that some blocks may be read/written more than once. If this option is
1268 used with :option:`verify` and multiple blocksizes (via :option:`bsrange`),
1269 only intact blocks are verified, i.e., partially-overwritten blocks are
1272 .. option:: softrandommap=bool
1274 See :option:`norandommap`. If fio runs with the random block map enabled and
1275 it fails to allocate the map, if this option is set it will continue without
1276 a random block map. As coverage will not be as complete as with random maps,
1277 this option is disabled by default.
1279 .. option:: random_generator=str
1281 Fio supports the following engines for generating
1282 I/O offsets for random I/O:
1285 Strong 2^88 cycle random number generator
1287 Linear feedback shift register generator
1289 Strong 64-bit 2^258 cycle random number generator
1291 **tausworthe** is a strong random number generator, but it requires tracking
1292 on the side if we want to ensure that blocks are only read or written
1293 once. **LFSR** guarantees that we never generate the same offset twice, and
1294 it's also less computationally expensive. It's not a true random generator,
1295 however, though for I/O purposes it's typically good enough. **LFSR** only
1296 works with single block sizes, not with workloads that use multiple block
1297 sizes. If used with such a workload, fio may read or write some blocks
1298 multiple times. The default value is **tausworthe**, unless the required
1299 space exceeds 2^32 blocks. If it does, then **tausworthe64** is
1300 selected automatically.
1306 .. option:: blocksize=int[,int][,int], bs=int[,int][,int]
1308 The block size in bytes used for I/O units. Default: 4096. A single value
1309 applies to reads, writes, and trims. Comma-separated values may be
1310 specified for reads, writes, and trims. A value not terminated in a comma
1311 applies to subsequent types.
1316 means 256k for reads, writes and trims.
1319 means 8k for reads, 32k for writes and trims.
1322 means 8k for reads, 32k for writes, and default for trims.
1325 means default for reads, 8k for writes and trims.
1328 means default for reads, 8k for writes, and default for trims.
1330 .. option:: blocksize_range=irange[,irange][,irange], bsrange=irange[,irange][,irange]
1332 A range of block sizes in bytes for I/O units. The issued I/O unit will
1333 always be a multiple of the minimum size, unless
1334 :option:`blocksize_unaligned` is set.
1336 Comma-separated ranges may be specified for reads, writes, and trims as
1337 described in :option:`blocksize`.
1339 Example: ``bsrange=1k-4k,2k-8k``.
1341 .. option:: bssplit=str[,str][,str]
1343 Sometimes you want even finer grained control of the block sizes issued, not
1344 just an even split between them. This option allows you to weight various
1345 block sizes, so that you are able to define a specific amount of block sizes
1346 issued. The format for this option is::
1348 bssplit=blocksize/percentage:blocksize/percentage
1350 for as many block sizes as needed. So if you want to define a workload that
1351 has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would write::
1353 bssplit=4k/10:64k/50:32k/40
1355 Ordering does not matter. If the percentage is left blank, fio will fill in
1356 the remaining values evenly. So a bssplit option like this one::
1358 bssplit=4k/50:1k/:32k/
1360 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always add up
1361 to 100, if bssplit is given a range that adds up to more, it will error out.
1363 Comma-separated values may be specified for reads, writes, and trims as
1364 described in :option:`blocksize`.
1366 If you want a workload that has 50% 2k reads and 50% 4k reads, while having
1367 90% 4k writes and 10% 8k writes, you would specify::
1369 bssplit=2k/50:4k/50,4k/90,8k/10
1371 .. option:: blocksize_unaligned, bs_unaligned
1373 If set, fio will issue I/O units with any size within
1374 :option:`blocksize_range`, not just multiples of the minimum size. This
1375 typically won't work with direct I/O, as that normally requires sector
1378 .. option:: bs_is_seq_rand
1380 If this option is set, fio will use the normal read,write blocksize settings
1381 as sequential,random blocksize settings instead. Any random read or write
1382 will use the WRITE blocksize settings, and any sequential read or write will
1383 use the READ blocksize settings.
1385 .. option:: blockalign=int[,int][,int], ba=int[,int][,int]
1387 Boundary to which fio will align random I/O units. Default:
1388 :option:`blocksize`. Minimum alignment is typically 512b for using direct
1389 I/O, though it usually depends on the hardware block size. This option is
1390 mutually exclusive with using a random map for files, so it will turn off
1391 that option. Comma-separated values may be specified for reads, writes, and
1392 trims as described in :option:`blocksize`.
1398 .. option:: zero_buffers
1400 Initialize buffers with all zeros. Default: fill buffers with random data.
1402 .. option:: refill_buffers
1404 If this option is given, fio will refill the I/O buffers on every
1405 submit. The default is to only fill it at init time and reuse that
1406 data. Only makes sense if zero_buffers isn't specified, naturally. If data
1407 verification is enabled, `refill_buffers` is also automatically enabled.
1409 .. option:: scramble_buffers=bool
1411 If :option:`refill_buffers` is too costly and the target is using data
1412 deduplication, then setting this option will slightly modify the I/O buffer
1413 contents to defeat normal de-dupe attempts. This is not enough to defeat
1414 more clever block compression attempts, but it will stop naive dedupe of
1415 blocks. Default: true.
1417 .. option:: buffer_compress_percentage=int
1419 If this is set, then fio will attempt to provide I/O buffer content (on
1420 WRITEs) that compresses to the specified level. Fio does this by providing a
1421 mix of random data and a fixed pattern. The fixed pattern is either zeros,
1422 or the pattern specified by :option:`buffer_pattern`. If the pattern option
1423 is used, it might skew the compression ratio slightly. Note that this is per
1424 block size unit, for file/disk wide compression level that matches this
1425 setting, you'll also want to set :option:`refill_buffers`.
1427 .. option:: buffer_compress_chunk=int
1429 See :option:`buffer_compress_percentage`. This setting allows fio to manage
1430 how big the ranges of random data and zeroed data is. Without this set, fio
1431 will provide :option:`buffer_compress_percentage` of blocksize random data,
1432 followed by the remaining zeroed. With this set to some chunk size smaller
1433 than the block size, fio can alternate random and zeroed data throughout the
1436 .. option:: buffer_pattern=str
1438 If set, fio will fill the I/O buffers with this pattern or with the contents
1439 of a file. If not set, the contents of I/O buffers are defined by the other
1440 options related to buffer contents. The setting can be any pattern of bytes,
1441 and can be prefixed with 0x for hex values. It may also be a string, where
1442 the string must then be wrapped with ``""``. Or it may also be a filename,
1443 where the filename must be wrapped with ``''`` in which case the file is
1444 opened and read. Note that not all the file contents will be read if that
1445 would cause the buffers to overflow. So, for example::
1447 buffer_pattern='filename'
1451 buffer_pattern="abcd"
1459 buffer_pattern=0xdeadface
1461 Also you can combine everything together in any order::
1463 buffer_pattern=0xdeadface"abcd"-12'filename'
1465 .. option:: dedupe_percentage=int
1467 If set, fio will generate this percentage of identical buffers when
1468 writing. These buffers will be naturally dedupable. The contents of the
1469 buffers depend on what other buffer compression settings have been set. It's
1470 possible to have the individual buffers either fully compressible, or not at
1471 all. This option only controls the distribution of unique buffers.
1473 .. option:: invalidate=bool
1475 Invalidate the buffer/page cache parts of the files to be used prior to
1476 starting I/O if the platform and file type support it. Defaults to true.
1477 This will be ignored if :option:`pre_read` is also specified for the
1480 .. option:: sync=bool
1482 Use synchronous I/O for buffered writes. For the majority of I/O engines,
1483 this means using O_SYNC. Default: false.
1485 .. option:: iomem=str, mem=str
1487 Fio can use various types of memory as the I/O unit buffer. The allowed
1491 Use memory from :manpage:`malloc(3)` as the buffers. Default memory
1495 Use shared memory as the buffers. Allocated through
1496 :manpage:`shmget(2)`.
1499 Same as shm, but use huge pages as backing.
1502 Use :manpage:`mmap(2)` to allocate buffers. May either be anonymous memory, or can
1503 be file backed if a filename is given after the option. The format
1504 is `mem=mmap:/path/to/file`.
1507 Use a memory mapped huge file as the buffer backing. Append filename
1508 after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file`.
1511 Same as mmap, but use a MMAP_SHARED mapping.
1514 Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1516 The area allocated is a function of the maximum allowed bs size for the job,
1517 multiplied by the I/O depth given. Note that for **shmhuge** and
1518 **mmaphuge** to work, the system must have free huge pages allocated. This
1519 can normally be checked and set by reading/writing
1520 :file:`/proc/sys/vm/nr_hugepages` on a Linux system. Fio assumes a huge page
1521 is 4MiB in size. So to calculate the number of huge pages you need for a
1522 given job file, add up the I/O depth of all jobs (normally one unless
1523 :option:`iodepth` is used) and multiply by the maximum bs set. Then divide
1524 that number by the huge page size. You can see the size of the huge pages in
1525 :file:`/proc/meminfo`. If no huge pages are allocated by having a non-zero
1526 number in `nr_hugepages`, using **mmaphuge** or **shmhuge** will fail. Also
1527 see :option:`hugepage-size`.
1529 **mmaphuge** also needs to have hugetlbfs mounted and the file location
1530 should point there. So if it's mounted in :file:`/huge`, you would use
1531 `mem=mmaphuge:/huge/somefile`.
1533 .. option:: iomem_align=int
1535 This indicates the memory alignment of the I/O memory buffers. Note that
1536 the given alignment is applied to the first I/O unit buffer, if using
1537 :option:`iodepth` the alignment of the following buffers are given by the
1538 :option:`bs` used. In other words, if using a :option:`bs` that is a
1539 multiple of the page sized in the system, all buffers will be aligned to
1540 this value. If using a :option:`bs` that is not page aligned, the alignment
1541 of subsequent I/O memory buffers is the sum of the :option:`iomem_align` and
1544 .. option:: hugepage-size=int
1546 Defines the size of a huge page. Must at least be equal to the system
1547 setting, see :file:`/proc/meminfo`. Defaults to 4MiB. Should probably
1548 always be a multiple of megabytes, so using ``hugepage-size=Xm`` is the
1549 preferred way to set this to avoid setting a non-pow-2 bad value.
1551 .. option:: lockmem=int
1553 Pin the specified amount of memory with :manpage:`mlock(2)`. Can be used to
1554 simulate a smaller amount of memory. The amount specified is per worker.
1560 .. option:: size=int
1562 The total size of file I/O for each thread of this job. Fio will run until
1563 this many bytes has been transferred, unless runtime is limited by other options
1564 (such as :option:`runtime`, for instance, or increased/decreased by :option:`io_size`).
1565 Fio will divide this size between the available files determined by options
1566 such as :option:`nrfiles`, :option:`filename`, unless :option:`filesize` is
1567 specified by the job. If the result of division happens to be 0, the size is
1568 set to the physical size of the given files or devices if they exist.
1569 If this option is not specified, fio will use the full size of the given
1570 files or devices. If the files do not exist, size must be given. It is also
1571 possible to give size as a percentage between 1 and 100. If ``size=20%`` is
1572 given, fio will use 20% of the full size of the given files or devices.
1573 Can be combined with :option:`offset` to constrain the start and end range
1574 that I/O will be done within.
1576 .. option:: io_size=int, io_limit=int
1578 Normally fio operates within the region set by :option:`size`, which means
1579 that the :option:`size` option sets both the region and size of I/O to be
1580 performed. Sometimes that is not what you want. With this option, it is
1581 possible to define just the amount of I/O that fio should do. For instance,
1582 if :option:`size` is set to 20GiB and :option:`io_size` is set to 5GiB, fio
1583 will perform I/O within the first 20GiB but exit when 5GiB have been
1584 done. The opposite is also possible -- if :option:`size` is set to 20GiB,
1585 and :option:`io_size` is set to 40GiB, then fio will do 40GiB of I/O within
1586 the 0..20GiB region.
1588 .. option:: filesize=irange(int)
1590 Individual file sizes. May be a range, in which case fio will select sizes
1591 for files at random within the given range and limited to :option:`size` in
1592 total (if that is given). If not given, each created file is the same size.
1593 This option overrides :option:`size` in terms of file size, which means
1594 this value is used as a fixed size or possible range of each file.
1596 .. option:: file_append=bool
1598 Perform I/O after the end of the file. Normally fio will operate within the
1599 size of a file. If this option is set, then fio will append to the file
1600 instead. This has identical behavior to setting :option:`offset` to the size
1601 of a file. This option is ignored on non-regular files.
1603 .. option:: fill_device=bool, fill_fs=bool
1605 Sets size to something really large and waits for ENOSPC (no space left on
1606 device) as the terminating condition. Only makes sense with sequential
1607 write. For a read workload, the mount point will be filled first then I/O
1608 started on the result. This option doesn't make sense if operating on a raw
1609 device node, since the size of that is already known by the file system.
1610 Additionally, writing beyond end-of-device will not return ENOSPC there.
1616 .. option:: ioengine=str
1618 Defines how the job issues I/O to the file. The following types are defined:
1621 Basic :manpage:`read(2)` or :manpage:`write(2)`
1622 I/O. :manpage:`lseek(2)` is used to position the I/O location.
1623 See :option:`fsync` and :option:`fdatasync` for syncing write I/Os.
1626 Basic :manpage:`pread(2)` or :manpage:`pwrite(2)` I/O. Default on
1627 all supported operating systems except for Windows.
1630 Basic :manpage:`readv(2)` or :manpage:`writev(2)` I/O. Will emulate
1631 queuing by coalescing adjacent I/Os into a single submission.
1634 Basic :manpage:`preadv(2)` or :manpage:`pwritev(2)` I/O.
1637 Basic :manpage:`preadv2(2)` or :manpage:`pwritev2(2)` I/O.
1640 Linux native asynchronous I/O. Note that Linux may only support
1641 queued behavior with non-buffered I/O (set ``direct=1`` or
1643 This engine defines engine specific options.
1646 POSIX asynchronous I/O using :manpage:`aio_read(3)` and
1647 :manpage:`aio_write(3)`.
1650 Solaris native asynchronous I/O.
1653 Windows native asynchronous I/O. Default on Windows.
1656 File is memory mapped with :manpage:`mmap(2)` and data copied
1657 to/from using :manpage:`memcpy(3)`.
1660 :manpage:`splice(2)` is used to transfer the data and
1661 :manpage:`vmsplice(2)` to transfer data from user space to the
1665 SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1666 ioctl, or if the target is an sg character device we use
1667 :manpage:`read(2)` and :manpage:`write(2)` for asynchronous
1668 I/O. Requires filename option to specify either block or character
1672 Doesn't transfer any data, just pretends to. This is mainly used to
1673 exercise fio itself and for debugging/testing purposes.
1676 Transfer over the network to given ``host:port``. Depending on the
1677 :option:`protocol` used, the :option:`hostname`, :option:`port`,
1678 :option:`listen` and :option:`filename` options are used to specify
1679 what sort of connection to make, while the :option:`protocol` option
1680 determines which protocol will be used. This engine defines engine
1684 Like **net**, but uses :manpage:`splice(2)` and
1685 :manpage:`vmsplice(2)` to map data and send/receive.
1686 This engine defines engine specific options.
1689 Doesn't transfer any data, but burns CPU cycles according to the
1690 :option:`cpuload` and :option:`cpuchunks` options. Setting
1691 :option:`cpuload`\=85 will cause that job to do nothing but burn 85%
1692 of the CPU. In case of SMP machines, use :option:`numjobs`
1693 =<no_of_cpu> to get desired CPU usage, as the cpuload only loads a
1694 single CPU at the desired rate. A job never finishes unless there is
1695 at least one non-cpuio job.
1698 The GUASI I/O engine is the Generic Userspace Asyncronous Syscall
1699 Interface approach to async I/O. See
1701 http://www.xmailserver.org/guasi-lib.html
1703 for more info on GUASI.
1706 The RDMA I/O engine supports both RDMA memory semantics
1707 (RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
1708 InfiniBand, RoCE and iWARP protocols.
1711 I/O engine that does regular fallocate to simulate data transfer as
1715 does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1718 does fallocate(,mode = 0).
1721 does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1724 I/O engine that sends :manpage:`ftruncate(2)` operations in response
1725 to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
1726 size to the current block offset. Block size is ignored.
1729 I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1730 defragment activity in request to DDIR_WRITE event.
1733 I/O engine supporting direct access to Ceph Rados Block Devices
1734 (RBD) via librbd without the need to use the kernel rbd driver. This
1735 ioengine defines engine specific options.
1738 Using GlusterFS libgfapi sync interface to direct access to
1739 GlusterFS volumes without having to go through FUSE. This ioengine
1740 defines engine specific options.
1743 Using GlusterFS libgfapi async interface to direct access to
1744 GlusterFS volumes without having to go through FUSE. This ioengine
1745 defines engine specific options.
1748 Read and write through Hadoop (HDFS). The :file:`filename` option
1749 is used to specify host,port of the hdfs name-node to connect. This
1750 engine interprets offsets a little differently. In HDFS, files once
1751 created cannot be modified so random writes are not possible. To
1752 imitate this the libhdfs engine expects a bunch of small files to be
1753 created over HDFS and will randomly pick a file from them
1754 based on the offset generated by fio backend (see the example
1755 job file to create such files, use ``rw=write`` option). Please
1756 note, it may be necessary to set environment variables to work
1757 with HDFS/libhdfs properly. Each job uses its own connection to
1761 Read, write and erase an MTD character device (e.g.,
1762 :file:`/dev/mtd0`). Discards are treated as erases. Depending on the
1763 underlying device type, the I/O may have to go in a certain pattern,
1764 e.g., on NAND, writing sequentially to erase blocks and discarding
1765 before overwriting. The `trimwrite` mode works well for this
1769 Read and write using filesystem DAX to a file on a filesystem
1770 mounted with DAX on a persistent memory device through the NVML
1774 Read and write using device DAX to a persistent memory device (e.g.,
1775 /dev/dax0.0) through the NVML libpmem library.
1778 Prefix to specify loading an external I/O engine object file. Append
1779 the engine filename, e.g. ``ioengine=external:/tmp/foo.o`` to load
1780 ioengine :file:`foo.o` in :file:`/tmp`.
1783 I/O engine specific parameters
1784 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1786 In addition, there are some parameters which are only valid when a specific
1787 ioengine is in use. These are used identically to normal parameters, with the
1788 caveat that when used on the command line, they must come after the
1789 :option:`ioengine` that defines them is selected.
1791 .. option:: userspace_reap : [libaio]
1793 Normally, with the libaio engine in use, fio will use the
1794 :manpage:`io_getevents(2)` system call to reap newly returned events. With
1795 this flag turned on, the AIO ring will be read directly from user-space to
1796 reap events. The reaping mode is only enabled when polling for a minimum of
1797 0 events (e.g. when :option:`iodepth_batch_complete` `=0`).
1799 .. option:: hipri : [pvsync2]
1801 Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
1804 .. option:: cpuload=int : [cpuio]
1806 Attempt to use the specified percentage of CPU cycles. This is a mandatory
1807 option when using cpuio I/O engine.
1809 .. option:: cpuchunks=int : [cpuio]
1811 Split the load into cycles of the given time. In microseconds.
1813 .. option:: exit_on_io_done=bool : [cpuio]
1815 Detect when I/O threads are done, then exit.
1817 .. option:: hostname=str : [netsplice] [net]
1819 The hostname or IP address to use for TCP or UDP based I/O. If the job is
1820 a TCP listener or UDP reader, the hostname is not used and must be omitted
1821 unless it is a valid UDP multicast address.
1823 .. option:: namenode=str : [libhdfs]
1825 The hostname or IP address of a HDFS cluster namenode to contact.
1827 .. option:: port=int
1831 The TCP or UDP port to bind to or connect to. If this is used with
1832 :option:`numjobs` to spawn multiple instances of the same job type, then
1833 this will be the starting port number since fio will use a range of
1838 the listening port of the HFDS cluster namenode.
1840 .. option:: interface=str : [netsplice] [net]
1842 The IP address of the network interface used to send or receive UDP
1845 .. option:: ttl=int : [netsplice] [net]
1847 Time-to-live value for outgoing UDP multicast packets. Default: 1.
1849 .. option:: nodelay=bool : [netsplice] [net]
1851 Set TCP_NODELAY on TCP connections.
1853 .. option:: protocol=str : [netsplice] [net]
1855 .. option:: proto=str : [netsplice] [net]
1857 The network protocol to use. Accepted values are:
1860 Transmission control protocol.
1862 Transmission control protocol V6.
1864 User datagram protocol.
1866 User datagram protocol V6.
1870 When the protocol is TCP or UDP, the port must also be given, as well as the
1871 hostname if the job is a TCP listener or UDP reader. For unix sockets, the
1872 normal filename option should be used and the port is invalid.
1874 .. option:: listen : [net]
1876 For TCP network connections, tell fio to listen for incoming connections
1877 rather than initiating an outgoing connection. The :option:`hostname` must
1878 be omitted if this option is used.
1880 .. option:: pingpong : [net]
1882 Normally a network writer will just continue writing data, and a network
1883 reader will just consume packages. If ``pingpong=1`` is set, a writer will
1884 send its normal payload to the reader, then wait for the reader to send the
1885 same payload back. This allows fio to measure network latencies. The
1886 submission and completion latencies then measure local time spent sending or
1887 receiving, and the completion latency measures how long it took for the
1888 other end to receive and send back. For UDP multicast traffic
1889 ``pingpong=1`` should only be set for a single reader when multiple readers
1890 are listening to the same address.
1892 .. option:: window_size : [net]
1894 Set the desired socket buffer size for the connection.
1896 .. option:: mss : [net]
1898 Set the TCP maximum segment size (TCP_MAXSEG).
1900 .. option:: donorname=str : [e4defrag]
1902 File will be used as a block donor (swap extents between files).
1904 .. option:: inplace=int : [e4defrag]
1906 Configure donor file blocks allocation strategy:
1909 Default. Preallocate donor's file on init.
1911 Allocate space immediately inside defragment event, and free right
1914 .. option:: clustername=str : [rbd]
1916 Specifies the name of the Ceph cluster.
1918 .. option:: rbdname=str : [rbd]
1920 Specifies the name of the RBD.
1922 .. option:: pool=str : [rbd]
1924 Specifies the name of the Ceph pool containing RBD.
1926 .. option:: clientname=str : [rbd]
1928 Specifies the username (without the 'client.' prefix) used to access the
1929 Ceph cluster. If the *clustername* is specified, the *clientname* shall be
1930 the full *type.id* string. If no type. prefix is given, fio will add
1931 'client.' by default.
1933 .. option:: skip_bad=bool : [mtd]
1935 Skip operations against known bad blocks.
1937 .. option:: hdfsdirectory : [libhdfs]
1939 libhdfs will create chunk in this HDFS directory.
1941 .. option:: chunk_size : [libhdfs]
1943 the size of the chunk to use for each file.
1949 .. option:: iodepth=int
1951 Number of I/O units to keep in flight against the file. Note that
1952 increasing *iodepth* beyond 1 will not affect synchronous ioengines (except
1953 for small degrees when :option:`verify_async` is in use). Even async
1954 engines may impose OS restrictions causing the desired depth not to be
1955 achieved. This may happen on Linux when using libaio and not setting
1956 :option:`direct`\=1, since buffered I/O is not async on that OS. Keep an
1957 eye on the I/O depth distribution in the fio output to verify that the
1958 achieved depth is as expected. Default: 1.
1960 .. option:: iodepth_batch_submit=int, iodepth_batch=int
1962 This defines how many pieces of I/O to submit at once. It defaults to 1
1963 which means that we submit each I/O as soon as it is available, but can be
1964 raised to submit bigger batches of I/O at the time. If it is set to 0 the
1965 :option:`iodepth` value will be used.
1967 .. option:: iodepth_batch_complete_min=int, iodepth_batch_complete=int
1969 This defines how many pieces of I/O to retrieve at once. It defaults to 1
1970 which means that we'll ask for a minimum of 1 I/O in the retrieval process
1971 from the kernel. The I/O retrieval will go on until we hit the limit set by
1972 :option:`iodepth_low`. If this variable is set to 0, then fio will always
1973 check for completed events before queuing more I/O. This helps reduce I/O
1974 latency, at the cost of more retrieval system calls.
1976 .. option:: iodepth_batch_complete_max=int
1978 This defines maximum pieces of I/O to retrieve at once. This variable should
1979 be used along with :option:`iodepth_batch_complete_min`\=int variable,
1980 specifying the range of min and max amount of I/O which should be
1981 retrieved. By default it is equal to the :option:`iodepth_batch_complete_min`
1986 iodepth_batch_complete_min=1
1987 iodepth_batch_complete_max=<iodepth>
1989 which means that we will retrieve at least 1 I/O and up to the whole
1990 submitted queue depth. If none of I/O has been completed yet, we will wait.
1994 iodepth_batch_complete_min=0
1995 iodepth_batch_complete_max=<iodepth>
1997 which means that we can retrieve up to the whole submitted queue depth, but
1998 if none of I/O has been completed yet, we will NOT wait and immediately exit
1999 the system call. In this example we simply do polling.
2001 .. option:: iodepth_low=int
2003 The low water mark indicating when to start filling the queue
2004 again. Defaults to the same as :option:`iodepth`, meaning that fio will
2005 attempt to keep the queue full at all times. If :option:`iodepth` is set to
2006 e.g. 16 and *iodepth_low* is set to 4, then after fio has filled the queue of
2007 16 requests, it will let the depth drain down to 4 before starting to fill
2010 .. option:: io_submit_mode=str
2012 This option controls how fio submits the I/O to the I/O engine. The default
2013 is `inline`, which means that the fio job threads submit and reap I/O
2014 directly. If set to `offload`, the job threads will offload I/O submission
2015 to a dedicated pool of I/O threads. This requires some coordination and thus
2016 has a bit of extra overhead, especially for lower queue depth I/O where it
2017 can increase latencies. The benefit is that fio can manage submission rates
2018 independently of the device completion rates. This avoids skewed latency
2019 reporting if I/O gets backed up on the device side (the coordinated omission
2026 .. option:: thinktime=time
2028 Stall the job for the specified period of time after an I/O has completed before issuing the
2029 next. May be used to simulate processing being done by an application.
2030 When the unit is omitted, the value is interpreted in microseconds. See
2031 :option:`thinktime_blocks` and :option:`thinktime_spin`.
2033 .. option:: thinktime_spin=time
2035 Only valid if :option:`thinktime` is set - pretend to spend CPU time doing
2036 something with the data received, before falling back to sleeping for the
2037 rest of the period specified by :option:`thinktime`. When the unit is
2038 omitted, the value is interpreted in microseconds.
2040 .. option:: thinktime_blocks=int
2042 Only valid if :option:`thinktime` is set - control how many blocks to issue,
2043 before waiting `thinktime` usecs. If not set, defaults to 1 which will make
2044 fio wait `thinktime` usecs after every block. This effectively makes any
2045 queue depth setting redundant, since no more than 1 I/O will be queued
2046 before we have to complete it and do our thinktime. In other words, this
2047 setting effectively caps the queue depth if the latter is larger.
2049 .. option:: rate=int[,int][,int]
2051 Cap the bandwidth used by this job. The number is in bytes/sec, the normal
2052 suffix rules apply. Comma-separated values may be specified for reads,
2053 writes, and trims as described in :option:`blocksize`.
2055 For example, using `rate=1m,500k` would limit reads to 1MiB/sec and writes to
2056 500KiB/sec. Capping only reads or writes can be done with `rate=,500k` or
2057 `rate=500k,` where the former will only limit writes (to 500KiB/sec) and the
2058 latter will only limit reads.
2060 .. option:: rate_min=int[,int][,int]
2062 Tell fio to do whatever it can to maintain at least this bandwidth. Failing
2063 to meet this requirement will cause the job to exit. Comma-separated values
2064 may be specified for reads, writes, and trims as described in
2065 :option:`blocksize`.
2067 .. option:: rate_iops=int[,int][,int]
2069 Cap the bandwidth to this number of IOPS. Basically the same as
2070 :option:`rate`, just specified independently of bandwidth. If the job is
2071 given a block size range instead of a fixed value, the smallest block size
2072 is used as the metric. Comma-separated values may be specified for reads,
2073 writes, and trims as described in :option:`blocksize`.
2075 .. option:: rate_iops_min=int[,int][,int]
2077 If fio doesn't meet this rate of I/O, it will cause the job to exit.
2078 Comma-separated values may be specified for reads, writes, and trims as
2079 described in :option:`blocksize`.
2081 .. option:: rate_process=str
2083 This option controls how fio manages rated I/O submissions. The default is
2084 `linear`, which submits I/O in a linear fashion with fixed delays between
2085 I/Os that gets adjusted based on I/O completion rates. If this is set to
2086 `poisson`, fio will submit I/O based on a more real world random request
2087 flow, known as the Poisson process
2088 (https://en.wikipedia.org/wiki/Poisson_point_process). The lambda will be
2089 10^6 / IOPS for the given workload.
2095 .. option:: latency_target=time
2097 If set, fio will attempt to find the max performance point that the given
2098 workload will run at while maintaining a latency below this target. When
2099 the unit is omitted, the value is interpreted in microseconds. See
2100 :option:`latency_window` and :option:`latency_percentile`.
2102 .. option:: latency_window=time
2104 Used with :option:`latency_target` to specify the sample window that the job
2105 is run at varying queue depths to test the performance. When the unit is
2106 omitted, the value is interpreted in microseconds.
2108 .. option:: latency_percentile=float
2110 The percentage of I/Os that must fall within the criteria specified by
2111 :option:`latency_target` and :option:`latency_window`. If not set, this
2112 defaults to 100.0, meaning that all I/Os must be equal or below to the value
2113 set by :option:`latency_target`.
2115 .. option:: max_latency=time
2117 If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
2118 maximum latency. When the unit is omitted, the value is interpreted in
2121 .. option:: rate_cycle=int
2123 Average bandwidth for :option:`rate` and :option:`rate_min` over this number
2124 of milliseconds. Defaults to 1000.
2130 .. option:: write_iolog=str
2132 Write the issued I/O patterns to the specified file. See
2133 :option:`read_iolog`. Specify a separate file for each job, otherwise the
2134 iologs will be interspersed and the file may be corrupt.
2136 .. option:: read_iolog=str
2138 Open an iolog with the specified filename and replay the I/O patterns it
2139 contains. This can be used to store a workload and replay it sometime
2140 later. The iolog given may also be a blktrace binary file, which allows fio
2141 to replay a workload captured by :command:`blktrace`. See
2142 :manpage:`blktrace(8)` for how to capture such logging data. For blktrace
2143 replay, the file needs to be turned into a blkparse binary data file first
2144 (``blkparse <device> -o /dev/null -d file_for_fio.bin``).
2146 .. option:: replay_no_stall=int
2148 When replaying I/O with :option:`read_iolog` the default behavior is to
2149 attempt to respect the timestamps within the log and replay them with the
2150 appropriate delay between IOPS. By setting this variable fio will not
2151 respect the timestamps and attempt to replay them as fast as possible while
2152 still respecting ordering. The result is the same I/O pattern to a given
2153 device, but different timings.
2155 .. option:: replay_redirect=str
2157 While replaying I/O patterns using :option:`read_iolog` the default behavior
2158 is to replay the IOPS onto the major/minor device that each IOP was recorded
2159 from. This is sometimes undesirable because on a different machine those
2160 major/minor numbers can map to a different device. Changing hardware on the
2161 same system can also result in a different major/minor mapping.
2162 ``replay_redirect`` causes all I/Os to be replayed onto the single specified
2163 device regardless of the device it was recorded
2164 from. i.e. :option:`replay_redirect`\= :file:`/dev/sdc` would cause all I/O
2165 in the blktrace or iolog to be replayed onto :file:`/dev/sdc`. This means
2166 multiple devices will be replayed onto a single device, if the trace
2167 contains multiple devices. If you want multiple devices to be replayed
2168 concurrently to multiple redirected devices you must blkparse your trace
2169 into separate traces and replay them with independent fio invocations.
2170 Unfortunately this also breaks the strict time ordering between multiple
2173 .. option:: replay_align=int
2175 Force alignment of I/O offsets and lengths in a trace to this power of 2
2178 .. option:: replay_scale=int
2180 Scale sector offsets down by this factor when replaying traces.
2183 Threads, processes and job synchronization
2184 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2188 Fio defaults to creating jobs by using fork, however if this option is
2189 given, fio will create jobs by using POSIX Threads' function
2190 :manpage:`pthread_create(3)` to create threads instead.
2192 .. option:: wait_for=str
2194 If set, the current job won't be started until all workers of the specified
2195 waitee job are done.
2197 ``wait_for`` operates on the job name basis, so there are a few
2198 limitations. First, the waitee must be defined prior to the waiter job
2199 (meaning no forward references). Second, if a job is being referenced as a
2200 waitee, it must have a unique name (no duplicate waitees).
2202 .. option:: nice=int
2204 Run the job with the given nice value. See man :manpage:`nice(2)`.
2206 On Windows, values less than -15 set the process class to "High"; -1 through
2207 -15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2210 .. option:: prio=int
2212 Set the I/O priority value of this job. Linux limits us to a positive value
2213 between 0 and 7, with 0 being the highest. See man
2214 :manpage:`ionice(1)`. Refer to an appropriate manpage for other operating
2215 systems since meaning of priority may differ.
2217 .. option:: prioclass=int
2219 Set the I/O priority class. See man :manpage:`ionice(1)`.
2221 .. option:: cpumask=int
2223 Set the CPU affinity of this job. The parameter given is a bit mask of
2224 allowed CPUs the job may run on. So if you want the allowed CPUs to be 1
2225 and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2226 :manpage:`sched_setaffinity(2)`. This may not work on all supported
2227 operating systems or kernel versions. This option doesn't work well for a
2228 higher CPU count than what you can store in an integer mask, so it can only
2229 control cpus 1-32. For boxes with larger CPU counts, use
2230 :option:`cpus_allowed`.
2232 .. option:: cpus_allowed=str
2234 Controls the same options as :option:`cpumask`, but accepts a textual
2235 specification of the permitted CPUs instead. So to use CPUs 1 and 5 you
2236 would specify ``cpus_allowed=1,5``. This option also allows a range of CPUs
2237 to be specified -- say you wanted a binding to CPUs 1, 5, and 8 to 15, you
2238 would set ``cpus_allowed=1,5,8-15``.
2240 .. option:: cpus_allowed_policy=str
2242 Set the policy of how fio distributes the CPUs specified by
2243 :option:`cpus_allowed` or :option:`cpumask`. Two policies are supported:
2246 All jobs will share the CPU set specified.
2248 Each job will get a unique CPU from the CPU set.
2250 **shared** is the default behavior, if the option isn't specified. If
2251 **split** is specified, then fio will will assign one cpu per job. If not
2252 enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2255 .. option:: numa_cpu_nodes=str
2257 Set this job running on specified NUMA nodes' CPUs. The arguments allow
2258 comma delimited list of cpu numbers, A-B ranges, or `all`. Note, to enable
2259 NUMA options support, fio must be built on a system with libnuma-dev(el)
2262 .. option:: numa_mem_policy=str
2264 Set this job's memory policy and corresponding NUMA nodes. Format of the
2269 ``mode`` is one of the following memory poicies: ``default``, ``prefer``,
2270 ``bind``, ``interleave`` or ``local``. For ``default`` and ``local`` memory
2271 policies, no node needs to be specified. For ``prefer``, only one node is
2272 allowed. For ``bind`` and ``interleave`` the ``nodelist`` may be as
2273 follows: a comma delimited list of numbers, A-B ranges, or `all`.
2275 .. option:: cgroup=str
2277 Add job to this control group. If it doesn't exist, it will be created. The
2278 system must have a mounted cgroup blkio mount point for this to work. If
2279 your system doesn't have it mounted, you can do so with::
2281 # mount -t cgroup -o blkio none /cgroup
2283 .. option:: cgroup_weight=int
2285 Set the weight of the cgroup to this value. See the documentation that comes
2286 with the kernel, allowed values are in the range of 100..1000.
2288 .. option:: cgroup_nodelete=bool
2290 Normally fio will delete the cgroups it has created after the job
2291 completion. To override this behavior and to leave cgroups around after the
2292 job completion, set ``cgroup_nodelete=1``. This can be useful if one wants
2293 to inspect various cgroup files after job completion. Default: false.
2295 .. option:: flow_id=int
2297 The ID of the flow. If not specified, it defaults to being a global
2298 flow. See :option:`flow`.
2300 .. option:: flow=int
2302 Weight in token-based flow control. If this value is used, then there is a
2303 'flow counter' which is used to regulate the proportion of activity between
2304 two or more jobs. Fio attempts to keep this flow counter near zero. The
2305 ``flow`` parameter stands for how much should be added or subtracted to the
2306 flow counter on each iteration of the main I/O loop. That is, if one job has
2307 ``flow=8`` and another job has ``flow=-1``, then there will be a roughly 1:8
2308 ratio in how much one runs vs the other.
2310 .. option:: flow_watermark=int
2312 The maximum value that the absolute value of the flow counter is allowed to
2313 reach before the job must wait for a lower value of the counter.
2315 .. option:: flow_sleep=int
2317 The period of time, in microseconds, to wait after the flow watermark has
2318 been exceeded before retrying operations.
2320 .. option:: stonewall, wait_for_previous
2322 Wait for preceding jobs in the job file to exit, before starting this
2323 one. Can be used to insert serialization points in the job file. A stone
2324 wall also implies starting a new reporting group, see
2325 :option:`group_reporting`.
2329 By default, fio will continue running all other jobs when one job finishes
2330 but sometimes this is not the desired action. Setting ``exitall`` will
2331 instead make fio terminate all other jobs when one job finishes.
2333 .. option:: exec_prerun=str
2335 Before running this job, issue the command specified through
2336 :manpage:`system(3)`. Output is redirected in a file called
2337 :file:`jobname.prerun.txt`.
2339 .. option:: exec_postrun=str
2341 After the job completes, issue the command specified though
2342 :manpage:`system(3)`. Output is redirected in a file called
2343 :file:`jobname.postrun.txt`.
2347 Instead of running as the invoking user, set the user ID to this value
2348 before the thread/process does any work.
2352 Set group ID, see :option:`uid`.
2358 .. option:: verify_only
2360 Do not perform specified workload, only verify data still matches previous
2361 invocation of this workload. This option allows one to check data multiple
2362 times at a later date without overwriting it. This option makes sense only
2363 for workloads that write data, and does not support workloads with the
2364 :option:`time_based` option set.
2366 .. option:: do_verify=bool
2368 Run the verify phase after a write phase. Only valid if :option:`verify` is
2371 .. option:: verify=str
2373 If writing to a file, fio can verify the file contents after each iteration
2374 of the job. Each verification method also implies verification of special
2375 header, which is written to the beginning of each block. This header also
2376 includes meta information, like offset of the block, block number, timestamp
2377 when block was written, etc. :option:`verify` can be combined with
2378 :option:`verify_pattern` option. The allowed values are:
2381 Use an md5 sum of the data area and store it in the header of
2385 Use an experimental crc64 sum of the data area and store it in the
2386 header of each block.
2389 Use a crc32c sum of the data area and store it in the header of
2390 each block. This will automatically use hardware acceleration
2391 (e.g. SSE4.2 on an x86 or CRC crypto extensions on ARM64) but will
2392 fall back to software crc32c if none is found. Generally the
2393 fatest checksum fio supports when hardware accelerated.
2399 Use a crc32 sum of the data area and store it in the header of each
2403 Use a crc16 sum of the data area and store it in the header of each
2407 Use a crc7 sum of the data area and store it in the header of each
2411 Use xxhash as the checksum function. Generally the fastest software
2412 checksum that fio supports.
2415 Use sha512 as the checksum function.
2418 Use sha256 as the checksum function.
2421 Use optimized sha1 as the checksum function.
2424 Use optimized sha3-224 as the checksum function.
2427 Use optimized sha3-256 as the checksum function.
2430 Use optimized sha3-384 as the checksum function.
2433 Use optimized sha3-512 as the checksum function.
2436 This option is deprecated, since now meta information is included in
2437 generic verification header and meta verification happens by
2438 default. For detailed information see the description of the
2439 :option:`verify` setting. This option is kept because of
2440 compatibility's sake with old configurations. Do not use it.
2443 Verify a strict pattern. Normally fio includes a header with some
2444 basic information and checksumming, but if this option is set, only
2445 the specific pattern set with :option:`verify_pattern` is verified.
2448 Only pretend to verify. Useful for testing internals with
2449 :option:`ioengine`\=null, not for much else.
2451 This option can be used for repeated burn-in tests of a system to make sure
2452 that the written data is also correctly read back. If the data direction
2453 given is a read or random read, fio will assume that it should verify a
2454 previously written file. If the data direction includes any form of write,
2455 the verify will be of the newly written data.
2457 .. option:: verifysort=bool
2459 If true, fio will sort written verify blocks when it deems it faster to read
2460 them back in a sorted manner. This is often the case when overwriting an
2461 existing file, since the blocks are already laid out in the file system. You
2462 can ignore this option unless doing huge amounts of really fast I/O where
2463 the red-black tree sorting CPU time becomes significant. Default: true.
2465 .. option:: verifysort_nr=int
2467 Pre-load and sort verify blocks for a read workload.
2469 .. option:: verify_offset=int
2471 Swap the verification header with data somewhere else in the block before
2472 writing. It is swapped back before verifying.
2474 .. option:: verify_interval=int
2476 Write the verification header at a finer granularity than the
2477 :option:`blocksize`. It will be written for chunks the size of
2478 ``verify_interval``. :option:`blocksize` should divide this evenly.
2480 .. option:: verify_pattern=str
2482 If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2483 filling with totally random bytes, but sometimes it's interesting to fill
2484 with a known pattern for I/O verification purposes. Depending on the width
2485 of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time (it can
2486 be either a decimal or a hex number). The ``verify_pattern`` if larger than
2487 a 32-bit quantity has to be a hex number that starts with either "0x" or
2488 "0X". Use with :option:`verify`. Also, ``verify_pattern`` supports %o
2489 format, which means that for each block offset will be written and then
2490 verified back, e.g.::
2494 Or use combination of everything::
2496 verify_pattern=0xff%o"abcd"-12
2498 .. option:: verify_fatal=bool
2500 Normally fio will keep checking the entire contents before quitting on a
2501 block verification failure. If this option is set, fio will exit the job on
2502 the first observed failure. Default: false.
2504 .. option:: verify_dump=bool
2506 If set, dump the contents of both the original data block and the data block
2507 we read off disk to files. This allows later analysis to inspect just what
2508 kind of data corruption occurred. Off by default.
2510 .. option:: verify_async=int
2512 Fio will normally verify I/O inline from the submitting thread. This option
2513 takes an integer describing how many async offload threads to create for I/O
2514 verification instead, causing fio to offload the duty of verifying I/O
2515 contents to one or more separate threads. If using this offload option, even
2516 sync I/O engines can benefit from using an :option:`iodepth` setting higher
2517 than 1, as it allows them to have I/O in flight while verifies are running.
2518 Defaults to 0 async threads, i.e. verification is not asynchronous.
2520 .. option:: verify_async_cpus=str
2522 Tell fio to set the given CPU affinity on the async I/O verification
2523 threads. See :option:`cpus_allowed` for the format used.
2525 .. option:: verify_backlog=int
2527 Fio will normally verify the written contents of a job that utilizes verify
2528 once that job has completed. In other words, everything is written then
2529 everything is read back and verified. You may want to verify continually
2530 instead for a variety of reasons. Fio stores the meta data associated with
2531 an I/O block in memory, so for large verify workloads, quite a bit of memory
2532 would be used up holding this meta data. If this option is enabled, fio will
2533 write only N blocks before verifying these blocks.
2535 .. option:: verify_backlog_batch=int
2537 Control how many blocks fio will verify if :option:`verify_backlog` is
2538 set. If not set, will default to the value of :option:`verify_backlog`
2539 (meaning the entire queue is read back and verified). If
2540 ``verify_backlog_batch`` is less than :option:`verify_backlog` then not all
2541 blocks will be verified, if ``verify_backlog_batch`` is larger than
2542 :option:`verify_backlog`, some blocks will be verified more than once.
2544 .. option:: verify_state_save=bool
2546 When a job exits during the write phase of a verify workload, save its
2547 current state. This allows fio to replay up until that point, if the verify
2548 state is loaded for the verify read phase. The format of the filename is,
2551 <type>-<jobname>-<jobindex>-verify.state.
2553 <type> is "local" for a local run, "sock" for a client/server socket
2554 connection, and "ip" (192.168.0.1, for instance) for a networked
2555 client/server connection. Defaults to true.
2557 .. option:: verify_state_load=bool
2559 If a verify termination trigger was used, fio stores the current write state
2560 of each thread. This can be used at verification time so that fio knows how
2561 far it should verify. Without this information, fio will run a full
2562 verification pass, according to the settings in the job file used. Default
2565 .. option:: trim_percentage=int
2567 Number of verify blocks to discard/trim.
2569 .. option:: trim_verify_zero=bool
2571 Verify that trim/discarded blocks are returned as zeros.
2573 .. option:: trim_backlog=int
2575 Verify that trim/discarded blocks are returned as zeros.
2577 .. option:: trim_backlog_batch=int
2579 Trim this number of I/O blocks.
2581 .. option:: experimental_verify=bool
2583 Enable experimental verification.
2589 .. option:: steadystate=str:float, ss=str:float
2591 Define the criterion and limit for assessing steady state performance. The
2592 first parameter designates the criterion whereas the second parameter sets
2593 the threshold. When the criterion falls below the threshold for the
2594 specified duration, the job will stop. For example, `iops_slope:0.1%` will
2595 direct fio to terminate the job when the least squares regression slope
2596 falls below 0.1% of the mean IOPS. If :option:`group_reporting` is enabled
2597 this will apply to all jobs in the group. Below is the list of available
2598 steady state assessment criteria. All assessments are carried out using only
2599 data from the rolling collection window. Threshold limits can be expressed
2600 as a fixed value or as a percentage of the mean in the collection window.
2603 Collect IOPS data. Stop the job if all individual IOPS measurements
2604 are within the specified limit of the mean IOPS (e.g., ``iops:2``
2605 means that all individual IOPS values must be within 2 of the mean,
2606 whereas ``iops:0.2%`` means that all individual IOPS values must be
2607 within 0.2% of the mean IOPS to terminate the job).
2610 Collect IOPS data and calculate the least squares regression
2611 slope. Stop the job if the slope falls below the specified limit.
2614 Collect bandwidth data. Stop the job if all individual bandwidth
2615 measurements are within the specified limit of the mean bandwidth.
2618 Collect bandwidth data and calculate the least squares regression
2619 slope. Stop the job if the slope falls below the specified limit.
2621 .. option:: steadystate_duration=time, ss_dur=time
2623 A rolling window of this duration will be used to judge whether steady state
2624 has been reached. Data will be collected once per second. The default is 0
2625 which disables steady state detection. When the unit is omitted, the
2626 value is interpreted in seconds.
2628 .. option:: steadystate_ramp_time=time, ss_ramp=time
2630 Allow the job to run for the specified duration before beginning data
2631 collection for checking the steady state job termination criterion. The
2632 default is 0. When the unit is omitted, the value is interpreted in seconds.
2635 Measurements and reporting
2636 ~~~~~~~~~~~~~~~~~~~~~~~~~~
2638 .. option:: per_job_logs=bool
2640 If set, this generates bw/clat/iops log with per file private filenames. If
2641 not set, jobs with identical names will share the log filename. Default:
2644 .. option:: group_reporting
2646 It may sometimes be interesting to display statistics for groups of jobs as
2647 a whole instead of for each individual job. This is especially true if
2648 :option:`numjobs` is used; looking at individual thread/process output
2649 quickly becomes unwieldy. To see the final report per-group instead of
2650 per-job, use :option:`group_reporting`. Jobs in a file will be part of the
2651 same reporting group, unless if separated by a :option:`stonewall`, or by
2652 using :option:`new_group`.
2654 .. option:: new_group
2656 Start a new reporting group. See: :option:`group_reporting`. If not given,
2657 all jobs in a file will be part of the same reporting group, unless
2658 separated by a :option:`stonewall`.
2662 By default, fio collects and shows final output results for all jobs
2663 that run. If this option is set to 0, then fio will ignore it in
2664 the final stat output.
2666 .. option:: write_bw_log=str
2668 If given, write a bandwidth log for this job. Can be used to store data of
2669 the bandwidth of the jobs in their lifetime. The included
2670 :command:`fio_generate_plots` script uses :command:`gnuplot` to turn these
2671 text files into nice graphs. See :option:`write_lat_log` for behavior of
2672 given filename. For this option, the postfix is :file:`_bw.x.log`, where `x`
2673 is the index of the job (`1..N`, where `N` is the number of jobs). If
2674 :option:`per_job_logs` is false, then the filename will not include the job
2675 index. See `Log File Formats`_.
2677 .. option:: write_lat_log=str
2679 Same as :option:`write_bw_log`, except that this option stores I/O
2680 submission, completion, and total latencies instead. If no filename is given
2681 with this option, the default filename of :file:`jobname_type.log` is
2682 used. Even if the filename is given, fio will still append the type of
2683 log. So if one specifies::
2687 The actual log names will be :file:`foo_slat.x.log`, :file:`foo_clat.x.log`,
2688 and :file:`foo_lat.x.log`, where `x` is the index of the job (1..N, where N
2689 is the number of jobs). This helps :command:`fio_generate_plot` find the
2690 logs automatically. If :option:`per_job_logs` is false, then the filename
2691 will not include the job index. See `Log File Formats`_.
2693 .. option:: write_hist_log=str
2695 Same as :option:`write_lat_log`, but writes I/O completion latency
2696 histograms. If no filename is given with this option, the default filename
2697 of :file:`jobname_clat_hist.x.log` is used, where `x` is the index of the
2698 job (1..N, where `N` is the number of jobs). Even if the filename is given,
2699 fio will still append the type of log. If :option:`per_job_logs` is false,
2700 then the filename will not include the job index. See `Log File Formats`_.
2702 .. option:: write_iops_log=str
2704 Same as :option:`write_bw_log`, but writes IOPS. If no filename is given
2705 with this option, the default filename of :file:`jobname_type.x.log` is
2706 used,where `x` is the index of the job (1..N, where `N` is the number of
2707 jobs). Even if the filename is given, fio will still append the type of
2708 log. If :option:`per_job_logs` is false, then the filename will not include
2709 the job index. See `Log File Formats`_.
2711 .. option:: log_avg_msec=int
2713 By default, fio will log an entry in the iops, latency, or bw log for every
2714 I/O that completes. When writing to the disk log, that can quickly grow to a
2715 very large size. Setting this option makes fio average the each log entry
2716 over the specified period of time, reducing the resolution of the log. See
2717 :option:`log_max_value` as well. Defaults to 0, logging all entries.
2718 Also see `Log File Formats`_.
2720 .. option:: log_hist_msec=int
2722 Same as :option:`log_avg_msec`, but logs entries for completion latency
2723 histograms. Computing latency percentiles from averages of intervals using
2724 :option:`log_avg_msec` is inaccurate. Setting this option makes fio log
2725 histogram entries over the specified period of time, reducing log sizes for
2726 high IOPS devices while retaining percentile accuracy. See
2727 :option:`log_hist_coarseness` as well. Defaults to 0, meaning histogram
2728 logging is disabled.
2730 .. option:: log_hist_coarseness=int
2732 Integer ranging from 0 to 6, defining the coarseness of the resolution of
2733 the histogram logs enabled with :option:`log_hist_msec`. For each increment
2734 in coarseness, fio outputs half as many bins. Defaults to 0, for which
2735 histogram logs contain 1216 latency bins. See `Log File Formats`_.
2737 .. option:: log_max_value=bool
2739 If :option:`log_avg_msec` is set, fio logs the average over that window. If
2740 you instead want to log the maximum value, set this option to 1. Defaults to
2741 0, meaning that averaged values are logged.
2743 .. option:: log_offset=int
2745 If this is set, the iolog options will include the byte offset for the I/O
2746 entry as well as the other data values.
2748 .. option:: log_compression=int
2750 If this is set, fio will compress the I/O logs as it goes, to keep the
2751 memory footprint lower. When a log reaches the specified size, that chunk is
2752 removed and compressed in the background. Given that I/O logs are fairly
2753 highly compressible, this yields a nice memory savings for longer runs. The
2754 downside is that the compression will consume some background CPU cycles, so
2755 it may impact the run. This, however, is also true if the logging ends up
2756 consuming most of the system memory. So pick your poison. The I/O logs are
2757 saved normally at the end of a run, by decompressing the chunks and storing
2758 them in the specified log file. This feature depends on the availability of
2761 .. option:: log_compression_cpus=str
2763 Define the set of CPUs that are allowed to handle online log compression for
2764 the I/O jobs. This can provide better isolation between performance
2765 sensitive jobs, and background compression work.
2767 .. option:: log_store_compressed=bool
2769 If set, fio will store the log files in a compressed format. They can be
2770 decompressed with fio, using the :option:`--inflate-log` command line
2771 parameter. The files will be stored with a :file:`.fz` suffix.
2773 .. option:: log_unix_epoch=bool
2775 If set, fio will log Unix timestamps to the log files produced by enabling
2776 write_type_log for each log type, instead of the default zero-based
2779 .. option:: block_error_percentiles=bool
2781 If set, record errors in trim block-sized units from writes and trims and
2782 output a histogram of how many trims it took to get to errors, and what kind
2783 of error was encountered.
2785 .. option:: bwavgtime=int
2787 Average the calculated bandwidth over the given time. Value is specified in
2788 milliseconds. If the job also does bandwidth logging through
2789 :option:`write_bw_log`, then the minimum of this option and
2790 :option:`log_avg_msec` will be used. Default: 500ms.
2792 .. option:: iopsavgtime=int
2794 Average the calculated IOPS over the given time. Value is specified in
2795 milliseconds. If the job also does IOPS logging through
2796 :option:`write_iops_log`, then the minimum of this option and
2797 :option:`log_avg_msec` will be used. Default: 500ms.
2799 .. option:: disk_util=bool
2801 Generate disk utilization statistics, if the platform supports it.
2804 .. option:: disable_lat=bool
2806 Disable measurements of total latency numbers. Useful only for cutting back
2807 the number of calls to :manpage:`gettimeofday(2)`, as that does impact
2808 performance at really high IOPS rates. Note that to really get rid of a
2809 large amount of these calls, this option must be used with
2810 :option:`disable_slat` and :option:`disable_bw_measurement` as well.
2812 .. option:: disable_clat=bool
2814 Disable measurements of completion latency numbers. See
2815 :option:`disable_lat`.
2817 .. option:: disable_slat=bool
2819 Disable measurements of submission latency numbers. See
2820 :option:`disable_slat`.
2822 .. option:: disable_bw_measurement=bool, disable_bw=bool
2824 Disable measurements of throughput/bandwidth numbers. See
2825 :option:`disable_lat`.
2827 .. option:: clat_percentiles=bool
2829 Enable the reporting of percentiles of completion latencies.
2831 .. option:: percentile_list=float_list
2833 Overwrite the default list of percentiles for completion latencies and the
2834 block error histogram. Each number is a floating number in the range
2835 (0,100], and the maximum length of the list is 20. Use ``:`` to separate the
2836 numbers, and list the numbers in ascending order. For example,
2837 ``--percentile_list=99.5:99.9`` will cause fio to report the values of
2838 completion latency below which 99.5% and 99.9% of the observed latencies
2845 .. option:: exitall_on_error
2847 When one job finishes in error, terminate the rest. The default is to wait
2848 for each job to finish.
2850 .. option:: continue_on_error=str
2852 Normally fio will exit the job on the first observed failure. If this option
2853 is set, fio will continue the job when there is a 'non-fatal error' (EIO or
2854 EILSEQ) until the runtime is exceeded or the I/O size specified is
2855 completed. If this option is used, there are two more stats that are
2856 appended, the total error count and the first error. The error field given
2857 in the stats is the first error that was hit during the run.
2859 The allowed values are:
2862 Exit on any I/O or verify errors.
2865 Continue on read errors, exit on all others.
2868 Continue on write errors, exit on all others.
2871 Continue on any I/O error, exit on all others.
2874 Continue on verify errors, exit on all others.
2877 Continue on all errors.
2880 Backward-compatible alias for 'none'.
2883 Backward-compatible alias for 'all'.
2885 .. option:: ignore_error=str
2887 Sometimes you want to ignore some errors during test in that case you can
2888 specify error list for each error type, instead of only being able to
2889 ignore the default 'non-fatal error' using :option:`continue_on_error`.
2890 ``ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST`` errors for
2891 given error type is separated with ':'. Error may be symbol ('ENOSPC',
2892 'ENOMEM') or integer. Example::
2894 ignore_error=EAGAIN,ENOSPC:122
2896 This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
2897 WRITE. This option works by overriding :option:`continue_on_error` with
2898 the list of errors for each error type if any.
2900 .. option:: error_dump=bool
2902 If set dump every error even if it is non fatal, true by default. If
2903 disabled only fatal error will be dumped.
2905 Running predefined workloads
2906 ----------------------------
2908 Fio includes predefined profiles that mimic the I/O workloads generated by
2911 .. option:: profile=str
2913 The predefined workload to run. Current profiles are:
2916 Threaded I/O bench (tiotest/tiobench) like workload.
2919 Aerospike Certification Tool (ACT) like workload.
2921 To view a profile's additional options use :option:`--cmdhelp` after specifying
2922 the profile. For example::
2924 $ fio --profile=act --cmdhelp
2929 .. option:: device-names=str
2934 .. option:: load=int
2937 ACT load multiplier. Default: 1.
2939 .. option:: test-duration=time
2942 How long the entire test takes to run. When the unit is omitted, the value
2943 is given in seconds. Default: 24h.
2945 .. option:: threads-per-queue=int
2948 Number of read IO threads per device. Default: 8.
2950 .. option:: read-req-num-512-blocks=int
2953 Number of 512B blocks to read at the time. Default: 3.
2955 .. option:: large-block-op-kbytes=int
2958 Size of large block ops in KiB (writes). Default: 131072.
2963 Set to run ACT prep phase.
2965 Tiobench profile options
2966 ~~~~~~~~~~~~~~~~~~~~~~~~
2968 .. option:: size=str
2973 .. option:: block=int
2976 Block size in bytes. Default: 4096.
2978 .. option:: numruns=int
2988 .. option:: threads=int
2993 Interpreting the output
2994 -----------------------
2997 Example output was based on the following:
2998 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --time_based \
2999 --rate=1256k --bs=14K --name=quick --runtime=1s --name=mixed \
3000 --runtime=2m --rw=rw
3002 Fio spits out a lot of output. While running, fio will display the status of the
3003 jobs created. An example of that would be::
3005 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]
3007 The characters inside the first set of square brackets denote the current status of
3008 each thread. The first character is the first job defined in the job file, and so
3009 forth. The possible values (in typical life cycle order) are:
3011 +------+-----+-----------------------------------------------------------+
3013 +======+=====+===========================================================+
3014 | P | | Thread setup, but not started. |
3015 +------+-----+-----------------------------------------------------------+
3016 | C | | Thread created. |
3017 +------+-----+-----------------------------------------------------------+
3018 | I | | Thread initialized, waiting or generating necessary data. |
3019 +------+-----+-----------------------------------------------------------+
3020 | | p | Thread running pre-reading file(s). |
3021 +------+-----+-----------------------------------------------------------+
3022 | | / | Thread is in ramp period. |
3023 +------+-----+-----------------------------------------------------------+
3024 | | R | Running, doing sequential reads. |
3025 +------+-----+-----------------------------------------------------------+
3026 | | r | Running, doing random reads. |
3027 +------+-----+-----------------------------------------------------------+
3028 | | W | Running, doing sequential writes. |
3029 +------+-----+-----------------------------------------------------------+
3030 | | w | Running, doing random writes. |
3031 +------+-----+-----------------------------------------------------------+
3032 | | M | Running, doing mixed sequential reads/writes. |
3033 +------+-----+-----------------------------------------------------------+
3034 | | m | Running, doing mixed random reads/writes. |
3035 +------+-----+-----------------------------------------------------------+
3036 | | D | Running, doing sequential trims. |
3037 +------+-----+-----------------------------------------------------------+
3038 | | d | Running, doing random trims. |
3039 +------+-----+-----------------------------------------------------------+
3040 | | F | Running, currently waiting for :manpage:`fsync(2)`. |
3041 +------+-----+-----------------------------------------------------------+
3042 | | V | Running, doing verification of written data. |
3043 +------+-----+-----------------------------------------------------------+
3044 | f | | Thread finishing. |
3045 +------+-----+-----------------------------------------------------------+
3046 | E | | Thread exited, not reaped by main thread yet. |
3047 +------+-----+-----------------------------------------------------------+
3048 | _ | | Thread reaped. |
3049 +------+-----+-----------------------------------------------------------+
3050 | X | | Thread reaped, exited with an error. |
3051 +------+-----+-----------------------------------------------------------+
3052 | K | | Thread reaped, exited due to signal. |
3053 +------+-----+-----------------------------------------------------------+
3056 Example output was based on the following:
3057 TZ=UTC fio --iodepth=8 --ioengine=null --size=100M --runtime=58m \
3058 --time_based --rate=2512k --bs=256K --numjobs=10 \
3059 --name=readers --rw=read --name=writers --rw=write
3061 Fio will condense the thread string as not to take up more space on the command
3062 line than needed. For instance, if you have 10 readers and 10 writers running,
3063 the output would look like this::
3065 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]
3067 Note that the status string is displayed in order, so it's possible to tell which of
3068 the jobs are currently doing what. In the example above this means that jobs 1--10
3069 are readers and 11--20 are writers.
3071 The other values are fairly self explanatory -- number of threads currently
3072 running and doing I/O, the number of currently open files (f=), the estimated
3073 completion percentage, the rate of I/O since last check (read speed listed first,
3074 then write speed and optionally trim speed) in terms of bandwidth and IOPS, and time to completion for the current
3075 running group. It's impossible to estimate runtime of the following groups (if
3079 Example output was based on the following:
3080 TZ=UTC fio --iodepth=16 --ioengine=posixaio --filename=/tmp/fiofile \
3081 --direct=1 --size=100M --time_based --runtime=50s --rate_iops=89 \
3082 --bs=7K --name=Client1 --rw=write
3084 When fio is done (or interrupted by :kbd:`Ctrl-C`), it will show the data for
3085 each thread, group of threads, and disks in that order. For each overall thread (or
3086 group) the output looks like::
3088 Client1: (groupid=0, jobs=1): err= 0: pid=16109: Sat Jun 24 12:07:54 2017
3089 write: IOPS=88, BW=623KiB/s (638kB/s)(30.4MiB/50032msec)
3090 slat (nsec): min=500, max=145500, avg=8318.00, stdev=4781.50
3091 clat (usec): min=170, max=78367, avg=4019.02, stdev=8293.31
3092 lat (usec): min=174, max=78375, avg=4027.34, stdev=8291.79
3093 clat percentiles (usec):
3094 | 1.00th=[ 302], 5.00th=[ 326], 10.00th=[ 343], 20.00th=[ 363],
3095 | 30.00th=[ 392], 40.00th=[ 404], 50.00th=[ 416], 60.00th=[ 445],
3096 | 70.00th=[ 816], 80.00th=[ 6718], 90.00th=[12911], 95.00th=[21627],
3097 | 99.00th=[43779], 99.50th=[51643], 99.90th=[68682], 99.95th=[72877],
3099 bw ( KiB/s): min= 532, max= 686, per=0.10%, avg=622.87, stdev=24.82, samples= 100
3100 iops : min= 76, max= 98, avg=88.98, stdev= 3.54, samples= 100
3101 lat (usec) : 250=0.04%, 500=64.11%, 750=4.81%, 1000=2.79%
3102 lat (msec) : 2=4.16%, 4=1.84%, 10=4.90%, 20=11.33%, 50=5.37%
3103 lat (msec) : 100=0.65%
3104 cpu : usr=0.27%, sys=0.18%, ctx=12072, majf=0, minf=21
3105 IO depths : 1=85.0%, 2=13.1%, 4=1.8%, 8=0.1%, 16=0.0%, 32=0.0%, >=64=0.0%
3106 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3107 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3108 issued rwt: total=0,4450,0, short=0,0,0, dropped=0,0,0
3109 latency : target=0, window=0, percentile=100.00%, depth=8
3111 The job name (or first job's name when using :option:`group_reporting`) is printed,
3112 along with the group id, count of jobs being aggregated, last error id seen (which
3113 is 0 when there are no errors), pid/tid of that thread and the time the job/group
3114 completed. Below are the I/O statistics for each data direction performed (showing
3115 writes in the example above). In the order listed, they denote:
3118 The string before the colon shows the I/O direction the statistics
3119 are for. **IOPS** is the average I/Os performed per second. **BW**
3120 is the average bandwidth rate shown as: value in power of 2 format
3121 (value in power of 10 format). The last two values show: (**total
3122 I/O performed** in power of 2 format / **runtime** of that thread).
3125 Submission latency (**min** being the minimum, **max** being the
3126 maximum, **avg** being the average, **stdev** being the standard
3127 deviation). This is the time it took to submit the I/O. For
3128 sync I/O this row is not displayed as the slat is really the
3129 completion latency (since queue/complete is one operation there).
3130 This value can be in nanoseconds, microseconds or milliseconds ---
3131 fio will choose the most appropriate base and print that (in the
3132 example above nanoseconds was the best scale). Note: in :option:`--minimal` mode
3133 latencies are always expressed in microseconds.
3136 Completion latency. Same names as slat, this denotes the time from
3137 submission to completion of the I/O pieces. For sync I/O, clat will
3138 usually be equal (or very close) to 0, as the time from submit to
3139 complete is basically just CPU time (I/O has already been done, see slat
3143 Bandwidth statistics based on samples. Same names as the xlat stats,
3144 but also includes the number of samples taken (**samples**) and an
3145 approximate percentage of total aggregate bandwidth this thread
3146 received in its group (**per**). This last value is only really
3147 useful if the threads in this group are on the same disk, since they
3148 are then competing for disk access.
3151 IOPS statistics based on samples. Same names as bw.
3154 CPU usage. User and system time, along with the number of context
3155 switches this thread went through, usage of system and user time, and
3156 finally the number of major and minor page faults. The CPU utilization
3157 numbers are averages for the jobs in that reporting group, while the
3158 context and fault counters are summed.
3161 The distribution of I/O depths over the job lifetime. The numbers are
3162 divided into powers of 2 and each entry covers depths from that value
3163 up to those that are lower than the next entry -- e.g., 16= covers
3164 depths from 16 to 31. Note that the range covered by a depth
3165 distribution entry can be different to the range covered by the
3166 equivalent submit/complete distribution entry.
3169 How many pieces of I/O were submitting in a single submit call. Each
3170 entry denotes that amount and below, until the previous entry -- e.g.,
3171 16=100% means that we submitted anywhere between 9 to 16 I/Os per submit
3172 call. Note that the range covered by a submit distribution entry can
3173 be different to the range covered by the equivalent depth distribution
3177 Like the above submit number, but for completions instead.
3180 The number of read/write/trim requests issued, and how many of them were
3184 The distribution of I/O completion latencies. This is the time from when
3185 I/O leaves fio and when it gets completed. The numbers follow the same
3186 pattern as the I/O depths, meaning that 2=1.6% means that 1.6% of the
3187 I/O completed within 2 msecs, 20=12.8% means that 12.8% of the I/O took
3188 more than 10 msecs, but less than (or equal to) 20 msecs.
3191 Example output was based on the following:
3192 TZ=UTC fio --ioengine=null --iodepth=2 --size=100M --numjobs=2 \
3193 --rate_process=poisson --io_limit=32M --name=read --bs=128k \
3194 --rate=11M --name=write --rw=write --bs=2k --rate=700k
3196 After each client has been listed, the group statistics are printed. They
3197 will look like this::
3199 Run status group 0 (all jobs):
3200 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
3201 WRITE: bw=1231KiB/s (1261kB/s), 616KiB/s-621KiB/s (630kB/s-636kB/s), io=64.0MiB (67.1MB), run=52747-53223msec
3203 For each data direction it prints:
3206 Aggregate bandwidth of threads in this group followed by the
3207 minimum and maximum bandwidth of all the threads in this group.
3208 Values outside of brackets are power-of-2 format and those
3209 within are the equivalent value in a power-of-10 format.
3211 Aggregate I/O performed of all threads in this group. The
3212 format is the same as bw.
3214 The smallest and longest runtimes of the threads in this group.
3216 And finally, the disk statistics are printed. They will look like this::
3218 Disk stats (read/write):
3219 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3221 Each value is printed for both reads and writes, with reads first. The
3225 Number of I/Os performed by all groups.
3227 Number of merges I/O the I/O scheduler.
3229 Number of ticks we kept the disk busy.
3231 Total time spent in the disk queue.
3233 The disk utilization. A value of 100% means we kept the disk
3234 busy constantly, 50% would be a disk idling half of the time.
3236 It is also possible to get fio to dump the current output while it is running,
3237 without terminating the job. To do that, send fio the **USR1** signal. You can
3238 also get regularly timed dumps by using the :option:`--status-interval`
3239 parameter, or by creating a file in :file:`/tmp` named
3240 :file:`fio-dump-status`. If fio sees this file, it will unlink it and dump the
3241 current output status.
3247 For scripted usage where you typically want to generate tables or graphs of the
3248 results, fio can output the results in a semicolon separated format. The format
3249 is one long line of values, such as::
3251 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%
3252 A description of this job goes here.
3254 The job description (if provided) follows on a second line.
3256 To enable terse output, use the :option:`--minimal` or
3257 :option:`--output-format`\=terse command line options. The
3258 first value is the version of the terse output format. If the output has to be
3259 changed for some reason, this number will be incremented by 1 to signify that
3262 Split up, the format is as follows (comments in brackets denote when a
3263 field was introduced or whether its specific to some terse version):
3267 terse version, fio version [v3], jobname, groupid, error
3271 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3272 Submission latency: min, max, mean, stdev (usec)
3273 Completion latency: min, max, mean, stdev (usec)
3274 Completion latency percentiles: 20 fields (see below)
3275 Total latency: min, max, mean, stdev (usec)
3276 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3277 IOPS [v5]: min, max, mean, stdev, number of samples
3283 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3284 Submission latency: min, max, mean, stdev (usec)
3285 Completion latency: min, max, mean, stdev (usec)
3286 Completion latency percentiles: 20 fields (see below)
3287 Total latency: min, max, mean, stdev (usec)
3288 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3289 IOPS [v5]: min, max, mean, stdev, number of samples
3291 TRIM status [all but version 3]:
3293 Fields are similar to READ/WRITE status.
3297 user, system, context switches, major faults, minor faults
3301 <=1, 2, 4, 8, 16, 32, >=64
3303 I/O latencies microseconds::
3305 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3307 I/O latencies milliseconds::
3309 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3311 Disk utilization [v3]::
3313 Disk name, Read ios, write ios,
3314 Read merges, write merges,
3315 Read ticks, write ticks,
3316 Time spent in queue, disk utilization percentage
3318 Additional Info (dependent on continue_on_error, default off)::
3320 total # errors, first error code
3322 Additional Info (dependent on description being set)::
3326 Completion latency percentiles can be a grouping of up to 20 sets, so for the
3327 terse output fio writes all of them. Each field will look like this::
3331 which is the Xth percentile, and the `usec` latency associated with it.
3333 For disk utilization, all disks used by fio are shown. So for each disk there
3334 will be a disk utilization section.
3336 Below is a single line containing short names for each of the fields in the
3337 minimal output v3, separated by semicolons::
3339 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
3345 There are two trace file format that you can encounter. The older (v1) format is
3346 unsupported since version 1.20-rc3 (March 2008). It will still be described
3347 below in case that you get an old trace and want to understand it.
3349 In any case the trace is a simple text file with a single action per line.
3352 Trace file format v1
3353 ~~~~~~~~~~~~~~~~~~~~
3355 Each line represents a single I/O action in the following format::
3359 where `rw=0/1` for read/write, and the offset and length entries being in bytes.
3361 This format is not supported in fio versions >= 1.20-rc3.
3364 Trace file format v2
3365 ~~~~~~~~~~~~~~~~~~~~
3367 The second version of the trace file format was added in fio version 1.17. It
3368 allows to access more then one file per trace and has a bigger set of possible
3371 The first line of the trace file has to be::
3375 Following this can be lines in two different formats, which are described below.
3377 The file management format::
3381 The filename is given as an absolute path. The action can be one of these:
3384 Add the given filename to the trace.
3386 Open the file with the given filename. The filename has to have
3387 been added with the **add** action before.
3389 Close the file with the given filename. The file has to have been
3393 The file I/O action format::
3395 filename action offset length
3397 The `filename` is given as an absolute path, and has to have been added and
3398 opened before it can be used with this format. The `offset` and `length` are
3399 given in bytes. The `action` can be one of these:
3402 Wait for `offset` microseconds. Everything below 100 is discarded.
3403 The time is relative to the previous `wait` statement.
3405 Read `length` bytes beginning from `offset`.
3407 Write `length` bytes beginning from `offset`.
3409 :manpage:`fsync(2)` the file.
3411 :manpage:`fdatasync(2)` the file.
3413 Trim the given file from the given `offset` for `length` bytes.
3415 CPU idleness profiling
3416 ----------------------
3418 In some cases, we want to understand CPU overhead in a test. For example, we
3419 test patches for the specific goodness of whether they reduce CPU usage.
3420 Fio implements a balloon approach to create a thread per CPU that runs at idle
3421 priority, meaning that it only runs when nobody else needs the cpu.
3422 By measuring the amount of work completed by the thread, idleness of each CPU
3423 can be derived accordingly.
3425 An unit work is defined as touching a full page of unsigned characters. Mean and
3426 standard deviation of time to complete an unit work is reported in "unit work"
3427 section. Options can be chosen to report detailed percpu idleness or overall
3428 system idleness by aggregating percpu stats.
3431 Verification and triggers
3432 -------------------------
3434 Fio is usually run in one of two ways, when data verification is done. The first
3435 is a normal write job of some sort with verify enabled. When the write phase has
3436 completed, fio switches to reads and verifies everything it wrote. The second
3437 model is running just the write phase, and then later on running the same job
3438 (but with reads instead of writes) to repeat the same I/O patterns and verify
3439 the contents. Both of these methods depend on the write phase being completed,
3440 as fio otherwise has no idea how much data was written.
3442 With verification triggers, fio supports dumping the current write state to
3443 local files. Then a subsequent read verify workload can load this state and know
3444 exactly where to stop. This is useful for testing cases where power is cut to a
3445 server in a managed fashion, for instance.
3447 A verification trigger consists of two things:
3449 1) Storing the write state of each job.
3450 2) Executing a trigger command.
3452 The write state is relatively small, on the order of hundreds of bytes to single
3453 kilobytes. It contains information on the number of completions done, the last X
3456 A trigger is invoked either through creation ('touch') of a specified file in
3457 the system, or through a timeout setting. If fio is run with
3458 :option:`--trigger-file`\= :file:`/tmp/trigger-file`, then it will continually
3459 check for the existence of :file:`/tmp/trigger-file`. When it sees this file, it
3460 will fire off the trigger (thus saving state, and executing the trigger
3463 For client/server runs, there's both a local and remote trigger. If fio is
3464 running as a server backend, it will send the job states back to the client for
3465 safe storage, then execute the remote trigger, if specified. If a local trigger
3466 is specified, the server will still send back the write state, but the client
3467 will then execute the trigger.
3469 Verification trigger example
3470 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3472 Let's say we want to run a powercut test on the remote machine 'server'. Our
3473 write workload is in :file:`write-test.fio`. We want to cut power to 'server' at
3474 some point during the run, and we'll run this test from the safety or our local
3475 machine, 'localbox'. On the server, we'll start the fio backend normally::
3477 server# fio --server
3479 and on the client, we'll fire off the workload::
3481 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
3483 We set :file:`/tmp/my-trigger` as the trigger file, and we tell fio to execute::
3485 echo b > /proc/sysrq-trigger
3487 on the server once it has received the trigger and sent us the write state. This
3488 will work, but it's not **really** cutting power to the server, it's merely
3489 abruptly rebooting it. If we have a remote way of cutting power to the server
3490 through IPMI or similar, we could do that through a local trigger command
3491 instead. Let's assume we have a script that does IPMI reboot of a given hostname,
3492 ipmi-reboot. On localbox, we could then have run fio with a local trigger
3495 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
3497 For this case, fio would wait for the server to send us the write state, then
3498 execute ``ipmi-reboot server`` when that happened.
3500 Loading verify state
3501 ~~~~~~~~~~~~~~~~~~~~
3503 To load stored write state, a read verification job file must contain the
3504 :option:`verify_state_load` option. If that is set, fio will load the previously
3505 stored state. For a local fio run this is done by loading the files directly,
3506 and on a client/server run, the server backend will ask the client to send the
3507 files over and load them from there.
3513 Fio supports a variety of log file formats, for logging latencies, bandwidth,
3514 and IOPS. The logs share a common format, which looks like this:
3516 *time* (`msec`), *value*, *data direction*, *offset*
3518 Time for the log entry is always in milliseconds. The *value* logged depends
3519 on the type of log, it will be one of the following:
3522 Value is latency in usecs
3528 *Data direction* is one of the following:
3537 The *offset* is the offset, in bytes, from the start of the file, for that
3538 particular I/O. The logging of the offset can be toggled with
3539 :option:`log_offset`.
3541 Fio defaults to logging every individual I/O. When IOPS are logged for individual
3542 I/Os the value entry will always be 1. If windowed logging is enabled through
3543 :option:`log_avg_msec`, fio logs the average values over the specified period of time.
3544 If windowed logging is enabled and :option:`log_max_value` is set, then fio logs
3545 maximum values in that window instead of averages. Since 'data direction' and
3546 'offset' are per-I/O values, they aren't applicable if windowed logging is enabled.
3551 Normally fio is invoked as a stand-alone application on the machine where the
3552 I/O workload should be generated. However, the backend and frontend of fio can
3553 be run separately i.e., the fio server can generate an I/O workload on the "Device
3554 Under Test" while being controlled by a client on another machine.
3556 Start the server on the machine which has access to the storage DUT::
3560 where `args` defines what fio listens to. The arguments are of the form
3561 ``type,hostname`` or ``IP,port``. *type* is either ``ip`` (or ip4) for TCP/IP
3562 v4, ``ip6`` for TCP/IP v6, or ``sock`` for a local unix domain socket.
3563 *hostname* is either a hostname or IP address, and *port* is the port to listen
3564 to (only valid for TCP/IP, not a local socket). Some examples:
3568 Start a fio server, listening on all interfaces on the default port (8765).
3570 2) ``fio --server=ip:hostname,4444``
3572 Start a fio server, listening on IP belonging to hostname and on port 4444.
3574 3) ``fio --server=ip6:::1,4444``
3576 Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
3578 4) ``fio --server=,4444``
3580 Start a fio server, listening on all interfaces on port 4444.
3582 5) ``fio --server=1.2.3.4``
3584 Start a fio server, listening on IP 1.2.3.4 on the default port.
3586 6) ``fio --server=sock:/tmp/fio.sock``
3588 Start a fio server, listening on the local socket :file:`/tmp/fio.sock`.
3590 Once a server is running, a "client" can connect to the fio server with::
3592 fio <local-args> --client=<server> <remote-args> <job file(s)>
3594 where `local-args` are arguments for the client where it is running, `server`
3595 is the connect string, and `remote-args` and `job file(s)` are sent to the
3596 server. The `server` string follows the same format as it does on the server
3597 side, to allow IP/hostname/socket and port strings.
3599 Fio can connect to multiple servers this way::
3601 fio --client=<server1> <job file(s)> --client=<server2> <job file(s)>
3603 If the job file is located on the fio server, then you can tell the server to
3604 load a local file as well. This is done by using :option:`--remote-config` ::
3606 fio --client=server --remote-config /path/to/file.fio
3608 Then fio will open this local (to the server) job file instead of being passed
3609 one from the client.
3611 If you have many servers (example: 100 VMs/containers), you can input a pathname
3612 of a file containing host IPs/names as the parameter value for the
3613 :option:`--client` option. For example, here is an example :file:`host.list`
3614 file containing 2 hostnames::
3616 host1.your.dns.domain
3617 host2.your.dns.domain
3619 The fio command would then be::
3621 fio --client=host.list <job file(s)>
3623 In this mode, you cannot input server-specific parameters or job files -- all
3624 servers receive the same job file.
3626 In order to let ``fio --client`` runs use a shared filesystem from multiple
3627 hosts, ``fio --client`` now prepends the IP address of the server to the
3628 filename. For example, if fio is using the directory :file:`/mnt/nfs/fio` and is
3629 writing filename :file:`fileio.tmp`, with a :option:`--client` `hostfile`
3630 containing two hostnames ``h1`` and ``h2`` with IP addresses 192.168.10.120 and
3631 192.168.10.121, then fio will create two files::
3633 /mnt/nfs/fio/192.168.10.120.fileio.tmp
3634 /mnt/nfs/fio/192.168.10.121.fileio.tmp