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, separate 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).
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.
182 .. option:: --status-interval=time
184 Force full status dump every `time` period passed.
186 .. option:: --section=name
188 Only run specified section in job file. Multiple sections can be specified.
189 The ``--section`` option allows one to combine related jobs into one file.
190 E.g. one job file could define light, moderate, and heavy sections. Tell
191 fio to run only the "heavy" section by giving ``--section=heavy``
192 command line option. One can also specify the "write" operations in one
193 section and "verify" operation in another section. The ``--section`` option
194 only applies to job sections. The reserved *global* section is always
197 .. option:: --alloc-size=kb
199 Set the internal smalloc pool to this size in kb (def 1024). The
200 ``--alloc-size`` switch allows one to use a larger pool size for smalloc.
201 If running large jobs with randommap enabled, fio can run out of memory.
202 Smalloc is an internal allocator for shared structures from a fixed size
203 memory pool. The pool size defaults to 16M and can grow to 8 pools.
205 NOTE: While running :file:`.fio_smalloc.*` backing store files are visible
208 .. option:: --warnings-fatal
210 All fio parser warnings are fatal, causing fio to exit with an
213 .. option:: --max-jobs=nr
215 Maximum number of threads/processes to support.
217 .. option:: --server=args
219 Start a backend server, with `args` specifying what to listen to.
220 See `Client/Server`_ section.
222 .. option:: --daemonize=pidfile
224 Background a fio server, writing the pid to the given `pidfile` file.
226 .. option:: --client=hostname
228 Instead of running the jobs locally, send and run them on the given host or
229 set of hosts. See `Client/Server`_ section.
231 .. option:: --remote-config=file
233 Tell fio server to load this local file.
235 .. option:: --idle-prof=option
237 Report cpu idleness on a system or percpu basis
238 ``--idle-prof=system,percpu`` or
239 run unit work calibration only ``--idle-prof=calibrate``.
241 .. option:: --inflate-log=log
243 Inflate and output compressed log.
245 .. option:: --trigger-file=file
247 Execute trigger cmd when file exists.
249 .. option:: --trigger-timeout=t
251 Execute trigger at this time.
253 .. option:: --trigger=cmd
255 Set this command as local trigger.
257 .. option:: --trigger-remote=cmd
259 Set this command as remote trigger.
261 .. option:: --aux-path=path
263 Use this path for fio state generated files.
265 Any parameters following the options will be assumed to be job files, unless
266 they match a job file parameter. Multiple job files can be listed and each job
267 file will be regarded as a separate group. Fio will :option:`stonewall`
268 execution between each group.
274 As previously described, fio accepts one or more job files describing what it is
275 supposed to do. The job file format is the classic ini file, where the names
276 enclosed in [] brackets define the job name. You are free to use any ASCII name
277 you want, except *global* which has special meaning. Following the job name is
278 a sequence of zero or more parameters, one per line, that define the behavior of
279 the job. If the first character in a line is a ';' or a '#', the entire line is
280 discarded as a comment.
282 A *global* section sets defaults for the jobs described in that file. A job may
283 override a *global* section parameter, and a job file may even have several
284 *global* sections if so desired. A job is only affected by a *global* section
287 The :option:`--cmdhelp` option also lists all options. If used with an `option`
288 argument, :option:`--cmdhelp` will detail the given `option`.
290 See the `examples/` directory for inspiration on how to write job files. Note
291 the copyright and license requirements currently apply to `examples/` files.
293 So let's look at a really simple job file that defines two processes, each
294 randomly reading from a 128MiB file:
298 ; -- start job file --
309 As you can see, the job file sections themselves are empty as all the described
310 parameters are shared. As no :option:`filename` option is given, fio makes up a
311 `filename` for each of the jobs as it sees fit. On the command line, this job
312 would look as follows::
314 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
317 Let's look at an example that has a number of processes writing randomly to
322 ; -- start job file --
333 Here we have no *global* section, as we only have one job defined anyway. We
334 want to use async I/O here, with a depth of 4 for each file. We also increased
335 the buffer size used to 32KiB and define numjobs to 4 to fork 4 identical
336 jobs. The result is 4 processes each randomly writing to their own 64MiB
337 file. Instead of using the above job file, you could have given the parameters
338 on the command line. For this case, you would specify::
340 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
342 When fio is utilized as a basis of any reasonably large test suite, it might be
343 desirable to share a set of standardized settings across multiple job files.
344 Instead of copy/pasting such settings, any section may pull in an external
345 :file:`filename.fio` file with *include filename* directive, as in the following
348 ; -- start job file including.fio --
352 include glob-include.fio
359 include test-include.fio
360 ; -- end job file including.fio --
364 ; -- start job file glob-include.fio --
367 ; -- end job file glob-include.fio --
371 ; -- start job file test-include.fio --
374 ; -- end job file test-include.fio --
376 Settings pulled into a section apply to that section only (except *global*
377 section). Include directives may be nested in that any included file may contain
378 further include directive(s). Include files may not contain [] sections.
381 Environment variables
382 ~~~~~~~~~~~~~~~~~~~~~
384 Fio also supports environment variable expansion in job files. Any sub-string of
385 the form ``${VARNAME}`` as part of an option value (in other words, on the right
386 of the '='), will be expanded to the value of the environment variable called
387 `VARNAME`. If no such environment variable is defined, or `VARNAME` is the
388 empty string, the empty string will be substituted.
390 As an example, let's look at a sample fio invocation and job file::
392 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
396 ; -- start job file --
403 This will expand to the following equivalent job file at runtime:
407 ; -- start job file --
414 Fio ships with a few example job files, you can also look there for inspiration.
419 Additionally, fio has a set of reserved keywords that will be replaced
420 internally with the appropriate value. Those keywords are:
424 The architecture page size of the running system.
428 Megabytes of total memory in the system.
432 Number of online available CPUs.
434 These can be used on the command line or in the job file, and will be
435 automatically substituted with the current system values when the job is
436 run. Simple math is also supported on these keywords, so you can perform actions
441 and get that properly expanded to 8 times the size of memory in the machine.
447 This section describes in details each parameter associated with a job. Some
448 parameters take an option of a given type, such as an integer or a
449 string. Anywhere a numeric value is required, an arithmetic expression may be
450 used, provided it is surrounded by parentheses. Supported operators are:
459 For time values in expressions, units are microseconds by default. This is
460 different than for time values not in expressions (not enclosed in
461 parentheses). The following types are used:
468 String. This is a sequence of alpha characters.
471 Integer with possible time suffix. In seconds unless otherwise
472 specified, use e.g. 10m for 10 minutes. Accepts s/m/h for seconds, minutes,
473 and hours, and accepts 'ms' (or 'msec') for milliseconds, and 'us' (or
474 'usec') for microseconds.
479 Integer. A whole number value, which may contain an integer prefix
480 and an integer suffix:
482 [*integer prefix*] **number** [*integer suffix*]
484 The optional *integer prefix* specifies the number's base. The default
485 is decimal. *0x* specifies hexadecimal.
487 The optional *integer suffix* specifies the number's units, and includes an
488 optional unit prefix and an optional unit. For quantities of data, the
489 default unit is bytes. For quantities of time, the default unit is seconds.
491 With :option:`kb_base` =1000, fio follows international standards for unit
492 prefixes. To specify power-of-10 decimal values defined in the
493 International System of Units (SI):
495 * *Ki* -- means kilo (K) or 1000
496 * *Mi* -- means mega (M) or 1000**2
497 * *Gi* -- means giga (G) or 1000**3
498 * *Ti* -- means tera (T) or 1000**4
499 * *Pi* -- means peta (P) or 1000**5
501 To specify power-of-2 binary values defined in IEC 80000-13:
503 * *k* -- means kibi (Ki) or 1024
504 * *M* -- means mebi (Mi) or 1024**2
505 * *G* -- means gibi (Gi) or 1024**3
506 * *T* -- means tebi (Ti) or 1024**4
507 * *P* -- means pebi (Pi) or 1024**5
509 With :option:`kb_base` =1024 (the default), the unit prefixes are opposite
510 from those specified in the SI and IEC 80000-13 standards to provide
511 compatibility with old scripts. For example, 4k means 4096.
513 For quantities of data, an optional unit of 'B' may be included
514 (e.g., 'kB' is the same as 'k').
516 The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
517 not milli). 'b' and 'B' both mean byte, not bit.
519 Examples with :option:`kb_base` =1000:
521 * *4 KiB*: 4096, 4096b, 4096B, 4ki, 4kib, 4kiB, 4Ki, 4KiB
522 * *1 MiB*: 1048576, 1mi, 1024ki
523 * *1 MB*: 1000000, 1m, 1000k
524 * *1 TiB*: 1099511627776, 1ti, 1024gi, 1048576mi
525 * *1 TB*: 1000000000, 1t, 1000m, 1000000k
527 Examples with :option:`kb_base` =1024 (default):
529 * *4 KiB*: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
530 * *1 MiB*: 1048576, 1m, 1024k
531 * *1 MB*: 1000000, 1mi, 1000ki
532 * *1 TiB*: 1099511627776, 1t, 1024g, 1048576m
533 * *1 TB*: 1000000000, 1ti, 1000mi, 1000000ki
535 To specify times (units are not case sensitive):
539 * *M* -- mean minutes
540 * *s* -- or sec means seconds (default)
541 * *ms* -- or *msec* means milliseconds
542 * *us* -- or *usec* means microseconds
544 If the option accepts an upper and lower range, use a colon ':' or
545 minus '-' to separate such values. See :ref:`irange <irange>`.
550 Boolean. Usually parsed as an integer, however only defined for
551 true and false (1 and 0).
556 Integer range with suffix. Allows value range to be given, such as
557 1024-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
558 option allows two sets of ranges, they can be specified with a ',' or '/'
559 delimiter: 1k-4k/8k-32k. Also see :ref:`int <int>`.
562 A list of floating point numbers, separated by a ':' character.
568 .. option:: kb_base=int
570 Select the interpretation of unit prefixes in input parameters.
573 Inputs comply with IEC 80000-13 and the International
574 System of Units (SI). Use:
576 - power-of-2 values with IEC prefixes (e.g., KiB)
577 - power-of-10 values with SI prefixes (e.g., kB)
580 Compatibility mode (default). To avoid breaking old scripts:
582 - power-of-2 values with SI prefixes
583 - power-of-10 values with IEC prefixes
585 See :option:`bs` for more details on input parameters.
587 Outputs always use correct prefixes. Most outputs include both
590 bw=2383.3kB/s (2327.4KiB/s)
592 If only one value is reported, then kb_base selects the one to use:
594 **1000** -- SI prefixes
596 **1024** -- IEC prefixes
598 .. option:: unit_base=int
600 Base unit for reporting. Allowed values are:
603 Use auto-detection (default).
610 With the above in mind, here follows the complete list of fio job parameters.
618 ASCII name of the job. This may be used to override the name printed by fio
619 for this job. Otherwise the job name is used. On the command line this
620 parameter has the special purpose of also signaling the start of a new job.
622 .. option:: description=str
624 Text description of the job. Doesn't do anything except dump this text
625 description when this job is run. It's not parsed.
627 .. option:: loops=int
629 Run the specified number of iterations of this job. Used to repeat the same
630 workload a given number of times. Defaults to 1.
632 .. option:: numjobs=int
634 Create the specified number of clones of this job. Each clone of job
635 is spawned as an independent thread or process. May be used to setup a
636 larger number of threads/processes doing the same thing. Each thread is
637 reported separately; to see statistics for all clones as a whole, use
638 :option:`group_reporting` in conjunction with :option:`new_group`.
639 See :option:`--max-jobs`.
642 Time related parameters
643 ~~~~~~~~~~~~~~~~~~~~~~~
645 .. option:: runtime=time
647 Tell fio to terminate processing after the specified period of time. It
648 can be quite hard to determine for how long a specified job will run, so
649 this parameter is handy to cap the total runtime to a given time. When
650 the unit is omitted, the value is given in seconds.
652 .. option:: time_based
654 If set, fio will run for the duration of the :option:`runtime` specified
655 even if the file(s) are completely read or written. It will simply loop over
656 the same workload as many times as the :option:`runtime` allows.
658 .. option:: startdelay=irange(time)
660 Delay start of job for the specified number of seconds. Supports all time
661 suffixes to allow specification of hours, minutes, seconds and milliseconds
662 -- seconds are the default if a unit is omitted. Can be given as a range
663 which causes each thread to choose randomly out of the range.
665 .. option:: ramp_time=time
667 If set, fio will run the specified workload for this amount of time before
668 logging any performance numbers. Useful for letting performance settle
669 before logging results, thus minimizing the runtime required for stable
670 results. Note that the ``ramp_time`` is considered lead in time for a job,
671 thus it will increase the total runtime if a special timeout or
672 :option:`runtime` is specified. When the unit is omitted, the value is
675 .. option:: clocksource=str
677 Use the given clocksource as the base of timing. The supported options are:
680 :manpage:`gettimeofday(2)`
683 :manpage:`clock_gettime(2)`
686 Internal CPU clock source
688 cpu is the preferred clocksource if it is reliable, as it is very fast (and
689 fio is heavy on time calls). Fio will automatically use this clocksource if
690 it's supported and considered reliable on the system it is running on,
691 unless another clocksource is specifically set. For x86/x86-64 CPUs, this
692 means supporting TSC Invariant.
694 .. option:: gtod_reduce=bool
696 Enable all of the :manpage:`gettimeofday(2)` reducing options
697 (:option:`disable_clat`, :option:`disable_slat`, :option:`disable_bw_measurement`) plus
698 reduce precision of the timeout somewhat to really shrink the
699 :manpage:`gettimeofday(2)` call count. With this option enabled, we only do
700 about 0.4% of the :manpage:`gettimeofday(2)` calls we would have done if all
701 time keeping was enabled.
703 .. option:: gtod_cpu=int
705 Sometimes it's cheaper to dedicate a single thread of execution to just
706 getting the current time. Fio (and databases, for instance) are very
707 intensive on :manpage:`gettimeofday(2)` calls. With this option, you can set
708 one CPU aside for doing nothing but logging current time to a shared memory
709 location. Then the other threads/processes that run I/O workloads need only
710 copy that segment, instead of entering the kernel with a
711 :manpage:`gettimeofday(2)` call. The CPU set aside for doing these time
712 calls will be excluded from other uses. Fio will manually clear it from the
713 CPU mask of other jobs.
719 .. option:: directory=str
721 Prefix filenames with this directory. Used to place files in a different
722 location than :file:`./`. You can specify a number of directories by
723 separating the names with a ':' character. These directories will be
724 assigned equally distributed to job clones creates with :option:`numjobs` as
725 long as they are using generated filenames. If specific `filename(s)` are
726 set fio will use the first listed directory, and thereby matching the
727 `filename` semantic which generates a file each clone if not specified, but
728 let all clones use the same if set.
730 See the :option:`filename` option for escaping certain characters.
732 .. option:: filename=str
734 Fio normally makes up a `filename` based on the job name, thread number, and
735 file number. If you want to share files between threads in a job or several
736 jobs with fixed file paths, specify a `filename` for each of them to override
737 the default. If the ioengine is file based, you can specify a number of files
738 by separating the names with a ':' colon. So if you wanted a job to open
739 :file:`/dev/sda` and :file:`/dev/sdb` as the two working files, you would use
740 ``filename=/dev/sda:/dev/sdb``. This also means that whenever this option is
741 specified, :option:`nrfiles` is ignored. The size of regular files specified
742 by this option will be :option:`size` divided by number of files unless
743 explicit size is specified by :option:`filesize`.
745 On Windows, disk devices are accessed as :file:`\\\\.\\PhysicalDrive0` for
746 the first device, :file:`\\\\.\\PhysicalDrive1` for the second etc.
747 Note: Windows and FreeBSD prevent write access to areas
748 of the disk containing in-use data (e.g. filesystems). If the wanted
749 `filename` does need to include a colon, then escape that with a ``\``
750 character. For instance, if the `filename` is :file:`/dev/dsk/foo@3,0:c`,
751 then you would use ``filename="/dev/dsk/foo@3,0\:c"``. The
752 :file:`-` is a reserved name, meaning stdin or stdout. Which of the two
753 depends on the read/write direction set.
755 .. option:: filename_format=str
757 If sharing multiple files between jobs, it is usually necessary to have fio
758 generate the exact names that you want. By default, fio will name a file
759 based on the default file format specification of
760 :file:`jobname.jobnumber.filenumber`. With this option, that can be
761 customized. Fio will recognize and replace the following keywords in this
765 The name of the worker thread or process.
767 The incremental number of the worker thread or process.
769 The incremental number of the file for that worker thread or
772 To have dependent jobs share a set of files, this option can be set to have
773 fio generate filenames that are shared between the two. For instance, if
774 :file:`testfiles.$filenum` is specified, file number 4 for any job will be
775 named :file:`testfiles.4`. The default of :file:`$jobname.$jobnum.$filenum`
776 will be used if no other format specifier is given.
778 .. option:: unique_filename=bool
780 To avoid collisions between networked clients, fio defaults to prefixing any
781 generated filenames (with a directory specified) with the source of the
782 client connecting. To disable this behavior, set this option to 0.
784 .. option:: opendir=str
786 Recursively open any files below directory `str`.
788 .. option:: lockfile=str
790 Fio defaults to not locking any files before it does I/O to them. If a file
791 or file descriptor is shared, fio can serialize I/O to that file to make the
792 end result consistent. This is usual for emulating real workloads that share
793 files. The lock modes are:
796 No locking. The default.
798 Only one thread or process may do I/O at a time, excluding all
801 Read-write locking on the file. Many readers may
802 access the file at the same time, but writes get exclusive access.
804 .. option:: nrfiles=int
806 Number of files to use for this job. Defaults to 1. The size of files
807 will be :option:`size` divided by this unless explicit size is specified by
808 :option:`filesize`. Files are created for each thread separately, and each
809 file will have a file number within its name by default, as explained in
810 :option:`filename` section.
813 .. option:: openfiles=int
815 Number of files to keep open at the same time. Defaults to the same as
816 :option:`nrfiles`, can be set smaller to limit the number simultaneous
819 .. option:: file_service_type=str
821 Defines how fio decides which file from a job to service next. The following
825 Choose a file at random.
828 Round robin over opened files. This is the default.
831 Finish one file before moving on to the next. Multiple files can
832 still be open depending on 'openfiles'.
835 Use a *Zipf* distribution to decide what file to access.
838 Use a *Pareto* distribution to decide what file to access.
841 Use a *Gaussian* (normal) distribution to decide what file to
844 For *random*, *roundrobin*, and *sequential*, a postfix can be appended to
845 tell fio how many I/Os to issue before switching to a new file. For example,
846 specifying ``file_service_type=random:8`` would cause fio to issue
847 8 I/Os before selecting a new file at random. For the non-uniform
848 distributions, a floating point postfix can be given to influence how the
849 distribution is skewed. See :option:`random_distribution` for a description
850 of how that would work.
852 .. option:: ioscheduler=str
854 Attempt to switch the device hosting the file to the specified I/O scheduler
857 .. option:: create_serialize=bool
859 If true, serialize the file creation for the jobs. This may be handy to
860 avoid interleaving of data files, which may greatly depend on the filesystem
861 used and even the number of processors in the system.
863 .. option:: create_fsync=bool
865 fsync the data file after creation. This is the default.
867 .. option:: create_on_open=bool
869 Don't pre-setup the files for I/O, just create open() when it's time to do
872 .. option:: create_only=bool
874 If true, fio will only run the setup phase of the job. If files need to be
875 laid out or updated on disk, only that will be done. The actual job contents
878 .. option:: allow_file_create=bool
880 If true, fio is permitted to create files as part of its workload. This is
881 the default behavior. If this option is false, then fio will error out if
882 the files it needs to use don't already exist. Default: true.
884 .. option:: allow_mounted_write=bool
886 If this isn't set, fio will abort jobs that are destructive (e.g. that write)
887 to what appears to be a mounted device or partition. This should help catch
888 creating inadvertently destructive tests, not realizing that the test will
889 destroy data on the mounted file system. Default: false.
891 .. option:: pre_read=bool
893 If this is given, files will be pre-read into memory before starting the
894 given I/O operation. This will also clear the :option:`invalidate` flag,
895 since it is pointless to pre-read and then drop the cache. This will only
896 work for I/O engines that are seek-able, since they allow you to read the
897 same data multiple times. Thus it will not work on e.g. network or splice I/O.
899 .. option:: unlink=bool
901 Unlink the job files when done. Not the default, as repeated runs of that
902 job would then waste time recreating the file set again and again.
904 .. option:: unlink_each_loop=bool
906 Unlink job files after each iteration or loop.
908 .. option:: zonesize=int
910 Divide a file into zones of the specified size. See :option:`zoneskip`.
912 .. option:: zonerange=int
914 Give size of an I/O zone. See :option:`zoneskip`.
916 .. option:: zoneskip=int
918 Skip the specified number of bytes when :option:`zonesize` data has been
919 read. The two zone options can be used to only do I/O on zones of a file.
925 .. option:: direct=bool
927 If value is true, use non-buffered I/O. This is usually O_DIRECT. Note that
928 ZFS on Solaris doesn't support direct I/O. On Windows the synchronous
929 ioengines don't support direct I/O. Default: false.
931 .. option:: atomic=bool
933 If value is true, attempt to use atomic direct I/O. Atomic writes are
934 guaranteed to be stable once acknowledged by the operating system. Only
935 Linux supports O_ATOMIC right now.
937 .. option:: buffered=bool
939 If value is true, use buffered I/O. This is the opposite of the
940 :option:`direct` option. Defaults to true.
942 .. option:: readwrite=str, rw=str
944 Type of I/O pattern. Accepted values are:
951 Sequential trims (Linux block devices only).
957 Random trims (Linux block devices only).
959 Sequential mixed reads and writes.
961 Random mixed reads and writes.
963 Sequential trim+write sequences. Blocks will be trimmed first,
964 then the same blocks will be written to.
966 Fio defaults to read if the option is not specified. For the mixed I/O
967 types, the default is to split them 50/50. For certain types of I/O the
968 result may still be skewed a bit, since the speed may be different. It is
969 possible to specify a number of I/O's to do before getting a new offset,
970 this is done by appending a ``:<nr>`` to the end of the string given. For a
971 random read, it would look like ``rw=randread:8`` for passing in an offset
972 modifier with a value of 8. If the suffix is used with a sequential I/O
973 pattern, then the value specified will be added to the generated offset for
974 each I/O. For instance, using ``rw=write:4k`` will skip 4k for every
975 write. It turns sequential I/O into sequential I/O with holes. See the
976 :option:`rw_sequencer` option.
978 .. option:: rw_sequencer=str
980 If an offset modifier is given by appending a number to the ``rw=<str>``
981 line, then this option controls how that number modifies the I/O offset
982 being generated. Accepted values are:
985 Generate sequential offset.
987 Generate the same offset.
989 ``sequential`` is only useful for random I/O, where fio would normally
990 generate a new random offset for every I/O. If you append e.g. 8 to randread,
991 you would get a new random offset for every 8 I/O's. The result would be a
992 seek for only every 8 I/O's, instead of for every I/O. Use ``rw=randread:8``
993 to specify that. As sequential I/O is already sequential, setting
994 ``sequential`` for that would not result in any differences. ``identical``
995 behaves in a similar fashion, except it sends the same offset 8 number of
996 times before generating a new offset.
998 .. option:: unified_rw_reporting=bool
1000 Fio normally reports statistics on a per data direction basis, meaning that
1001 reads, writes, and trims are accounted and reported separately. If this
1002 option is set fio sums the results and report them as "mixed" instead.
1004 .. option:: randrepeat=bool
1006 Seed the random number generator used for random I/O patterns in a
1007 predictable way so the pattern is repeatable across runs. Default: true.
1009 .. option:: allrandrepeat=bool
1011 Seed all random number generators in a predictable way so results are
1012 repeatable across runs. Default: false.
1014 .. option:: randseed=int
1016 Seed the random number generators based on this seed value, to be able to
1017 control what sequence of output is being generated. If not set, the random
1018 sequence depends on the :option:`randrepeat` setting.
1020 .. option:: fallocate=str
1022 Whether pre-allocation is performed when laying down files.
1023 Accepted values are:
1026 Do not pre-allocate space.
1029 Pre-allocate via :manpage:`posix_fallocate(3)`.
1032 Pre-allocate via :manpage:`fallocate(2)` with
1033 FALLOC_FL_KEEP_SIZE set.
1036 Backward-compatible alias for **none**.
1039 Backward-compatible alias for **posix**.
1041 May not be available on all supported platforms. **keep** is only available
1042 on Linux. If using ZFS on Solaris this must be set to **none** because ZFS
1043 doesn't support it. Default: **posix**.
1045 .. option:: fadvise_hint=str
1047 Use :manpage:`posix_fadvise(2)` to advise the kernel on what I/O patterns
1048 are likely to be issued. Accepted values are:
1051 Backwards-compatible hint for "no hint".
1054 Backwards compatible hint for "advise with fio workload type". This
1055 uses **FADV_RANDOM** for a random workload, and **FADV_SEQUENTIAL**
1056 for a sequential workload.
1059 Advise using **FADV_SEQUENTIAL**.
1062 Advise using **FADV_RANDOM**.
1064 .. option:: fadvise_stream=int
1066 Use :manpage:`posix_fadvise(2)` to advise the kernel what stream ID the
1067 writes issued belong to. Only supported on Linux. Note, this option may
1068 change going forward.
1070 .. option:: offset=int
1072 Start I/O at the given offset in the file. The data before the given offset
1073 will not be touched. This effectively caps the file size at `real_size -
1074 offset`. Can be combined with :option:`size` to constrain the start and
1075 end range that I/O will be done within.
1077 .. option:: offset_increment=int
1079 If this is provided, then the real offset becomes `offset + offset_increment
1080 * thread_number`, where the thread number is a counter that starts at 0 and
1081 is incremented for each sub-job (i.e. when :option:`numjobs` option is
1082 specified). This option is useful if there are several jobs which are
1083 intended to operate on a file in parallel disjoint segments, with even
1084 spacing between the starting points.
1086 .. option:: number_ios=int
1088 Fio will normally perform I/Os until it has exhausted the size of the region
1089 set by :option:`size`, or if it exhaust the allocated time (or hits an error
1090 condition). With this setting, the range/size can be set independently of
1091 the number of I/Os to perform. When fio reaches this number, it will exit
1092 normally and report status. Note that this does not extend the amount of I/O
1093 that will be done, it will only stop fio if this condition is met before
1094 other end-of-job criteria.
1096 .. option:: fsync=int
1098 If writing to a file, issue a sync of the dirty data for every number of
1099 blocks given. For example, if you give 32 as a parameter, fio will sync the
1100 file for every 32 writes issued. If fio is using non-buffered I/O, we may
1101 not sync the file. The exception is the sg I/O engine, which synchronizes
1102 the disk cache anyway.
1104 .. option:: fdatasync=int
1106 Like :option:`fsync` but uses :manpage:`fdatasync(2)` to only sync data and
1107 not metadata blocks. In FreeBSD and Windows there is no
1108 :manpage:`fdatasync(2)`, this falls back to using :manpage:`fsync(2)`.
1110 .. option:: write_barrier=int
1112 Make every `N-th` write a barrier write.
1114 .. option:: sync_file_range=str:val
1116 Use :manpage:`sync_file_range(2)` for every `val` number of write
1117 operations. Fio will track range of writes that have happened since the last
1118 :manpage:`sync_file_range(2)` call. `str` can currently be one or more of:
1121 SYNC_FILE_RANGE_WAIT_BEFORE
1123 SYNC_FILE_RANGE_WRITE
1125 SYNC_FILE_RANGE_WAIT_AFTER
1127 So if you do ``sync_file_range=wait_before,write:8``, fio would use
1128 ``SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE`` for every 8
1129 writes. Also see the :manpage:`sync_file_range(2)` man page. This option is
1132 .. option:: overwrite=bool
1134 If true, writes to a file will always overwrite existing data. If the file
1135 doesn't already exist, it will be created before the write phase begins. If
1136 the file exists and is large enough for the specified write phase, nothing
1139 .. option:: end_fsync=bool
1141 If true, fsync file contents when a write stage has completed.
1143 .. option:: fsync_on_close=bool
1145 If true, fio will :manpage:`fsync(2)` a dirty file on close. This differs
1146 from end_fsync in that it will happen on every file close, not just at the
1149 .. option:: rwmixread=int
1151 Percentage of a mixed workload that should be reads. Default: 50.
1153 .. option:: rwmixwrite=int
1155 Percentage of a mixed workload that should be writes. If both
1156 :option:`rwmixread` and :option:`rwmixwrite` is given and the values do not
1157 add up to 100%, the latter of the two will be used to override the
1158 first. This may interfere with a given rate setting, if fio is asked to
1159 limit reads or writes to a certain rate. If that is the case, then the
1160 distribution may be skewed. Default: 50.
1162 .. option:: random_distribution=str:float[,str:float][,str:float]
1164 By default, fio will use a completely uniform random distribution when asked
1165 to perform random I/O. Sometimes it is useful to skew the distribution in
1166 specific ways, ensuring that some parts of the data is more hot than others.
1167 fio includes the following distribution models:
1170 Uniform random distribution
1179 Normal (Gaussian) distribution
1182 Zoned random distribution
1184 When using a **zipf** or **pareto** distribution, an input value is also
1185 needed to define the access pattern. For **zipf**, this is the `zipf
1186 theta`. For **pareto**, it's the `Pareto power`. Fio includes a test
1187 program, :command:`genzipf`, that can be used visualize what the given input
1188 values will yield in terms of hit rates. If you wanted to use **zipf** with
1189 a `theta` of 1.2, you would use ``random_distribution=zipf:1.2`` as the
1190 option. If a non-uniform model is used, fio will disable use of the random
1191 map. For the **gauss** distribution, a normal deviation is supplied as a
1192 value between 0 and 100.
1194 For a **zoned** distribution, fio supports specifying percentages of I/O
1195 access that should fall within what range of the file or device. For
1196 example, given a criteria of:
1198 * 60% of accesses should be to the first 10%
1199 * 30% of accesses should be to the next 20%
1200 * 8% of accesses should be to to the next 30%
1201 * 2% of accesses should be to the next 40%
1203 we can define that through zoning of the random accesses. For the above
1204 example, the user would do::
1206 random_distribution=zoned:60/10:30/20:8/30:2/40
1208 similarly to how :option:`bssplit` works for setting ranges and percentages
1209 of block sizes. Like :option:`bssplit`, it's possible to specify separate
1210 zones for reads, writes, and trims. If just one set is given, it'll apply to
1213 .. option:: percentage_random=int[,int][,int]
1215 For a random workload, set how big a percentage should be random. This
1216 defaults to 100%, in which case the workload is fully random. It can be set
1217 from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1218 sequential. Any setting in between will result in a random mix of sequential
1219 and random I/O, at the given percentages. Comma-separated values may be
1220 specified for reads, writes, and trims as described in :option:`blocksize`.
1222 .. option:: norandommap
1224 Normally fio will cover every block of the file when doing random I/O. If
1225 this option is given, fio will just get a new random offset without looking
1226 at past I/O history. This means that some blocks may not be read or written,
1227 and that some blocks may be read/written more than once. If this option is
1228 used with :option:`verify` and multiple blocksizes (via :option:`bsrange`),
1229 only intact blocks are verified, i.e., partially-overwritten blocks are
1232 .. option:: softrandommap=bool
1234 See :option:`norandommap`. If fio runs with the random block map enabled and
1235 it fails to allocate the map, if this option is set it will continue without
1236 a random block map. As coverage will not be as complete as with random maps,
1237 this option is disabled by default.
1239 .. option:: random_generator=str
1241 Fio supports the following engines for generating
1242 I/O offsets for random I/O:
1245 Strong 2^88 cycle random number generator
1247 Linear feedback shift register generator
1249 Strong 64-bit 2^258 cycle random number generator
1251 **tausworthe** is a strong random number generator, but it requires tracking
1252 on the side if we want to ensure that blocks are only read or written
1253 once. **LFSR** guarantees that we never generate the same offset twice, and
1254 it's also less computationally expensive. It's not a true random generator,
1255 however, though for I/O purposes it's typically good enough. **LFSR** only
1256 works with single block sizes, not with workloads that use multiple block
1257 sizes. If used with such a workload, fio may read or write some blocks
1258 multiple times. The default value is **tausworthe**, unless the required
1259 space exceeds 2^32 blocks. If it does, then **tausworthe64** is
1260 selected automatically.
1266 .. option:: blocksize=int[,int][,int], bs=int[,int][,int]
1268 The block size in bytes used for I/O units. Default: 4096. A single value
1269 applies to reads, writes, and trims. Comma-separated values may be
1270 specified for reads, writes, and trims. A value not terminated in a comma
1271 applies to subsequent types.
1276 means 256k for reads, writes and trims.
1279 means 8k for reads, 32k for writes and trims.
1282 means 8k for reads, 32k for writes, and default for trims.
1285 means default for reads, 8k for writes and trims.
1288 means default for reads, 8k for writes, and default for writes.
1290 .. option:: blocksize_range=irange[,irange][,irange], bsrange=irange[,irange][,irange]
1292 A range of block sizes in bytes for I/O units. The issued I/O unit will
1293 always be a multiple of the minimum size, unless
1294 :option:`blocksize_unaligned` is set.
1296 Comma-separated ranges may be specified for reads, writes, and trims as
1297 described in :option:`blocksize`.
1299 Example: ``bsrange=1k-4k,2k-8k``.
1301 .. option:: bssplit=str[,str][,str]
1303 Sometimes you want even finer grained control of the block sizes issued, not
1304 just an even split between them. This option allows you to weight various
1305 block sizes, so that you are able to define a specific amount of block sizes
1306 issued. The format for this option is::
1308 bssplit=blocksize/percentage:blocksize/percentage
1310 for as many block sizes as needed. So if you want to define a workload that
1311 has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would write::
1313 bssplit=4k/10:64k/50:32k/40
1315 Ordering does not matter. If the percentage is left blank, fio will fill in
1316 the remaining values evenly. So a bssplit option like this one::
1318 bssplit=4k/50:1k/:32k/
1320 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always add up
1321 to 100, if bssplit is given a range that adds up to more, it will error out.
1323 Comma-separated values may be specified for reads, writes, and trims as
1324 described in :option:`blocksize`.
1326 If you want a workload that has 50% 2k reads and 50% 4k reads, while having
1327 90% 4k writes and 10% 8k writes, you would specify::
1329 bssplit=2k/50:4k/50,4k/90,8k/10
1331 .. option:: blocksize_unaligned, bs_unaligned
1333 If set, fio will issue I/O units with any size within
1334 :option:`blocksize_range`, not just multiples of the minimum size. This
1335 typically won't work with direct I/O, as that normally requires sector
1338 .. option:: bs_is_seq_rand
1340 If this option is set, fio will use the normal read,write blocksize settings
1341 as sequential,random blocksize settings instead. Any random read or write
1342 will use the WRITE blocksize settings, and any sequential read or write will
1343 use the READ blocksize settings.
1345 .. option:: blockalign=int[,int][,int], ba=int[,int][,int]
1347 Boundary to which fio will align random I/O units. Default:
1348 :option:`blocksize`. Minimum alignment is typically 512b for using direct
1349 I/O, though it usually depends on the hardware block size. This option is
1350 mutually exclusive with using a random map for files, so it will turn off
1351 that option. Comma-separated values may be specified for reads, writes, and
1352 trims as described in :option:`blocksize`.
1358 .. option:: zero_buffers
1360 Initialize buffers with all zeros. Default: fill buffers with random data.
1362 .. option:: refill_buffers
1364 If this option is given, fio will refill the I/O buffers on every
1365 submit. The default is to only fill it at init time and reuse that
1366 data. Only makes sense if zero_buffers isn't specified, naturally. If data
1367 verification is enabled, `refill_buffers` is also automatically enabled.
1369 .. option:: scramble_buffers=bool
1371 If :option:`refill_buffers` is too costly and the target is using data
1372 deduplication, then setting this option will slightly modify the I/O buffer
1373 contents to defeat normal de-dupe attempts. This is not enough to defeat
1374 more clever block compression attempts, but it will stop naive dedupe of
1375 blocks. Default: true.
1377 .. option:: buffer_compress_percentage=int
1379 If this is set, then fio will attempt to provide I/O buffer content (on
1380 WRITEs) that compress to the specified level. Fio does this by providing a
1381 mix of random data and a fixed pattern. The fixed pattern is either zeroes,
1382 or the pattern specified by :option:`buffer_pattern`. If the pattern option
1383 is used, it might skew the compression ratio slightly. Note that this is per
1384 block size unit, for file/disk wide compression level that matches this
1385 setting, you'll also want to set :option:`refill_buffers`.
1387 .. option:: buffer_compress_chunk=int
1389 See :option:`buffer_compress_percentage`. This setting allows fio to manage
1390 how big the ranges of random data and zeroed data is. Without this set, fio
1391 will provide :option:`buffer_compress_percentage` of blocksize random data,
1392 followed by the remaining zeroed. With this set to some chunk size smaller
1393 than the block size, fio can alternate random and zeroed data throughout the
1396 .. option:: buffer_pattern=str
1398 If set, fio will fill the I/O buffers with this pattern. If not set, the
1399 contents of I/O buffers is defined by the other options related to buffer
1400 contents. The setting can be any pattern of bytes, and can be prefixed with
1401 0x for hex values. It may also be a string, where the string must then be
1402 wrapped with ``""``, e.g.::
1404 buffer_pattern="abcd"
1412 buffer_pattern=0xdeadface
1414 Also you can combine everything together in any order::
1416 buffer_pattern=0xdeadface"abcd"-12
1418 .. option:: dedupe_percentage=int
1420 If set, fio will generate this percentage of identical buffers when
1421 writing. These buffers will be naturally dedupable. The contents of the
1422 buffers depend on what other buffer compression settings have been set. It's
1423 possible to have the individual buffers either fully compressible, or not at
1424 all. This option only controls the distribution of unique buffers.
1426 .. option:: invalidate=bool
1428 Invalidate the buffer/page cache parts for this file prior to starting
1429 I/O. Defaults to true.
1431 .. option:: sync=bool
1433 Use synchronous I/O for buffered writes. For the majority of I/O engines,
1434 this means using O_SYNC. Default: false.
1436 .. option:: iomem=str, mem=str
1438 Fio can use various types of memory as the I/O unit buffer. The allowed
1442 Use memory from :manpage:`malloc(3)` as the buffers. Default memory
1446 Use shared memory as the buffers. Allocated through
1447 :manpage:`shmget(2)`.
1450 Same as shm, but use huge pages as backing.
1453 Use mmap to allocate buffers. May either be anonymous memory, or can
1454 be file backed if a filename is given after the option. The format
1455 is `mem=mmap:/path/to/file`.
1458 Use a memory mapped huge file as the buffer backing. Append filename
1459 after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file`.
1462 Same as mmap, but use a MMAP_SHARED mapping.
1464 The area allocated is a function of the maximum allowed bs size for the job,
1465 multiplied by the I/O depth given. Note that for **shmhuge** and
1466 **mmaphuge** to work, the system must have free huge pages allocated. This
1467 can normally be checked and set by reading/writing
1468 :file:`/proc/sys/vm/nr_hugepages` on a Linux system. Fio assumes a huge page
1469 is 4MiB in size. So to calculate the number of huge pages you need for a
1470 given job file, add up the I/O depth of all jobs (normally one unless
1471 :option:`iodepth` is used) and multiply by the maximum bs set. Then divide
1472 that number by the huge page size. You can see the size of the huge pages in
1473 :file:`/proc/meminfo`. If no huge pages are allocated by having a non-zero
1474 number in `nr_hugepages`, using **mmaphuge** or **shmhuge** will fail. Also
1475 see :option:`hugepage-size`.
1477 **mmaphuge** also needs to have hugetlbfs mounted and the file location
1478 should point there. So if it's mounted in :file:`/huge`, you would use
1479 `mem=mmaphuge:/huge/somefile`.
1481 .. option:: iomem_align=int
1483 This indicates the memory alignment of the I/O memory buffers. Note that
1484 the given alignment is applied to the first I/O unit buffer, if using
1485 :option:`iodepth` the alignment of the following buffers are given by the
1486 :option:`bs` used. In other words, if using a :option:`bs` that is a
1487 multiple of the page sized in the system, all buffers will be aligned to
1488 this value. If using a :option:`bs` that is not page aligned, the alignment
1489 of subsequent I/O memory buffers is the sum of the :option:`iomem_align` and
1492 .. option:: hugepage-size=int
1494 Defines the size of a huge page. Must at least be equal to the system
1495 setting, see :file:`/proc/meminfo`. Defaults to 4MiB. Should probably
1496 always be a multiple of megabytes, so using ``hugepage-size=Xm`` is the
1497 preferred way to set this to avoid setting a non-pow-2 bad value.
1499 .. option:: lockmem=int
1501 Pin the specified amount of memory with :manpage:`mlock(2)`. Can be used to
1502 simulate a smaller amount of memory. The amount specified is per worker.
1508 .. option:: size=int
1510 The total size of file I/O for each thread of this job. Fio will run until
1511 this many bytes has been transferred, unless runtime is limited by other options
1512 (such as :option:`runtime`, for instance, or increased/decreased by :option:`io_size`).
1513 Fio will divide this size between the available files determined by options
1514 such as :option:`nrfiles`, :option:`filename`, unless :option:`filesize` is
1515 specified by the job. If the result of division happens to be 0, the size is
1516 set to the physical size of the given files or devices.
1517 If this option is not specified, fio will use the full size of the given
1518 files or devices. If the files do not exist, size must be given. It is also
1519 possible to give size as a percentage between 1 and 100. If ``size=20%`` is
1520 given, fio will use 20% of the full size of the given files or devices.
1521 Can be combined with :option:`offset` to constrain the start and end range
1522 that I/O will be done within.
1524 .. option:: io_size=int, io_limit=int
1526 Normally fio operates within the region set by :option:`size`, which means
1527 that the :option:`size` option sets both the region and size of I/O to be
1528 performed. Sometimes that is not what you want. With this option, it is
1529 possible to define just the amount of I/O that fio should do. For instance,
1530 if :option:`size` is set to 20GiB and :option:`io_size` is set to 5GiB, fio
1531 will perform I/O within the first 20GiB but exit when 5GiB have been
1532 done. The opposite is also possible -- if :option:`size` is set to 20GiB,
1533 and :option:`io_size` is set to 40GiB, then fio will do 40GiB of I/O within
1534 the 0..20GiB region.
1536 .. option:: filesize=int
1538 Individual file sizes. May be a range, in which case fio will select sizes
1539 for files at random within the given range and limited to :option:`size` in
1540 total (if that is given). If not given, each created file is the same size.
1541 This option overrides :option:`size` in terms of file size, which means
1542 this value is used as a fixed size or possible range of each file.
1544 .. option:: file_append=bool
1546 Perform I/O after the end of the file. Normally fio will operate within the
1547 size of a file. If this option is set, then fio will append to the file
1548 instead. This has identical behavior to setting :option:`offset` to the size
1549 of a file. This option is ignored on non-regular files.
1551 .. option:: fill_device=bool, fill_fs=bool
1553 Sets size to something really large and waits for ENOSPC (no space left on
1554 device) as the terminating condition. Only makes sense with sequential
1555 write. For a read workload, the mount point will be filled first then I/O
1556 started on the result. This option doesn't make sense if operating on a raw
1557 device node, since the size of that is already known by the file system.
1558 Additionally, writing beyond end-of-device will not return ENOSPC there.
1564 .. option:: ioengine=str
1566 Defines how the job issues I/O to the file. The following types are defined:
1569 Basic :manpage:`read(2)` or :manpage:`write(2)`
1570 I/O. :manpage:`lseek(2)` is used to position the I/O location.
1573 Basic :manpage:`pread(2)` or :manpage:`pwrite(2)` I/O. Default on
1574 all supported operating systems except for Windows.
1577 Basic :manpage:`readv(2)` or :manpage:`writev(2)` I/O. Will emulate
1578 queuing by coalescing adjacent I/Os into a single submission.
1581 Basic :manpage:`preadv(2)` or :manpage:`pwritev(2)` I/O.
1584 Basic :manpage:`preadv2(2)` or :manpage:`pwritev2(2)` I/O.
1587 Linux native asynchronous I/O. Note that Linux may only support
1588 queued behaviour with non-buffered I/O (set ``direct=1`` or
1590 This engine defines engine specific options.
1593 POSIX asynchronous I/O using :manpage:`aio_read(3)` and
1594 :manpage:`aio_write(3)`.
1597 Solaris native asynchronous I/O.
1600 Windows native asynchronous I/O. Default on Windows.
1603 File is memory mapped with :manpage:`mmap(2)` and data copied
1604 to/from using :manpage:`memcpy(3)`.
1607 :manpage:`splice(2)` is used to transfer the data and
1608 :manpage:`vmsplice(2)` to transfer data from user space to the
1612 SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1613 ioctl, or if the target is an sg character device we use
1614 :manpage:`read(2)` and :manpage:`write(2)` for asynchronous
1615 I/O. Requires filename option to specify either block or character
1619 Doesn't transfer any data, just pretends to. This is mainly used to
1620 exercise fio itself and for debugging/testing purposes.
1623 Transfer over the network to given ``host:port``. Depending on the
1624 :option:`protocol` used, the :option:`hostname`, :option:`port`,
1625 :option:`listen` and :option:`filename` options are used to specify
1626 what sort of connection to make, while the :option:`protocol` option
1627 determines which protocol will be used. This engine defines engine
1631 Like **net**, but uses :manpage:`splice(2)` and
1632 :manpage:`vmsplice(2)` to map data and send/receive.
1633 This engine defines engine specific options.
1636 Doesn't transfer any data, but burns CPU cycles according to the
1637 :option:`cpuload` and :option:`cpuchunks` options. Setting
1638 :option:`cpuload` =85 will cause that job to do nothing but burn 85%
1639 of the CPU. In case of SMP machines, use :option:`numjobs`
1640 =<no_of_cpu> to get desired CPU usage, as the cpuload only loads a
1641 single CPU at the desired rate. A job never finishes unless there is
1642 at least one non-cpuio job.
1645 The GUASI I/O engine is the Generic Userspace Asyncronous Syscall
1646 Interface approach to async I/O. See
1648 http://www.xmailserver.org/guasi-lib.html
1650 for more info on GUASI.
1653 The RDMA I/O engine supports both RDMA memory semantics
1654 (RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
1655 InfiniBand, RoCE and iWARP protocols.
1658 I/O engine that does regular fallocate to simulate data transfer as
1662 does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1665 does fallocate(,mode = 0).
1668 does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1671 I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1672 defragment activity in request to DDIR_WRITE event.
1675 I/O engine supporting direct access to Ceph Rados Block Devices
1676 (RBD) via librbd without the need to use the kernel rbd driver. This
1677 ioengine defines engine specific options.
1680 Using Glusterfs libgfapi sync interface to direct access to
1681 Glusterfs volumes without having to go through FUSE. This ioengine
1682 defines engine specific options.
1685 Using Glusterfs libgfapi async interface to direct access to
1686 Glusterfs volumes without having to go through FUSE. This ioengine
1687 defines engine specific options.
1690 Read and write through Hadoop (HDFS). The :file:`filename` option
1691 is used to specify host,port of the hdfs name-node to connect. This
1692 engine interprets offsets a little differently. In HDFS, files once
1693 created cannot be modified. So random writes are not possible. To
1694 imitate this, libhdfs engine expects bunch of small files to be
1695 created over HDFS, and engine will randomly pick a file out of those
1696 files based on the offset generated by fio backend. (see the example
1697 job file to create such files, use ``rw=write`` option). Please
1698 note, you might want to set necessary environment variables to work
1699 with hdfs/libhdfs properly. Each job uses its own connection to
1703 Read, write and erase an MTD character device (e.g.,
1704 :file:`/dev/mtd0`). Discards are treated as erases. Depending on the
1705 underlying device type, the I/O may have to go in a certain pattern,
1706 e.g., on NAND, writing sequentially to erase blocks and discarding
1707 before overwriting. The writetrim mode works well for this
1711 Read and write using filesystem DAX to a file on a filesystem
1712 mounted with DAX on a persistent memory device through the NVML
1716 Read and write using device DAX to a persistent memory device (e.g.,
1717 /dev/dax0.0) through the NVML libpmem library.
1720 Prefix to specify loading an external I/O engine object file. Append
1721 the engine filename, e.g. ``ioengine=external:/tmp/foo.o`` to load
1722 ioengine :file:`foo.o` in :file:`/tmp`.
1725 I/O engine specific parameters
1726 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1728 In addition, there are some parameters which are only valid when a specific
1729 ioengine is in use. These are used identically to normal parameters, with the
1730 caveat that when used on the command line, they must come after the
1731 :option:`ioengine` that defines them is selected.
1733 .. option:: userspace_reap : [libaio]
1735 Normally, with the libaio engine in use, fio will use the
1736 :manpage:`io_getevents(2)` system call to reap newly returned events. With
1737 this flag turned on, the AIO ring will be read directly from user-space to
1738 reap events. The reaping mode is only enabled when polling for a minimum of
1739 0 events (e.g. when :option:`iodepth_batch_complete` `=0`).
1741 .. option:: hipri : [pvsync2]
1743 Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
1746 .. option:: cpuload=int : [cpuio]
1748 Attempt to use the specified percentage of CPU cycles.
1750 .. option:: cpuchunks=int : [cpuio]
1752 Split the load into cycles of the given time. In microseconds.
1754 .. option:: exit_on_io_done=bool : [cpuio]
1756 Detect when I/O threads are done, then exit.
1758 .. option:: hostname=str : [netsplice] [net]
1760 The host name or IP address to use for TCP or UDP based I/O. If the job is
1761 a TCP listener or UDP reader, the host name is not used and must be omitted
1762 unless it is a valid UDP multicast address.
1764 .. option:: namenode=str : [libhdfs]
1766 The host name or IP address of a HDFS cluster namenode to contact.
1768 .. option:: port=int
1772 The TCP or UDP port to bind to or connect to. If this is used with
1773 :option:`numjobs` to spawn multiple instances of the same job type, then
1774 this will be the starting port number since fio will use a range of
1779 the listening port of the HFDS cluster namenode.
1781 .. option:: interface=str : [netsplice] [net]
1783 The IP address of the network interface used to send or receive UDP
1786 .. option:: ttl=int : [netsplice] [net]
1788 Time-to-live value for outgoing UDP multicast packets. Default: 1.
1790 .. option:: nodelay=bool : [netsplice] [net]
1792 Set TCP_NODELAY on TCP connections.
1794 .. option:: protocol=str : [netsplice] [net]
1796 .. option:: proto=str : [netsplice] [net]
1798 The network protocol to use. Accepted values are:
1801 Transmission control protocol.
1803 Transmission control protocol V6.
1805 User datagram protocol.
1807 User datagram protocol V6.
1811 When the protocol is TCP or UDP, the port must also be given, as well as the
1812 hostname if the job is a TCP listener or UDP reader. For unix sockets, the
1813 normal filename option should be used and the port is invalid.
1815 .. option:: listen : [net]
1817 For TCP network connections, tell fio to listen for incoming connections
1818 rather than initiating an outgoing connection. The :option:`hostname` must
1819 be omitted if this option is used.
1821 .. option:: pingpong : [net]
1823 Normally a network writer will just continue writing data, and a network
1824 reader will just consume packages. If ``pingpong=1`` is set, a writer will
1825 send its normal payload to the reader, then wait for the reader to send the
1826 same payload back. This allows fio to measure network latencies. The
1827 submission and completion latencies then measure local time spent sending or
1828 receiving, and the completion latency measures how long it took for the
1829 other end to receive and send back. For UDP multicast traffic
1830 ``pingpong=1`` should only be set for a single reader when multiple readers
1831 are listening to the same address.
1833 .. option:: window_size : [net]
1835 Set the desired socket buffer size for the connection.
1837 .. option:: mss : [net]
1839 Set the TCP maximum segment size (TCP_MAXSEG).
1841 .. option:: donorname=str : [e4defrag]
1843 File will be used as a block donor(swap extents between files).
1845 .. option:: inplace=int : [e4defrag]
1847 Configure donor file blocks allocation strategy:
1850 Default. Preallocate donor's file on init.
1852 Allocate space immediately inside defragment event, and free right
1855 .. option:: clustername=str : [rbd]
1857 Specifies the name of the Ceph cluster.
1859 .. option:: rbdname=str : [rbd]
1861 Specifies the name of the RBD.
1863 .. option:: pool=str : [rbd]
1865 Specifies the name of the Ceph pool containing RBD.
1867 .. option:: clientname=str : [rbd]
1869 Specifies the username (without the 'client.' prefix) used to access the
1870 Ceph cluster. If the *clustername* is specified, the *clientname* shall be
1871 the full *type.id* string. If no type. prefix is given, fio will add
1872 'client.' by default.
1874 .. option:: skip_bad=bool : [mtd]
1876 Skip operations against known bad blocks.
1878 .. option:: hdfsdirectory : [libhdfs]
1880 libhdfs will create chunk in this HDFS directory.
1882 .. option:: chunk_size : [libhdfs]
1884 the size of the chunk to use for each file.
1890 .. option:: iodepth=int
1892 Number of I/O units to keep in flight against the file. Note that
1893 increasing *iodepth* beyond 1 will not affect synchronous ioengines (except
1894 for small degrees when :option:`verify_async` is in use). Even async
1895 engines may impose OS restrictions causing the desired depth not to be
1896 achieved. This may happen on Linux when using libaio and not setting
1897 :option:`direct` =1, since buffered I/O is not async on that OS. Keep an
1898 eye on the I/O depth distribution in the fio output to verify that the
1899 achieved depth is as expected. Default: 1.
1901 .. option:: iodepth_batch_submit=int, iodepth_batch=int
1903 This defines how many pieces of I/O to submit at once. It defaults to 1
1904 which means that we submit each I/O as soon as it is available, but can be
1905 raised to submit bigger batches of I/O at the time. If it is set to 0 the
1906 :option:`iodepth` value will be used.
1908 .. option:: iodepth_batch_complete_min=int, iodepth_batch_complete=int
1910 This defines how many pieces of I/O to retrieve at once. It defaults to 1
1911 which means that we'll ask for a minimum of 1 I/O in the retrieval process
1912 from the kernel. The I/O retrieval will go on until we hit the limit set by
1913 :option:`iodepth_low`. If this variable is set to 0, then fio will always
1914 check for completed events before queuing more I/O. This helps reduce I/O
1915 latency, at the cost of more retrieval system calls.
1917 .. option:: iodepth_batch_complete_max=int
1919 This defines maximum pieces of I/O to retrieve at once. This variable should
1920 be used along with :option:`iodepth_batch_complete_min` =int variable,
1921 specifying the range of min and max amount of I/O which should be
1922 retrieved. By default it is equal to :option:`iodepth_batch_complete_min`
1927 iodepth_batch_complete_min=1
1928 iodepth_batch_complete_max=<iodepth>
1930 which means that we will retrieve at least 1 I/O and up to the whole
1931 submitted queue depth. If none of I/O has been completed yet, we will wait.
1935 iodepth_batch_complete_min=0
1936 iodepth_batch_complete_max=<iodepth>
1938 which means that we can retrieve up to the whole submitted queue depth, but
1939 if none of I/O has been completed yet, we will NOT wait and immediately exit
1940 the system call. In this example we simply do polling.
1942 .. option:: iodepth_low=int
1944 The low water mark indicating when to start filling the queue
1945 again. Defaults to the same as :option:`iodepth`, meaning that fio will
1946 attempt to keep the queue full at all times. If :option:`iodepth` is set to
1947 e.g. 16 and *iodepth_low* is set to 4, then after fio has filled the queue of
1948 16 requests, it will let the depth drain down to 4 before starting to fill
1951 .. option:: io_submit_mode=str
1953 This option controls how fio submits the I/O to the I/O engine. The default
1954 is `inline`, which means that the fio job threads submit and reap I/O
1955 directly. If set to `offload`, the job threads will offload I/O submission
1956 to a dedicated pool of I/O threads. This requires some coordination and thus
1957 has a bit of extra overhead, especially for lower queue depth I/O where it
1958 can increase latencies. The benefit is that fio can manage submission rates
1959 independently of the device completion rates. This avoids skewed latency
1960 reporting if I/O gets back up on the device side (the coordinated omission
1967 .. option:: thinktime=time
1969 Stall the job for the specified period of time after an I/O has completed before issuing the
1970 next. May be used to simulate processing being done by an application.
1971 When the unit is omitted, the value is given in microseconds. See
1972 :option:`thinktime_blocks` and :option:`thinktime_spin`.
1974 .. option:: thinktime_spin=time
1976 Only valid if :option:`thinktime` is set - pretend to spend CPU time doing
1977 something with the data received, before falling back to sleeping for the
1978 rest of the period specified by :option:`thinktime`. When the unit is
1979 omitted, the value is given in microseconds.
1981 .. option:: thinktime_blocks=int
1983 Only valid if :option:`thinktime` is set - control how many blocks to issue,
1984 before waiting `thinktime` usecs. If not set, defaults to 1 which will make
1985 fio wait `thinktime` usecs after every block. This effectively makes any
1986 queue depth setting redundant, since no more than 1 I/O will be queued
1987 before we have to complete it and do our thinktime. In other words, this
1988 setting effectively caps the queue depth if the latter is larger.
1990 .. option:: rate=int[,int][,int]
1992 Cap the bandwidth used by this job. The number is in bytes/sec, the normal
1993 suffix rules apply. Comma-separated values may be specified for reads,
1994 writes, and trims as described in :option:`blocksize`.
1996 .. option:: rate_min=int[,int][,int]
1998 Tell fio to do whatever it can to maintain at least this bandwidth. Failing
1999 to meet this requirement will cause the job to exit. Comma-separated values
2000 may be specified for reads, writes, and trims as described in
2001 :option:`blocksize`.
2003 .. option:: rate_iops=int[,int][,int]
2005 Cap the bandwidth to this number of IOPS. Basically the same as
2006 :option:`rate`, just specified independently of bandwidth. If the job is
2007 given a block size range instead of a fixed value, the smallest block size
2008 is used as the metric. Comma-separated values may be specified for reads,
2009 writes, and trims as described in :option:`blocksize`.
2011 .. option:: rate_iops_min=int[,int][,int]
2013 If fio doesn't meet this rate of I/O, it will cause the job to exit.
2014 Comma-separated values may be specified for reads, writes, and trims as
2015 described in :option:`blocksize`.
2017 .. option:: rate_process=str
2019 This option controls how fio manages rated I/O submissions. The default is
2020 `linear`, which submits I/O in a linear fashion with fixed delays between
2021 I/Os that gets adjusted based on I/O completion rates. If this is set to
2022 `poisson`, fio will submit I/O based on a more real world random request
2023 flow, known as the Poisson process
2024 (https://en.wikipedia.org/wiki/Poisson_point_process). The lambda will be
2025 10^6 / IOPS for the given workload.
2031 .. option:: latency_target=time
2033 If set, fio will attempt to find the max performance point that the given
2034 workload will run at while maintaining a latency below this target. When
2035 the unit is omitted, the value is given in microseconds. See
2036 :option:`latency_window` and :option:`latency_percentile`.
2038 .. option:: latency_window=time
2040 Used with :option:`latency_target` to specify the sample window that the job
2041 is run at varying queue depths to test the performance. When the unit is
2042 omitted, the value is given in microseconds.
2044 .. option:: latency_percentile=float
2046 The percentage of I/Os that must fall within the criteria specified by
2047 :option:`latency_target` and :option:`latency_window`. If not set, this
2048 defaults to 100.0, meaning that all I/Os must be equal or below to the value
2049 set by :option:`latency_target`.
2051 .. option:: max_latency=time
2053 If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
2054 maximum latency. When the unit is omitted, the value is given in
2057 .. option:: rate_cycle=int
2059 Average bandwidth for :option:`rate` and :option:`rate_min` over this number
2066 .. option:: write_iolog=str
2068 Write the issued I/O patterns to the specified file. See
2069 :option:`read_iolog`. Specify a separate file for each job, otherwise the
2070 iologs will be interspersed and the file may be corrupt.
2072 .. option:: read_iolog=str
2074 Open an iolog with the specified file name and replay the I/O patterns it
2075 contains. This can be used to store a workload and replay it sometime
2076 later. The iolog given may also be a blktrace binary file, which allows fio
2077 to replay a workload captured by :command:`blktrace`. See
2078 :manpage:`blktrace(8)` for how to capture such logging data. For blktrace
2079 replay, the file needs to be turned into a blkparse binary data file first
2080 (``blkparse <device> -o /dev/null -d file_for_fio.bin``).
2082 .. option:: replay_no_stall=int
2084 When replaying I/O with :option:`read_iolog` the default behavior is to
2085 attempt to respect the time stamps within the log and replay them with the
2086 appropriate delay between IOPS. By setting this variable fio will not
2087 respect the timestamps and attempt to replay them as fast as possible while
2088 still respecting ordering. The result is the same I/O pattern to a given
2089 device, but different timings.
2091 .. option:: replay_redirect=str
2093 While replaying I/O patterns using :option:`read_iolog` the default behavior
2094 is to replay the IOPS onto the major/minor device that each IOP was recorded
2095 from. This is sometimes undesirable because on a different machine those
2096 major/minor numbers can map to a different device. Changing hardware on the
2097 same system can also result in a different major/minor mapping.
2098 ``replay_redirect`` causes all IOPS to be replayed onto the single specified
2099 device regardless of the device it was recorded
2100 from. i.e. :option:`replay_redirect` = :file:`/dev/sdc` would cause all I/O
2101 in the blktrace or iolog to be replayed onto :file:`/dev/sdc`. This means
2102 multiple devices will be replayed onto a single device, if the trace
2103 contains multiple devices. If you want multiple devices to be replayed
2104 concurrently to multiple redirected devices you must blkparse your trace
2105 into separate traces and replay them with independent fio invocations.
2106 Unfortunately this also breaks the strict time ordering between multiple
2109 .. option:: replay_align=int
2111 Force alignment of I/O offsets and lengths in a trace to this power of 2
2114 .. option:: replay_scale=int
2116 Scale sector offsets down by this factor when replaying traces.
2119 Threads, processes and job synchronization
2120 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2124 Fio defaults to forking jobs, however if this option is given, fio will use
2125 POSIX Threads function :manpage:`pthread_create(3)` to create threads instead
2126 of forking processes.
2128 .. option:: wait_for=str
2130 Specifies the name of the already defined job to wait for. Single waitee
2131 name only may be specified. If set, the job won't be started until all
2132 workers of the waitee job are done.
2134 ``wait_for`` operates on the job name basis, so there are a few
2135 limitations. First, the waitee must be defined prior to the waiter job
2136 (meaning no forward references). Second, if a job is being referenced as a
2137 waitee, it must have a unique name (no duplicate waitees).
2139 .. option:: nice=int
2141 Run the job with the given nice value. See man :manpage:`nice(2)`.
2143 On Windows, values less than -15 set the process class to "High"; -1 through
2144 -15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2147 .. option:: prio=int
2149 Set the I/O priority value of this job. Linux limits us to a positive value
2150 between 0 and 7, with 0 being the highest. See man
2151 :manpage:`ionice(1)`. Refer to an appropriate manpage for other operating
2152 systems since meaning of priority may differ.
2154 .. option:: prioclass=int
2156 Set the I/O priority class. See man :manpage:`ionice(1)`.
2158 .. option:: cpumask=int
2160 Set the CPU affinity of this job. The parameter given is a bitmask of
2161 allowed CPU's the job may run on. So if you want the allowed CPUs to be 1
2162 and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2163 :manpage:`sched_setaffinity(2)`. This may not work on all supported
2164 operating systems or kernel versions. This option doesn't work well for a
2165 higher CPU count than what you can store in an integer mask, so it can only
2166 control cpus 1-32. For boxes with larger CPU counts, use
2167 :option:`cpus_allowed`.
2169 .. option:: cpus_allowed=str
2171 Controls the same options as :option:`cpumask`, but it allows a text setting
2172 of the permitted CPUs instead. So to use CPUs 1 and 5, you would specify
2173 ``cpus_allowed=1,5``. This options also allows a range of CPUs. Say you
2174 wanted a binding to CPUs 1, 5, and 8-15, you would set
2175 ``cpus_allowed=1,5,8-15``.
2177 .. option:: cpus_allowed_policy=str
2179 Set the policy of how fio distributes the CPUs specified by
2180 :option:`cpus_allowed` or cpumask. Two policies are supported:
2183 All jobs will share the CPU set specified.
2185 Each job will get a unique CPU from the CPU set.
2187 **shared** is the default behaviour, if the option isn't specified. If
2188 **split** is specified, then fio will will assign one cpu per job. If not
2189 enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2192 .. option:: numa_cpu_nodes=str
2194 Set this job running on specified NUMA nodes' CPUs. The arguments allow
2195 comma delimited list of cpu numbers, A-B ranges, or `all`. Note, to enable
2196 numa options support, fio must be built on a system with libnuma-dev(el)
2199 .. option:: numa_mem_policy=str
2201 Set this job's memory policy and corresponding NUMA nodes. Format of the
2206 ``mode`` is one of the following memory policy: ``default``, ``prefer``,
2207 ``bind``, ``interleave``, ``local`` For ``default`` and ``local`` memory
2208 policy, no node is needed to be specified. For ``prefer``, only one node is
2209 allowed. For ``bind`` and ``interleave``, it allow comma delimited list of
2210 numbers, A-B ranges, or `all`.
2212 .. option:: cgroup=str
2214 Add job to this control group. If it doesn't exist, it will be created. The
2215 system must have a mounted cgroup blkio mount point for this to work. If
2216 your system doesn't have it mounted, you can do so with::
2218 # mount -t cgroup -o blkio none /cgroup
2220 .. option:: cgroup_weight=int
2222 Set the weight of the cgroup to this value. See the documentation that comes
2223 with the kernel, allowed values are in the range of 100..1000.
2225 .. option:: cgroup_nodelete=bool
2227 Normally fio will delete the cgroups it has created after the job
2228 completion. To override this behavior and to leave cgroups around after the
2229 job completion, set ``cgroup_nodelete=1``. This can be useful if one wants
2230 to inspect various cgroup files after job completion. Default: false.
2232 .. option:: flow_id=int
2234 The ID of the flow. If not specified, it defaults to being a global
2235 flow. See :option:`flow`.
2237 .. option:: flow=int
2239 Weight in token-based flow control. If this value is used, then there is a
2240 'flow counter' which is used to regulate the proportion of activity between
2241 two or more jobs. Fio attempts to keep this flow counter near zero. The
2242 ``flow`` parameter stands for how much should be added or subtracted to the
2243 flow counter on each iteration of the main I/O loop. That is, if one job has
2244 ``flow=8`` and another job has ``flow=-1``, then there will be a roughly 1:8
2245 ratio in how much one runs vs the other.
2247 .. option:: flow_watermark=int
2249 The maximum value that the absolute value of the flow counter is allowed to
2250 reach before the job must wait for a lower value of the counter.
2252 .. option:: flow_sleep=int
2254 The period of time, in microseconds, to wait after the flow watermark has
2255 been exceeded before retrying operations.
2257 .. option:: stonewall, wait_for_previous
2259 Wait for preceding jobs in the job file to exit, before starting this
2260 one. Can be used to insert serialization points in the job file. A stone
2261 wall also implies starting a new reporting group, see
2262 :option:`group_reporting`.
2266 When one job finishes, terminate the rest. The default is to wait for each
2267 job to finish, sometimes that is not the desired action.
2269 .. option:: exec_prerun=str
2271 Before running this job, issue the command specified through
2272 :manpage:`system(3)`. Output is redirected in a file called
2273 :file:`jobname.prerun.txt`.
2275 .. option:: exec_postrun=str
2277 After the job completes, issue the command specified though
2278 :manpage:`system(3)`. Output is redirected in a file called
2279 :file:`jobname.postrun.txt`.
2283 Instead of running as the invoking user, set the user ID to this value
2284 before the thread/process does any work.
2288 Set group ID, see :option:`uid`.
2294 .. option:: verify_only
2296 Do not perform specified workload, only verify data still matches previous
2297 invocation of this workload. This option allows one to check data multiple
2298 times at a later date without overwriting it. This option makes sense only
2299 for workloads that write data, and does not support workloads with the
2300 :option:`time_based` option set.
2302 .. option:: do_verify=bool
2304 Run the verify phase after a write phase. Only valid if :option:`verify` is
2307 .. option:: verify=str
2309 If writing to a file, fio can verify the file contents after each iteration
2310 of the job. Each verification method also implies verification of special
2311 header, which is written to the beginning of each block. This header also
2312 includes meta information, like offset of the block, block number, timestamp
2313 when block was written, etc. :option:`verify` can be combined with
2314 :option:`verify_pattern` option. The allowed values are:
2317 Use an md5 sum of the data area and store it in the header of
2321 Use an experimental crc64 sum of the data area and store it in the
2322 header of each block.
2325 Use a crc32c sum of the data area and store it in the header of each
2329 Use hardware assisted crc32c calculation provided on SSE4.2 enabled
2330 processors. Falls back to regular software crc32c, if not supported
2334 Use a crc32 sum of the data area and store it in the header of each
2338 Use a crc16 sum of the data area and store it in the header of each
2342 Use a crc7 sum of the data area and store it in the header of each
2346 Use xxhash as the checksum function. Generally the fastest software
2347 checksum that fio supports.
2350 Use sha512 as the checksum function.
2353 Use sha256 as the checksum function.
2356 Use optimized sha1 as the checksum function.
2359 This option is deprecated, since now meta information is included in
2360 generic verification header and meta verification happens by
2361 default. For detailed information see the description of the
2362 :option:`verify` setting. This option is kept because of
2363 compatibility's sake with old configurations. Do not use it.
2366 Verify a strict pattern. Normally fio includes a header with some
2367 basic information and checksumming, but if this option is set, only
2368 the specific pattern set with :option:`verify_pattern` is verified.
2371 Only pretend to verify. Useful for testing internals with
2372 :option:`ioengine` `=null`, not for much else.
2374 This option can be used for repeated burn-in tests of a system to make sure
2375 that the written data is also correctly read back. If the data direction
2376 given is a read or random read, fio will assume that it should verify a
2377 previously written file. If the data direction includes any form of write,
2378 the verify will be of the newly written data.
2380 .. option:: verifysort=bool
2382 If true, fio will sort written verify blocks when it deems it faster to read
2383 them back in a sorted manner. This is often the case when overwriting an
2384 existing file, since the blocks are already laid out in the file system. You
2385 can ignore this option unless doing huge amounts of really fast I/O where
2386 the red-black tree sorting CPU time becomes significant. Default: true.
2388 .. option:: verifysort_nr=int
2390 Pre-load and sort verify blocks for a read workload.
2392 .. option:: verify_offset=int
2394 Swap the verification header with data somewhere else in the block before
2395 writing. It is swapped back before verifying.
2397 .. option:: verify_interval=int
2399 Write the verification header at a finer granularity than the
2400 :option:`blocksize`. It will be written for chunks the size of
2401 ``verify_interval``. :option:`blocksize` should divide this evenly.
2403 .. option:: verify_pattern=str
2405 If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2406 filling with totally random bytes, but sometimes it's interesting to fill
2407 with a known pattern for I/O verification purposes. Depending on the width
2408 of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time(it can
2409 be either a decimal or a hex number). The ``verify_pattern`` if larger than
2410 a 32-bit quantity has to be a hex number that starts with either "0x" or
2411 "0X". Use with :option:`verify`. Also, ``verify_pattern`` supports %o
2412 format, which means that for each block offset will be written and then
2413 verified back, e.g.::
2417 Or use combination of everything::
2419 verify_pattern=0xff%o"abcd"-12
2421 .. option:: verify_fatal=bool
2423 Normally fio will keep checking the entire contents before quitting on a
2424 block verification failure. If this option is set, fio will exit the job on
2425 the first observed failure. Default: false.
2427 .. option:: verify_dump=bool
2429 If set, dump the contents of both the original data block and the data block
2430 we read off disk to files. This allows later analysis to inspect just what
2431 kind of data corruption occurred. Off by default.
2433 .. option:: verify_async=int
2435 Fio will normally verify I/O inline from the submitting thread. This option
2436 takes an integer describing how many async offload threads to create for I/O
2437 verification instead, causing fio to offload the duty of verifying I/O
2438 contents to one or more separate threads. If using this offload option, even
2439 sync I/O engines can benefit from using an :option:`iodepth` setting higher
2440 than 1, as it allows them to have I/O in flight while verifies are running.
2442 .. option:: verify_async_cpus=str
2444 Tell fio to set the given CPU affinity on the async I/O verification
2445 threads. See :option:`cpus_allowed` for the format used.
2447 .. option:: verify_backlog=int
2449 Fio will normally verify the written contents of a job that utilizes verify
2450 once that job has completed. In other words, everything is written then
2451 everything is read back and verified. You may want to verify continually
2452 instead for a variety of reasons. Fio stores the meta data associated with
2453 an I/O block in memory, so for large verify workloads, quite a bit of memory
2454 would be used up holding this meta data. If this option is enabled, fio will
2455 write only N blocks before verifying these blocks.
2457 .. option:: verify_backlog_batch=int
2459 Control how many blocks fio will verify if :option:`verify_backlog` is
2460 set. If not set, will default to the value of :option:`verify_backlog`
2461 (meaning the entire queue is read back and verified). If
2462 ``verify_backlog_batch`` is less than :option:`verify_backlog` then not all
2463 blocks will be verified, if ``verify_backlog_batch`` is larger than
2464 :option:`verify_backlog`, some blocks will be verified more than once.
2466 .. option:: verify_state_save=bool
2468 When a job exits during the write phase of a verify workload, save its
2469 current state. This allows fio to replay up until that point, if the verify
2470 state is loaded for the verify read phase. The format of the filename is,
2473 <type>-<jobname>-<jobindex>-verify.state.
2475 <type> is "local" for a local run, "sock" for a client/server socket
2476 connection, and "ip" (192.168.0.1, for instance) for a networked
2477 client/server connection.
2479 .. option:: verify_state_load=bool
2481 If a verify termination trigger was used, fio stores the current write state
2482 of each thread. This can be used at verification time so that fio knows how
2483 far it should verify. Without this information, fio will run a full
2484 verification pass, according to the settings in the job file used.
2486 .. option:: trim_percentage=int
2488 Number of verify blocks to discard/trim.
2490 .. option:: trim_verify_zero=bool
2492 Verify that trim/discarded blocks are returned as zeroes.
2494 .. option:: trim_backlog=int
2496 Verify that trim/discarded blocks are returned as zeroes.
2498 .. option:: trim_backlog_batch=int
2500 Trim this number of I/O blocks.
2502 .. option:: experimental_verify=bool
2504 Enable experimental verification.
2510 .. option:: steadystate=str:float, ss=str:float
2512 Define the criterion and limit for assessing steady state performance. The
2513 first parameter designates the criterion whereas the second parameter sets
2514 the threshold. When the criterion falls below the threshold for the
2515 specified duration, the job will stop. For example, `iops_slope:0.1%` will
2516 direct fio to terminate the job when the least squares regression slope
2517 falls below 0.1% of the mean IOPS. If :option:`group_reporting` is enabled
2518 this will apply to all jobs in the group. Below is the list of available
2519 steady state assessment criteria. All assessments are carried out using only
2520 data from the rolling collection window. Threshold limits can be expressed
2521 as a fixed value or as a percentage of the mean in the collection window.
2524 Collect IOPS data. Stop the job if all individual IOPS measurements
2525 are within the specified limit of the mean IOPS (e.g., ``iops:2``
2526 means that all individual IOPS values must be within 2 of the mean,
2527 whereas ``iops:0.2%`` means that all individual IOPS values must be
2528 within 0.2% of the mean IOPS to terminate the job).
2531 Collect IOPS data and calculate the least squares regression
2532 slope. Stop the job if the slope falls below the specified limit.
2535 Collect bandwidth data. Stop the job if all individual bandwidth
2536 measurements are within the specified limit of the mean bandwidth.
2539 Collect bandwidth data and calculate the least squares regression
2540 slope. Stop the job if the slope falls below the specified limit.
2542 .. option:: steadystate_duration=time, ss_dur=time
2544 A rolling window of this duration will be used to judge whether steady state
2545 has been reached. Data will be collected once per second. The default is 0
2546 which disables steady state detection. When the unit is omitted, the
2547 value is given in seconds.
2549 .. option:: steadystate_ramp_time=time, ss_ramp=time
2551 Allow the job to run for the specified duration before beginning data
2552 collection for checking the steady state job termination criterion. The
2553 default is 0. When the unit is omitted, the value is given in seconds.
2556 Measurements and reporting
2557 ~~~~~~~~~~~~~~~~~~~~~~~~~~
2559 .. option:: per_job_logs=bool
2561 If set, this generates bw/clat/iops log with per file private filenames. If
2562 not set, jobs with identical names will share the log filename. Default:
2565 .. option:: group_reporting
2567 It may sometimes be interesting to display statistics for groups of jobs as
2568 a whole instead of for each individual job. This is especially true if
2569 :option:`numjobs` is used; looking at individual thread/process output
2570 quickly becomes unwieldy. To see the final report per-group instead of
2571 per-job, use :option:`group_reporting`. Jobs in a file will be part of the
2572 same reporting group, unless if separated by a :option:`stonewall`, or by
2573 using :option:`new_group`.
2575 .. option:: new_group
2577 Start a new reporting group. See: :option:`group_reporting`. If not given,
2578 all jobs in a file will be part of the same reporting group, unless
2579 separated by a :option:`stonewall`.
2581 .. option:: write_bw_log=str
2583 If given, write a bandwidth log for this job. Can be used to store data of
2584 the bandwidth of the jobs in their lifetime. The included
2585 :command:`fio_generate_plots` script uses :command:`gnuplot` to turn these
2586 text files into nice graphs. See :option:`write_lat_log` for behaviour of
2587 given filename. For this option, the postfix is :file:`_bw.x.log`, where `x`
2588 is the index of the job (`1..N`, where `N` is the number of jobs). If
2589 :option:`per_job_logs` is false, then the filename will not include the job
2590 index. See `Log File Formats`_.
2592 .. option:: write_lat_log=str
2594 Same as :option:`write_bw_log`, except that this option stores I/O
2595 submission, completion, and total latencies instead. If no filename is given
2596 with this option, the default filename of :file:`jobname_type.log` is
2597 used. Even if the filename is given, fio will still append the type of
2598 log. So if one specifies::
2602 The actual log names will be :file:`foo_slat.x.log`, :file:`foo_clat.x.log`,
2603 and :file:`foo_lat.x.log`, where `x` is the index of the job (1..N, where N
2604 is the number of jobs). This helps :command:`fio_generate_plot` find the
2605 logs automatically. If :option:`per_job_logs` is false, then the filename
2606 will not include the job index. See `Log File Formats`_.
2608 .. option:: write_hist_log=str
2610 Same as :option:`write_lat_log`, but writes I/O completion latency
2611 histograms. If no filename is given with this option, the default filename
2612 of :file:`jobname_clat_hist.x.log` is used, where `x` is the index of the
2613 job (1..N, where `N` is the number of jobs). Even if the filename is given,
2614 fio will still append the type of log. If :option:`per_job_logs` is false,
2615 then the filename will not include the job index. See `Log File Formats`_.
2617 .. option:: write_iops_log=str
2619 Same as :option:`write_bw_log`, but writes IOPS. If no filename is given
2620 with this option, the default filename of :file:`jobname_type.x.log` is
2621 used,where `x` is the index of the job (1..N, where `N` is the number of
2622 jobs). Even if the filename is given, fio will still append the type of
2623 log. If :option:`per_job_logs` is false, then the filename will not include
2624 the job index. See `Log File Formats`_.
2626 .. option:: log_avg_msec=int
2628 By default, fio will log an entry in the iops, latency, or bw log for every
2629 I/O that completes. When writing to the disk log, that can quickly grow to a
2630 very large size. Setting this option makes fio average the each log entry
2631 over the specified period of time, reducing the resolution of the log. See
2632 :option:`log_max_value` as well. Defaults to 0, logging all entries.
2634 .. option:: log_hist_msec=int
2636 Same as :option:`log_avg_msec`, but logs entries for completion latency
2637 histograms. Computing latency percentiles from averages of intervals using
2638 :option:`log_avg_msec` is inaccurate. Setting this option makes fio log
2639 histogram entries over the specified period of time, reducing log sizes for
2640 high IOPS devices while retaining percentile accuracy. See
2641 :option:`log_hist_coarseness` as well. Defaults to 0, meaning histogram
2642 logging is disabled.
2644 .. option:: log_hist_coarseness=int
2646 Integer ranging from 0 to 6, defining the coarseness of the resolution of
2647 the histogram logs enabled with :option:`log_hist_msec`. For each increment
2648 in coarseness, fio outputs half as many bins. Defaults to 0, for which
2649 histogram logs contain 1216 latency bins. See `Log File Formats`_.
2651 .. option:: log_max_value=bool
2653 If :option:`log_avg_msec` is set, fio logs the average over that window. If
2654 you instead want to log the maximum value, set this option to 1. Defaults to
2655 0, meaning that averaged values are logged.
2657 .. option:: log_offset=int
2659 If this is set, the iolog options will include the byte offset for the I/O
2660 entry as well as the other data values.
2662 .. option:: log_compression=int
2664 If this is set, fio will compress the I/O logs as it goes, to keep the
2665 memory footprint lower. When a log reaches the specified size, that chunk is
2666 removed and compressed in the background. Given that I/O logs are fairly
2667 highly compressible, this yields a nice memory savings for longer runs. The
2668 downside is that the compression will consume some background CPU cycles, so
2669 it may impact the run. This, however, is also true if the logging ends up
2670 consuming most of the system memory. So pick your poison. The I/O logs are
2671 saved normally at the end of a run, by decompressing the chunks and storing
2672 them in the specified log file. This feature depends on the availability of
2675 .. option:: log_compression_cpus=str
2677 Define the set of CPUs that are allowed to handle online log compression for
2678 the I/O jobs. This can provide better isolation between performance
2679 sensitive jobs, and background compression work.
2681 .. option:: log_store_compressed=bool
2683 If set, fio will store the log files in a compressed format. They can be
2684 decompressed with fio, using the :option:`--inflate-log` command line
2685 parameter. The files will be stored with a :file:`.fz` suffix.
2687 .. option:: log_unix_epoch=bool
2689 If set, fio will log Unix timestamps to the log files produced by enabling
2690 write_type_log for each log type, instead of the default zero-based
2693 .. option:: block_error_percentiles=bool
2695 If set, record errors in trim block-sized units from writes and trims and
2696 output a histogram of how many trims it took to get to errors, and what kind
2697 of error was encountered.
2699 .. option:: bwavgtime=int
2701 Average the calculated bandwidth over the given time. Value is specified in
2702 milliseconds. If the job also does bandwidth logging through
2703 :option:`write_bw_log`, then the minimum of this option and
2704 :option:`log_avg_msec` will be used. Default: 500ms.
2706 .. option:: iopsavgtime=int
2708 Average the calculated IOPS over the given time. Value is specified in
2709 milliseconds. If the job also does IOPS logging through
2710 :option:`write_iops_log`, then the minimum of this option and
2711 :option:`log_avg_msec` will be used. Default: 500ms.
2713 .. option:: disk_util=bool
2715 Generate disk utilization statistics, if the platform supports it.
2718 .. option:: disable_lat=bool
2720 Disable measurements of total latency numbers. Useful only for cutting back
2721 the number of calls to :manpage:`gettimeofday(2)`, as that does impact
2722 performance at really high IOPS rates. Note that to really get rid of a
2723 large amount of these calls, this option must be used with
2724 :option:`disable_slat` and :option:`disable_bw_measurement` as well.
2726 .. option:: disable_clat=bool
2728 Disable measurements of completion latency numbers. See
2729 :option:`disable_lat`.
2731 .. option:: disable_slat=bool
2733 Disable measurements of submission latency numbers. See
2734 :option:`disable_slat`.
2736 .. option:: disable_bw_measurement=bool, disable_bw=bool
2738 Disable measurements of throughput/bandwidth numbers. See
2739 :option:`disable_lat`.
2741 .. option:: clat_percentiles=bool
2743 Enable the reporting of percentiles of completion latencies.
2745 .. option:: percentile_list=float_list
2747 Overwrite the default list of percentiles for completion latencies and the
2748 block error histogram. Each number is a floating number in the range
2749 (0,100], and the maximum length of the list is 20. Use ``:`` to separate the
2750 numbers, and list the numbers in ascending order. For example,
2751 ``--percentile_list=99.5:99.9`` will cause fio to report the values of
2752 completion latency below which 99.5% and 99.9% of the observed latencies
2759 .. option:: exitall_on_error
2761 When one job finishes in error, terminate the rest. The default is to wait
2762 for each job to finish.
2764 .. option:: continue_on_error=str
2766 Normally fio will exit the job on the first observed failure. If this option
2767 is set, fio will continue the job when there is a 'non-fatal error' (EIO or
2768 EILSEQ) until the runtime is exceeded or the I/O size specified is
2769 completed. If this option is used, there are two more stats that are
2770 appended, the total error count and the first error. The error field given
2771 in the stats is the first error that was hit during the run.
2773 The allowed values are:
2776 Exit on any I/O or verify errors.
2779 Continue on read errors, exit on all others.
2782 Continue on write errors, exit on all others.
2785 Continue on any I/O error, exit on all others.
2788 Continue on verify errors, exit on all others.
2791 Continue on all errors.
2794 Backward-compatible alias for 'none'.
2797 Backward-compatible alias for 'all'.
2799 .. option:: ignore_error=str
2801 Sometimes you want to ignore some errors during test in that case you can
2802 specify error list for each error type.
2803 ``ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST`` errors for
2804 given error type is separated with ':'. Error may be symbol ('ENOSPC',
2805 'ENOMEM') or integer. Example::
2807 ignore_error=EAGAIN,ENOSPC:122
2809 This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
2812 .. option:: error_dump=bool
2814 If set dump every error even if it is non fatal, true by default. If
2815 disabled only fatal error will be dumped.
2817 Running predefined workloads
2818 ----------------------------
2820 Fio includes predefined profiles that mimic the I/O workloads generated by
2823 .. option:: profile=str
2825 The predefined workload to run. Current profiles are:
2828 Threaded I/O bench (tiotest/tiobench) like workload.
2831 Aerospike Certification Tool (ACT) like workload.
2833 To view a profile's additional options use :option:`--cmdhelp` after specifying
2834 the profile. For example::
2836 $ fio --profile=act --cmdhelp
2841 .. option:: device-names=str
2846 .. option:: load=int
2849 ACT load multiplier. Default: 1.
2851 .. option:: test-duration=time
2854 How long the entire test takes to run. Default: 24h.
2856 .. option:: threads-per-queue=int
2859 Number of read IO threads per device. Default: 8.
2861 .. option:: read-req-num-512-blocks=int
2864 Number of 512B blocks to read at the time. Default: 3.
2866 .. option:: large-block-op-kbytes=int
2869 Size of large block ops in KiB (writes). Default: 131072.
2874 Set to run ACT prep phase.
2876 Tiobench profile options
2877 ~~~~~~~~~~~~~~~~~~~~~~~~
2879 .. option:: size=str
2884 .. option:: block=int
2887 Block size in bytes. Default: 4096.
2889 .. option:: numruns=int
2899 .. option:: threads=int
2904 Interpreting the output
2905 -----------------------
2907 Fio spits out a lot of output. While running, fio will display the status of the
2908 jobs created. An example of that would be::
2910 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]
2912 The characters inside the square brackets denote the current status of each
2913 thread. The possible values (in typical life cycle order) are:
2915 +------+-----+-----------------------------------------------------------+
2917 +======+=====+===========================================================+
2918 | P | | Thread setup, but not started. |
2919 +------+-----+-----------------------------------------------------------+
2920 | C | | Thread created. |
2921 +------+-----+-----------------------------------------------------------+
2922 | I | | Thread initialized, waiting or generating necessary data. |
2923 +------+-----+-----------------------------------------------------------+
2924 | | p | Thread running pre-reading file(s). |
2925 +------+-----+-----------------------------------------------------------+
2926 | | R | Running, doing sequential reads. |
2927 +------+-----+-----------------------------------------------------------+
2928 | | r | Running, doing random reads. |
2929 +------+-----+-----------------------------------------------------------+
2930 | | W | Running, doing sequential writes. |
2931 +------+-----+-----------------------------------------------------------+
2932 | | w | Running, doing random writes. |
2933 +------+-----+-----------------------------------------------------------+
2934 | | M | Running, doing mixed sequential reads/writes. |
2935 +------+-----+-----------------------------------------------------------+
2936 | | m | Running, doing mixed random reads/writes. |
2937 +------+-----+-----------------------------------------------------------+
2938 | | F | Running, currently waiting for :manpage:`fsync(2)` |
2939 +------+-----+-----------------------------------------------------------+
2940 | | V | Running, doing verification of written data. |
2941 +------+-----+-----------------------------------------------------------+
2942 | E | | Thread exited, not reaped by main thread yet. |
2943 +------+-----+-----------------------------------------------------------+
2944 | _ | | Thread reaped, or |
2945 +------+-----+-----------------------------------------------------------+
2946 | X | | Thread reaped, exited with an error. |
2947 +------+-----+-----------------------------------------------------------+
2948 | K | | Thread reaped, exited due to signal. |
2949 +------+-----+-----------------------------------------------------------+
2951 Fio will condense the thread string as not to take up more space on the command
2952 line as is needed. For instance, if you have 10 readers and 10 writers running,
2953 the output would look like this::
2955 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]
2957 Fio will still maintain the ordering, though. So the above means that jobs 1..10
2958 are readers, and 11..20 are writers.
2960 The other values are fairly self explanatory -- number of threads currently
2961 running and doing I/O, the number of currently open files (f=), the rate of I/O
2962 since last check (read speed listed first, then write speed and optionally trim
2963 speed), and the estimated completion percentage and time for the current
2964 running group. It's impossible to estimate runtime of the following groups (if
2965 any). Note that the string is displayed in order, so it's possible to tell which
2966 of the jobs are currently doing what. The first character is the first job
2967 defined in the job file, and so forth.
2969 When fio is done (or interrupted by :kbd:`ctrl-c`), it will show the data for
2970 each thread, group of threads, and disks in that order. For each data direction,
2971 the output looks like::
2973 Client1 (g=0): err= 0:
2974 write: io= 32MiB, bw= 666KiB/s, iops=89 , runt= 50320msec
2975 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
2976 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
2977 bw (KiB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
2978 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
2979 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
2980 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2981 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2982 issued r/w: total=0/32768, short=0/0
2983 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
2984 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
2986 The client number is printed, along with the group id and error of that
2987 thread. Below is the I/O statistics, here for writes. In the order listed, they
2991 Number of megabytes I/O performed.
2994 Average bandwidth rate.
2997 Average I/Os performed per second.
3000 The runtime of that thread.
3003 Submission latency (avg being the average, stdev being the standard
3004 deviation). This is the time it took to submit the I/O. For sync I/O,
3005 the slat is really the completion latency, since queue/complete is one
3006 operation there. This value can be in milliseconds or microseconds, fio
3007 will choose the most appropriate base and print that. In the example
3008 above, milliseconds is the best scale. Note: in :option:`--minimal` mode
3009 latencies are always expressed in microseconds.
3012 Completion latency. Same names as slat, this denotes the time from
3013 submission to completion of the I/O pieces. For sync I/O, clat will
3014 usually be equal (or very close) to 0, as the time from submit to
3015 complete is basically just CPU time (I/O has already been done, see slat
3019 Bandwidth. Same names as the xlat stats, but also includes an
3020 approximate percentage of total aggregate bandwidth this thread received
3021 in this group. This last value is only really useful if the threads in
3022 this group are on the same disk, since they are then competing for disk
3026 CPU usage. User and system time, along with the number of context
3027 switches this thread went through, usage of system and user time, and
3028 finally the number of major and minor page faults. The CPU utilization
3029 numbers are averages for the jobs in that reporting group, while the
3030 context and fault counters are summed.
3033 The distribution of I/O depths over the job life time. The numbers are
3034 divided into powers of 2, so for example the 16= entries includes depths
3035 up to that value but higher than the previous entry. In other words, it
3036 covers the range from 16 to 31.
3039 How many pieces of I/O were submitting in a single submit call. Each
3040 entry denotes that amount and below, until the previous entry -- e.g.,
3041 8=100% mean that we submitted anywhere in between 5-8 I/Os per submit
3045 Like the above submit number, but for completions instead.
3048 The number of read/write requests issued, and how many of them were
3052 The distribution of I/O completion latencies. This is the time from when
3053 I/O leaves fio and when it gets completed. The numbers follow the same
3054 pattern as the I/O depths, meaning that 2=1.6% means that 1.6% of the
3055 I/O completed within 2 msecs, 20=12.8% means that 12.8% of the I/O took
3056 more than 10 msecs, but less than (or equal to) 20 msecs.
3058 After each client has been listed, the group statistics are printed. They
3059 will look like this::
3061 Run status group 0 (all jobs):
3062 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
3063 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
3065 For each data direction, it prints:
3068 Number of megabytes I/O performed.
3070 Aggregate bandwidth of threads in this group.
3072 The minimum average bandwidth a thread saw.
3074 The maximum average bandwidth a thread saw.
3076 The smallest runtime of the threads in that group.
3078 The longest runtime of the threads in that group.
3080 And finally, the disk statistics are printed. They will look like this::
3082 Disk stats (read/write):
3083 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3085 Each value is printed for both reads and writes, with reads first. The
3089 Number of I/Os performed by all groups.
3091 Number of merges I/O the I/O scheduler.
3093 Number of ticks we kept the disk busy.
3095 Total time spent in the disk queue.
3097 The disk utilization. A value of 100% means we kept the disk
3098 busy constantly, 50% would be a disk idling half of the time.
3100 It is also possible to get fio to dump the current output while it is running,
3101 without terminating the job. To do that, send fio the **USR1** signal. You can
3102 also get regularly timed dumps by using the :option:`--status-interval`
3103 parameter, or by creating a file in :file:`/tmp` named
3104 :file:`fio-dump-status`. If fio sees this file, it will unlink it and dump the
3105 current output status.
3111 For scripted usage where you typically want to generate tables or graphs of the
3112 results, fio can output the results in a semicolon separated format. The format
3113 is one long line of values, such as::
3115 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%
3116 A description of this job goes here.
3118 The job description (if provided) follows on a second line.
3120 To enable terse output, use the :option:`--minimal` command line option. The
3121 first value is the version of the terse output format. If the output has to be
3122 changed for some reason, this number will be incremented by 1 to signify that
3125 Split up, the format is as follows:
3129 terse version, fio version, jobname, groupid, error
3133 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3134 Submission latency: min, max, mean, stdev (usec)
3135 Completion latency: min, max, mean, stdev (usec)
3136 Completion latency percentiles: 20 fields (see below)
3137 Total latency: min, max, mean, stdev (usec)
3138 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev
3144 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3145 Submission latency: min, max, mean, stdev (usec)
3146 Completion latency: min, max, mean, stdev(usec)
3147 Completion latency percentiles: 20 fields (see below)
3148 Total latency: min, max, mean, stdev (usec)
3149 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev
3153 user, system, context switches, major faults, minor faults
3157 <=1, 2, 4, 8, 16, 32, >=64
3159 I/O latencies microseconds::
3161 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3163 I/O latencies milliseconds::
3165 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3169 Disk name, Read ios, write ios,
3170 Read merges, write merges,
3171 Read ticks, write ticks,
3172 Time spent in queue, disk utilization percentage
3174 Additional Info (dependent on continue_on_error, default off)::
3176 total # errors, first error code
3178 Additional Info (dependent on description being set)::
3182 Completion latency percentiles can be a grouping of up to 20 sets, so for the
3183 terse output fio writes all of them. Each field will look like this::
3187 which is the Xth percentile, and the `usec` latency associated with it.
3189 For disk utilization, all disks used by fio are shown. So for each disk there
3190 will be a disk utilization section.
3196 There are two trace file format that you can encounter. The older (v1) format is
3197 unsupported since version 1.20-rc3 (March 2008). It will still be described
3198 below in case that you get an old trace and want to understand it.
3200 In any case the trace is a simple text file with a single action per line.
3203 Trace file format v1
3204 ~~~~~~~~~~~~~~~~~~~~
3206 Each line represents a single I/O action in the following format::
3210 where `rw=0/1` for read/write, and the offset and length entries being in bytes.
3212 This format is not supported in fio versions => 1.20-rc3.
3215 Trace file format v2
3216 ~~~~~~~~~~~~~~~~~~~~
3218 The second version of the trace file format was added in fio version 1.17. It
3219 allows to access more then one file per trace and has a bigger set of possible
3222 The first line of the trace file has to be::
3226 Following this can be lines in two different formats, which are described below.
3228 The file management format::
3232 The filename is given as an absolute path. The action can be one of these:
3235 Add the given filename to the trace.
3237 Open the file with the given filename. The filename has to have
3238 been added with the **add** action before.
3240 Close the file with the given filename. The file has to have been
3244 The file I/O action format::
3246 filename action offset length
3248 The `filename` is given as an absolute path, and has to have been added and
3249 opened before it can be used with this format. The `offset` and `length` are
3250 given in bytes. The `action` can be one of these:
3253 Wait for `offset` microseconds. Everything below 100 is discarded.
3254 The time is relative to the previous `wait` statement.
3256 Read `length` bytes beginning from `offset`.
3258 Write `length` bytes beginning from `offset`.
3260 :manpage:`fsync(2)` the file.
3262 :manpage:`fdatasync(2)` the file.
3264 Trim the given file from the given `offset` for `length` bytes.
3266 CPU idleness profiling
3267 ----------------------
3269 In some cases, we want to understand CPU overhead in a test. For example, we
3270 test patches for the specific goodness of whether they reduce CPU usage.
3271 Fio implements a balloon approach to create a thread per CPU that runs at idle
3272 priority, meaning that it only runs when nobody else needs the cpu.
3273 By measuring the amount of work completed by the thread, idleness of each CPU
3274 can be derived accordingly.
3276 An unit work is defined as touching a full page of unsigned characters. Mean and
3277 standard deviation of time to complete an unit work is reported in "unit work"
3278 section. Options can be chosen to report detailed percpu idleness or overall
3279 system idleness by aggregating percpu stats.
3282 Verification and triggers
3283 -------------------------
3285 Fio is usually run in one of two ways, when data verification is done. The first
3286 is a normal write job of some sort with verify enabled. When the write phase has
3287 completed, fio switches to reads and verifies everything it wrote. The second
3288 model is running just the write phase, and then later on running the same job
3289 (but with reads instead of writes) to repeat the same I/O patterns and verify
3290 the contents. Both of these methods depend on the write phase being completed,
3291 as fio otherwise has no idea how much data was written.
3293 With verification triggers, fio supports dumping the current write state to
3294 local files. Then a subsequent read verify workload can load this state and know
3295 exactly where to stop. This is useful for testing cases where power is cut to a
3296 server in a managed fashion, for instance.
3298 A verification trigger consists of two things:
3300 1) Storing the write state of each job.
3301 2) Executing a trigger command.
3303 The write state is relatively small, on the order of hundreds of bytes to single
3304 kilobytes. It contains information on the number of completions done, the last X
3307 A trigger is invoked either through creation ('touch') of a specified file in
3308 the system, or through a timeout setting. If fio is run with
3309 :option:`--trigger-file` = :file:`/tmp/trigger-file`, then it will continually
3310 check for the existence of :file:`/tmp/trigger-file`. When it sees this file, it
3311 will fire off the trigger (thus saving state, and executing the trigger
3314 For client/server runs, there's both a local and remote trigger. If fio is
3315 running as a server backend, it will send the job states back to the client for
3316 safe storage, then execute the remote trigger, if specified. If a local trigger
3317 is specified, the server will still send back the write state, but the client
3318 will then execute the trigger.
3320 Verification trigger example
3321 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3323 Lets say we want to run a powercut test on the remote machine 'server'. Our
3324 write workload is in :file:`write-test.fio`. We want to cut power to 'server' at
3325 some point during the run, and we'll run this test from the safety or our local
3326 machine, 'localbox'. On the server, we'll start the fio backend normally::
3328 server# fio --server
3330 and on the client, we'll fire off the workload::
3332 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
3334 We set :file:`/tmp/my-trigger` as the trigger file, and we tell fio to execute::
3336 echo b > /proc/sysrq-trigger
3338 on the server once it has received the trigger and sent us the write state. This
3339 will work, but it's not **really** cutting power to the server, it's merely
3340 abruptly rebooting it. If we have a remote way of cutting power to the server
3341 through IPMI or similar, we could do that through a local trigger command
3342 instead. Lets assume we have a script that does IPMI reboot of a given hostname,
3343 ipmi-reboot. On localbox, we could then have run fio with a local trigger
3346 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
3348 For this case, fio would wait for the server to send us the write state, then
3349 execute ``ipmi-reboot server`` when that happened.
3351 Loading verify state
3352 ~~~~~~~~~~~~~~~~~~~~
3354 To load store write state, read verification job file must contain the
3355 :option:`verify_state_load` option. If that is set, fio will load the previously
3356 stored state. For a local fio run this is done by loading the files directly,
3357 and on a client/server run, the server backend will ask the client to send the
3358 files over and load them from there.
3364 Fio supports a variety of log file formats, for logging latencies, bandwidth,
3365 and IOPS. The logs share a common format, which looks like this:
3367 *time* (`msec`), *value*, *data direction*, *offset*
3369 Time for the log entry is always in milliseconds. The *value* logged depends
3370 on the type of log, it will be one of the following:
3373 Value is latency in usecs
3379 *Data direction* is one of the following:
3388 The *offset* is the offset, in bytes, from the start of the file, for that
3389 particular I/O. The logging of the offset can be toggled with
3390 :option:`log_offset`.
3392 If windowed logging is enabled through :option:`log_avg_msec` then fio doesn't
3393 log individual I/Os. Instead of logs the average values over the specified period
3394 of time. Since 'data direction' and 'offset' are per-I/O values, they aren't
3395 applicable if windowed logging is enabled. If windowed logging is enabled and
3396 :option:`log_max_value` is set, then fio logs maximum values in that window
3397 instead of averages.
3403 Normally fio is invoked as a stand-alone application on the machine where the
3404 I/O workload should be generated. However, the frontend and backend of fio can
3405 be run separately. Ie the fio server can generate an I/O workload on the "Device
3406 Under Test" while being controlled from another machine.
3408 Start the server on the machine which has access to the storage DUT::
3412 where args defines what fio listens to. The arguments are of the form
3413 ``type,hostname`` or ``IP,port``. *type* is either ``ip`` (or ip4) for TCP/IP
3414 v4, ``ip6`` for TCP/IP v6, or ``sock`` for a local unix domain socket.
3415 *hostname* is either a hostname or IP address, and *port* is the port to listen
3416 to (only valid for TCP/IP, not a local socket). Some examples:
3420 Start a fio server, listening on all interfaces on the default port (8765).
3422 2) ``fio --server=ip:hostname,4444``
3424 Start a fio server, listening on IP belonging to hostname and on port 4444.
3426 3) ``fio --server=ip6:::1,4444``
3428 Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
3430 4) ``fio --server=,4444``
3432 Start a fio server, listening on all interfaces on port 4444.
3434 5) ``fio --server=1.2.3.4``
3436 Start a fio server, listening on IP 1.2.3.4 on the default port.
3438 6) ``fio --server=sock:/tmp/fio.sock``
3440 Start a fio server, listening on the local socket /tmp/fio.sock.
3442 Once a server is running, a "client" can connect to the fio server with::
3444 fio <local-args> --client=<server> <remote-args> <job file(s)>
3446 where `local-args` are arguments for the client where it is running, `server`
3447 is the connect string, and `remote-args` and `job file(s)` are sent to the
3448 server. The `server` string follows the same format as it does on the server
3449 side, to allow IP/hostname/socket and port strings.
3451 Fio can connect to multiple servers this way::
3453 fio --client=<server1> <job file(s)> --client=<server2> <job file(s)>
3455 If the job file is located on the fio server, then you can tell the server to
3456 load a local file as well. This is done by using :option:`--remote-config` ::
3458 fio --client=server --remote-config /path/to/file.fio
3460 Then fio will open this local (to the server) job file instead of being passed
3461 one from the client.
3463 If you have many servers (example: 100 VMs/containers), you can input a pathname
3464 of a file containing host IPs/names as the parameter value for the
3465 :option:`--client` option. For example, here is an example :file:`host.list`
3466 file containing 2 hostnames::
3468 host1.your.dns.domain
3469 host2.your.dns.domain
3471 The fio command would then be::
3473 fio --client=host.list <job file(s)>
3475 In this mode, you cannot input server-specific parameters or job files -- all
3476 servers receive the same job file.
3478 In order to let ``fio --client`` runs use a shared filesystem from multiple
3479 hosts, ``fio --client`` now prepends the IP address of the server to the
3480 filename. For example, if fio is using directory :file:`/mnt/nfs/fio` and is
3481 writing filename :file:`fileio.tmp`, with a :option:`--client` `hostfile`
3482 containing two hostnames ``h1`` and ``h2`` with IP addresses 192.168.10.120 and
3483 192.168.10.121, then fio will create two files::
3485 /mnt/nfs/fio/192.168.10.120.fileio.tmp
3486 /mnt/nfs/fio/192.168.10.121.fileio.tmp