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:: --bandwidth-log
109 Generate aggregate bandwidth logs.
111 .. option:: --minimal
113 Print statistics in a terse, semicolon-delimited format.
115 .. option:: --append-terse
117 Print statistics in selected mode AND terse, semicolon-delimited format.
118 **deprecated**, use :option:`--output-format` instead to select multiple
121 .. option:: --output-format=type
123 Set the reporting format to `normal`, `terse`, `json`, or `json+`. Multiple
124 formats can be selected, separated by a comma. `terse` is a CSV based
125 format. `json+` is like `json`, except it adds a full dump of the latency
128 .. option:: --terse-version=type
130 Set terse version output format (default 3, or 2 or 4 or 5).
132 .. option:: --version
134 Print version info and exit.
140 .. option:: --cpuclock-test
142 Perform test and validation of internal CPU clock.
144 .. option:: --crctest=test
146 Test the speed of the builtin checksumming functions. If no argument is
147 given, all of them are tested. Or a comma separated list can be passed, in
148 which case the given ones are tested.
150 .. option:: --cmdhelp=command
152 Print help information for `command`. May be ``all`` for all commands.
154 .. option:: --enghelp=[ioengine[,command]]
156 List all commands defined by :option:`ioengine`, or print help for `command`
157 defined by :option:`ioengine`. If no :option:`ioengine` is given, list all
160 .. option:: --showcmd=jobfile
162 Turn a job file into command line options.
164 .. option:: --readonly
166 Turn on safety read-only checks, preventing writes. The ``--readonly``
167 option is an extra safety guard to prevent users from accidentally starting
168 a write workload when that is not desired. Fio will only write if
169 `rw=write/randwrite/rw/randrw` is given. This extra safety net can be used
170 as an extra precaution as ``--readonly`` will also enable a write check in
171 the I/O engine core to prevent writes due to unknown user space bug(s).
173 .. option:: --eta=when
175 When real-time ETA estimate should be printed. May be `always`, `never` or
178 .. option:: --eta-newline=time
180 Force a new line for every `time` period passed. When the unit is omitted,
181 the value is interpreted in seconds.
183 .. option:: --status-interval=time
185 Force full status dump every `time` period passed. When the unit is
186 omitted, the value is interpreted in seconds.
188 .. option:: --section=name
190 Only run specified section in job file. Multiple sections can be specified.
191 The ``--section`` option allows one to combine related jobs into one file.
192 E.g. one job file could define light, moderate, and heavy sections. Tell
193 fio to run only the "heavy" section by giving ``--section=heavy``
194 command line option. One can also specify the "write" operations in one
195 section and "verify" operation in another section. The ``--section`` option
196 only applies to job sections. The reserved *global* section is always
199 .. option:: --alloc-size=kb
201 Set the internal smalloc pool to this size in kb (def 1024). The
202 ``--alloc-size`` switch allows one to use a larger pool size for smalloc.
203 If running large jobs with randommap enabled, fio can run out of memory.
204 Smalloc is an internal allocator for shared structures from a fixed size
205 memory pool. The pool size defaults to 16M and can grow to 8 pools.
207 NOTE: While running :file:`.fio_smalloc.*` backing store files are visible
210 .. option:: --warnings-fatal
212 All fio parser warnings are fatal, causing fio to exit with an
215 .. option:: --max-jobs=nr
217 Maximum number of threads/processes to support.
219 .. option:: --server=args
221 Start a backend server, with `args` specifying what to listen to.
222 See `Client/Server`_ section.
224 .. option:: --daemonize=pidfile
226 Background a fio server, writing the pid to the given `pidfile` file.
228 .. option:: --client=hostname
230 Instead of running the jobs locally, send and run them on the given host or
231 set of hosts. See `Client/Server`_ section.
233 .. option:: --remote-config=file
235 Tell fio server to load this local file.
237 .. option:: --idle-prof=option
239 Report cpu idleness on a system or percpu basis
240 ``--idle-prof=system,percpu`` or
241 run unit work calibration only ``--idle-prof=calibrate``.
243 .. option:: --inflate-log=log
245 Inflate and output compressed log.
247 .. option:: --trigger-file=file
249 Execute trigger cmd when file exists.
251 .. option:: --trigger-timeout=t
253 Execute trigger at this time.
255 .. option:: --trigger=cmd
257 Set this command as local trigger.
259 .. option:: --trigger-remote=cmd
261 Set this command as remote trigger.
263 .. option:: --aux-path=path
265 Use this path for fio state generated files.
267 Any parameters following the options will be assumed to be job files, unless
268 they match a job file parameter. Multiple job files can be listed and each job
269 file will be regarded as a separate group. Fio will :option:`stonewall`
270 execution between each group.
276 As previously described, fio accepts one or more job files describing what it is
277 supposed to do. The job file format is the classic ini file, where the names
278 enclosed in [] brackets define the job name. You are free to use any ASCII name
279 you want, except *global* which has special meaning. Following the job name is
280 a sequence of zero or more parameters, one per line, that define the behavior of
281 the job. If the first character in a line is a ';' or a '#', the entire line is
282 discarded as a comment.
284 A *global* section sets defaults for the jobs described in that file. A job may
285 override a *global* section parameter, and a job file may even have several
286 *global* sections if so desired. A job is only affected by a *global* section
289 The :option:`--cmdhelp` option also lists all options. If used with an `option`
290 argument, :option:`--cmdhelp` will detail the given `option`.
292 See the `examples/` directory for inspiration on how to write job files. Note
293 the copyright and license requirements currently apply to `examples/` files.
295 So let's look at a really simple job file that defines two processes, each
296 randomly reading from a 128MiB file:
300 ; -- start job file --
311 As you can see, the job file sections themselves are empty as all the described
312 parameters are shared. As no :option:`filename` option is given, fio makes up a
313 `filename` for each of the jobs as it sees fit. On the command line, this job
314 would look as follows::
316 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
319 Let's look at an example that has a number of processes writing randomly to
324 ; -- start job file --
335 Here we have no *global* section, as we only have one job defined anyway. We
336 want to use async I/O here, with a depth of 4 for each file. We also increased
337 the buffer size used to 32KiB and define numjobs to 4 to fork 4 identical
338 jobs. The result is 4 processes each randomly writing to their own 64MiB
339 file. Instead of using the above job file, you could have given the parameters
340 on the command line. For this case, you would specify::
342 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
344 When fio is utilized as a basis of any reasonably large test suite, it might be
345 desirable to share a set of standardized settings across multiple job files.
346 Instead of copy/pasting such settings, any section may pull in an external
347 :file:`filename.fio` file with *include filename* directive, as in the following
350 ; -- start job file including.fio --
354 include glob-include.fio
361 include test-include.fio
362 ; -- end job file including.fio --
366 ; -- start job file glob-include.fio --
369 ; -- end job file glob-include.fio --
373 ; -- start job file test-include.fio --
376 ; -- end job file test-include.fio --
378 Settings pulled into a section apply to that section only (except *global*
379 section). Include directives may be nested in that any included file may contain
380 further include directive(s). Include files may not contain [] sections.
383 Environment variables
384 ~~~~~~~~~~~~~~~~~~~~~
386 Fio also supports environment variable expansion in job files. Any sub-string of
387 the form ``${VARNAME}`` as part of an option value (in other words, on the right
388 of the '='), will be expanded to the value of the environment variable called
389 `VARNAME`. If no such environment variable is defined, or `VARNAME` is the
390 empty string, the empty string will be substituted.
392 As an example, let's look at a sample fio invocation and job file::
394 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
398 ; -- start job file --
405 This will expand to the following equivalent job file at runtime:
409 ; -- start job file --
416 Fio ships with a few example job files, you can also look there for inspiration.
421 Additionally, fio has a set of reserved keywords that will be replaced
422 internally with the appropriate value. Those keywords are:
426 The architecture page size of the running system.
430 Megabytes of total memory in the system.
434 Number of online available CPUs.
436 These can be used on the command line or in the job file, and will be
437 automatically substituted with the current system values when the job is
438 run. Simple math is also supported on these keywords, so you can perform actions
443 and get that properly expanded to 8 times the size of memory in the machine.
449 This section describes in details each parameter associated with a job. Some
450 parameters take an option of a given type, such as an integer or a
451 string. Anywhere a numeric value is required, an arithmetic expression may be
452 used, provided it is surrounded by parentheses. Supported operators are:
461 For time values in expressions, units are microseconds by default. This is
462 different than for time values not in expressions (not enclosed in
463 parentheses). The following types are used:
470 String. This is a sequence of alpha characters.
473 Integer with possible time suffix. Without a unit value is interpreted as
474 seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
475 hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and
476 'us' (or 'usec') for microseconds. For example, use 10m for 10 minutes.
481 Integer. A whole number value, which may contain an integer prefix
482 and an integer suffix:
484 [*integer prefix*] **number** [*integer suffix*]
486 The optional *integer prefix* specifies the number's base. The default
487 is decimal. *0x* specifies hexadecimal.
489 The optional *integer suffix* specifies the number's units, and includes an
490 optional unit prefix and an optional unit. For quantities of data, the
491 default unit is bytes. For quantities of time, the default unit is seconds
492 unless otherwise specified.
494 With :option:`kb_base`\=1000, fio follows international standards for unit
495 prefixes. To specify power-of-10 decimal values defined in the
496 International System of Units (SI):
498 * *Ki* -- means kilo (K) or 1000
499 * *Mi* -- means mega (M) or 1000**2
500 * *Gi* -- means giga (G) or 1000**3
501 * *Ti* -- means tera (T) or 1000**4
502 * *Pi* -- means peta (P) or 1000**5
504 To specify power-of-2 binary values defined in IEC 80000-13:
506 * *k* -- means kibi (Ki) or 1024
507 * *M* -- means mebi (Mi) or 1024**2
508 * *G* -- means gibi (Gi) or 1024**3
509 * *T* -- means tebi (Ti) or 1024**4
510 * *P* -- means pebi (Pi) or 1024**5
512 With :option:`kb_base`\=1024 (the default), the unit prefixes are opposite
513 from those specified in the SI and IEC 80000-13 standards to provide
514 compatibility with old scripts. For example, 4k means 4096.
516 For quantities of data, an optional unit of 'B' may be included
517 (e.g., 'kB' is the same as 'k').
519 The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
520 not milli). 'b' and 'B' both mean byte, not bit.
522 Examples with :option:`kb_base`\=1000:
524 * *4 KiB*: 4096, 4096b, 4096B, 4ki, 4kib, 4kiB, 4Ki, 4KiB
525 * *1 MiB*: 1048576, 1mi, 1024ki
526 * *1 MB*: 1000000, 1m, 1000k
527 * *1 TiB*: 1099511627776, 1ti, 1024gi, 1048576mi
528 * *1 TB*: 1000000000, 1t, 1000m, 1000000k
530 Examples with :option:`kb_base`\=1024 (default):
532 * *4 KiB*: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
533 * *1 MiB*: 1048576, 1m, 1024k
534 * *1 MB*: 1000000, 1mi, 1000ki
535 * *1 TiB*: 1099511627776, 1t, 1024g, 1048576m
536 * *1 TB*: 1000000000, 1ti, 1000mi, 1000000ki
538 To specify times (units are not case sensitive):
542 * *M* -- means minutes
543 * *s* -- or sec means seconds (default)
544 * *ms* -- or *msec* means milliseconds
545 * *us* -- or *usec* means microseconds
547 If the option accepts an upper and lower range, use a colon ':' or
548 minus '-' to separate such values. See :ref:`irange <irange>`.
549 If the lower value specified happens to be larger than the upper value
550 the two values are swapped.
555 Boolean. Usually parsed as an integer, however only defined for
556 true and false (1 and 0).
561 Integer range with suffix. Allows value range to be given, such as
562 1024-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
563 option allows two sets of ranges, they can be specified with a ',' or '/'
564 delimiter: 1k-4k/8k-32k. Also see :ref:`int <int>`.
567 A list of floating point numbers, separated by a ':' character.
573 .. option:: kb_base=int
575 Select the interpretation of unit prefixes in input parameters.
578 Inputs comply with IEC 80000-13 and the International
579 System of Units (SI). Use:
581 - power-of-2 values with IEC prefixes (e.g., KiB)
582 - power-of-10 values with SI prefixes (e.g., kB)
585 Compatibility mode (default). To avoid breaking old scripts:
587 - power-of-2 values with SI prefixes
588 - power-of-10 values with IEC prefixes
590 See :option:`bs` for more details on input parameters.
592 Outputs always use correct prefixes. Most outputs include both
595 bw=2383.3kB/s (2327.4KiB/s)
597 If only one value is reported, then kb_base selects the one to use:
599 **1000** -- SI prefixes
601 **1024** -- IEC prefixes
603 .. option:: unit_base=int
605 Base unit for reporting. Allowed values are:
608 Use auto-detection (default).
615 With the above in mind, here follows the complete list of fio job parameters.
623 ASCII name of the job. This may be used to override the name printed by fio
624 for this job. Otherwise the job name is used. On the command line this
625 parameter has the special purpose of also signaling the start of a new job.
627 .. option:: description=str
629 Text description of the job. Doesn't do anything except dump this text
630 description when this job is run. It's not parsed.
632 .. option:: loops=int
634 Run the specified number of iterations of this job. Used to repeat the same
635 workload a given number of times. Defaults to 1.
637 .. option:: numjobs=int
639 Create the specified number of clones of this job. Each clone of job
640 is spawned as an independent thread or process. May be used to setup a
641 larger number of threads/processes doing the same thing. Each thread is
642 reported separately; to see statistics for all clones as a whole, use
643 :option:`group_reporting` in conjunction with :option:`new_group`.
644 See :option:`--max-jobs`. Default: 1.
647 Time related parameters
648 ~~~~~~~~~~~~~~~~~~~~~~~
650 .. option:: runtime=time
652 Tell fio to terminate processing after the specified period of time. It
653 can be quite hard to determine for how long a specified job will run, so
654 this parameter is handy to cap the total runtime to a given time. When
655 the unit is omitted, the value is intepreted in seconds.
657 .. option:: time_based
659 If set, fio will run for the duration of the :option:`runtime` specified
660 even if the file(s) are completely read or written. It will simply loop over
661 the same workload as many times as the :option:`runtime` allows.
663 .. option:: startdelay=irange(time)
665 Delay the start of job for the specified amount of time. Can be a single
666 value or a range. When given as a range, each thread will choose a value
667 randomly from within the range. Value is in seconds if a unit is omitted.
669 .. option:: ramp_time=time
671 If set, fio will run the specified workload for this amount of time before
672 logging any performance numbers. Useful for letting performance settle
673 before logging results, thus minimizing the runtime required for stable
674 results. Note that the ``ramp_time`` is considered lead in time for a job,
675 thus it will increase the total runtime if a special timeout or
676 :option:`runtime` is specified. When the unit is omitted, the value is
679 .. option:: clocksource=str
681 Use the given clocksource as the base of timing. The supported options are:
684 :manpage:`gettimeofday(2)`
687 :manpage:`clock_gettime(2)`
690 Internal CPU clock source
692 cpu is the preferred clocksource if it is reliable, as it is very fast (and
693 fio is heavy on time calls). Fio will automatically use this clocksource if
694 it's supported and considered reliable on the system it is running on,
695 unless another clocksource is specifically set. For x86/x86-64 CPUs, this
696 means supporting TSC Invariant.
698 .. option:: gtod_reduce=bool
700 Enable all of the :manpage:`gettimeofday(2)` reducing options
701 (:option:`disable_clat`, :option:`disable_slat`, :option:`disable_bw_measurement`) plus
702 reduce precision of the timeout somewhat to really shrink the
703 :manpage:`gettimeofday(2)` call count. With this option enabled, we only do
704 about 0.4% of the :manpage:`gettimeofday(2)` calls we would have done if all
705 time keeping was enabled.
707 .. option:: gtod_cpu=int
709 Sometimes it's cheaper to dedicate a single thread of execution to just
710 getting the current time. Fio (and databases, for instance) are very
711 intensive on :manpage:`gettimeofday(2)` calls. With this option, you can set
712 one CPU aside for doing nothing but logging current time to a shared memory
713 location. Then the other threads/processes that run I/O workloads need only
714 copy that segment, instead of entering the kernel with a
715 :manpage:`gettimeofday(2)` call. The CPU set aside for doing these time
716 calls will be excluded from other uses. Fio will manually clear it from the
717 CPU mask of other jobs.
723 .. option:: directory=str
725 Prefix filenames with this directory. Used to place files in a different
726 location than :file:`./`. You can specify a number of directories by
727 separating the names with a ':' character. These directories will be
728 assigned equally distributed to job clones creates with :option:`numjobs` as
729 long as they are using generated filenames. If specific `filename(s)` are
730 set fio will use the first listed directory, and thereby matching the
731 `filename` semantic which generates a file each clone if not specified, but
732 let all clones use the same if set.
734 See the :option:`filename` option for escaping certain characters.
736 .. option:: filename=str
738 Fio normally makes up a `filename` based on the job name, thread number, and
739 file number. If you want to share files between threads in a job or several
740 jobs with fixed file paths, specify a `filename` for each of them to override
741 the default. If the ioengine is file based, you can specify a number of files
742 by separating the names with a ':' colon. So if you wanted a job to open
743 :file:`/dev/sda` and :file:`/dev/sdb` as the two working files, you would use
744 ``filename=/dev/sda:/dev/sdb``. This also means that whenever this option is
745 specified, :option:`nrfiles` is ignored. The size of regular files specified
746 by this option will be :option:`size` divided by number of files unless
747 explicit size is specified by :option:`filesize`.
749 On Windows, disk devices are accessed as :file:`\\\\.\\PhysicalDrive0` for
750 the first device, :file:`\\\\.\\PhysicalDrive1` for the second etc.
751 Note: Windows and FreeBSD prevent write access to areas
752 of the disk containing in-use data (e.g. filesystems). If the wanted
753 `filename` does need to include a colon, then escape that with a ``\``
754 character. For instance, if the `filename` is :file:`/dev/dsk/foo@3,0:c`,
755 then you would use ``filename="/dev/dsk/foo@3,0\:c"``. The
756 :file:`-` is a reserved name, meaning stdin or stdout. Which of the two
757 depends on the read/write direction set.
759 .. option:: filename_format=str
761 If sharing multiple files between jobs, it is usually necessary to have fio
762 generate the exact names that you want. By default, fio will name a file
763 based on the default file format specification of
764 :file:`jobname.jobnumber.filenumber`. With this option, that can be
765 customized. Fio will recognize and replace the following keywords in this
769 The name of the worker thread or process.
771 The incremental number of the worker thread or process.
773 The incremental number of the file for that worker thread or
776 To have dependent jobs share a set of files, this option can be set to have
777 fio generate filenames that are shared between the two. For instance, if
778 :file:`testfiles.$filenum` is specified, file number 4 for any job will be
779 named :file:`testfiles.4`. The default of :file:`$jobname.$jobnum.$filenum`
780 will be used if no other format specifier is given.
782 .. option:: unique_filename=bool
784 To avoid collisions between networked clients, fio defaults to prefixing any
785 generated filenames (with a directory specified) with the source of the
786 client connecting. To disable this behavior, set this option to 0.
788 .. option:: opendir=str
790 Recursively open any files below directory `str`.
792 .. option:: lockfile=str
794 Fio defaults to not locking any files before it does I/O to them. If a file
795 or file descriptor is shared, fio can serialize I/O to that file to make the
796 end result consistent. This is usual for emulating real workloads that share
797 files. The lock modes are:
800 No locking. The default.
802 Only one thread or process may do I/O at a time, excluding all
805 Read-write locking on the file. Many readers may
806 access the file at the same time, but writes get exclusive access.
808 .. option:: nrfiles=int
810 Number of files to use for this job. Defaults to 1. The size of files
811 will be :option:`size` divided by this unless explicit size is specified by
812 :option:`filesize`. Files are created for each thread separately, and each
813 file will have a file number within its name by default, as explained in
814 :option:`filename` section.
817 .. option:: openfiles=int
819 Number of files to keep open at the same time. Defaults to the same as
820 :option:`nrfiles`, can be set smaller to limit the number simultaneous
823 .. option:: file_service_type=str
825 Defines how fio decides which file from a job to service next. The following
829 Choose a file at random.
832 Round robin over opened files. This is the default.
835 Finish one file before moving on to the next. Multiple files can
836 still be open depending on 'openfiles'.
839 Use a *Zipf* distribution to decide what file to access.
842 Use a *Pareto* distribution to decide what file to access.
845 Use a *Gaussian* (normal) distribution to decide what file to
848 For *random*, *roundrobin*, and *sequential*, a postfix can be appended to
849 tell fio how many I/Os to issue before switching to a new file. For example,
850 specifying ``file_service_type=random:8`` would cause fio to issue
851 8 I/Os before selecting a new file at random. For the non-uniform
852 distributions, a floating point postfix can be given to influence how the
853 distribution is skewed. See :option:`random_distribution` for a description
854 of how that would work.
856 .. option:: ioscheduler=str
858 Attempt to switch the device hosting the file to the specified I/O scheduler
861 .. option:: create_serialize=bool
863 If true, serialize the file creation for the jobs. This may be handy to
864 avoid interleaving of data files, which may greatly depend on the filesystem
865 used and even the number of processors in the system. Default: true.
867 .. option:: create_fsync=bool
869 fsync the data file after creation. This is the default.
871 .. option:: create_on_open=bool
873 Don't pre-setup the files for I/O, just create open() when it's time to do
874 I/O to that file. Default: false.
876 .. option:: create_only=bool
878 If true, fio will only run the setup phase of the job. If files need to be
879 laid out or updated on disk, only that will be done -- the actual job contents
880 are not executed. Default: false.
882 .. option:: allow_file_create=bool
884 If true, fio is permitted to create files as part of its workload. This is
885 the default behavior. If this option is false, then fio will error out if
886 the files it needs to use don't already exist. Default: true.
888 .. option:: allow_mounted_write=bool
890 If this isn't set, fio will abort jobs that are destructive (e.g. that write)
891 to what appears to be a mounted device or partition. This should help catch
892 creating inadvertently destructive tests, not realizing that the test will
893 destroy data on the mounted file system. Note that some platforms don't allow
894 writing against a mounted device regardless of this option. Default: false.
896 .. option:: pre_read=bool
898 If this is given, files will be pre-read into memory before starting the
899 given I/O operation. This will also clear the :option:`invalidate` flag,
900 since it is pointless to pre-read and then drop the cache. This will only
901 work for I/O engines that are seek-able, since they allow you to read the
902 same data multiple times. Thus it will not work on non-seekable I/O engines
903 (e.g. network, splice). Default: false.
905 .. option:: unlink=bool
907 Unlink the job files when done. Not the default, as repeated runs of that
908 job would then waste time recreating the file set again and again. Default:
911 .. option:: unlink_each_loop=bool
913 Unlink job files after each iteration or loop. Default: false.
915 .. option:: zonesize=int
917 Divide a file into zones of the specified size. See :option:`zoneskip`.
919 .. option:: zonerange=int
921 Give size of an I/O zone. See :option:`zoneskip`.
923 .. option:: zoneskip=int
925 Skip the specified number of bytes when :option:`zonesize` data has been
926 read. The two zone options can be used to only do I/O on zones of a file.
932 .. option:: direct=bool
934 If value is true, use non-buffered I/O. This is usually O_DIRECT. Note that
935 ZFS on Solaris doesn't support direct I/O. On Windows the synchronous
936 ioengines don't support direct I/O. Default: false.
938 .. option:: atomic=bool
940 If value is true, attempt to use atomic direct I/O. Atomic writes are
941 guaranteed to be stable once acknowledged by the operating system. Only
942 Linux supports O_ATOMIC right now.
944 .. option:: buffered=bool
946 If value is true, use buffered I/O. This is the opposite of the
947 :option:`direct` option. Defaults to true.
949 .. option:: readwrite=str, rw=str
951 Type of I/O pattern. Accepted values are:
958 Sequential trims (Linux block devices only).
964 Random trims (Linux block devices only).
966 Sequential mixed reads and writes.
968 Random mixed reads and writes.
970 Sequential trim+write sequences. Blocks will be trimmed first,
971 then the same blocks will be written to.
973 Fio defaults to read if the option is not specified. For the mixed I/O
974 types, the default is to split them 50/50. For certain types of I/O the
975 result may still be skewed a bit, since the speed may be different. It is
976 possible to specify a number of I/O's to do before getting a new offset,
977 this is done by appending a ``:<nr>`` to the end of the string given. For a
978 random read, it would look like ``rw=randread:8`` for passing in an offset
979 modifier with a value of 8. If the suffix is used with a sequential I/O
980 pattern, then the value specified will be added to the generated offset for
981 each I/O. For instance, using ``rw=write:4k`` will skip 4k for every
982 write. It turns sequential I/O into sequential I/O with holes. See the
983 :option:`rw_sequencer` option.
985 .. option:: rw_sequencer=str
987 If an offset modifier is given by appending a number to the ``rw=<str>``
988 line, then this option controls how that number modifies the I/O offset
989 being generated. Accepted values are:
992 Generate sequential offset.
994 Generate the same offset.
996 ``sequential`` is only useful for random I/O, where fio would normally
997 generate a new random offset for every I/O. If you append e.g. 8 to randread,
998 you would get a new random offset for every 8 I/O's. The result would be a
999 seek for only every 8 I/O's, instead of for every I/O. Use ``rw=randread:8``
1000 to specify that. As sequential I/O is already sequential, setting
1001 ``sequential`` for that would not result in any differences. ``identical``
1002 behaves in a similar fashion, except it sends the same offset 8 number of
1003 times before generating a new offset.
1005 .. option:: unified_rw_reporting=bool
1007 Fio normally reports statistics on a per data direction basis, meaning that
1008 reads, writes, and trims are accounted and reported separately. If this
1009 option is set fio sums the results and report them as "mixed" instead.
1011 .. option:: randrepeat=bool
1013 Seed the random number generator used for random I/O patterns in a
1014 predictable way so the pattern is repeatable across runs. Default: true.
1016 .. option:: allrandrepeat=bool
1018 Seed all random number generators in a predictable way so results are
1019 repeatable across runs. Default: false.
1021 .. option:: randseed=int
1023 Seed the random number generators based on this seed value, to be able to
1024 control what sequence of output is being generated. If not set, the random
1025 sequence depends on the :option:`randrepeat` setting.
1027 .. option:: fallocate=str
1029 Whether pre-allocation is performed when laying down files.
1030 Accepted values are:
1033 Do not pre-allocate space.
1036 Pre-allocate via :manpage:`posix_fallocate(3)`.
1039 Pre-allocate via :manpage:`fallocate(2)` with
1040 FALLOC_FL_KEEP_SIZE set.
1043 Backward-compatible alias for **none**.
1046 Backward-compatible alias for **posix**.
1048 May not be available on all supported platforms. **keep** is only available
1049 on Linux. If using ZFS on Solaris this must be set to **none** because ZFS
1050 doesn't support it. Default: **posix**.
1052 .. option:: fadvise_hint=str
1054 Use :manpage:`posix_fadvise(2)` to advise the kernel on what I/O patterns
1055 are likely to be issued. Accepted values are:
1058 Backwards-compatible hint for "no hint".
1061 Backwards compatible hint for "advise with fio workload type". This
1062 uses **FADV_RANDOM** for a random workload, and **FADV_SEQUENTIAL**
1063 for a sequential workload.
1066 Advise using **FADV_SEQUENTIAL**.
1069 Advise using **FADV_RANDOM**.
1071 .. option:: fadvise_stream=int
1073 Use :manpage:`posix_fadvise(2)` to advise the kernel what stream ID the
1074 writes issued belong to. Only supported on Linux. Note, this option may
1075 change going forward.
1077 .. option:: offset=int
1079 Start I/O at the provided offset in the file, given as either a fixed size or
1080 a percentage. If a percentage is given, the next ``blockalign``-ed offset
1081 will be used. Data before the given offset will not be touched. This
1082 effectively caps the file size at `real_size - offset`. Can be combined with
1083 :option:`size` to constrain the start and end range of the I/O workload.
1085 .. option:: offset_increment=int
1087 If this is provided, then the real offset becomes `offset + offset_increment
1088 * thread_number`, where the thread number is a counter that starts at 0 and
1089 is incremented for each sub-job (i.e. when :option:`numjobs` option is
1090 specified). This option is useful if there are several jobs which are
1091 intended to operate on a file in parallel disjoint segments, with even
1092 spacing between the starting points.
1094 .. option:: number_ios=int
1096 Fio will normally perform I/Os until it has exhausted the size of the region
1097 set by :option:`size`, or if it exhaust the allocated time (or hits an error
1098 condition). With this setting, the range/size can be set independently of
1099 the number of I/Os to perform. When fio reaches this number, it will exit
1100 normally and report status. Note that this does not extend the amount of I/O
1101 that will be done, it will only stop fio if this condition is met before
1102 other end-of-job criteria.
1104 .. option:: fsync=int
1106 If writing to a file, issue a sync of the dirty data for every number of
1107 blocks given. For example, if you give 32 as a parameter, fio will sync the
1108 file for every 32 writes issued. If fio is using non-buffered I/O, we may
1109 not sync the file. The exception is the sg I/O engine, which synchronizes
1110 the disk cache anyway. Defaults to 0, which means no sync every certain
1113 .. option:: fdatasync=int
1115 Like :option:`fsync` but uses :manpage:`fdatasync(2)` to only sync data and
1116 not metadata blocks. In Windows, FreeBSD, and DragonFlyBSD there is no
1117 :manpage:`fdatasync(2)`, this falls back to using :manpage:`fsync(2)`.
1118 Defaults to 0, which means no sync data every certain number of writes.
1120 .. option:: write_barrier=int
1122 Make every `N-th` write a barrier write.
1124 .. option:: sync_file_range=str:val
1126 Use :manpage:`sync_file_range(2)` for every `val` number of write
1127 operations. Fio will track range of writes that have happened since the last
1128 :manpage:`sync_file_range(2)` call. `str` can currently be one or more of:
1131 SYNC_FILE_RANGE_WAIT_BEFORE
1133 SYNC_FILE_RANGE_WRITE
1135 SYNC_FILE_RANGE_WAIT_AFTER
1137 So if you do ``sync_file_range=wait_before,write:8``, fio would use
1138 ``SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE`` for every 8
1139 writes. Also see the :manpage:`sync_file_range(2)` man page. This option is
1142 .. option:: overwrite=bool
1144 If true, writes to a file will always overwrite existing data. If the file
1145 doesn't already exist, it will be created before the write phase begins. If
1146 the file exists and is large enough for the specified write phase, nothing
1147 will be done. Default: false.
1149 .. option:: end_fsync=bool
1151 If true, :manpage:`fsync(2)` file contents when a write stage has completed.
1154 .. option:: fsync_on_close=bool
1156 If true, fio will :manpage:`fsync(2)` a dirty file on close. This differs
1157 from :option:`end_fsync` in that it will happen on every file close, not
1158 just at the end of the job. Default: false.
1160 .. option:: rwmixread=int
1162 Percentage of a mixed workload that should be reads. Default: 50.
1164 .. option:: rwmixwrite=int
1166 Percentage of a mixed workload that should be writes. If both
1167 :option:`rwmixread` and :option:`rwmixwrite` is given and the values do not
1168 add up to 100%, the latter of the two will be used to override the
1169 first. This may interfere with a given rate setting, if fio is asked to
1170 limit reads or writes to a certain rate. If that is the case, then the
1171 distribution may be skewed. Default: 50.
1173 .. option:: random_distribution=str:float[,str:float][,str:float]
1175 By default, fio will use a completely uniform random distribution when asked
1176 to perform random I/O. Sometimes it is useful to skew the distribution in
1177 specific ways, ensuring that some parts of the data is more hot than others.
1178 fio includes the following distribution models:
1181 Uniform random distribution
1190 Normal (Gaussian) distribution
1193 Zoned random distribution
1195 When using a **zipf** or **pareto** distribution, an input value is also
1196 needed to define the access pattern. For **zipf**, this is the `zipf
1197 theta`. For **pareto**, it's the `Pareto power`. Fio includes a test
1198 program, :command:`genzipf`, that can be used visualize what the given input
1199 values will yield in terms of hit rates. If you wanted to use **zipf** with
1200 a `theta` of 1.2, you would use ``random_distribution=zipf:1.2`` as the
1201 option. If a non-uniform model is used, fio will disable use of the random
1202 map. For the **gauss** distribution, a normal deviation is supplied as a
1203 value between 0 and 100.
1205 For a **zoned** distribution, fio supports specifying percentages of I/O
1206 access that should fall within what range of the file or device. For
1207 example, given a criteria of:
1209 * 60% of accesses should be to the first 10%
1210 * 30% of accesses should be to the next 20%
1211 * 8% of accesses should be to to the next 30%
1212 * 2% of accesses should be to the next 40%
1214 we can define that through zoning of the random accesses. For the above
1215 example, the user would do::
1217 random_distribution=zoned:60/10:30/20:8/30:2/40
1219 similarly to how :option:`bssplit` works for setting ranges and percentages
1220 of block sizes. Like :option:`bssplit`, it's possible to specify separate
1221 zones for reads, writes, and trims. If just one set is given, it'll apply to
1224 .. option:: percentage_random=int[,int][,int]
1226 For a random workload, set how big a percentage should be random. This
1227 defaults to 100%, in which case the workload is fully random. It can be set
1228 from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1229 sequential. Any setting in between will result in a random mix of sequential
1230 and random I/O, at the given percentages. Comma-separated values may be
1231 specified for reads, writes, and trims as described in :option:`blocksize`.
1233 .. option:: norandommap
1235 Normally fio will cover every block of the file when doing random I/O. If
1236 this option is given, fio will just get a new random offset without looking
1237 at past I/O history. This means that some blocks may not be read or written,
1238 and that some blocks may be read/written more than once. If this option is
1239 used with :option:`verify` and multiple blocksizes (via :option:`bsrange`),
1240 only intact blocks are verified, i.e., partially-overwritten blocks are
1243 .. option:: softrandommap=bool
1245 See :option:`norandommap`. If fio runs with the random block map enabled and
1246 it fails to allocate the map, if this option is set it will continue without
1247 a random block map. As coverage will not be as complete as with random maps,
1248 this option is disabled by default.
1250 .. option:: random_generator=str
1252 Fio supports the following engines for generating
1253 I/O offsets for random I/O:
1256 Strong 2^88 cycle random number generator
1258 Linear feedback shift register generator
1260 Strong 64-bit 2^258 cycle random number generator
1262 **tausworthe** is a strong random number generator, but it requires tracking
1263 on the side if we want to ensure that blocks are only read or written
1264 once. **LFSR** guarantees that we never generate the same offset twice, and
1265 it's also less computationally expensive. It's not a true random generator,
1266 however, though for I/O purposes it's typically good enough. **LFSR** only
1267 works with single block sizes, not with workloads that use multiple block
1268 sizes. If used with such a workload, fio may read or write some blocks
1269 multiple times. The default value is **tausworthe**, unless the required
1270 space exceeds 2^32 blocks. If it does, then **tausworthe64** is
1271 selected automatically.
1277 .. option:: blocksize=int[,int][,int], bs=int[,int][,int]
1279 The block size in bytes used for I/O units. Default: 4096. A single value
1280 applies to reads, writes, and trims. Comma-separated values may be
1281 specified for reads, writes, and trims. A value not terminated in a comma
1282 applies to subsequent types.
1287 means 256k for reads, writes and trims.
1290 means 8k for reads, 32k for writes and trims.
1293 means 8k for reads, 32k for writes, and default for trims.
1296 means default for reads, 8k for writes and trims.
1299 means default for reads, 8k for writes, and default for trims.
1301 .. option:: blocksize_range=irange[,irange][,irange], bsrange=irange[,irange][,irange]
1303 A range of block sizes in bytes for I/O units. The issued I/O unit will
1304 always be a multiple of the minimum size, unless
1305 :option:`blocksize_unaligned` is set.
1307 Comma-separated ranges may be specified for reads, writes, and trims as
1308 described in :option:`blocksize`.
1310 Example: ``bsrange=1k-4k,2k-8k``.
1312 .. option:: bssplit=str[,str][,str]
1314 Sometimes you want even finer grained control of the block sizes issued, not
1315 just an even split between them. This option allows you to weight various
1316 block sizes, so that you are able to define a specific amount of block sizes
1317 issued. The format for this option is::
1319 bssplit=blocksize/percentage:blocksize/percentage
1321 for as many block sizes as needed. So if you want to define a workload that
1322 has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would write::
1324 bssplit=4k/10:64k/50:32k/40
1326 Ordering does not matter. If the percentage is left blank, fio will fill in
1327 the remaining values evenly. So a bssplit option like this one::
1329 bssplit=4k/50:1k/:32k/
1331 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always add up
1332 to 100, if bssplit is given a range that adds up to more, it will error out.
1334 Comma-separated values may be specified for reads, writes, and trims as
1335 described in :option:`blocksize`.
1337 If you want a workload that has 50% 2k reads and 50% 4k reads, while having
1338 90% 4k writes and 10% 8k writes, you would specify::
1340 bssplit=2k/50:4k/50,4k/90,8k/10
1342 .. option:: blocksize_unaligned, bs_unaligned
1344 If set, fio will issue I/O units with any size within
1345 :option:`blocksize_range`, not just multiples of the minimum size. This
1346 typically won't work with direct I/O, as that normally requires sector
1349 .. option:: bs_is_seq_rand
1351 If this option is set, fio will use the normal read,write blocksize settings
1352 as sequential,random blocksize settings instead. Any random read or write
1353 will use the WRITE blocksize settings, and any sequential read or write will
1354 use the READ blocksize settings.
1356 .. option:: blockalign=int[,int][,int], ba=int[,int][,int]
1358 Boundary to which fio will align random I/O units. Default:
1359 :option:`blocksize`. Minimum alignment is typically 512b for using direct
1360 I/O, though it usually depends on the hardware block size. This option is
1361 mutually exclusive with using a random map for files, so it will turn off
1362 that option. Comma-separated values may be specified for reads, writes, and
1363 trims as described in :option:`blocksize`.
1369 .. option:: zero_buffers
1371 Initialize buffers with all zeros. Default: fill buffers with random data.
1373 .. option:: refill_buffers
1375 If this option is given, fio will refill the I/O buffers on every
1376 submit. The default is to only fill it at init time and reuse that
1377 data. Only makes sense if zero_buffers isn't specified, naturally. If data
1378 verification is enabled, `refill_buffers` is also automatically enabled.
1380 .. option:: scramble_buffers=bool
1382 If :option:`refill_buffers` is too costly and the target is using data
1383 deduplication, then setting this option will slightly modify the I/O buffer
1384 contents to defeat normal de-dupe attempts. This is not enough to defeat
1385 more clever block compression attempts, but it will stop naive dedupe of
1386 blocks. Default: true.
1388 .. option:: buffer_compress_percentage=int
1390 If this is set, then fio will attempt to provide I/O buffer content (on
1391 WRITEs) that compress to the specified level. Fio does this by providing a
1392 mix of random data and a fixed pattern. The fixed pattern is either zeroes,
1393 or the pattern specified by :option:`buffer_pattern`. If the pattern option
1394 is used, it might skew the compression ratio slightly. Note that this is per
1395 block size unit, for file/disk wide compression level that matches this
1396 setting, you'll also want to set :option:`refill_buffers`.
1398 .. option:: buffer_compress_chunk=int
1400 See :option:`buffer_compress_percentage`. This setting allows fio to manage
1401 how big the ranges of random data and zeroed data is. Without this set, fio
1402 will provide :option:`buffer_compress_percentage` of blocksize random data,
1403 followed by the remaining zeroed. With this set to some chunk size smaller
1404 than the block size, fio can alternate random and zeroed data throughout the
1407 .. option:: buffer_pattern=str
1409 If set, fio will fill the I/O buffers with this pattern or with the contents
1410 of a file. If not set, the contents of I/O buffers are defined by the other
1411 options related to buffer contents. The setting can be any pattern of bytes,
1412 and can be prefixed with 0x for hex values. It may also be a string, where
1413 the string must then be wrapped with ``""``. Or it may also be a filename,
1414 where the filename must be wrapped with ``''`` in which case the file is
1415 opened and read. Note that not all the file contents will be read if that
1416 would cause the buffers to overflow. So, for example::
1418 buffer_pattern='filename'
1422 buffer_pattern="abcd"
1430 buffer_pattern=0xdeadface
1432 Also you can combine everything together in any order::
1434 buffer_pattern=0xdeadface"abcd"-12'filename'
1436 .. option:: dedupe_percentage=int
1438 If set, fio will generate this percentage of identical buffers when
1439 writing. These buffers will be naturally dedupable. The contents of the
1440 buffers depend on what other buffer compression settings have been set. It's
1441 possible to have the individual buffers either fully compressible, or not at
1442 all. This option only controls the distribution of unique buffers.
1444 .. option:: invalidate=bool
1446 Invalidate the buffer/page cache parts for this file prior to starting
1447 I/O if the platform and file type support it. Defaults to true.
1448 This will be ignored if :option:`pre_read` is also specified for the
1451 .. option:: sync=bool
1453 Use synchronous I/O for buffered writes. For the majority of I/O engines,
1454 this means using O_SYNC. Default: false.
1456 .. option:: iomem=str, mem=str
1458 Fio can use various types of memory as the I/O unit buffer. The allowed
1462 Use memory from :manpage:`malloc(3)` as the buffers. Default memory
1466 Use shared memory as the buffers. Allocated through
1467 :manpage:`shmget(2)`.
1470 Same as shm, but use huge pages as backing.
1473 Use mmap to allocate buffers. May either be anonymous memory, or can
1474 be file backed if a filename is given after the option. The format
1475 is `mem=mmap:/path/to/file`.
1478 Use a memory mapped huge file as the buffer backing. Append filename
1479 after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file`.
1482 Same as mmap, but use a MMAP_SHARED mapping.
1485 Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1487 The area allocated is a function of the maximum allowed bs size for the job,
1488 multiplied by the I/O depth given. Note that for **shmhuge** and
1489 **mmaphuge** to work, the system must have free huge pages allocated. This
1490 can normally be checked and set by reading/writing
1491 :file:`/proc/sys/vm/nr_hugepages` on a Linux system. Fio assumes a huge page
1492 is 4MiB in size. So to calculate the number of huge pages you need for a
1493 given job file, add up the I/O depth of all jobs (normally one unless
1494 :option:`iodepth` is used) and multiply by the maximum bs set. Then divide
1495 that number by the huge page size. You can see the size of the huge pages in
1496 :file:`/proc/meminfo`. If no huge pages are allocated by having a non-zero
1497 number in `nr_hugepages`, using **mmaphuge** or **shmhuge** will fail. Also
1498 see :option:`hugepage-size`.
1500 **mmaphuge** also needs to have hugetlbfs mounted and the file location
1501 should point there. So if it's mounted in :file:`/huge`, you would use
1502 `mem=mmaphuge:/huge/somefile`.
1504 .. option:: iomem_align=int
1506 This indicates the memory alignment of the I/O memory buffers. Note that
1507 the given alignment is applied to the first I/O unit buffer, if using
1508 :option:`iodepth` the alignment of the following buffers are given by the
1509 :option:`bs` used. In other words, if using a :option:`bs` that is a
1510 multiple of the page sized in the system, all buffers will be aligned to
1511 this value. If using a :option:`bs` that is not page aligned, the alignment
1512 of subsequent I/O memory buffers is the sum of the :option:`iomem_align` and
1515 .. option:: hugepage-size=int
1517 Defines the size of a huge page. Must at least be equal to the system
1518 setting, see :file:`/proc/meminfo`. Defaults to 4MiB. Should probably
1519 always be a multiple of megabytes, so using ``hugepage-size=Xm`` is the
1520 preferred way to set this to avoid setting a non-pow-2 bad value.
1522 .. option:: lockmem=int
1524 Pin the specified amount of memory with :manpage:`mlock(2)`. Can be used to
1525 simulate a smaller amount of memory. The amount specified is per worker.
1531 .. option:: size=int
1533 The total size of file I/O for each thread of this job. Fio will run until
1534 this many bytes has been transferred, unless runtime is limited by other options
1535 (such as :option:`runtime`, for instance, or increased/decreased by :option:`io_size`).
1536 Fio will divide this size between the available files determined by options
1537 such as :option:`nrfiles`, :option:`filename`, unless :option:`filesize` is
1538 specified by the job. If the result of division happens to be 0, the size is
1539 set to the physical size of the given files or devices if they exist.
1540 If this option is not specified, fio will use the full size of the given
1541 files or devices. If the files do not exist, size must be given. It is also
1542 possible to give size as a percentage between 1 and 100. If ``size=20%`` is
1543 given, fio will use 20% of the full size of the given files or devices.
1544 Can be combined with :option:`offset` to constrain the start and end range
1545 that I/O will be done within.
1547 .. option:: io_size=int, io_limit=int
1549 Normally fio operates within the region set by :option:`size`, which means
1550 that the :option:`size` option sets both the region and size of I/O to be
1551 performed. Sometimes that is not what you want. With this option, it is
1552 possible to define just the amount of I/O that fio should do. For instance,
1553 if :option:`size` is set to 20GiB and :option:`io_size` is set to 5GiB, fio
1554 will perform I/O within the first 20GiB but exit when 5GiB have been
1555 done. The opposite is also possible -- if :option:`size` is set to 20GiB,
1556 and :option:`io_size` is set to 40GiB, then fio will do 40GiB of I/O within
1557 the 0..20GiB region.
1559 .. option:: filesize=int
1561 Individual file sizes. May be a range, in which case fio will select sizes
1562 for files at random within the given range and limited to :option:`size` in
1563 total (if that is given). If not given, each created file is the same size.
1564 This option overrides :option:`size` in terms of file size, which means
1565 this value is used as a fixed size or possible range of each file.
1567 .. option:: file_append=bool
1569 Perform I/O after the end of the file. Normally fio will operate within the
1570 size of a file. If this option is set, then fio will append to the file
1571 instead. This has identical behavior to setting :option:`offset` to the size
1572 of a file. This option is ignored on non-regular files.
1574 .. option:: fill_device=bool, fill_fs=bool
1576 Sets size to something really large and waits for ENOSPC (no space left on
1577 device) as the terminating condition. Only makes sense with sequential
1578 write. For a read workload, the mount point will be filled first then I/O
1579 started on the result. This option doesn't make sense if operating on a raw
1580 device node, since the size of that is already known by the file system.
1581 Additionally, writing beyond end-of-device will not return ENOSPC there.
1587 .. option:: ioengine=str
1589 Defines how the job issues I/O to the file. The following types are defined:
1592 Basic :manpage:`read(2)` or :manpage:`write(2)`
1593 I/O. :manpage:`lseek(2)` is used to position the I/O location.
1594 See :option:`fsync` and :option:`fdatasync` for syncing write I/Os.
1597 Basic :manpage:`pread(2)` or :manpage:`pwrite(2)` I/O. Default on
1598 all supported operating systems except for Windows.
1601 Basic :manpage:`readv(2)` or :manpage:`writev(2)` I/O. Will emulate
1602 queuing by coalescing adjacent I/Os into a single submission.
1605 Basic :manpage:`preadv(2)` or :manpage:`pwritev(2)` I/O.
1608 Basic :manpage:`preadv2(2)` or :manpage:`pwritev2(2)` I/O.
1611 Linux native asynchronous I/O. Note that Linux may only support
1612 queued behaviour with non-buffered I/O (set ``direct=1`` or
1614 This engine defines engine specific options.
1617 POSIX asynchronous I/O using :manpage:`aio_read(3)` and
1618 :manpage:`aio_write(3)`.
1621 Solaris native asynchronous I/O.
1624 Windows native asynchronous I/O. Default on Windows.
1627 File is memory mapped with :manpage:`mmap(2)` and data copied
1628 to/from using :manpage:`memcpy(3)`.
1631 :manpage:`splice(2)` is used to transfer the data and
1632 :manpage:`vmsplice(2)` to transfer data from user space to the
1636 SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1637 ioctl, or if the target is an sg character device we use
1638 :manpage:`read(2)` and :manpage:`write(2)` for asynchronous
1639 I/O. Requires filename option to specify either block or character
1643 Doesn't transfer any data, just pretends to. This is mainly used to
1644 exercise fio itself and for debugging/testing purposes.
1647 Transfer over the network to given ``host:port``. Depending on the
1648 :option:`protocol` used, the :option:`hostname`, :option:`port`,
1649 :option:`listen` and :option:`filename` options are used to specify
1650 what sort of connection to make, while the :option:`protocol` option
1651 determines which protocol will be used. This engine defines engine
1655 Like **net**, but uses :manpage:`splice(2)` and
1656 :manpage:`vmsplice(2)` to map data and send/receive.
1657 This engine defines engine specific options.
1660 Doesn't transfer any data, but burns CPU cycles according to the
1661 :option:`cpuload` and :option:`cpuchunks` options. Setting
1662 :option:`cpuload`\=85 will cause that job to do nothing but burn 85%
1663 of the CPU. In case of SMP machines, use :option:`numjobs`
1664 =<no_of_cpu> to get desired CPU usage, as the cpuload only loads a
1665 single CPU at the desired rate. A job never finishes unless there is
1666 at least one non-cpuio job.
1669 The GUASI I/O engine is the Generic Userspace Asyncronous Syscall
1670 Interface approach to async I/O. See
1672 http://www.xmailserver.org/guasi-lib.html
1674 for more info on GUASI.
1677 The RDMA I/O engine supports both RDMA memory semantics
1678 (RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
1679 InfiniBand, RoCE and iWARP protocols.
1682 I/O engine that does regular fallocate to simulate data transfer as
1686 does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1689 does fallocate(,mode = 0).
1692 does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1695 I/O engine that sends :manpage:`ftruncate(2)` operations in response
1696 to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
1697 size to the current block offset. Block size is ignored.
1700 I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1701 defragment activity in request to DDIR_WRITE event.
1704 I/O engine supporting direct access to Ceph Rados Block Devices
1705 (RBD) via librbd without the need to use the kernel rbd driver. This
1706 ioengine defines engine specific options.
1709 Using Glusterfs libgfapi sync interface to direct access to
1710 Glusterfs volumes without having to go through FUSE. This ioengine
1711 defines engine specific options.
1714 Using Glusterfs libgfapi async interface to direct access to
1715 Glusterfs volumes without having to go through FUSE. This ioengine
1716 defines engine specific options.
1719 Read and write through Hadoop (HDFS). The :file:`filename` option
1720 is used to specify host,port of the hdfs name-node to connect. This
1721 engine interprets offsets a little differently. In HDFS, files once
1722 created cannot be modified. So random writes are not possible. To
1723 imitate this, libhdfs engine expects bunch of small files to be
1724 created over HDFS, and engine will randomly pick a file out of those
1725 files based on the offset generated by fio backend. (see the example
1726 job file to create such files, use ``rw=write`` option). Please
1727 note, you might want to set necessary environment variables to work
1728 with hdfs/libhdfs properly. Each job uses its own connection to
1732 Read, write and erase an MTD character device (e.g.,
1733 :file:`/dev/mtd0`). Discards are treated as erases. Depending on the
1734 underlying device type, the I/O may have to go in a certain pattern,
1735 e.g., on NAND, writing sequentially to erase blocks and discarding
1736 before overwriting. The writetrim mode works well for this
1740 Read and write using filesystem DAX to a file on a filesystem
1741 mounted with DAX on a persistent memory device through the NVML
1745 Read and write using device DAX to a persistent memory device (e.g.,
1746 /dev/dax0.0) through the NVML libpmem library.
1749 Prefix to specify loading an external I/O engine object file. Append
1750 the engine filename, e.g. ``ioengine=external:/tmp/foo.o`` to load
1751 ioengine :file:`foo.o` in :file:`/tmp`.
1754 I/O engine specific parameters
1755 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1757 In addition, there are some parameters which are only valid when a specific
1758 ioengine is in use. These are used identically to normal parameters, with the
1759 caveat that when used on the command line, they must come after the
1760 :option:`ioengine` that defines them is selected.
1762 .. option:: userspace_reap : [libaio]
1764 Normally, with the libaio engine in use, fio will use the
1765 :manpage:`io_getevents(2)` system call to reap newly returned events. With
1766 this flag turned on, the AIO ring will be read directly from user-space to
1767 reap events. The reaping mode is only enabled when polling for a minimum of
1768 0 events (e.g. when :option:`iodepth_batch_complete` `=0`).
1770 .. option:: hipri : [pvsync2]
1772 Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
1775 .. option:: cpuload=int : [cpuio]
1777 Attempt to use the specified percentage of CPU cycles. This is a mandatory
1778 option when using cpuio I/O engine.
1780 .. option:: cpuchunks=int : [cpuio]
1782 Split the load into cycles of the given time. In microseconds.
1784 .. option:: exit_on_io_done=bool : [cpuio]
1786 Detect when I/O threads are done, then exit.
1788 .. option:: hostname=str : [netsplice] [net]
1790 The host name or IP address to use for TCP or UDP based I/O. If the job is
1791 a TCP listener or UDP reader, the host name is not used and must be omitted
1792 unless it is a valid UDP multicast address.
1794 .. option:: namenode=str : [libhdfs]
1796 The host name or IP address of a HDFS cluster namenode to contact.
1798 .. option:: port=int
1802 The TCP or UDP port to bind to or connect to. If this is used with
1803 :option:`numjobs` to spawn multiple instances of the same job type, then
1804 this will be the starting port number since fio will use a range of
1809 the listening port of the HFDS cluster namenode.
1811 .. option:: interface=str : [netsplice] [net]
1813 The IP address of the network interface used to send or receive UDP
1816 .. option:: ttl=int : [netsplice] [net]
1818 Time-to-live value for outgoing UDP multicast packets. Default: 1.
1820 .. option:: nodelay=bool : [netsplice] [net]
1822 Set TCP_NODELAY on TCP connections.
1824 .. option:: protocol=str : [netsplice] [net]
1826 .. option:: proto=str : [netsplice] [net]
1828 The network protocol to use. Accepted values are:
1831 Transmission control protocol.
1833 Transmission control protocol V6.
1835 User datagram protocol.
1837 User datagram protocol V6.
1841 When the protocol is TCP or UDP, the port must also be given, as well as the
1842 hostname if the job is a TCP listener or UDP reader. For unix sockets, the
1843 normal filename option should be used and the port is invalid.
1845 .. option:: listen : [net]
1847 For TCP network connections, tell fio to listen for incoming connections
1848 rather than initiating an outgoing connection. The :option:`hostname` must
1849 be omitted if this option is used.
1851 .. option:: pingpong : [net]
1853 Normally a network writer will just continue writing data, and a network
1854 reader will just consume packages. If ``pingpong=1`` is set, a writer will
1855 send its normal payload to the reader, then wait for the reader to send the
1856 same payload back. This allows fio to measure network latencies. The
1857 submission and completion latencies then measure local time spent sending or
1858 receiving, and the completion latency measures how long it took for the
1859 other end to receive and send back. For UDP multicast traffic
1860 ``pingpong=1`` should only be set for a single reader when multiple readers
1861 are listening to the same address.
1863 .. option:: window_size : [net]
1865 Set the desired socket buffer size for the connection.
1867 .. option:: mss : [net]
1869 Set the TCP maximum segment size (TCP_MAXSEG).
1871 .. option:: donorname=str : [e4defrag]
1873 File will be used as a block donor(swap extents between files).
1875 .. option:: inplace=int : [e4defrag]
1877 Configure donor file blocks allocation strategy:
1880 Default. Preallocate donor's file on init.
1882 Allocate space immediately inside defragment event, and free right
1885 .. option:: clustername=str : [rbd]
1887 Specifies the name of the Ceph cluster.
1889 .. option:: rbdname=str : [rbd]
1891 Specifies the name of the RBD.
1893 .. option:: pool=str : [rbd]
1895 Specifies the name of the Ceph pool containing RBD.
1897 .. option:: clientname=str : [rbd]
1899 Specifies the username (without the 'client.' prefix) used to access the
1900 Ceph cluster. If the *clustername* is specified, the *clientname* shall be
1901 the full *type.id* string. If no type. prefix is given, fio will add
1902 'client.' by default.
1904 .. option:: skip_bad=bool : [mtd]
1906 Skip operations against known bad blocks.
1908 .. option:: hdfsdirectory : [libhdfs]
1910 libhdfs will create chunk in this HDFS directory.
1912 .. option:: chunk_size : [libhdfs]
1914 the size of the chunk to use for each file.
1920 .. option:: iodepth=int
1922 Number of I/O units to keep in flight against the file. Note that
1923 increasing *iodepth* beyond 1 will not affect synchronous ioengines (except
1924 for small degrees when :option:`verify_async` is in use). Even async
1925 engines may impose OS restrictions causing the desired depth not to be
1926 achieved. This may happen on Linux when using libaio and not setting
1927 :option:`direct`\=1, since buffered I/O is not async on that OS. Keep an
1928 eye on the I/O depth distribution in the fio output to verify that the
1929 achieved depth is as expected. Default: 1.
1931 .. option:: iodepth_batch_submit=int, iodepth_batch=int
1933 This defines how many pieces of I/O to submit at once. It defaults to 1
1934 which means that we submit each I/O as soon as it is available, but can be
1935 raised to submit bigger batches of I/O at the time. If it is set to 0 the
1936 :option:`iodepth` value will be used.
1938 .. option:: iodepth_batch_complete_min=int, iodepth_batch_complete=int
1940 This defines how many pieces of I/O to retrieve at once. It defaults to 1
1941 which means that we'll ask for a minimum of 1 I/O in the retrieval process
1942 from the kernel. The I/O retrieval will go on until we hit the limit set by
1943 :option:`iodepth_low`. If this variable is set to 0, then fio will always
1944 check for completed events before queuing more I/O. This helps reduce I/O
1945 latency, at the cost of more retrieval system calls.
1947 .. option:: iodepth_batch_complete_max=int
1949 This defines maximum pieces of I/O to retrieve at once. This variable should
1950 be used along with :option:`iodepth_batch_complete_min`\=int variable,
1951 specifying the range of min and max amount of I/O which should be
1952 retrieved. By default it is equal to :option:`iodepth_batch_complete_min`
1957 iodepth_batch_complete_min=1
1958 iodepth_batch_complete_max=<iodepth>
1960 which means that we will retrieve at least 1 I/O and up to the whole
1961 submitted queue depth. If none of I/O has been completed yet, we will wait.
1965 iodepth_batch_complete_min=0
1966 iodepth_batch_complete_max=<iodepth>
1968 which means that we can retrieve up to the whole submitted queue depth, but
1969 if none of I/O has been completed yet, we will NOT wait and immediately exit
1970 the system call. In this example we simply do polling.
1972 .. option:: iodepth_low=int
1974 The low water mark indicating when to start filling the queue
1975 again. Defaults to the same as :option:`iodepth`, meaning that fio will
1976 attempt to keep the queue full at all times. If :option:`iodepth` is set to
1977 e.g. 16 and *iodepth_low* is set to 4, then after fio has filled the queue of
1978 16 requests, it will let the depth drain down to 4 before starting to fill
1981 .. option:: io_submit_mode=str
1983 This option controls how fio submits the I/O to the I/O engine. The default
1984 is `inline`, which means that the fio job threads submit and reap I/O
1985 directly. If set to `offload`, the job threads will offload I/O submission
1986 to a dedicated pool of I/O threads. This requires some coordination and thus
1987 has a bit of extra overhead, especially for lower queue depth I/O where it
1988 can increase latencies. The benefit is that fio can manage submission rates
1989 independently of the device completion rates. This avoids skewed latency
1990 reporting if I/O gets back up on the device side (the coordinated omission
1997 .. option:: thinktime=time
1999 Stall the job for the specified period of time after an I/O has completed before issuing the
2000 next. May be used to simulate processing being done by an application.
2001 When the unit is omitted, the value is interpreted in microseconds. See
2002 :option:`thinktime_blocks` and :option:`thinktime_spin`.
2004 .. option:: thinktime_spin=time
2006 Only valid if :option:`thinktime` is set - pretend to spend CPU time doing
2007 something with the data received, before falling back to sleeping for the
2008 rest of the period specified by :option:`thinktime`. When the unit is
2009 omitted, the value is interpreted in microseconds.
2011 .. option:: thinktime_blocks=int
2013 Only valid if :option:`thinktime` is set - control how many blocks to issue,
2014 before waiting `thinktime` usecs. If not set, defaults to 1 which will make
2015 fio wait `thinktime` usecs after every block. This effectively makes any
2016 queue depth setting redundant, since no more than 1 I/O will be queued
2017 before we have to complete it and do our thinktime. In other words, this
2018 setting effectively caps the queue depth if the latter is larger.
2020 .. option:: rate=int[,int][,int]
2022 Cap the bandwidth used by this job. The number is in bytes/sec, the normal
2023 suffix rules apply. Comma-separated values may be specified for reads,
2024 writes, and trims as described in :option:`blocksize`.
2026 .. option:: rate_min=int[,int][,int]
2028 Tell fio to do whatever it can to maintain at least this bandwidth. Failing
2029 to meet this requirement will cause the job to exit. Comma-separated values
2030 may be specified for reads, writes, and trims as described in
2031 :option:`blocksize`.
2033 .. option:: rate_iops=int[,int][,int]
2035 Cap the bandwidth to this number of IOPS. Basically the same as
2036 :option:`rate`, just specified independently of bandwidth. If the job is
2037 given a block size range instead of a fixed value, the smallest block size
2038 is used as the metric. Comma-separated values may be specified for reads,
2039 writes, and trims as described in :option:`blocksize`.
2041 .. option:: rate_iops_min=int[,int][,int]
2043 If fio doesn't meet this rate of I/O, it will cause the job to exit.
2044 Comma-separated values may be specified for reads, writes, and trims as
2045 described in :option:`blocksize`.
2047 .. option:: rate_process=str
2049 This option controls how fio manages rated I/O submissions. The default is
2050 `linear`, which submits I/O in a linear fashion with fixed delays between
2051 I/Os that gets adjusted based on I/O completion rates. If this is set to
2052 `poisson`, fio will submit I/O based on a more real world random request
2053 flow, known as the Poisson process
2054 (https://en.wikipedia.org/wiki/Poisson_point_process). The lambda will be
2055 10^6 / IOPS for the given workload.
2061 .. option:: latency_target=time
2063 If set, fio will attempt to find the max performance point that the given
2064 workload will run at while maintaining a latency below this target. When
2065 the unit is omitted, the value is interpreted in microseconds. See
2066 :option:`latency_window` and :option:`latency_percentile`.
2068 .. option:: latency_window=time
2070 Used with :option:`latency_target` to specify the sample window that the job
2071 is run at varying queue depths to test the performance. When the unit is
2072 omitted, the value is interpreted in microseconds.
2074 .. option:: latency_percentile=float
2076 The percentage of I/Os that must fall within the criteria specified by
2077 :option:`latency_target` and :option:`latency_window`. If not set, this
2078 defaults to 100.0, meaning that all I/Os must be equal or below to the value
2079 set by :option:`latency_target`.
2081 .. option:: max_latency=time
2083 If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
2084 maximum latency. When the unit is omitted, the value is interpreted in
2087 .. option:: rate_cycle=int
2089 Average bandwidth for :option:`rate` and :option:`rate_min` over this number
2090 of milliseconds. Defaults to 1000.
2096 .. option:: write_iolog=str
2098 Write the issued I/O patterns to the specified file. See
2099 :option:`read_iolog`. Specify a separate file for each job, otherwise the
2100 iologs will be interspersed and the file may be corrupt.
2102 .. option:: read_iolog=str
2104 Open an iolog with the specified file name and replay the I/O patterns it
2105 contains. This can be used to store a workload and replay it sometime
2106 later. The iolog given may also be a blktrace binary file, which allows fio
2107 to replay a workload captured by :command:`blktrace`. See
2108 :manpage:`blktrace(8)` for how to capture such logging data. For blktrace
2109 replay, the file needs to be turned into a blkparse binary data file first
2110 (``blkparse <device> -o /dev/null -d file_for_fio.bin``).
2112 .. option:: replay_no_stall=int
2114 When replaying I/O with :option:`read_iolog` the default behavior is to
2115 attempt to respect the time stamps within the log and replay them with the
2116 appropriate delay between IOPS. By setting this variable fio will not
2117 respect the timestamps and attempt to replay them as fast as possible while
2118 still respecting ordering. The result is the same I/O pattern to a given
2119 device, but different timings.
2121 .. option:: replay_redirect=str
2123 While replaying I/O patterns using :option:`read_iolog` the default behavior
2124 is to replay the IOPS onto the major/minor device that each IOP was recorded
2125 from. This is sometimes undesirable because on a different machine those
2126 major/minor numbers can map to a different device. Changing hardware on the
2127 same system can also result in a different major/minor mapping.
2128 ``replay_redirect`` causes all IOPS to be replayed onto the single specified
2129 device regardless of the device it was recorded
2130 from. i.e. :option:`replay_redirect`\= :file:`/dev/sdc` would cause all I/O
2131 in the blktrace or iolog to be replayed onto :file:`/dev/sdc`. This means
2132 multiple devices will be replayed onto a single device, if the trace
2133 contains multiple devices. If you want multiple devices to be replayed
2134 concurrently to multiple redirected devices you must blkparse your trace
2135 into separate traces and replay them with independent fio invocations.
2136 Unfortunately this also breaks the strict time ordering between multiple
2139 .. option:: replay_align=int
2141 Force alignment of I/O offsets and lengths in a trace to this power of 2
2144 .. option:: replay_scale=int
2146 Scale sector offsets down by this factor when replaying traces.
2149 Threads, processes and job synchronization
2150 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2154 Fio defaults to forking jobs, however if this option is given, fio will use
2155 POSIX Threads function :manpage:`pthread_create(3)` to create threads instead
2156 of forking processes.
2158 .. option:: wait_for=str
2160 Specifies the name of the already defined job to wait for. Single waitee
2161 name only may be specified. If set, the job won't be started until all
2162 workers of the waitee job are done.
2164 ``wait_for`` operates on the job name basis, so there are a few
2165 limitations. First, the waitee must be defined prior to the waiter job
2166 (meaning no forward references). Second, if a job is being referenced as a
2167 waitee, it must have a unique name (no duplicate waitees).
2169 .. option:: nice=int
2171 Run the job with the given nice value. See man :manpage:`nice(2)`.
2173 On Windows, values less than -15 set the process class to "High"; -1 through
2174 -15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2177 .. option:: prio=int
2179 Set the I/O priority value of this job. Linux limits us to a positive value
2180 between 0 and 7, with 0 being the highest. See man
2181 :manpage:`ionice(1)`. Refer to an appropriate manpage for other operating
2182 systems since meaning of priority may differ.
2184 .. option:: prioclass=int
2186 Set the I/O priority class. See man :manpage:`ionice(1)`.
2188 .. option:: cpumask=int
2190 Set the CPU affinity of this job. The parameter given is a bitmask of
2191 allowed CPU's the job may run on. So if you want the allowed CPUs to be 1
2192 and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2193 :manpage:`sched_setaffinity(2)`. This may not work on all supported
2194 operating systems or kernel versions. This option doesn't work well for a
2195 higher CPU count than what you can store in an integer mask, so it can only
2196 control cpus 1-32. For boxes with larger CPU counts, use
2197 :option:`cpus_allowed`.
2199 .. option:: cpus_allowed=str
2201 Controls the same options as :option:`cpumask`, but it allows a text setting
2202 of the permitted CPUs instead. So to use CPUs 1 and 5, you would specify
2203 ``cpus_allowed=1,5``. This options also allows a range of CPUs. Say you
2204 wanted a binding to CPUs 1, 5, and 8-15, you would set
2205 ``cpus_allowed=1,5,8-15``.
2207 .. option:: cpus_allowed_policy=str
2209 Set the policy of how fio distributes the CPUs specified by
2210 :option:`cpus_allowed` or cpumask. Two policies are supported:
2213 All jobs will share the CPU set specified.
2215 Each job will get a unique CPU from the CPU set.
2217 **shared** is the default behaviour, if the option isn't specified. If
2218 **split** is specified, then fio will will assign one cpu per job. If not
2219 enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2222 .. option:: numa_cpu_nodes=str
2224 Set this job running on specified NUMA nodes' CPUs. The arguments allow
2225 comma delimited list of cpu numbers, A-B ranges, or `all`. Note, to enable
2226 numa options support, fio must be built on a system with libnuma-dev(el)
2229 .. option:: numa_mem_policy=str
2231 Set this job's memory policy and corresponding NUMA nodes. Format of the
2236 ``mode`` is one of the following memory policy: ``default``, ``prefer``,
2237 ``bind``, ``interleave``, ``local`` For ``default`` and ``local`` memory
2238 policy, no node is needed to be specified. For ``prefer``, only one node is
2239 allowed. For ``bind`` and ``interleave``, it allow comma delimited list of
2240 numbers, A-B ranges, or `all`.
2242 .. option:: cgroup=str
2244 Add job to this control group. If it doesn't exist, it will be created. The
2245 system must have a mounted cgroup blkio mount point for this to work. If
2246 your system doesn't have it mounted, you can do so with::
2248 # mount -t cgroup -o blkio none /cgroup
2250 .. option:: cgroup_weight=int
2252 Set the weight of the cgroup to this value. See the documentation that comes
2253 with the kernel, allowed values are in the range of 100..1000.
2255 .. option:: cgroup_nodelete=bool
2257 Normally fio will delete the cgroups it has created after the job
2258 completion. To override this behavior and to leave cgroups around after the
2259 job completion, set ``cgroup_nodelete=1``. This can be useful if one wants
2260 to inspect various cgroup files after job completion. Default: false.
2262 .. option:: flow_id=int
2264 The ID of the flow. If not specified, it defaults to being a global
2265 flow. See :option:`flow`.
2267 .. option:: flow=int
2269 Weight in token-based flow control. If this value is used, then there is a
2270 'flow counter' which is used to regulate the proportion of activity between
2271 two or more jobs. Fio attempts to keep this flow counter near zero. The
2272 ``flow`` parameter stands for how much should be added or subtracted to the
2273 flow counter on each iteration of the main I/O loop. That is, if one job has
2274 ``flow=8`` and another job has ``flow=-1``, then there will be a roughly 1:8
2275 ratio in how much one runs vs the other.
2277 .. option:: flow_watermark=int
2279 The maximum value that the absolute value of the flow counter is allowed to
2280 reach before the job must wait for a lower value of the counter.
2282 .. option:: flow_sleep=int
2284 The period of time, in microseconds, to wait after the flow watermark has
2285 been exceeded before retrying operations.
2287 .. option:: stonewall, wait_for_previous
2289 Wait for preceding jobs in the job file to exit, before starting this
2290 one. Can be used to insert serialization points in the job file. A stone
2291 wall also implies starting a new reporting group, see
2292 :option:`group_reporting`.
2296 When one job finishes, terminate the rest. The default is to wait for each
2297 job to finish, sometimes that is not the desired action.
2299 .. option:: exec_prerun=str
2301 Before running this job, issue the command specified through
2302 :manpage:`system(3)`. Output is redirected in a file called
2303 :file:`jobname.prerun.txt`.
2305 .. option:: exec_postrun=str
2307 After the job completes, issue the command specified though
2308 :manpage:`system(3)`. Output is redirected in a file called
2309 :file:`jobname.postrun.txt`.
2313 Instead of running as the invoking user, set the user ID to this value
2314 before the thread/process does any work.
2318 Set group ID, see :option:`uid`.
2324 .. option:: verify_only
2326 Do not perform specified workload, only verify data still matches previous
2327 invocation of this workload. This option allows one to check data multiple
2328 times at a later date without overwriting it. This option makes sense only
2329 for workloads that write data, and does not support workloads with the
2330 :option:`time_based` option set.
2332 .. option:: do_verify=bool
2334 Run the verify phase after a write phase. Only valid if :option:`verify` is
2337 .. option:: verify=str
2339 If writing to a file, fio can verify the file contents after each iteration
2340 of the job. Each verification method also implies verification of special
2341 header, which is written to the beginning of each block. This header also
2342 includes meta information, like offset of the block, block number, timestamp
2343 when block was written, etc. :option:`verify` can be combined with
2344 :option:`verify_pattern` option. The allowed values are:
2347 Use an md5 sum of the data area and store it in the header of
2351 Use an experimental crc64 sum of the data area and store it in the
2352 header of each block.
2355 Use a crc32c sum of the data area and store it in the header of each
2359 Use hardware assisted crc32c calculation provided on SSE4.2 enabled
2360 processors. Falls back to regular software crc32c, if not supported
2364 Use a crc32 sum of the data area and store it in the header of each
2368 Use a crc16 sum of the data area and store it in the header of each
2372 Use a crc7 sum of the data area and store it in the header of each
2376 Use xxhash as the checksum function. Generally the fastest software
2377 checksum that fio supports.
2380 Use sha512 as the checksum function.
2383 Use sha256 as the checksum function.
2386 Use optimized sha1 as the checksum function.
2389 Use optimized sha3-224 as the checksum function.
2392 Use optimized sha3-256 as the checksum function.
2395 Use optimized sha3-384 as the checksum function.
2398 Use optimized sha3-512 as the checksum function.
2401 This option is deprecated, since now meta information is included in
2402 generic verification header and meta verification happens by
2403 default. For detailed information see the description of the
2404 :option:`verify` setting. This option is kept because of
2405 compatibility's sake with old configurations. Do not use it.
2408 Verify a strict pattern. Normally fio includes a header with some
2409 basic information and checksumming, but if this option is set, only
2410 the specific pattern set with :option:`verify_pattern` is verified.
2413 Only pretend to verify. Useful for testing internals with
2414 :option:`ioengine`\=null, not for much else.
2416 This option can be used for repeated burn-in tests of a system to make sure
2417 that the written data is also correctly read back. If the data direction
2418 given is a read or random read, fio will assume that it should verify a
2419 previously written file. If the data direction includes any form of write,
2420 the verify will be of the newly written data.
2422 .. option:: verifysort=bool
2424 If true, fio will sort written verify blocks when it deems it faster to read
2425 them back in a sorted manner. This is often the case when overwriting an
2426 existing file, since the blocks are already laid out in the file system. You
2427 can ignore this option unless doing huge amounts of really fast I/O where
2428 the red-black tree sorting CPU time becomes significant. Default: true.
2430 .. option:: verifysort_nr=int
2432 Pre-load and sort verify blocks for a read workload.
2434 .. option:: verify_offset=int
2436 Swap the verification header with data somewhere else in the block before
2437 writing. It is swapped back before verifying.
2439 .. option:: verify_interval=int
2441 Write the verification header at a finer granularity than the
2442 :option:`blocksize`. It will be written for chunks the size of
2443 ``verify_interval``. :option:`blocksize` should divide this evenly.
2445 .. option:: verify_pattern=str
2447 If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2448 filling with totally random bytes, but sometimes it's interesting to fill
2449 with a known pattern for I/O verification purposes. Depending on the width
2450 of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time(it can
2451 be either a decimal or a hex number). The ``verify_pattern`` if larger than
2452 a 32-bit quantity has to be a hex number that starts with either "0x" or
2453 "0X". Use with :option:`verify`. Also, ``verify_pattern`` supports %o
2454 format, which means that for each block offset will be written and then
2455 verified back, e.g.::
2459 Or use combination of everything::
2461 verify_pattern=0xff%o"abcd"-12
2463 .. option:: verify_fatal=bool
2465 Normally fio will keep checking the entire contents before quitting on a
2466 block verification failure. If this option is set, fio will exit the job on
2467 the first observed failure. Default: false.
2469 .. option:: verify_dump=bool
2471 If set, dump the contents of both the original data block and the data block
2472 we read off disk to files. This allows later analysis to inspect just what
2473 kind of data corruption occurred. Off by default.
2475 .. option:: verify_async=int
2477 Fio will normally verify I/O inline from the submitting thread. This option
2478 takes an integer describing how many async offload threads to create for I/O
2479 verification instead, causing fio to offload the duty of verifying I/O
2480 contents to one or more separate threads. If using this offload option, even
2481 sync I/O engines can benefit from using an :option:`iodepth` setting higher
2482 than 1, as it allows them to have I/O in flight while verifies are running.
2483 Defaults to 0 async threads, i.e. verification is not asynchronous.
2485 .. option:: verify_async_cpus=str
2487 Tell fio to set the given CPU affinity on the async I/O verification
2488 threads. See :option:`cpus_allowed` for the format used.
2490 .. option:: verify_backlog=int
2492 Fio will normally verify the written contents of a job that utilizes verify
2493 once that job has completed. In other words, everything is written then
2494 everything is read back and verified. You may want to verify continually
2495 instead for a variety of reasons. Fio stores the meta data associated with
2496 an I/O block in memory, so for large verify workloads, quite a bit of memory
2497 would be used up holding this meta data. If this option is enabled, fio will
2498 write only N blocks before verifying these blocks.
2500 .. option:: verify_backlog_batch=int
2502 Control how many blocks fio will verify if :option:`verify_backlog` is
2503 set. If not set, will default to the value of :option:`verify_backlog`
2504 (meaning the entire queue is read back and verified). If
2505 ``verify_backlog_batch`` is less than :option:`verify_backlog` then not all
2506 blocks will be verified, if ``verify_backlog_batch`` is larger than
2507 :option:`verify_backlog`, some blocks will be verified more than once.
2509 .. option:: verify_state_save=bool
2511 When a job exits during the write phase of a verify workload, save its
2512 current state. This allows fio to replay up until that point, if the verify
2513 state is loaded for the verify read phase. The format of the filename is,
2516 <type>-<jobname>-<jobindex>-verify.state.
2518 <type> is "local" for a local run, "sock" for a client/server socket
2519 connection, and "ip" (192.168.0.1, for instance) for a networked
2520 client/server connection. Defaults to true.
2522 .. option:: verify_state_load=bool
2524 If a verify termination trigger was used, fio stores the current write state
2525 of each thread. This can be used at verification time so that fio knows how
2526 far it should verify. Without this information, fio will run a full
2527 verification pass, according to the settings in the job file used. Default
2530 .. option:: trim_percentage=int
2532 Number of verify blocks to discard/trim.
2534 .. option:: trim_verify_zero=bool
2536 Verify that trim/discarded blocks are returned as zeroes.
2538 .. option:: trim_backlog=int
2540 Verify that trim/discarded blocks are returned as zeroes.
2542 .. option:: trim_backlog_batch=int
2544 Trim this number of I/O blocks.
2546 .. option:: experimental_verify=bool
2548 Enable experimental verification.
2554 .. option:: steadystate=str:float, ss=str:float
2556 Define the criterion and limit for assessing steady state performance. The
2557 first parameter designates the criterion whereas the second parameter sets
2558 the threshold. When the criterion falls below the threshold for the
2559 specified duration, the job will stop. For example, `iops_slope:0.1%` will
2560 direct fio to terminate the job when the least squares regression slope
2561 falls below 0.1% of the mean IOPS. If :option:`group_reporting` is enabled
2562 this will apply to all jobs in the group. Below is the list of available
2563 steady state assessment criteria. All assessments are carried out using only
2564 data from the rolling collection window. Threshold limits can be expressed
2565 as a fixed value or as a percentage of the mean in the collection window.
2568 Collect IOPS data. Stop the job if all individual IOPS measurements
2569 are within the specified limit of the mean IOPS (e.g., ``iops:2``
2570 means that all individual IOPS values must be within 2 of the mean,
2571 whereas ``iops:0.2%`` means that all individual IOPS values must be
2572 within 0.2% of the mean IOPS to terminate the job).
2575 Collect IOPS data and calculate the least squares regression
2576 slope. Stop the job if the slope falls below the specified limit.
2579 Collect bandwidth data. Stop the job if all individual bandwidth
2580 measurements are within the specified limit of the mean bandwidth.
2583 Collect bandwidth data and calculate the least squares regression
2584 slope. Stop the job if the slope falls below the specified limit.
2586 .. option:: steadystate_duration=time, ss_dur=time
2588 A rolling window of this duration will be used to judge whether steady state
2589 has been reached. Data will be collected once per second. The default is 0
2590 which disables steady state detection. When the unit is omitted, the
2591 value is interpreted in seconds.
2593 .. option:: steadystate_ramp_time=time, ss_ramp=time
2595 Allow the job to run for the specified duration before beginning data
2596 collection for checking the steady state job termination criterion. The
2597 default is 0. When the unit is omitted, the value is interpreted in seconds.
2600 Measurements and reporting
2601 ~~~~~~~~~~~~~~~~~~~~~~~~~~
2603 .. option:: per_job_logs=bool
2605 If set, this generates bw/clat/iops log with per file private filenames. If
2606 not set, jobs with identical names will share the log filename. Default:
2609 .. option:: group_reporting
2611 It may sometimes be interesting to display statistics for groups of jobs as
2612 a whole instead of for each individual job. This is especially true if
2613 :option:`numjobs` is used; looking at individual thread/process output
2614 quickly becomes unwieldy. To see the final report per-group instead of
2615 per-job, use :option:`group_reporting`. Jobs in a file will be part of the
2616 same reporting group, unless if separated by a :option:`stonewall`, or by
2617 using :option:`new_group`.
2619 .. option:: new_group
2621 Start a new reporting group. See: :option:`group_reporting`. If not given,
2622 all jobs in a file will be part of the same reporting group, unless
2623 separated by a :option:`stonewall`.
2627 By default, fio collects and shows final output results for all jobs
2628 that run. If this option is set to 0, then fio will ignore it in
2629 the final stat output.
2631 .. option:: write_bw_log=str
2633 If given, write a bandwidth log for this job. Can be used to store data of
2634 the bandwidth of the jobs in their lifetime. The included
2635 :command:`fio_generate_plots` script uses :command:`gnuplot` to turn these
2636 text files into nice graphs. See :option:`write_lat_log` for behaviour of
2637 given filename. For this option, the postfix is :file:`_bw.x.log`, where `x`
2638 is the index of the job (`1..N`, where `N` is the number of jobs). If
2639 :option:`per_job_logs` is false, then the filename will not include the job
2640 index. See `Log File Formats`_.
2642 .. option:: write_lat_log=str
2644 Same as :option:`write_bw_log`, except that this option stores I/O
2645 submission, completion, and total latencies instead. If no filename is given
2646 with this option, the default filename of :file:`jobname_type.log` is
2647 used. Even if the filename is given, fio will still append the type of
2648 log. So if one specifies::
2652 The actual log names will be :file:`foo_slat.x.log`, :file:`foo_clat.x.log`,
2653 and :file:`foo_lat.x.log`, where `x` is the index of the job (1..N, where N
2654 is the number of jobs). This helps :command:`fio_generate_plot` find the
2655 logs automatically. If :option:`per_job_logs` is false, then the filename
2656 will not include the job index. See `Log File Formats`_.
2658 .. option:: write_hist_log=str
2660 Same as :option:`write_lat_log`, but writes I/O completion latency
2661 histograms. If no filename is given with this option, the default filename
2662 of :file:`jobname_clat_hist.x.log` is used, where `x` is the index of the
2663 job (1..N, where `N` is the number of jobs). Even if the filename is given,
2664 fio will still append the type of log. If :option:`per_job_logs` is false,
2665 then the filename will not include the job index. See `Log File Formats`_.
2667 .. option:: write_iops_log=str
2669 Same as :option:`write_bw_log`, but writes IOPS. If no filename is given
2670 with this option, the default filename of :file:`jobname_type.x.log` is
2671 used,where `x` is the index of the job (1..N, where `N` is the number of
2672 jobs). Even if the filename is given, fio will still append the type of
2673 log. If :option:`per_job_logs` is false, then the filename will not include
2674 the job index. See `Log File Formats`_.
2676 .. option:: log_avg_msec=int
2678 By default, fio will log an entry in the iops, latency, or bw log for every
2679 I/O that completes. When writing to the disk log, that can quickly grow to a
2680 very large size. Setting this option makes fio average the each log entry
2681 over the specified period of time, reducing the resolution of the log. See
2682 :option:`log_max_value` as well. Defaults to 0, logging all entries.
2684 .. option:: log_hist_msec=int
2686 Same as :option:`log_avg_msec`, but logs entries for completion latency
2687 histograms. Computing latency percentiles from averages of intervals using
2688 :option:`log_avg_msec` is inaccurate. Setting this option makes fio log
2689 histogram entries over the specified period of time, reducing log sizes for
2690 high IOPS devices while retaining percentile accuracy. See
2691 :option:`log_hist_coarseness` as well. Defaults to 0, meaning histogram
2692 logging is disabled.
2694 .. option:: log_hist_coarseness=int
2696 Integer ranging from 0 to 6, defining the coarseness of the resolution of
2697 the histogram logs enabled with :option:`log_hist_msec`. For each increment
2698 in coarseness, fio outputs half as many bins. Defaults to 0, for which
2699 histogram logs contain 1216 latency bins. See `Log File Formats`_.
2701 .. option:: log_max_value=bool
2703 If :option:`log_avg_msec` is set, fio logs the average over that window. If
2704 you instead want to log the maximum value, set this option to 1. Defaults to
2705 0, meaning that averaged values are logged.
2707 .. option:: log_offset=int
2709 If this is set, the iolog options will include the byte offset for the I/O
2710 entry as well as the other data values.
2712 .. option:: log_compression=int
2714 If this is set, fio will compress the I/O logs as it goes, to keep the
2715 memory footprint lower. When a log reaches the specified size, that chunk is
2716 removed and compressed in the background. Given that I/O logs are fairly
2717 highly compressible, this yields a nice memory savings for longer runs. The
2718 downside is that the compression will consume some background CPU cycles, so
2719 it may impact the run. This, however, is also true if the logging ends up
2720 consuming most of the system memory. So pick your poison. The I/O logs are
2721 saved normally at the end of a run, by decompressing the chunks and storing
2722 them in the specified log file. This feature depends on the availability of
2725 .. option:: log_compression_cpus=str
2727 Define the set of CPUs that are allowed to handle online log compression for
2728 the I/O jobs. This can provide better isolation between performance
2729 sensitive jobs, and background compression work.
2731 .. option:: log_store_compressed=bool
2733 If set, fio will store the log files in a compressed format. They can be
2734 decompressed with fio, using the :option:`--inflate-log` command line
2735 parameter. The files will be stored with a :file:`.fz` suffix.
2737 .. option:: log_unix_epoch=bool
2739 If set, fio will log Unix timestamps to the log files produced by enabling
2740 write_type_log for each log type, instead of the default zero-based
2743 .. option:: block_error_percentiles=bool
2745 If set, record errors in trim block-sized units from writes and trims and
2746 output a histogram of how many trims it took to get to errors, and what kind
2747 of error was encountered.
2749 .. option:: bwavgtime=int
2751 Average the calculated bandwidth over the given time. Value is specified in
2752 milliseconds. If the job also does bandwidth logging through
2753 :option:`write_bw_log`, then the minimum of this option and
2754 :option:`log_avg_msec` will be used. Default: 500ms.
2756 .. option:: iopsavgtime=int
2758 Average the calculated IOPS over the given time. Value is specified in
2759 milliseconds. If the job also does IOPS logging through
2760 :option:`write_iops_log`, then the minimum of this option and
2761 :option:`log_avg_msec` will be used. Default: 500ms.
2763 .. option:: disk_util=bool
2765 Generate disk utilization statistics, if the platform supports it.
2768 .. option:: disable_lat=bool
2770 Disable measurements of total latency numbers. Useful only for cutting back
2771 the number of calls to :manpage:`gettimeofday(2)`, as that does impact
2772 performance at really high IOPS rates. Note that to really get rid of a
2773 large amount of these calls, this option must be used with
2774 :option:`disable_slat` and :option:`disable_bw_measurement` as well.
2776 .. option:: disable_clat=bool
2778 Disable measurements of completion latency numbers. See
2779 :option:`disable_lat`.
2781 .. option:: disable_slat=bool
2783 Disable measurements of submission latency numbers. See
2784 :option:`disable_slat`.
2786 .. option:: disable_bw_measurement=bool, disable_bw=bool
2788 Disable measurements of throughput/bandwidth numbers. See
2789 :option:`disable_lat`.
2791 .. option:: clat_percentiles=bool
2793 Enable the reporting of percentiles of completion latencies.
2795 .. option:: percentile_list=float_list
2797 Overwrite the default list of percentiles for completion latencies and the
2798 block error histogram. Each number is a floating number in the range
2799 (0,100], and the maximum length of the list is 20. Use ``:`` to separate the
2800 numbers, and list the numbers in ascending order. For example,
2801 ``--percentile_list=99.5:99.9`` will cause fio to report the values of
2802 completion latency below which 99.5% and 99.9% of the observed latencies
2809 .. option:: exitall_on_error
2811 When one job finishes in error, terminate the rest. The default is to wait
2812 for each job to finish.
2814 .. option:: continue_on_error=str
2816 Normally fio will exit the job on the first observed failure. If this option
2817 is set, fio will continue the job when there is a 'non-fatal error' (EIO or
2818 EILSEQ) until the runtime is exceeded or the I/O size specified is
2819 completed. If this option is used, there are two more stats that are
2820 appended, the total error count and the first error. The error field given
2821 in the stats is the first error that was hit during the run.
2823 The allowed values are:
2826 Exit on any I/O or verify errors.
2829 Continue on read errors, exit on all others.
2832 Continue on write errors, exit on all others.
2835 Continue on any I/O error, exit on all others.
2838 Continue on verify errors, exit on all others.
2841 Continue on all errors.
2844 Backward-compatible alias for 'none'.
2847 Backward-compatible alias for 'all'.
2849 .. option:: ignore_error=str
2851 Sometimes you want to ignore some errors during test in that case you can
2852 specify error list for each error type, instead of only being able to
2853 ignore the default 'non-fatal error' using :option:`continue_on_error`.
2854 ``ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST`` errors for
2855 given error type is separated with ':'. Error may be symbol ('ENOSPC',
2856 'ENOMEM') or integer. Example::
2858 ignore_error=EAGAIN,ENOSPC:122
2860 This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
2861 WRITE. This option works by overriding :option:`continue_on_error` with
2862 the list of errors for each error type if any.
2864 .. option:: error_dump=bool
2866 If set dump every error even if it is non fatal, true by default. If
2867 disabled only fatal error will be dumped.
2869 Running predefined workloads
2870 ----------------------------
2872 Fio includes predefined profiles that mimic the I/O workloads generated by
2875 .. option:: profile=str
2877 The predefined workload to run. Current profiles are:
2880 Threaded I/O bench (tiotest/tiobench) like workload.
2883 Aerospike Certification Tool (ACT) like workload.
2885 To view a profile's additional options use :option:`--cmdhelp` after specifying
2886 the profile. For example::
2888 $ fio --profile=act --cmdhelp
2893 .. option:: device-names=str
2898 .. option:: load=int
2901 ACT load multiplier. Default: 1.
2903 .. option:: test-duration=time
2906 How long the entire test takes to run. When the unit is omitted, the value
2907 is given in seconds. Default: 24h.
2909 .. option:: threads-per-queue=int
2912 Number of read IO threads per device. Default: 8.
2914 .. option:: read-req-num-512-blocks=int
2917 Number of 512B blocks to read at the time. Default: 3.
2919 .. option:: large-block-op-kbytes=int
2922 Size of large block ops in KiB (writes). Default: 131072.
2927 Set to run ACT prep phase.
2929 Tiobench profile options
2930 ~~~~~~~~~~~~~~~~~~~~~~~~
2932 .. option:: size=str
2937 .. option:: block=int
2940 Block size in bytes. Default: 4096.
2942 .. option:: numruns=int
2952 .. option:: threads=int
2957 Interpreting the output
2958 -----------------------
2960 Fio spits out a lot of output. While running, fio will display the status of the
2961 jobs created. An example of that would be::
2963 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]
2965 The characters inside the square brackets denote the current status of each
2966 thread. The possible values (in typical life cycle order) are:
2968 +------+-----+-----------------------------------------------------------+
2970 +======+=====+===========================================================+
2971 | P | | Thread setup, but not started. |
2972 +------+-----+-----------------------------------------------------------+
2973 | C | | Thread created. |
2974 +------+-----+-----------------------------------------------------------+
2975 | I | | Thread initialized, waiting or generating necessary data. |
2976 +------+-----+-----------------------------------------------------------+
2977 | | p | Thread running pre-reading file(s). |
2978 +------+-----+-----------------------------------------------------------+
2979 | | R | Running, doing sequential reads. |
2980 +------+-----+-----------------------------------------------------------+
2981 | | r | Running, doing random reads. |
2982 +------+-----+-----------------------------------------------------------+
2983 | | W | Running, doing sequential writes. |
2984 +------+-----+-----------------------------------------------------------+
2985 | | w | Running, doing random writes. |
2986 +------+-----+-----------------------------------------------------------+
2987 | | M | Running, doing mixed sequential reads/writes. |
2988 +------+-----+-----------------------------------------------------------+
2989 | | m | Running, doing mixed random reads/writes. |
2990 +------+-----+-----------------------------------------------------------+
2991 | | F | Running, currently waiting for :manpage:`fsync(2)` |
2992 +------+-----+-----------------------------------------------------------+
2993 | | V | Running, doing verification of written data. |
2994 +------+-----+-----------------------------------------------------------+
2995 | E | | Thread exited, not reaped by main thread yet. |
2996 +------+-----+-----------------------------------------------------------+
2997 | _ | | Thread reaped, or |
2998 +------+-----+-----------------------------------------------------------+
2999 | X | | Thread reaped, exited with an error. |
3000 +------+-----+-----------------------------------------------------------+
3001 | K | | Thread reaped, exited due to signal. |
3002 +------+-----+-----------------------------------------------------------+
3004 Fio will condense the thread string as not to take up more space on the command
3005 line as is needed. For instance, if you have 10 readers and 10 writers running,
3006 the output would look like this::
3008 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]
3010 Fio will still maintain the ordering, though. So the above means that jobs 1..10
3011 are readers, and 11..20 are writers.
3013 The other values are fairly self explanatory -- number of threads currently
3014 running and doing I/O, the number of currently open files (f=), the rate of I/O
3015 since last check (read speed listed first, then write speed and optionally trim
3016 speed), and the estimated completion percentage and time for the current
3017 running group. It's impossible to estimate runtime of the following groups (if
3018 any). Note that the string is displayed in order, so it's possible to tell which
3019 of the jobs are currently doing what. The first character is the first job
3020 defined in the job file, and so forth.
3022 When fio is done (or interrupted by :kbd:`ctrl-c`), it will show the data for
3023 each thread, group of threads, and disks in that order. For each data direction,
3024 the output looks like::
3026 Client1 (g=0): err= 0:
3027 write: io= 32MiB, bw= 666KiB/s, iops=89 , runt= 50320msec
3028 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
3029 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
3030 bw (KiB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
3031 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
3032 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
3033 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3034 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3035 issued r/w: total=0/32768, short=0/0
3036 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
3037 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
3039 The client number is printed, along with the group id and error of that
3040 thread. Below is the I/O statistics, here for writes. In the order listed, they
3044 Number of megabytes I/O performed.
3047 Average bandwidth rate.
3050 Average I/Os performed per second.
3053 The runtime of that thread.
3056 Submission latency (avg being the average, stdev being the standard
3057 deviation). This is the time it took to submit the I/O. For sync I/O,
3058 the slat is really the completion latency, since queue/complete is one
3059 operation there. This value can be in milliseconds or microseconds, fio
3060 will choose the most appropriate base and print that. In the example
3061 above, milliseconds is the best scale. Note: in :option:`--minimal` mode
3062 latencies are always expressed in microseconds.
3065 Completion latency. Same names as slat, this denotes the time from
3066 submission to completion of the I/O pieces. For sync I/O, clat will
3067 usually be equal (or very close) to 0, as the time from submit to
3068 complete is basically just CPU time (I/O has already been done, see slat
3072 Bandwidth. Same names as the xlat stats, but also includes an
3073 approximate percentage of total aggregate bandwidth this thread received
3074 in this group. This last value is only really useful if the threads in
3075 this group are on the same disk, since they are then competing for disk
3079 CPU usage. User and system time, along with the number of context
3080 switches this thread went through, usage of system and user time, and
3081 finally the number of major and minor page faults. The CPU utilization
3082 numbers are averages for the jobs in that reporting group, while the
3083 context and fault counters are summed.
3086 The distribution of I/O depths over the job life time. The numbers are
3087 divided into powers of 2, so for example the 16= entries includes depths
3088 up to that value but higher than the previous entry. In other words, it
3089 covers the range from 16 to 31.
3092 How many pieces of I/O were submitting in a single submit call. Each
3093 entry denotes that amount and below, until the previous entry -- e.g.,
3094 8=100% mean that we submitted anywhere in between 5-8 I/Os per submit
3098 Like the above submit number, but for completions instead.
3101 The number of read/write requests issued, and how many of them were
3105 The distribution of I/O completion latencies. This is the time from when
3106 I/O leaves fio and when it gets completed. The numbers follow the same
3107 pattern as the I/O depths, meaning that 2=1.6% means that 1.6% of the
3108 I/O completed within 2 msecs, 20=12.8% means that 12.8% of the I/O took
3109 more than 10 msecs, but less than (or equal to) 20 msecs.
3111 After each client has been listed, the group statistics are printed. They
3112 will look like this::
3114 Run status group 0 (all jobs):
3115 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
3116 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
3118 For each data direction, it prints:
3121 Number of megabytes I/O performed.
3123 Aggregate bandwidth of threads in this group.
3125 The minimum average bandwidth a thread saw.
3127 The maximum average bandwidth a thread saw.
3129 The smallest runtime of the threads in that group.
3131 The longest runtime of the threads in that group.
3133 And finally, the disk statistics are printed. They will look like this::
3135 Disk stats (read/write):
3136 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3138 Each value is printed for both reads and writes, with reads first. The
3142 Number of I/Os performed by all groups.
3144 Number of merges I/O the I/O scheduler.
3146 Number of ticks we kept the disk busy.
3148 Total time spent in the disk queue.
3150 The disk utilization. A value of 100% means we kept the disk
3151 busy constantly, 50% would be a disk idling half of the time.
3153 It is also possible to get fio to dump the current output while it is running,
3154 without terminating the job. To do that, send fio the **USR1** signal. You can
3155 also get regularly timed dumps by using the :option:`--status-interval`
3156 parameter, or by creating a file in :file:`/tmp` named
3157 :file:`fio-dump-status`. If fio sees this file, it will unlink it and dump the
3158 current output status.
3164 For scripted usage where you typically want to generate tables or graphs of the
3165 results, fio can output the results in a semicolon separated format. The format
3166 is one long line of values, such as::
3168 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%
3169 A description of this job goes here.
3171 The job description (if provided) follows on a second line.
3173 To enable terse output, use the :option:`--minimal` command line option. The
3174 first value is the version of the terse output format. If the output has to be
3175 changed for some reason, this number will be incremented by 1 to signify that
3178 Split up, the format is as follows (comments in brackets denote when a
3179 field was introduced or whether its specific to some terse version):
3183 terse version, fio version [v3], jobname, groupid, error
3187 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3188 Submission latency: min, max, mean, stdev (usec)
3189 Completion latency: min, max, mean, stdev (usec)
3190 Completion latency percentiles: 20 fields (see below)
3191 Total latency: min, max, mean, stdev (usec)
3192 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3193 IOPS [v5]: min, max, mean, stdev, number of samples
3199 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3200 Submission latency: min, max, mean, stdev (usec)
3201 Completion latency: min, max, mean, stdev (usec)
3202 Completion latency percentiles: 20 fields (see below)
3203 Total latency: min, max, mean, stdev (usec)
3204 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3205 IOPS [v5]: min, max, mean, stdev, number of samples
3207 TRIM status [all but version 3]:
3209 Fields are similar to READ/WRITE status.
3213 user, system, context switches, major faults, minor faults
3217 <=1, 2, 4, 8, 16, 32, >=64
3219 I/O latencies microseconds::
3221 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3223 I/O latencies milliseconds::
3225 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3227 Disk utilization [v3]::
3229 Disk name, Read ios, write ios,
3230 Read merges, write merges,
3231 Read ticks, write ticks,
3232 Time spent in queue, disk utilization percentage
3234 Additional Info (dependent on continue_on_error, default off)::
3236 total # errors, first error code
3238 Additional Info (dependent on description being set)::
3242 Completion latency percentiles can be a grouping of up to 20 sets, so for the
3243 terse output fio writes all of them. Each field will look like this::
3247 which is the Xth percentile, and the `usec` latency associated with it.
3249 For disk utilization, all disks used by fio are shown. So for each disk there
3250 will be a disk utilization section.
3252 Below is a single line containing short names for each of the fields in the
3253 minimal output v3, separated by semicolons:
3255 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_max;read_clat_min;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_max;write_clat_min;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;pu_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
3261 There are two trace file format that you can encounter. The older (v1) format is
3262 unsupported since version 1.20-rc3 (March 2008). It will still be described
3263 below in case that you get an old trace and want to understand it.
3265 In any case the trace is a simple text file with a single action per line.
3268 Trace file format v1
3269 ~~~~~~~~~~~~~~~~~~~~
3271 Each line represents a single I/O action in the following format::
3275 where `rw=0/1` for read/write, and the offset and length entries being in bytes.
3277 This format is not supported in fio versions => 1.20-rc3.
3280 Trace file format v2
3281 ~~~~~~~~~~~~~~~~~~~~
3283 The second version of the trace file format was added in fio version 1.17. It
3284 allows to access more then one file per trace and has a bigger set of possible
3287 The first line of the trace file has to be::
3291 Following this can be lines in two different formats, which are described below.
3293 The file management format::
3297 The filename is given as an absolute path. The action can be one of these:
3300 Add the given filename to the trace.
3302 Open the file with the given filename. The filename has to have
3303 been added with the **add** action before.
3305 Close the file with the given filename. The file has to have been
3309 The file I/O action format::
3311 filename action offset length
3313 The `filename` is given as an absolute path, and has to have been added and
3314 opened before it can be used with this format. The `offset` and `length` are
3315 given in bytes. The `action` can be one of these:
3318 Wait for `offset` microseconds. Everything below 100 is discarded.
3319 The time is relative to the previous `wait` statement.
3321 Read `length` bytes beginning from `offset`.
3323 Write `length` bytes beginning from `offset`.
3325 :manpage:`fsync(2)` the file.
3327 :manpage:`fdatasync(2)` the file.
3329 Trim the given file from the given `offset` for `length` bytes.
3331 CPU idleness profiling
3332 ----------------------
3334 In some cases, we want to understand CPU overhead in a test. For example, we
3335 test patches for the specific goodness of whether they reduce CPU usage.
3336 Fio implements a balloon approach to create a thread per CPU that runs at idle
3337 priority, meaning that it only runs when nobody else needs the cpu.
3338 By measuring the amount of work completed by the thread, idleness of each CPU
3339 can be derived accordingly.
3341 An unit work is defined as touching a full page of unsigned characters. Mean and
3342 standard deviation of time to complete an unit work is reported in "unit work"
3343 section. Options can be chosen to report detailed percpu idleness or overall
3344 system idleness by aggregating percpu stats.
3347 Verification and triggers
3348 -------------------------
3350 Fio is usually run in one of two ways, when data verification is done. The first
3351 is a normal write job of some sort with verify enabled. When the write phase has
3352 completed, fio switches to reads and verifies everything it wrote. The second
3353 model is running just the write phase, and then later on running the same job
3354 (but with reads instead of writes) to repeat the same I/O patterns and verify
3355 the contents. Both of these methods depend on the write phase being completed,
3356 as fio otherwise has no idea how much data was written.
3358 With verification triggers, fio supports dumping the current write state to
3359 local files. Then a subsequent read verify workload can load this state and know
3360 exactly where to stop. This is useful for testing cases where power is cut to a
3361 server in a managed fashion, for instance.
3363 A verification trigger consists of two things:
3365 1) Storing the write state of each job.
3366 2) Executing a trigger command.
3368 The write state is relatively small, on the order of hundreds of bytes to single
3369 kilobytes. It contains information on the number of completions done, the last X
3372 A trigger is invoked either through creation ('touch') of a specified file in
3373 the system, or through a timeout setting. If fio is run with
3374 :option:`--trigger-file`\= :file:`/tmp/trigger-file`, then it will continually
3375 check for the existence of :file:`/tmp/trigger-file`. When it sees this file, it
3376 will fire off the trigger (thus saving state, and executing the trigger
3379 For client/server runs, there's both a local and remote trigger. If fio is
3380 running as a server backend, it will send the job states back to the client for
3381 safe storage, then execute the remote trigger, if specified. If a local trigger
3382 is specified, the server will still send back the write state, but the client
3383 will then execute the trigger.
3385 Verification trigger example
3386 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3388 Let's say we want to run a powercut test on the remote machine 'server'. Our
3389 write workload is in :file:`write-test.fio`. We want to cut power to 'server' at
3390 some point during the run, and we'll run this test from the safety or our local
3391 machine, 'localbox'. On the server, we'll start the fio backend normally::
3393 server# fio --server
3395 and on the client, we'll fire off the workload::
3397 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
3399 We set :file:`/tmp/my-trigger` as the trigger file, and we tell fio to execute::
3401 echo b > /proc/sysrq-trigger
3403 on the server once it has received the trigger and sent us the write state. This
3404 will work, but it's not **really** cutting power to the server, it's merely
3405 abruptly rebooting it. If we have a remote way of cutting power to the server
3406 through IPMI or similar, we could do that through a local trigger command
3407 instead. Let's assume we have a script that does IPMI reboot of a given hostname,
3408 ipmi-reboot. On localbox, we could then have run fio with a local trigger
3411 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
3413 For this case, fio would wait for the server to send us the write state, then
3414 execute ``ipmi-reboot server`` when that happened.
3416 Loading verify state
3417 ~~~~~~~~~~~~~~~~~~~~
3419 To load stored write state, a read verification job file must contain the
3420 :option:`verify_state_load` option. If that is set, fio will load the previously
3421 stored state. For a local fio run this is done by loading the files directly,
3422 and on a client/server run, the server backend will ask the client to send the
3423 files over and load them from there.
3429 Fio supports a variety of log file formats, for logging latencies, bandwidth,
3430 and IOPS. The logs share a common format, which looks like this:
3432 *time* (`msec`), *value*, *data direction*, *offset*
3434 Time for the log entry is always in milliseconds. The *value* logged depends
3435 on the type of log, it will be one of the following:
3438 Value is latency in usecs
3444 *Data direction* is one of the following:
3453 The *offset* is the offset, in bytes, from the start of the file, for that
3454 particular I/O. The logging of the offset can be toggled with
3455 :option:`log_offset`.
3457 If windowed logging is enabled through :option:`log_avg_msec` then fio doesn't
3458 log individual I/Os. Instead of logs the average values over the specified period
3459 of time. Since 'data direction' and 'offset' are per-I/O values, they aren't
3460 applicable if windowed logging is enabled. If windowed logging is enabled and
3461 :option:`log_max_value` is set, then fio logs maximum values in that window
3462 instead of averages.
3468 Normally fio is invoked as a stand-alone application on the machine where the
3469 I/O workload should be generated. However, the frontend and backend of fio can
3470 be run separately. Ie the fio server can generate an I/O workload on the "Device
3471 Under Test" while being controlled from another machine.
3473 Start the server on the machine which has access to the storage DUT::
3477 where args defines what fio listens to. The arguments are of the form
3478 ``type,hostname`` or ``IP,port``. *type* is either ``ip`` (or ip4) for TCP/IP
3479 v4, ``ip6`` for TCP/IP v6, or ``sock`` for a local unix domain socket.
3480 *hostname* is either a hostname or IP address, and *port* is the port to listen
3481 to (only valid for TCP/IP, not a local socket). Some examples:
3485 Start a fio server, listening on all interfaces on the default port (8765).
3487 2) ``fio --server=ip:hostname,4444``
3489 Start a fio server, listening on IP belonging to hostname and on port 4444.
3491 3) ``fio --server=ip6:::1,4444``
3493 Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
3495 4) ``fio --server=,4444``
3497 Start a fio server, listening on all interfaces on port 4444.
3499 5) ``fio --server=1.2.3.4``
3501 Start a fio server, listening on IP 1.2.3.4 on the default port.
3503 6) ``fio --server=sock:/tmp/fio.sock``
3505 Start a fio server, listening on the local socket /tmp/fio.sock.
3507 Once a server is running, a "client" can connect to the fio server with::
3509 fio <local-args> --client=<server> <remote-args> <job file(s)>
3511 where `local-args` are arguments for the client where it is running, `server`
3512 is the connect string, and `remote-args` and `job file(s)` are sent to the
3513 server. The `server` string follows the same format as it does on the server
3514 side, to allow IP/hostname/socket and port strings.
3516 Fio can connect to multiple servers this way::
3518 fio --client=<server1> <job file(s)> --client=<server2> <job file(s)>
3520 If the job file is located on the fio server, then you can tell the server to
3521 load a local file as well. This is done by using :option:`--remote-config` ::
3523 fio --client=server --remote-config /path/to/file.fio
3525 Then fio will open this local (to the server) job file instead of being passed
3526 one from the client.
3528 If you have many servers (example: 100 VMs/containers), you can input a pathname
3529 of a file containing host IPs/names as the parameter value for the
3530 :option:`--client` option. For example, here is an example :file:`host.list`
3531 file containing 2 hostnames::
3533 host1.your.dns.domain
3534 host2.your.dns.domain
3536 The fio command would then be::
3538 fio --client=host.list <job file(s)>
3540 In this mode, you cannot input server-specific parameters or job files -- all
3541 servers receive the same job file.
3543 In order to let ``fio --client`` runs use a shared filesystem from multiple
3544 hosts, ``fio --client`` now prepends the IP address of the server to the
3545 filename. For example, if fio is using the directory :file:`/mnt/nfs/fio` and is
3546 writing filename :file:`fileio.tmp`, with a :option:`--client` `hostfile`
3547 containing two hostnames ``h1`` and ``h2`` with IP addresses 192.168.10.120 and
3548 192.168.10.121, then fio will create two files::
3550 /mnt/nfs/fio/192.168.10.120.fileio.tmp
3551 /mnt/nfs/fio/192.168.10.121.fileio.tmp