Refactor #includes and headers
[fio.git] / fio.1
CommitLineData
523bad63 1.TH fio 1 "August 2017" "User Manual"
d60e92d1
AC
2.SH NAME
3fio \- flexible I/O tester
4.SH SYNOPSIS
5.B fio
6[\fIoptions\fR] [\fIjobfile\fR]...
7.SH DESCRIPTION
8.B fio
9is a tool that will spawn a number of threads or processes doing a
10particular type of I/O action as specified by the user.
11The typical use of fio is to write a job file matching the I/O load
12one wants to simulate.
13.SH OPTIONS
14.TP
49da1240 15.BI \-\-debug \fR=\fPtype
7db7a5a0
TK
16Enable verbose tracing \fItype\fR of various fio actions. May be `all' for all \fItype\fRs
17or individual types separated by a comma (e.g. `\-\-debug=file,mem' will enable
bdd88be3
TK
18file and memory debugging). `help' will list all available tracing options.
19.TP
7db7a5a0 20.BI \-\-parse\-only
bdd88be3 21Parse options only, don't start any I/O.
49da1240 22.TP
d60e92d1
AC
23.BI \-\-output \fR=\fPfilename
24Write output to \fIfilename\fR.
25.TP
7db7a5a0
TK
26.BI \-\-output\-format \fR=\fPformat
27Set the reporting \fIformat\fR to `normal', `terse', `json', or
28`json+'. Multiple formats can be selected, separate by a comma. `terse'
29is a CSV based format. `json+' is like `json', except it adds a full
513e37ee 30dump of the latency buckets.
e28ee21d 31.TP
7db7a5a0 32.BI \-\-bandwidth\-log
d23ae827 33Generate aggregate bandwidth logs.
d60e92d1 34.TP
7db7a5a0
TK
35.BI \-\-minimal
36Print statistics in a terse, semicolon\-delimited format.
d60e92d1 37.TP
7db7a5a0
TK
38.BI \-\-append\-terse
39Print statistics in selected mode AND terse, semicolon\-delimited format.
40\fBDeprecated\fR, use \fB\-\-output\-format\fR instead to select multiple formats.
f6a7df53 41.TP
065248bf 42.BI \-\-terse\-version \fR=\fPversion
7db7a5a0 43Set terse \fIversion\fR output format (default `3', or `2', `4', `5').
49da1240 44.TP
7db7a5a0 45.BI \-\-version
bdd88be3
TK
46Print version information and exit.
47.TP
7db7a5a0 48.BI \-\-help
bdd88be3 49Print a summary of the command line options and exit.
49da1240 50.TP
7db7a5a0 51.BI \-\-cpuclock\-test
bdd88be3 52Perform test and validation of internal CPU clock.
fec0f21c 53.TP
bdd88be3 54.BI \-\-crctest \fR=\fP[test]
7db7a5a0 55Test the speed of the built\-in checksumming functions. If no argument is given,
bdd88be3 56all of them are tested. Alternatively, a comma separated list can be passed, in which
fec0f21c
JA
57case the given ones are tested.
58.TP
49da1240 59.BI \-\-cmdhelp \fR=\fPcommand
bdd88be3 60Print help information for \fIcommand\fR. May be `all' for all commands.
49da1240 61.TP
7db7a5a0
TK
62.BI \-\-enghelp \fR=\fP[ioengine[,command]]
63List all commands defined by \fIioengine\fR, or print help for \fIcommand\fR
64defined by \fIioengine\fR. If no \fIioengine\fR is given, list all
65available ioengines.
de890a1e 66.TP
d60e92d1 67.BI \-\-showcmd \fR=\fPjobfile
7db7a5a0 68Convert \fIjobfile\fR to a set of command\-line options.
d60e92d1 69.TP
bdd88be3 70.BI \-\-readonly
7db7a5a0 71Turn on safety read\-only checks, preventing writes. The \fB\-\-readonly\fR
bdd88be3
TK
72option is an extra safety guard to prevent users from accidentally starting
73a write workload when that is not desired. Fio will only write if
7db7a5a0
TK
74`rw=write/randwrite/rw/randrw' is given. This extra safety net can be used
75as an extra precaution as \fB\-\-readonly\fR will also enable a write check in
bdd88be3
TK
76the I/O engine core to prevent writes due to unknown user space bug(s).
77.TP
d60e92d1 78.BI \-\-eta \fR=\fPwhen
7db7a5a0 79Specifies when real\-time ETA estimate should be printed. \fIwhen\fR may
db37d890
JA
80be `always', `never' or `auto'. `auto' is the default, it prints ETA when
81requested if the output is a TTY. `always' disregards the output type, and
82prints ETA when requested. `never' never prints ETA.
83.TP
84.BI \-\-eta\-interval \fR=\fPtime
85By default, fio requests client ETA status roughly every second. With this
86option, the interval is configurable. Fio imposes a minimum allowed time to
87avoid flooding the console, less than 250 msec is not supported.
d60e92d1 88.TP
30b5d57f 89.BI \-\-eta\-newline \fR=\fPtime
bdd88be3
TK
90Force a new line for every \fItime\fR period passed. When the unit is omitted,
91the value is interpreted in seconds.
30b5d57f
JA
92.TP
93.BI \-\-status\-interval \fR=\fPtime
aa6cb459
VF
94Force a full status dump of cumulative (from job start) values at \fItime\fR
95intervals. This option does *not* provide per-period measurements. So
96values such as bandwidth are running averages. When the time unit is omitted,
97\fItime\fR is interpreted in seconds.
bdd88be3
TK
98.TP
99.BI \-\-section \fR=\fPname
100Only run specified section \fIname\fR in job file. Multiple sections can be specified.
7db7a5a0 101The \fB\-\-section\fR option allows one to combine related jobs into one file.
bdd88be3 102E.g. one job file could define light, moderate, and heavy sections. Tell
7db7a5a0 103fio to run only the "heavy" section by giving `\-\-section=heavy'
bdd88be3 104command line option. One can also specify the "write" operations in one
7db7a5a0 105section and "verify" operation in another section. The \fB\-\-section\fR option
bdd88be3
TK
106only applies to job sections. The reserved *global* section is always
107parsed and used.
c0a5d35e 108.TP
49da1240 109.BI \-\-alloc\-size \fR=\fPkb
7db7a5a0
TK
110Set the internal smalloc pool size to \fIkb\fR in KiB. The
111\fB\-\-alloc\-size\fR switch allows one to use a larger pool size for smalloc.
bdd88be3
TK
112If running large jobs with randommap enabled, fio can run out of memory.
113Smalloc is an internal allocator for shared structures from a fixed size
114memory pool and can grow to 16 pools. The pool size defaults to 16MiB.
7db7a5a0
TK
115NOTE: While running `.fio_smalloc.*' backing store files are visible
116in `/tmp'.
d60e92d1 117.TP
49da1240
JA
118.BI \-\-warnings\-fatal
119All fio parser warnings are fatal, causing fio to exit with an error.
9183788d 120.TP
49da1240 121.BI \-\-max\-jobs \fR=\fPnr
7db7a5a0 122Set the maximum number of threads/processes to support to \fInr\fR.
7f4811bb
RNS
123NOTE: On Linux, it may be necessary to increase the shared-memory limit
124(`/proc/sys/kernel/shmmax') if fio runs into errors while creating jobs.
d60e92d1 125.TP
49da1240 126.BI \-\-server \fR=\fPargs
7db7a5a0
TK
127Start a backend server, with \fIargs\fR specifying what to listen to.
128See \fBCLIENT/SERVER\fR section.
f57a9c59 129.TP
49da1240 130.BI \-\-daemonize \fR=\fPpidfile
7db7a5a0 131Background a fio server, writing the pid to the given \fIpidfile\fR file.
49da1240 132.TP
bdd88be3 133.BI \-\-client \fR=\fPhostname
7db7a5a0
TK
134Instead of running the jobs locally, send and run them on the given \fIhostname\fR
135or set of \fIhostname\fRs. See \fBCLIENT/SERVER\fR section.
bdd88be3 136.TP
7db7a5a0
TK
137.BI \-\-remote\-config \fR=\fPfile
138Tell fio server to load this local \fIfile\fR.
f2a2ce0e
HL
139.TP
140.BI \-\-idle\-prof \fR=\fPoption
7db7a5a0 141Report CPU idleness. \fIoption\fR is one of the following:
bdd88be3
TK
142.RS
143.RS
144.TP
145.B calibrate
146Run unit work calibration only and exit.
147.TP
148.B system
149Show aggregate system idleness and unit work.
150.TP
151.B percpu
7db7a5a0 152As \fBsystem\fR but also show per CPU idleness.
bdd88be3
TK
153.RE
154.RE
155.TP
7db7a5a0
TK
156.BI \-\-inflate\-log \fR=\fPlog
157Inflate and output compressed \fIlog\fR.
bdd88be3 158.TP
7db7a5a0
TK
159.BI \-\-trigger\-file \fR=\fPfile
160Execute trigger command when \fIfile\fR exists.
bdd88be3 161.TP
7db7a5a0
TK
162.BI \-\-trigger\-timeout \fR=\fPtime
163Execute trigger at this \fItime\fR.
bdd88be3 164.TP
7db7a5a0
TK
165.BI \-\-trigger \fR=\fPcommand
166Set this \fIcommand\fR as local trigger.
bdd88be3 167.TP
7db7a5a0
TK
168.BI \-\-trigger\-remote \fR=\fPcommand
169Set this \fIcommand\fR as remote trigger.
bdd88be3 170.TP
7db7a5a0
TK
171.BI \-\-aux\-path \fR=\fPpath
172Use this \fIpath\fR for fio state generated files.
d60e92d1 173.SH "JOB FILE FORMAT"
7a14cf18
TK
174Any parameters following the options will be assumed to be job files, unless
175they match a job file parameter. Multiple job files can be listed and each job
7db7a5a0 176file will be regarded as a separate group. Fio will \fBstonewall\fR execution
7a14cf18
TK
177between each group.
178
179Fio accepts one or more job files describing what it is
180supposed to do. The job file format is the classic ini file, where the names
181enclosed in [] brackets define the job name. You are free to use any ASCII name
182you want, except *global* which has special meaning. Following the job name is
183a sequence of zero or more parameters, one per line, that define the behavior of
184the job. If the first character in a line is a ';' or a '#', the entire line is
185discarded as a comment.
186
187A *global* section sets defaults for the jobs described in that file. A job may
188override a *global* section parameter, and a job file may even have several
189*global* sections if so desired. A job is only affected by a *global* section
190residing above it.
191
7db7a5a0
TK
192The \fB\-\-cmdhelp\fR option also lists all options. If used with an \fIcommand\fR
193argument, \fB\-\-cmdhelp\fR will detail the given \fIcommand\fR.
7a14cf18 194
7db7a5a0
TK
195See the `examples/' directory for inspiration on how to write job files. Note
196the copyright and license requirements currently apply to
197`examples/' files.
54eb4569
TK
198.SH "JOB FILE PARAMETERS"
199Some parameters take an option of a given type, such as an integer or a
200string. Anywhere a numeric value is required, an arithmetic expression may be
201used, provided it is surrounded by parentheses. Supported operators are:
d59aa780 202.RS
7db7a5a0 203.P
d59aa780 204.B addition (+)
7db7a5a0
TK
205.P
206.B subtraction (\-)
207.P
d59aa780 208.B multiplication (*)
7db7a5a0 209.P
d59aa780 210.B division (/)
7db7a5a0 211.P
d59aa780 212.B modulus (%)
7db7a5a0 213.P
d59aa780
JA
214.B exponentiation (^)
215.RE
d59aa780
JA
216.P
217For time values in expressions, units are microseconds by default. This is
218different than for time values not in expressions (not enclosed in
54eb4569
TK
219parentheses).
220.SH "PARAMETER TYPES"
221The following parameter types are used.
d60e92d1
AC
222.TP
223.I str
6b86fc18
TK
224String. A sequence of alphanumeric characters.
225.TP
226.I time
227Integer with possible time suffix. Without a unit value is interpreted as
228seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
229hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and 'us'
230(or 'usec') for microseconds. For example, use 10m for 10 minutes.
d60e92d1
AC
231.TP
232.I int
6d500c2e
RE
233Integer. A whole number value, which may contain an integer prefix
234and an integer suffix.
0b43a833
TK
235.RS
236.RS
237.P
6b86fc18 238[*integer prefix*] **number** [*integer suffix*]
0b43a833
TK
239.RE
240.P
6b86fc18
TK
241The optional *integer prefix* specifies the number's base. The default
242is decimal. *0x* specifies hexadecimal.
0b43a833 243.P
6b86fc18
TK
244The optional *integer suffix* specifies the number's units, and includes an
245optional unit prefix and an optional unit. For quantities of data, the
246default unit is bytes. For quantities of time, the default unit is seconds
247unless otherwise specified.
0b43a833
TK
248.P
249With `kb_base=1000', fio follows international standards for unit
7db7a5a0 250prefixes. To specify power\-of\-10 decimal values defined in the
6b86fc18 251International System of Units (SI):
0b43a833
TK
252.RS
253.P
7db7a5a0 254.PD 0
eccce61a 255K means kilo (K) or 1000
7db7a5a0 256.P
eccce61a 257M means mega (M) or 1000**2
7db7a5a0 258.P
eccce61a 259G means giga (G) or 1000**3
7db7a5a0 260.P
eccce61a 261T means tera (T) or 1000**4
7db7a5a0 262.P
eccce61a 263P means peta (P) or 1000**5
7db7a5a0 264.PD
0b43a833
TK
265.RE
266.P
7db7a5a0 267To specify power\-of\-2 binary values defined in IEC 80000\-13:
0b43a833
TK
268.RS
269.P
7db7a5a0 270.PD 0
eccce61a 271Ki means kibi (Ki) or 1024
7db7a5a0 272.P
eccce61a 273Mi means mebi (Mi) or 1024**2
7db7a5a0 274.P
eccce61a 275Gi means gibi (Gi) or 1024**3
7db7a5a0 276.P
eccce61a 277Ti means tebi (Ti) or 1024**4
7db7a5a0 278.P
eccce61a 279Pi means pebi (Pi) or 1024**5
7db7a5a0 280.PD
0b43a833
TK
281.RE
282.P
283With `kb_base=1024' (the default), the unit prefixes are opposite
7db7a5a0 284from those specified in the SI and IEC 80000\-13 standards to provide
6b86fc18 285compatibility with old scripts. For example, 4k means 4096.
0b43a833 286.P
6b86fc18
TK
287For quantities of data, an optional unit of 'B' may be included
288(e.g., 'kB' is the same as 'k').
0b43a833 289.P
6b86fc18
TK
290The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
291not milli). 'b' and 'B' both mean byte, not bit.
0b43a833
TK
292.P
293Examples with `kb_base=1000':
294.RS
295.P
7db7a5a0 296.PD 0
6d500c2e 2974 KiB: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
7db7a5a0 298.P
6d500c2e 2991 MiB: 1048576, 1m, 1024k
7db7a5a0 300.P
6d500c2e 3011 MB: 1000000, 1mi, 1000ki
7db7a5a0 302.P
6d500c2e 3031 TiB: 1073741824, 1t, 1024m, 1048576k
7db7a5a0 304.P
6d500c2e 3051 TB: 1000000000, 1ti, 1000mi, 1000000ki
7db7a5a0 306.PD
0b43a833
TK
307.RE
308.P
309Examples with `kb_base=1024' (default):
310.RS
311.P
7db7a5a0 312.PD 0
6d500c2e 3134 KiB: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
7db7a5a0 314.P
6d500c2e 3151 MiB: 1048576, 1m, 1024k
7db7a5a0 316.P
6d500c2e 3171 MB: 1000000, 1mi, 1000ki
7db7a5a0 318.P
6d500c2e 3191 TiB: 1073741824, 1t, 1024m, 1048576k
7db7a5a0 320.P
6d500c2e 3211 TB: 1000000000, 1ti, 1000mi, 1000000ki
7db7a5a0 322.PD
0b43a833
TK
323.RE
324.P
6d500c2e 325To specify times (units are not case sensitive):
0b43a833
TK
326.RS
327.P
7db7a5a0 328.PD 0
6d500c2e 329D means days
7db7a5a0 330.P
6d500c2e 331H means hours
7db7a5a0 332.P
6d500c2e 333M mean minutes
7db7a5a0 334.P
6d500c2e 335s or sec means seconds (default)
7db7a5a0 336.P
6d500c2e 337ms or msec means milliseconds
7db7a5a0 338.P
6d500c2e 339us or usec means microseconds
7db7a5a0 340.PD
0b43a833
TK
341.RE
342.P
6b86fc18 343If the option accepts an upper and lower range, use a colon ':' or
7db7a5a0 344minus '\-' to separate such values. See \fIirange\fR parameter type.
6b86fc18
TK
345If the lower value specified happens to be larger than the upper value
346the two values are swapped.
0b43a833 347.RE
d60e92d1
AC
348.TP
349.I bool
6b86fc18
TK
350Boolean. Usually parsed as an integer, however only defined for
351true and false (1 and 0).
d60e92d1
AC
352.TP
353.I irange
6b86fc18 354Integer range with suffix. Allows value range to be given, such as
7db7a5a0 3551024\-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
6b86fc18 356option allows two sets of ranges, they can be specified with a ',' or '/'
7db7a5a0 357delimiter: 1k\-4k/8k\-32k. Also see \fIint\fR parameter type.
83349190
YH
358.TP
359.I float_list
6b86fc18 360A list of floating point numbers, separated by a ':' character.
523bad63 361.SH "JOB PARAMETERS"
54eb4569 362With the above in mind, here follows the complete list of fio job parameters.
523bad63 363.SS "Units"
d60e92d1 364.TP
523bad63
TK
365.BI kb_base \fR=\fPint
366Select the interpretation of unit prefixes in input parameters.
367.RS
368.RS
d60e92d1 369.TP
523bad63
TK
370.B 1000
371Inputs comply with IEC 80000\-13 and the International
372System of Units (SI). Use:
373.RS
374.P
375.PD 0
376\- power\-of\-2 values with IEC prefixes (e.g., KiB)
377.P
378\- power\-of\-10 values with SI prefixes (e.g., kB)
379.PD
380.RE
381.TP
382.B 1024
383Compatibility mode (default). To avoid breaking old scripts:
384.P
385.RS
386.PD 0
387\- power\-of\-2 values with SI prefixes
388.P
389\- power\-of\-10 values with IEC prefixes
390.PD
391.RE
392.RE
393.P
394See \fBbs\fR for more details on input parameters.
395.P
396Outputs always use correct prefixes. Most outputs include both
397side\-by\-side, like:
398.P
399.RS
400bw=2383.3kB/s (2327.4KiB/s)
401.RE
402.P
403If only one value is reported, then kb_base selects the one to use:
404.P
405.RS
406.PD 0
4071000 \-\- SI prefixes
408.P
4091024 \-\- IEC prefixes
410.PD
411.RE
412.RE
413.TP
414.BI unit_base \fR=\fPint
415Base unit for reporting. Allowed values are:
416.RS
417.RS
418.TP
419.B 0
420Use auto\-detection (default).
421.TP
422.B 8
423Byte based.
424.TP
425.B 1
426Bit based.
427.RE
428.RE
429.SS "Job description"
430.TP
431.BI name \fR=\fPstr
432ASCII name of the job. This may be used to override the name printed by fio
433for this job. Otherwise the job name is used. On the command line this
434parameter has the special purpose of also signaling the start of a new job.
9cc8cb91 435.TP
d60e92d1 436.BI description \fR=\fPstr
523bad63
TK
437Text description of the job. Doesn't do anything except dump this text
438description when this job is run. It's not parsed.
439.TP
440.BI loops \fR=\fPint
441Run the specified number of iterations of this job. Used to repeat the same
442workload a given number of times. Defaults to 1.
443.TP
444.BI numjobs \fR=\fPint
445Create the specified number of clones of this job. Each clone of job
446is spawned as an independent thread or process. May be used to setup a
447larger number of threads/processes doing the same thing. Each thread is
448reported separately; to see statistics for all clones as a whole, use
449\fBgroup_reporting\fR in conjunction with \fBnew_group\fR.
450See \fB\-\-max\-jobs\fR. Default: 1.
451.SS "Time related parameters"
452.TP
453.BI runtime \fR=\fPtime
454Tell fio to terminate processing after the specified period of time. It
455can be quite hard to determine for how long a specified job will run, so
456this parameter is handy to cap the total runtime to a given time. When
f1dd3fb1 457the unit is omitted, the value is interpreted in seconds.
523bad63
TK
458.TP
459.BI time_based
460If set, fio will run for the duration of the \fBruntime\fR specified
461even if the file(s) are completely read or written. It will simply loop over
462the same workload as many times as the \fBruntime\fR allows.
463.TP
464.BI startdelay \fR=\fPirange(int)
465Delay the start of job for the specified amount of time. Can be a single
466value or a range. When given as a range, each thread will choose a value
467randomly from within the range. Value is in seconds if a unit is omitted.
468.TP
469.BI ramp_time \fR=\fPtime
470If set, fio will run the specified workload for this amount of time before
471logging any performance numbers. Useful for letting performance settle
472before logging results, thus minimizing the runtime required for stable
473results. Note that the \fBramp_time\fR is considered lead in time for a job,
474thus it will increase the total runtime if a special timeout or
475\fBruntime\fR is specified. When the unit is omitted, the value is
476given in seconds.
477.TP
478.BI clocksource \fR=\fPstr
479Use the given clocksource as the base of timing. The supported options are:
480.RS
481.RS
482.TP
483.B gettimeofday
484\fBgettimeofday\fR\|(2)
485.TP
486.B clock_gettime
487\fBclock_gettime\fR\|(2)
488.TP
489.B cpu
490Internal CPU clock source
491.RE
492.P
493\fBcpu\fR is the preferred clocksource if it is reliable, as it is very fast (and
494fio is heavy on time calls). Fio will automatically use this clocksource if
495it's supported and considered reliable on the system it is running on,
496unless another clocksource is specifically set. For x86/x86\-64 CPUs, this
497means supporting TSC Invariant.
498.RE
499.TP
500.BI gtod_reduce \fR=\fPbool
501Enable all of the \fBgettimeofday\fR\|(2) reducing options
502(\fBdisable_clat\fR, \fBdisable_slat\fR, \fBdisable_bw_measurement\fR) plus
503reduce precision of the timeout somewhat to really shrink the
504\fBgettimeofday\fR\|(2) call count. With this option enabled, we only do
505about 0.4% of the \fBgettimeofday\fR\|(2) calls we would have done if all
506time keeping was enabled.
507.TP
508.BI gtod_cpu \fR=\fPint
509Sometimes it's cheaper to dedicate a single thread of execution to just
510getting the current time. Fio (and databases, for instance) are very
511intensive on \fBgettimeofday\fR\|(2) calls. With this option, you can set
512one CPU aside for doing nothing but logging current time to a shared memory
513location. Then the other threads/processes that run I/O workloads need only
514copy that segment, instead of entering the kernel with a
515\fBgettimeofday\fR\|(2) call. The CPU set aside for doing these time
516calls will be excluded from other uses. Fio will manually clear it from the
517CPU mask of other jobs.
518.SS "Target file/device"
d60e92d1
AC
519.TP
520.BI directory \fR=\fPstr
523bad63
TK
521Prefix \fBfilename\fRs with this directory. Used to place files in a different
522location than `./'. You can specify a number of directories by
523separating the names with a ':' character. These directories will be
524assigned equally distributed to job clones created by \fBnumjobs\fR as
525long as they are using generated filenames. If specific \fBfilename\fR(s) are
526set fio will use the first listed directory, and thereby matching the
527\fBfilename\fR semantic which generates a file each clone if not specified, but
528let all clones use the same if set.
529.RS
530.P
531See the \fBfilename\fR option for information on how to escape ':' and '\'
532characters within the directory path itself.
533.RE
d60e92d1
AC
534.TP
535.BI filename \fR=\fPstr
523bad63
TK
536Fio normally makes up a \fBfilename\fR based on the job name, thread number, and
537file number (see \fBfilename_format\fR). If you want to share files
538between threads in a job or several
539jobs with fixed file paths, specify a \fBfilename\fR for each of them to override
540the default. If the ioengine is file based, you can specify a number of files
541by separating the names with a ':' colon. So if you wanted a job to open
542`/dev/sda' and `/dev/sdb' as the two working files, you would use
543`filename=/dev/sda:/dev/sdb'. This also means that whenever this option is
544specified, \fBnrfiles\fR is ignored. The size of regular files specified
545by this option will be \fBsize\fR divided by number of files unless an
546explicit size is specified by \fBfilesize\fR.
547.RS
548.P
549Each colon and backslash in the wanted path must be escaped with a '\'
550character. For instance, if the path is `/dev/dsk/foo@3,0:c' then you
551would use `filename=/dev/dsk/foo@3,0\\:c' and if the path is
552`F:\\\\filename' then you would use `filename=F\\:\\\\filename'.
553.P
554On Windows, disk devices are accessed as `\\\\\\\\.\\\\PhysicalDrive0' for
555the first device, `\\\\\\\\.\\\\PhysicalDrive1' for the second etc.
556Note: Windows and FreeBSD prevent write access to areas
557of the disk containing in\-use data (e.g. filesystems).
558.P
559The filename `\-' is a reserved name, meaning *stdin* or *stdout*. Which
560of the two depends on the read/write direction set.
561.RE
d60e92d1 562.TP
de98bd30 563.BI filename_format \fR=\fPstr
523bad63
TK
564If sharing multiple files between jobs, it is usually necessary to have fio
565generate the exact names that you want. By default, fio will name a file
de98bd30 566based on the default file format specification of
523bad63 567`jobname.jobnumber.filenumber'. With this option, that can be
de98bd30
JA
568customized. Fio will recognize and replace the following keywords in this
569string:
570.RS
571.RS
572.TP
573.B $jobname
574The name of the worker thread or process.
575.TP
576.B $jobnum
577The incremental number of the worker thread or process.
578.TP
579.B $filenum
580The incremental number of the file for that worker thread or process.
581.RE
582.P
523bad63
TK
583To have dependent jobs share a set of files, this option can be set to have
584fio generate filenames that are shared between the two. For instance, if
585`testfiles.$filenum' is specified, file number 4 for any job will be
586named `testfiles.4'. The default of `$jobname.$jobnum.$filenum'
de98bd30 587will be used if no other format specifier is given.
645943c0
JB
588.P
589If you specify a path then the directories will be created up to the main
590directory for the file. So for example if you specify `a/b/c/$jobnum` then the
591directories a/b/c will be created before the file setup part of the job. If you
592specify \fBdirectory\fR then the path will be relative that directory, otherwise
593it is treated as the absolute path.
de98bd30 594.RE
de98bd30 595.TP
922a5be8 596.BI unique_filename \fR=\fPbool
523bad63
TK
597To avoid collisions between networked clients, fio defaults to prefixing any
598generated filenames (with a directory specified) with the source of the
599client connecting. To disable this behavior, set this option to 0.
600.TP
601.BI opendir \fR=\fPstr
602Recursively open any files below directory \fIstr\fR.
922a5be8 603.TP
3ce9dcaf 604.BI lockfile \fR=\fPstr
523bad63
TK
605Fio defaults to not locking any files before it does I/O to them. If a file
606or file descriptor is shared, fio can serialize I/O to that file to make the
607end result consistent. This is usual for emulating real workloads that share
608files. The lock modes are:
3ce9dcaf
JA
609.RS
610.RS
611.TP
612.B none
523bad63 613No locking. The default.
3ce9dcaf
JA
614.TP
615.B exclusive
523bad63 616Only one thread or process may do I/O at a time, excluding all others.
3ce9dcaf
JA
617.TP
618.B readwrite
523bad63
TK
619Read\-write locking on the file. Many readers may
620access the file at the same time, but writes get exclusive access.
3ce9dcaf 621.RE
ce594fbe 622.RE
523bad63
TK
623.TP
624.BI nrfiles \fR=\fPint
625Number of files to use for this job. Defaults to 1. The size of files
626will be \fBsize\fR divided by this unless explicit size is specified by
627\fBfilesize\fR. Files are created for each thread separately, and each
628file will have a file number within its name by default, as explained in
629\fBfilename\fR section.
630.TP
631.BI openfiles \fR=\fPint
632Number of files to keep open at the same time. Defaults to the same as
633\fBnrfiles\fR, can be set smaller to limit the number simultaneous
634opens.
635.TP
636.BI file_service_type \fR=\fPstr
637Defines how fio decides which file from a job to service next. The following
638types are defined:
639.RS
640.RS
641.TP
642.B random
643Choose a file at random.
644.TP
645.B roundrobin
646Round robin over opened files. This is the default.
647.TP
648.B sequential
649Finish one file before moving on to the next. Multiple files can
650still be open depending on \fBopenfiles\fR.
651.TP
652.B zipf
653Use a Zipf distribution to decide what file to access.
654.TP
655.B pareto
656Use a Pareto distribution to decide what file to access.
657.TP
658.B normal
659Use a Gaussian (normal) distribution to decide what file to access.
660.TP
661.B gauss
662Alias for normal.
663.RE
3ce9dcaf 664.P
523bad63
TK
665For \fBrandom\fR, \fBroundrobin\fR, and \fBsequential\fR, a postfix can be appended to
666tell fio how many I/Os to issue before switching to a new file. For example,
667specifying `file_service_type=random:8' would cause fio to issue
6688 I/Os before selecting a new file at random. For the non\-uniform
669distributions, a floating point postfix can be given to influence how the
670distribution is skewed. See \fBrandom_distribution\fR for a description
671of how that would work.
672.RE
673.TP
674.BI ioscheduler \fR=\fPstr
675Attempt to switch the device hosting the file to the specified I/O scheduler
676before running.
677.TP
678.BI create_serialize \fR=\fPbool
679If true, serialize the file creation for the jobs. This may be handy to
680avoid interleaving of data files, which may greatly depend on the filesystem
681used and even the number of processors in the system. Default: true.
682.TP
683.BI create_fsync \fR=\fPbool
684\fBfsync\fR\|(2) the data file after creation. This is the default.
685.TP
686.BI create_on_open \fR=\fPbool
687If true, don't pre\-create files but allow the job's open() to create a file
688when it's time to do I/O. Default: false \-\- pre\-create all necessary files
689when the job starts.
690.TP
691.BI create_only \fR=\fPbool
692If true, fio will only run the setup phase of the job. If files need to be
693laid out or updated on disk, only that will be done \-\- the actual job contents
694are not executed. Default: false.
695.TP
696.BI allow_file_create \fR=\fPbool
697If true, fio is permitted to create files as part of its workload. If this
698option is false, then fio will error out if
699the files it needs to use don't already exist. Default: true.
700.TP
701.BI allow_mounted_write \fR=\fPbool
702If this isn't set, fio will abort jobs that are destructive (e.g. that write)
703to what appears to be a mounted device or partition. This should help catch
704creating inadvertently destructive tests, not realizing that the test will
705destroy data on the mounted file system. Note that some platforms don't allow
706writing against a mounted device regardless of this option. Default: false.
707.TP
708.BI pre_read \fR=\fPbool
709If this is given, files will be pre\-read into memory before starting the
710given I/O operation. This will also clear the \fBinvalidate\fR flag,
711since it is pointless to pre\-read and then drop the cache. This will only
712work for I/O engines that are seek\-able, since they allow you to read the
713same data multiple times. Thus it will not work on non\-seekable I/O engines
714(e.g. network, splice). Default: false.
715.TP
716.BI unlink \fR=\fPbool
717Unlink the job files when done. Not the default, as repeated runs of that
718job would then waste time recreating the file set again and again. Default:
719false.
720.TP
721.BI unlink_each_loop \fR=\fPbool
722Unlink job files after each iteration or loop. Default: false.
723.TP
724.BI zonesize \fR=\fPint
725Divide a file into zones of the specified size. See \fBzoneskip\fR.
726.TP
727.BI zonerange \fR=\fPint
728Give size of an I/O zone. See \fBzoneskip\fR.
729.TP
730.BI zoneskip \fR=\fPint
731Skip the specified number of bytes when \fBzonesize\fR data has been
732read. The two zone options can be used to only do I/O on zones of a file.
733.SS "I/O type"
734.TP
735.BI direct \fR=\fPbool
736If value is true, use non\-buffered I/O. This is usually O_DIRECT. Note that
8e889110 737OpenBSD and ZFS on Solaris don't support direct I/O. On Windows the synchronous
523bad63
TK
738ioengines don't support direct I/O. Default: false.
739.TP
740.BI atomic \fR=\fPbool
741If value is true, attempt to use atomic direct I/O. Atomic writes are
742guaranteed to be stable once acknowledged by the operating system. Only
743Linux supports O_ATOMIC right now.
744.TP
745.BI buffered \fR=\fPbool
746If value is true, use buffered I/O. This is the opposite of the
747\fBdirect\fR option. Defaults to true.
d60e92d1
AC
748.TP
749.BI readwrite \fR=\fPstr "\fR,\fP rw" \fR=\fPstr
523bad63 750Type of I/O pattern. Accepted values are:
d60e92d1
AC
751.RS
752.RS
753.TP
754.B read
d1429b5c 755Sequential reads.
d60e92d1
AC
756.TP
757.B write
d1429b5c 758Sequential writes.
d60e92d1 759.TP
fa769d44 760.B trim
169c098d 761Sequential trims (Linux block devices only).
fa769d44 762.TP
d60e92d1 763.B randread
d1429b5c 764Random reads.
d60e92d1
AC
765.TP
766.B randwrite
d1429b5c 767Random writes.
d60e92d1 768.TP
fa769d44 769.B randtrim
169c098d 770Random trims (Linux block devices only).
fa769d44 771.TP
523bad63
TK
772.B rw,readwrite
773Sequential mixed reads and writes.
d60e92d1 774.TP
ff6bb260 775.B randrw
523bad63 776Random mixed reads and writes.
82a90686
JA
777.TP
778.B trimwrite
523bad63
TK
779Sequential trim+write sequences. Blocks will be trimmed first,
780then the same blocks will be written to.
d60e92d1
AC
781.RE
782.P
523bad63
TK
783Fio defaults to read if the option is not specified. For the mixed I/O
784types, the default is to split them 50/50. For certain types of I/O the
785result may still be skewed a bit, since the speed may be different.
786.P
787It is possible to specify the number of I/Os to do before getting a new
788offset by appending `:<nr>' to the end of the string given. For a
789random read, it would look like `rw=randread:8' for passing in an offset
790modifier with a value of 8. If the suffix is used with a sequential I/O
791pattern, then the `<nr>' value specified will be added to the generated
792offset for each I/O turning sequential I/O into sequential I/O with holes.
793For instance, using `rw=write:4k' will skip 4k for every write. Also see
794the \fBrw_sequencer\fR option.
d60e92d1
AC
795.RE
796.TP
38dad62d 797.BI rw_sequencer \fR=\fPstr
523bad63
TK
798If an offset modifier is given by appending a number to the `rw=\fIstr\fR'
799line, then this option controls how that number modifies the I/O offset
800being generated. Accepted values are:
38dad62d
JA
801.RS
802.RS
803.TP
804.B sequential
523bad63 805Generate sequential offset.
38dad62d
JA
806.TP
807.B identical
523bad63 808Generate the same offset.
38dad62d
JA
809.RE
810.P
523bad63
TK
811\fBsequential\fR is only useful for random I/O, where fio would normally
812generate a new random offset for every I/O. If you append e.g. 8 to randread,
813you would get a new random offset for every 8 I/Os. The result would be a
814seek for only every 8 I/Os, instead of for every I/O. Use `rw=randread:8'
815to specify that. As sequential I/O is already sequential, setting
816\fBsequential\fR for that would not result in any differences. \fBidentical\fR
817behaves in a similar fashion, except it sends the same offset 8 number of
818times before generating a new offset.
38dad62d 819.RE
90fef2d1 820.TP
771e58be
JA
821.BI unified_rw_reporting \fR=\fPbool
822Fio normally reports statistics on a per data direction basis, meaning that
523bad63
TK
823reads, writes, and trims are accounted and reported separately. If this
824option is set fio sums the results and report them as "mixed" instead.
771e58be 825.TP
d60e92d1 826.BI randrepeat \fR=\fPbool
523bad63
TK
827Seed the random number generator used for random I/O patterns in a
828predictable way so the pattern is repeatable across runs. Default: true.
56e2a5fc
CE
829.TP
830.BI allrandrepeat \fR=\fPbool
831Seed all random number generators in a predictable way so results are
523bad63 832repeatable across runs. Default: false.
d60e92d1 833.TP
04778baf
JA
834.BI randseed \fR=\fPint
835Seed the random number generators based on this seed value, to be able to
836control what sequence of output is being generated. If not set, the random
837sequence depends on the \fBrandrepeat\fR setting.
838.TP
a596f047 839.BI fallocate \fR=\fPstr
523bad63
TK
840Whether pre\-allocation is performed when laying down files.
841Accepted values are:
a596f047
EG
842.RS
843.RS
844.TP
845.B none
523bad63 846Do not pre\-allocate space.
a596f047 847.TP
2c3e17be 848.B native
523bad63
TK
849Use a platform's native pre\-allocation call but fall back to
850\fBnone\fR behavior if it fails/is not implemented.
2c3e17be 851.TP
a596f047 852.B posix
523bad63 853Pre\-allocate via \fBposix_fallocate\fR\|(3).
a596f047
EG
854.TP
855.B keep
523bad63
TK
856Pre\-allocate via \fBfallocate\fR\|(2) with
857FALLOC_FL_KEEP_SIZE set.
a596f047
EG
858.TP
859.B 0
523bad63 860Backward\-compatible alias for \fBnone\fR.
a596f047
EG
861.TP
862.B 1
523bad63 863Backward\-compatible alias for \fBposix\fR.
a596f047
EG
864.RE
865.P
523bad63
TK
866May not be available on all supported platforms. \fBkeep\fR is only available
867on Linux. If using ZFS on Solaris this cannot be set to \fBposix\fR
868because ZFS doesn't support pre\-allocation. Default: \fBnative\fR if any
869pre\-allocation methods are available, \fBnone\fR if not.
a596f047 870.RE
7bc8c2cf 871.TP
ecb2083d 872.BI fadvise_hint \fR=\fPstr
c712c97a
JA
873Use \fBposix_fadvise\fR\|(2) or \fBposix_madvise\fR\|(2) to advise the kernel
874what I/O patterns are likely to be issued. Accepted values are:
ecb2083d
JA
875.RS
876.RS
877.TP
878.B 0
879Backwards compatible hint for "no hint".
880.TP
881.B 1
882Backwards compatible hint for "advise with fio workload type". This
523bad63 883uses FADV_RANDOM for a random workload, and FADV_SEQUENTIAL
ecb2083d
JA
884for a sequential workload.
885.TP
886.B sequential
523bad63 887Advise using FADV_SEQUENTIAL.
ecb2083d
JA
888.TP
889.B random
523bad63 890Advise using FADV_RANDOM.
ecb2083d
JA
891.RE
892.RE
d60e92d1 893.TP
8f4b9f24 894.BI write_hint \fR=\fPstr
523bad63
TK
895Use \fBfcntl\fR\|(2) to advise the kernel what life time to expect
896from a write. Only supported on Linux, as of version 4.13. Accepted
8f4b9f24
JA
897values are:
898.RS
899.RS
900.TP
901.B none
902No particular life time associated with this file.
903.TP
904.B short
905Data written to this file has a short life time.
906.TP
907.B medium
908Data written to this file has a medium life time.
909.TP
910.B long
911Data written to this file has a long life time.
912.TP
913.B extreme
914Data written to this file has a very long life time.
915.RE
523bad63
TK
916.P
917The values are all relative to each other, and no absolute meaning
918should be associated with them.
8f4b9f24 919.RE
37659335 920.TP
523bad63
TK
921.BI offset \fR=\fPint
922Start I/O at the provided offset in the file, given as either a fixed size in
83c8b093
JF
923bytes or a percentage. If a percentage is given, the generated offset will be
924aligned to the minimum \fBblocksize\fR or to the value of \fBoffset_align\fR if
925provided. Data before the given offset will not be touched. This
523bad63
TK
926effectively caps the file size at `real_size \- offset'. Can be combined with
927\fBsize\fR to constrain the start and end range of the I/O workload.
928A percentage can be specified by a number between 1 and 100 followed by '%',
929for example, `offset=20%' to specify 20%.
6d500c2e 930.TP
83c8b093
JF
931.BI offset_align \fR=\fPint
932If set to non-zero value, the byte offset generated by a percentage \fBoffset\fR
933is aligned upwards to this value. Defaults to 0 meaning that a percentage
934offset is aligned to the minimum block size.
935.TP
523bad63
TK
936.BI offset_increment \fR=\fPint
937If this is provided, then the real offset becomes `\fBoffset\fR + \fBoffset_increment\fR
938* thread_number', where the thread number is a counter that starts at 0 and
939is incremented for each sub\-job (i.e. when \fBnumjobs\fR option is
940specified). This option is useful if there are several jobs which are
941intended to operate on a file in parallel disjoint segments, with even
942spacing between the starting points.
6d500c2e 943.TP
523bad63
TK
944.BI number_ios \fR=\fPint
945Fio will normally perform I/Os until it has exhausted the size of the region
946set by \fBsize\fR, or if it exhaust the allocated time (or hits an error
947condition). With this setting, the range/size can be set independently of
948the number of I/Os to perform. When fio reaches this number, it will exit
949normally and report status. Note that this does not extend the amount of I/O
950that will be done, it will only stop fio if this condition is met before
951other end\-of\-job criteria.
d60e92d1 952.TP
523bad63
TK
953.BI fsync \fR=\fPint
954If writing to a file, issue an \fBfsync\fR\|(2) (or its equivalent) of
955the dirty data for every number of blocks given. For example, if you give 32
956as a parameter, fio will sync the file after every 32 writes issued. If fio is
957using non\-buffered I/O, we may not sync the file. The exception is the sg
958I/O engine, which synchronizes the disk cache anyway. Defaults to 0, which
959means fio does not periodically issue and wait for a sync to complete. Also
960see \fBend_fsync\fR and \fBfsync_on_close\fR.
6d500c2e 961.TP
523bad63
TK
962.BI fdatasync \fR=\fPint
963Like \fBfsync\fR but uses \fBfdatasync\fR\|(2) to only sync data and
964not metadata blocks. In Windows, FreeBSD, and DragonFlyBSD there is no
965\fBfdatasync\fR\|(2) so this falls back to using \fBfsync\fR\|(2).
966Defaults to 0, which means fio does not periodically issue and wait for a
967data\-only sync to complete.
d60e92d1 968.TP
523bad63
TK
969.BI write_barrier \fR=\fPint
970Make every N\-th write a barrier write.
901bb994 971.TP
523bad63
TK
972.BI sync_file_range \fR=\fPstr:int
973Use \fBsync_file_range\fR\|(2) for every \fIint\fR number of write
974operations. Fio will track range of writes that have happened since the last
975\fBsync_file_range\fR\|(2) call. \fIstr\fR can currently be one or more of:
976.RS
977.RS
fd68418e 978.TP
523bad63
TK
979.B wait_before
980SYNC_FILE_RANGE_WAIT_BEFORE
c5751c62 981.TP
523bad63
TK
982.B write
983SYNC_FILE_RANGE_WRITE
c5751c62 984.TP
523bad63
TK
985.B wait_after
986SYNC_FILE_RANGE_WRITE_AFTER
2fa5a241 987.RE
523bad63
TK
988.P
989So if you do `sync_file_range=wait_before,write:8', fio would use
990`SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE' for every 8
991writes. Also see the \fBsync_file_range\fR\|(2) man page. This option is
992Linux specific.
2fa5a241 993.RE
ce35b1ec 994.TP
523bad63
TK
995.BI overwrite \fR=\fPbool
996If true, writes to a file will always overwrite existing data. If the file
997doesn't already exist, it will be created before the write phase begins. If
998the file exists and is large enough for the specified write phase, nothing
999will be done. Default: false.
5c94b008 1000.TP
523bad63
TK
1001.BI end_fsync \fR=\fPbool
1002If true, \fBfsync\fR\|(2) file contents when a write stage has completed.
1003Default: false.
d60e92d1 1004.TP
523bad63
TK
1005.BI fsync_on_close \fR=\fPbool
1006If true, fio will \fBfsync\fR\|(2) a dirty file on close. This differs
1007from \fBend_fsync\fR in that it will happen on every file close, not
1008just at the end of the job. Default: false.
d60e92d1 1009.TP
523bad63
TK
1010.BI rwmixread \fR=\fPint
1011Percentage of a mixed workload that should be reads. Default: 50.
1012.TP
1013.BI rwmixwrite \fR=\fPint
1014Percentage of a mixed workload that should be writes. If both
1015\fBrwmixread\fR and \fBrwmixwrite\fR is given and the values do not
1016add up to 100%, the latter of the two will be used to override the
1017first. This may interfere with a given rate setting, if fio is asked to
1018limit reads or writes to a certain rate. If that is the case, then the
1019distribution may be skewed. Default: 50.
1020.TP
1021.BI random_distribution \fR=\fPstr:float[,str:float][,str:float]
1022By default, fio will use a completely uniform random distribution when asked
1023to perform random I/O. Sometimes it is useful to skew the distribution in
1024specific ways, ensuring that some parts of the data is more hot than others.
1025fio includes the following distribution models:
d60e92d1
AC
1026.RS
1027.RS
1028.TP
1029.B random
523bad63 1030Uniform random distribution
8c07860d
JA
1031.TP
1032.B zipf
523bad63 1033Zipf distribution
8c07860d
JA
1034.TP
1035.B pareto
523bad63 1036Pareto distribution
8c07860d 1037.TP
dd3503d3 1038.B normal
523bad63 1039Normal (Gaussian) distribution
dd3503d3 1040.TP
523bad63
TK
1041.B zoned
1042Zoned random distribution
59466396
JA
1043.B zoned_abs
1044Zoned absolute random distribution
d60e92d1
AC
1045.RE
1046.P
523bad63
TK
1047When using a \fBzipf\fR or \fBpareto\fR distribution, an input value is also
1048needed to define the access pattern. For \fBzipf\fR, this is the `Zipf theta'.
1049For \fBpareto\fR, it's the `Pareto power'. Fio includes a test
1050program, \fBfio\-genzipf\fR, that can be used visualize what the given input
1051values will yield in terms of hit rates. If you wanted to use \fBzipf\fR with
1052a `theta' of 1.2, you would use `random_distribution=zipf:1.2' as the
1053option. If a non\-uniform model is used, fio will disable use of the random
1054map. For the \fBnormal\fR distribution, a normal (Gaussian) deviation is
1055supplied as a value between 0 and 100.
1056.P
1057For a \fBzoned\fR distribution, fio supports specifying percentages of I/O
1058access that should fall within what range of the file or device. For
1059example, given a criteria of:
d60e92d1 1060.RS
523bad63
TK
1061.P
1062.PD 0
106360% of accesses should be to the first 10%
1064.P
106530% of accesses should be to the next 20%
1066.P
10678% of accesses should be to the next 30%
1068.P
10692% of accesses should be to the next 40%
1070.PD
1071.RE
1072.P
1073we can define that through zoning of the random accesses. For the above
1074example, the user would do:
1075.RS
1076.P
1077random_distribution=zoned:60/10:30/20:8/30:2/40
1078.RE
1079.P
59466396
JA
1080A \fBzoned_abs\fR distribution works exactly like the\fBzoned\fR, except that
1081it takes absolute sizes. For example, let's say you wanted to define access
1082according to the following criteria:
1083.RS
1084.P
1085.PD 0
108660% of accesses should be to the first 20G
1087.P
108830% of accesses should be to the next 100G
1089.P
109010% of accesses should be to the next 500G
1091.PD
1092.RE
1093.P
1094we can define an absolute zoning distribution with:
1095.RS
1096.P
1097random_distribution=zoned:60/10:30/20:8/30:2/40
1098.RE
1099.P
6a16ece8
JA
1100For both \fBzoned\fR and \fBzoned_abs\fR, fio supports defining up to 256
1101separate zones.
1102.P
59466396 1103Similarly to how \fBbssplit\fR works for setting ranges and percentages
523bad63
TK
1104of block sizes. Like \fBbssplit\fR, it's possible to specify separate
1105zones for reads, writes, and trims. If just one set is given, it'll apply to
1106all of them.
1107.RE
1108.TP
1109.BI percentage_random \fR=\fPint[,int][,int]
1110For a random workload, set how big a percentage should be random. This
1111defaults to 100%, in which case the workload is fully random. It can be set
1112from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1113sequential. Any setting in between will result in a random mix of sequential
1114and random I/O, at the given percentages. Comma\-separated values may be
1115specified for reads, writes, and trims as described in \fBblocksize\fR.
1116.TP
1117.BI norandommap
1118Normally fio will cover every block of the file when doing random I/O. If
1119this option is given, fio will just get a new random offset without looking
1120at past I/O history. This means that some blocks may not be read or written,
1121and that some blocks may be read/written more than once. If this option is
1122used with \fBverify\fR and multiple blocksizes (via \fBbsrange\fR),
1123only intact blocks are verified, i.e., partially\-overwritten blocks are
1124ignored.
1125.TP
1126.BI softrandommap \fR=\fPbool
1127See \fBnorandommap\fR. If fio runs with the random block map enabled and
1128it fails to allocate the map, if this option is set it will continue without
1129a random block map. As coverage will not be as complete as with random maps,
1130this option is disabled by default.
1131.TP
1132.BI random_generator \fR=\fPstr
1133Fio supports the following engines for generating I/O offsets for random I/O:
1134.RS
1135.RS
1136.TP
1137.B tausworthe
1138Strong 2^88 cycle random number generator.
1139.TP
1140.B lfsr
1141Linear feedback shift register generator.
1142.TP
1143.B tausworthe64
1144Strong 64\-bit 2^258 cycle random number generator.
1145.RE
1146.P
1147\fBtausworthe\fR is a strong random number generator, but it requires tracking
1148on the side if we want to ensure that blocks are only read or written
1149once. \fBlfsr\fR guarantees that we never generate the same offset twice, and
1150it's also less computationally expensive. It's not a true random generator,
1151however, though for I/O purposes it's typically good enough. \fBlfsr\fR only
1152works with single block sizes, not with workloads that use multiple block
1153sizes. If used with such a workload, fio may read or write some blocks
1154multiple times. The default value is \fBtausworthe\fR, unless the required
1155space exceeds 2^32 blocks. If it does, then \fBtausworthe64\fR is
1156selected automatically.
1157.RE
1158.SS "Block size"
1159.TP
1160.BI blocksize \fR=\fPint[,int][,int] "\fR,\fB bs" \fR=\fPint[,int][,int]
1161The block size in bytes used for I/O units. Default: 4096. A single value
1162applies to reads, writes, and trims. Comma\-separated values may be
1163specified for reads, writes, and trims. A value not terminated in a comma
1164applies to subsequent types. Examples:
1165.RS
1166.RS
1167.P
1168.PD 0
1169bs=256k means 256k for reads, writes and trims.
1170.P
1171bs=8k,32k means 8k for reads, 32k for writes and trims.
1172.P
1173bs=8k,32k, means 8k for reads, 32k for writes, and default for trims.
1174.P
1175bs=,8k means default for reads, 8k for writes and trims.
1176.P
1177bs=,8k, means default for reads, 8k for writes, and default for trims.
1178.PD
1179.RE
1180.RE
1181.TP
1182.BI blocksize_range \fR=\fPirange[,irange][,irange] "\fR,\fB bsrange" \fR=\fPirange[,irange][,irange]
1183A range of block sizes in bytes for I/O units. The issued I/O unit will
1184always be a multiple of the minimum size, unless
1185\fBblocksize_unaligned\fR is set.
1186Comma\-separated ranges may be specified for reads, writes, and trims as
1187described in \fBblocksize\fR. Example:
1188.RS
1189.RS
1190.P
1191bsrange=1k\-4k,2k\-8k
1192.RE
1193.RE
1194.TP
1195.BI bssplit \fR=\fPstr[,str][,str]
1196Sometimes you want even finer grained control of the block sizes issued, not
1197just an even split between them. This option allows you to weight various
1198block sizes, so that you are able to define a specific amount of block sizes
1199issued. The format for this option is:
1200.RS
1201.RS
1202.P
1203bssplit=blocksize/percentage:blocksize/percentage
1204.RE
1205.P
1206for as many block sizes as needed. So if you want to define a workload that
1207has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would write:
1208.RS
1209.P
1210bssplit=4k/10:64k/50:32k/40
1211.RE
1212.P
1213Ordering does not matter. If the percentage is left blank, fio will fill in
1214the remaining values evenly. So a bssplit option like this one:
1215.RS
1216.P
1217bssplit=4k/50:1k/:32k/
1218.RE
1219.P
1220would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always add up
1221to 100, if bssplit is given a range that adds up to more, it will error out.
1222.P
1223Comma\-separated values may be specified for reads, writes, and trims as
1224described in \fBblocksize\fR.
1225.P
1226If you want a workload that has 50% 2k reads and 50% 4k reads, while having
122790% 4k writes and 10% 8k writes, you would specify:
1228.RS
1229.P
1230bssplit=2k/50:4k/50,4k/90,8k/10
1231.RE
6a16ece8
JA
1232.P
1233Fio supports defining up to 64 different weights for each data direction.
523bad63
TK
1234.RE
1235.TP
1236.BI blocksize_unaligned "\fR,\fB bs_unaligned"
1237If set, fio will issue I/O units with any size within
1238\fBblocksize_range\fR, not just multiples of the minimum size. This
1239typically won't work with direct I/O, as that normally requires sector
1240alignment.
1241.TP
1242.BI bs_is_seq_rand \fR=\fPbool
1243If this option is set, fio will use the normal read,write blocksize settings
1244as sequential,random blocksize settings instead. Any random read or write
1245will use the WRITE blocksize settings, and any sequential read or write will
1246use the READ blocksize settings.
1247.TP
1248.BI blockalign \fR=\fPint[,int][,int] "\fR,\fB ba" \fR=\fPint[,int][,int]
1249Boundary to which fio will align random I/O units. Default:
1250\fBblocksize\fR. Minimum alignment is typically 512b for using direct
1251I/O, though it usually depends on the hardware block size. This option is
1252mutually exclusive with using a random map for files, so it will turn off
1253that option. Comma\-separated values may be specified for reads, writes, and
1254trims as described in \fBblocksize\fR.
1255.SS "Buffers and memory"
1256.TP
1257.BI zero_buffers
1258Initialize buffers with all zeros. Default: fill buffers with random data.
1259.TP
1260.BI refill_buffers
1261If this option is given, fio will refill the I/O buffers on every
1262submit. The default is to only fill it at init time and reuse that
1263data. Only makes sense if zero_buffers isn't specified, naturally. If data
1264verification is enabled, \fBrefill_buffers\fR is also automatically enabled.
1265.TP
1266.BI scramble_buffers \fR=\fPbool
1267If \fBrefill_buffers\fR is too costly and the target is using data
1268deduplication, then setting this option will slightly modify the I/O buffer
1269contents to defeat normal de\-dupe attempts. This is not enough to defeat
1270more clever block compression attempts, but it will stop naive dedupe of
1271blocks. Default: true.
1272.TP
1273.BI buffer_compress_percentage \fR=\fPint
72592780
SW
1274If this is set, then fio will attempt to provide I/O buffer content
1275(on WRITEs) that compresses to the specified level. Fio does this by
1276providing a mix of random data followed by fixed pattern data. The
1277fixed pattern is either zeros, or the pattern specified by
1278\fBbuffer_pattern\fR. If the \fBbuffer_pattern\fR option is used, it
1279might skew the compression ratio slightly. Setting
1280\fBbuffer_compress_percentage\fR to a value other than 100 will also
1281enable \fBrefill_buffers\fR in order to reduce the likelihood that
1282adjacent blocks are so similar that they over compress when seen
1283together. See \fBbuffer_compress_chunk\fR for how to set a finer or
1284coarser granularity of the random/fixed data regions. Defaults to unset
1285i.e., buffer data will not adhere to any compression level.
523bad63
TK
1286.TP
1287.BI buffer_compress_chunk \fR=\fPint
72592780
SW
1288This setting allows fio to manage how big the random/fixed data region
1289is when using \fBbuffer_compress_percentage\fR. When
1290\fBbuffer_compress_chunk\fR is set to some non-zero value smaller than the
1291block size, fio can repeat the random/fixed region throughout the I/O
1292buffer at the specified interval (which particularly useful when
1293bigger block sizes are used for a job). When set to 0, fio will use a
1294chunk size that matches the block size resulting in a single
1295random/fixed region within the I/O buffer. Defaults to 512. When the
1296unit is omitted, the value is interpreted in bytes.
523bad63
TK
1297.TP
1298.BI buffer_pattern \fR=\fPstr
1299If set, fio will fill the I/O buffers with this pattern or with the contents
1300of a file. If not set, the contents of I/O buffers are defined by the other
1301options related to buffer contents. The setting can be any pattern of bytes,
1302and can be prefixed with 0x for hex values. It may also be a string, where
1303the string must then be wrapped with "". Or it may also be a filename,
1304where the filename must be wrapped with '' in which case the file is
1305opened and read. Note that not all the file contents will be read if that
1306would cause the buffers to overflow. So, for example:
1307.RS
1308.RS
1309.P
1310.PD 0
1311buffer_pattern='filename'
1312.P
1313or:
1314.P
1315buffer_pattern="abcd"
1316.P
1317or:
1318.P
1319buffer_pattern=\-12
1320.P
1321or:
1322.P
1323buffer_pattern=0xdeadface
1324.PD
1325.RE
1326.P
1327Also you can combine everything together in any order:
1328.RS
1329.P
1330buffer_pattern=0xdeadface"abcd"\-12'filename'
1331.RE
1332.RE
1333.TP
1334.BI dedupe_percentage \fR=\fPint
1335If set, fio will generate this percentage of identical buffers when
1336writing. These buffers will be naturally dedupable. The contents of the
1337buffers depend on what other buffer compression settings have been set. It's
1338possible to have the individual buffers either fully compressible, or not at
72592780
SW
1339all \-\- this option only controls the distribution of unique buffers. Setting
1340this option will also enable \fBrefill_buffers\fR to prevent every buffer
1341being identical.
523bad63
TK
1342.TP
1343.BI invalidate \fR=\fPbool
1344Invalidate the buffer/page cache parts of the files to be used prior to
1345starting I/O if the platform and file type support it. Defaults to true.
1346This will be ignored if \fBpre_read\fR is also specified for the
1347same job.
1348.TP
1349.BI sync \fR=\fPbool
1350Use synchronous I/O for buffered writes. For the majority of I/O engines,
1351this means using O_SYNC. Default: false.
1352.TP
1353.BI iomem \fR=\fPstr "\fR,\fP mem" \fR=\fPstr
1354Fio can use various types of memory as the I/O unit buffer. The allowed
1355values are:
1356.RS
1357.RS
1358.TP
1359.B malloc
1360Use memory from \fBmalloc\fR\|(3) as the buffers. Default memory type.
1361.TP
1362.B shm
1363Use shared memory as the buffers. Allocated through \fBshmget\fR\|(2).
1364.TP
1365.B shmhuge
1366Same as \fBshm\fR, but use huge pages as backing.
1367.TP
1368.B mmap
1369Use \fBmmap\fR\|(2) to allocate buffers. May either be anonymous memory, or can
1370be file backed if a filename is given after the option. The format
1371is `mem=mmap:/path/to/file'.
1372.TP
1373.B mmaphuge
1374Use a memory mapped huge file as the buffer backing. Append filename
1375after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file'.
1376.TP
1377.B mmapshared
1378Same as \fBmmap\fR, but use a MMAP_SHARED mapping.
1379.TP
1380.B cudamalloc
1381Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1382The \fBioengine\fR must be \fBrdma\fR.
1383.RE
1384.P
1385The area allocated is a function of the maximum allowed bs size for the job,
1386multiplied by the I/O depth given. Note that for \fBshmhuge\fR and
1387\fBmmaphuge\fR to work, the system must have free huge pages allocated. This
1388can normally be checked and set by reading/writing
1389`/proc/sys/vm/nr_hugepages' on a Linux system. Fio assumes a huge page
1390is 4MiB in size. So to calculate the number of huge pages you need for a
1391given job file, add up the I/O depth of all jobs (normally one unless
1392\fBiodepth\fR is used) and multiply by the maximum bs set. Then divide
1393that number by the huge page size. You can see the size of the huge pages in
1394`/proc/meminfo'. If no huge pages are allocated by having a non\-zero
1395number in `nr_hugepages', using \fBmmaphuge\fR or \fBshmhuge\fR will fail. Also
1396see \fBhugepage\-size\fR.
1397.P
1398\fBmmaphuge\fR also needs to have hugetlbfs mounted and the file location
1399should point there. So if it's mounted in `/huge', you would use
1400`mem=mmaphuge:/huge/somefile'.
1401.RE
1402.TP
1403.BI iomem_align \fR=\fPint "\fR,\fP mem_align" \fR=\fPint
1404This indicates the memory alignment of the I/O memory buffers. Note that
1405the given alignment is applied to the first I/O unit buffer, if using
1406\fBiodepth\fR the alignment of the following buffers are given by the
1407\fBbs\fR used. In other words, if using a \fBbs\fR that is a
1408multiple of the page sized in the system, all buffers will be aligned to
1409this value. If using a \fBbs\fR that is not page aligned, the alignment
1410of subsequent I/O memory buffers is the sum of the \fBiomem_align\fR and
1411\fBbs\fR used.
1412.TP
1413.BI hugepage\-size \fR=\fPint
1414Defines the size of a huge page. Must at least be equal to the system
1415setting, see `/proc/meminfo'. Defaults to 4MiB. Should probably
1416always be a multiple of megabytes, so using `hugepage\-size=Xm' is the
1417preferred way to set this to avoid setting a non\-pow\-2 bad value.
1418.TP
1419.BI lockmem \fR=\fPint
1420Pin the specified amount of memory with \fBmlock\fR\|(2). Can be used to
1421simulate a smaller amount of memory. The amount specified is per worker.
1422.SS "I/O size"
1423.TP
1424.BI size \fR=\fPint
1425The total size of file I/O for each thread of this job. Fio will run until
1426this many bytes has been transferred, unless runtime is limited by other options
1427(such as \fBruntime\fR, for instance, or increased/decreased by \fBio_size\fR).
1428Fio will divide this size between the available files determined by options
1429such as \fBnrfiles\fR, \fBfilename\fR, unless \fBfilesize\fR is
1430specified by the job. If the result of division happens to be 0, the size is
1431set to the physical size of the given files or devices if they exist.
1432If this option is not specified, fio will use the full size of the given
1433files or devices. If the files do not exist, size must be given. It is also
1434possible to give size as a percentage between 1 and 100. If `size=20%' is
1435given, fio will use 20% of the full size of the given files or devices.
1436Can be combined with \fBoffset\fR to constrain the start and end range
1437that I/O will be done within.
1438.TP
1439.BI io_size \fR=\fPint "\fR,\fB io_limit" \fR=\fPint
1440Normally fio operates within the region set by \fBsize\fR, which means
1441that the \fBsize\fR option sets both the region and size of I/O to be
1442performed. Sometimes that is not what you want. With this option, it is
1443possible to define just the amount of I/O that fio should do. For instance,
1444if \fBsize\fR is set to 20GiB and \fBio_size\fR is set to 5GiB, fio
1445will perform I/O within the first 20GiB but exit when 5GiB have been
1446done. The opposite is also possible \-\- if \fBsize\fR is set to 20GiB,
1447and \fBio_size\fR is set to 40GiB, then fio will do 40GiB of I/O within
1448the 0..20GiB region.
1449.TP
1450.BI filesize \fR=\fPirange(int)
1451Individual file sizes. May be a range, in which case fio will select sizes
1452for files at random within the given range and limited to \fBsize\fR in
1453total (if that is given). If not given, each created file is the same size.
1454This option overrides \fBsize\fR in terms of file size, which means
1455this value is used as a fixed size or possible range of each file.
1456.TP
1457.BI file_append \fR=\fPbool
1458Perform I/O after the end of the file. Normally fio will operate within the
1459size of a file. If this option is set, then fio will append to the file
1460instead. This has identical behavior to setting \fBoffset\fR to the size
1461of a file. This option is ignored on non\-regular files.
1462.TP
1463.BI fill_device \fR=\fPbool "\fR,\fB fill_fs" \fR=\fPbool
1464Sets size to something really large and waits for ENOSPC (no space left on
1465device) as the terminating condition. Only makes sense with sequential
1466write. For a read workload, the mount point will be filled first then I/O
1467started on the result. This option doesn't make sense if operating on a raw
1468device node, since the size of that is already known by the file system.
1469Additionally, writing beyond end\-of\-device will not return ENOSPC there.
1470.SS "I/O engine"
1471.TP
1472.BI ioengine \fR=\fPstr
1473Defines how the job issues I/O to the file. The following types are defined:
1474.RS
1475.RS
1476.TP
1477.B sync
1478Basic \fBread\fR\|(2) or \fBwrite\fR\|(2)
1479I/O. \fBlseek\fR\|(2) is used to position the I/O location.
1480See \fBfsync\fR and \fBfdatasync\fR for syncing write I/Os.
1481.TP
1482.B psync
1483Basic \fBpread\fR\|(2) or \fBpwrite\fR\|(2) I/O. Default on
1484all supported operating systems except for Windows.
1485.TP
1486.B vsync
1487Basic \fBreadv\fR\|(2) or \fBwritev\fR\|(2) I/O. Will emulate
1488queuing by coalescing adjacent I/Os into a single submission.
1489.TP
1490.B pvsync
1491Basic \fBpreadv\fR\|(2) or \fBpwritev\fR\|(2) I/O.
a46c5e01 1492.TP
2cafffbe
JA
1493.B pvsync2
1494Basic \fBpreadv2\fR\|(2) or \fBpwritev2\fR\|(2) I/O.
1495.TP
d60e92d1 1496.B libaio
523bad63
TK
1497Linux native asynchronous I/O. Note that Linux may only support
1498queued behavior with non\-buffered I/O (set `direct=1' or
1499`buffered=0').
1500This engine defines engine specific options.
d60e92d1
AC
1501.TP
1502.B posixaio
523bad63
TK
1503POSIX asynchronous I/O using \fBaio_read\fR\|(3) and
1504\fBaio_write\fR\|(3).
03e20d68
BC
1505.TP
1506.B solarisaio
1507Solaris native asynchronous I/O.
1508.TP
1509.B windowsaio
38f8c318 1510Windows native asynchronous I/O. Default on Windows.
d60e92d1
AC
1511.TP
1512.B mmap
523bad63
TK
1513File is memory mapped with \fBmmap\fR\|(2) and data copied
1514to/from using \fBmemcpy\fR\|(3).
d60e92d1
AC
1515.TP
1516.B splice
523bad63
TK
1517\fBsplice\fR\|(2) is used to transfer the data and
1518\fBvmsplice\fR\|(2) to transfer data from user space to the
1519kernel.
d60e92d1 1520.TP
d60e92d1 1521.B sg
523bad63
TK
1522SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1523ioctl, or if the target is an sg character device we use
1524\fBread\fR\|(2) and \fBwrite\fR\|(2) for asynchronous
1525I/O. Requires \fBfilename\fR option to specify either block or
52b81b7c 1526character devices. The sg engine includes engine specific options.
d60e92d1
AC
1527.TP
1528.B null
523bad63
TK
1529Doesn't transfer any data, just pretends to. This is mainly used to
1530exercise fio itself and for debugging/testing purposes.
d60e92d1
AC
1531.TP
1532.B net
523bad63
TK
1533Transfer over the network to given `host:port'. Depending on the
1534\fBprotocol\fR used, the \fBhostname\fR, \fBport\fR,
1535\fBlisten\fR and \fBfilename\fR options are used to specify
1536what sort of connection to make, while the \fBprotocol\fR option
1537determines which protocol will be used. This engine defines engine
1538specific options.
d60e92d1
AC
1539.TP
1540.B netsplice
523bad63
TK
1541Like \fBnet\fR, but uses \fBsplice\fR\|(2) and
1542\fBvmsplice\fR\|(2) to map data and send/receive.
1543This engine defines engine specific options.
d60e92d1 1544.TP
53aec0a4 1545.B cpuio
523bad63
TK
1546Doesn't transfer any data, but burns CPU cycles according to the
1547\fBcpuload\fR and \fBcpuchunks\fR options. Setting
1548\fBcpuload\fR\=85 will cause that job to do nothing but burn 85%
1549of the CPU. In case of SMP machines, use `numjobs=<nr_of_cpu>'
1550to get desired CPU usage, as the cpuload only loads a
1551single CPU at the desired rate. A job never finishes unless there is
1552at least one non\-cpuio job.
d60e92d1
AC
1553.TP
1554.B guasi
f1dd3fb1 1555The GUASI I/O engine is the Generic Userspace Asynchronous Syscall
523bad63
TK
1556Interface approach to async I/O. See \fIhttp://www.xmailserver.org/guasi\-lib.html\fR
1557for more info on GUASI.
d60e92d1 1558.TP
21b8aee8 1559.B rdma
523bad63
TK
1560The RDMA I/O engine supports both RDMA memory semantics
1561(RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
609ac152
SB
1562InfiniBand, RoCE and iWARP protocols. This engine defines engine
1563specific options.
d54fce84
DM
1564.TP
1565.B falloc
523bad63
TK
1566I/O engine that does regular fallocate to simulate data transfer as
1567fio ioengine.
1568.RS
1569.P
1570.PD 0
1571DDIR_READ does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1572.P
1573DIR_WRITE does fallocate(,mode = 0).
1574.P
1575DDIR_TRIM does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1576.PD
1577.RE
1578.TP
1579.B ftruncate
1580I/O engine that sends \fBftruncate\fR\|(2) operations in response
1581to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
1582size to the current block offset. \fBblocksize\fR is ignored.
d54fce84
DM
1583.TP
1584.B e4defrag
523bad63
TK
1585I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1586defragment activity in request to DDIR_WRITE event.
0d978694 1587.TP
d5f9b0ea
IF
1588.B rados
1589I/O engine supporting direct access to Ceph Reliable Autonomic Distributed
1590Object Store (RADOS) via librados. This ioengine defines engine specific
1591options.
1592.TP
0d978694 1593.B rbd
523bad63
TK
1594I/O engine supporting direct access to Ceph Rados Block Devices
1595(RBD) via librbd without the need to use the kernel rbd driver. This
1596ioengine defines engine specific options.
a7c386f4 1597.TP
1598.B gfapi
523bad63
TK
1599Using GlusterFS libgfapi sync interface to direct access to
1600GlusterFS volumes without having to go through FUSE. This ioengine
1601defines engine specific options.
cc47f094 1602.TP
1603.B gfapi_async
523bad63
TK
1604Using GlusterFS libgfapi async interface to direct access to
1605GlusterFS volumes without having to go through FUSE. This ioengine
1606defines engine specific options.
1b10477b 1607.TP
b74e419e 1608.B libhdfs
523bad63
TK
1609Read and write through Hadoop (HDFS). The \fBfilename\fR option
1610is used to specify host,port of the hdfs name\-node to connect. This
1611engine interprets offsets a little differently. In HDFS, files once
1612created cannot be modified so random writes are not possible. To
1613imitate this the libhdfs engine expects a bunch of small files to be
1614created over HDFS and will randomly pick a file from them
1615based on the offset generated by fio backend (see the example
1616job file to create such files, use `rw=write' option). Please
1617note, it may be necessary to set environment variables to work
1618with HDFS/libhdfs properly. Each job uses its own connection to
1619HDFS.
65fa28ca
DE
1620.TP
1621.B mtd
523bad63
TK
1622Read, write and erase an MTD character device (e.g.,
1623`/dev/mtd0'). Discards are treated as erases. Depending on the
1624underlying device type, the I/O may have to go in a certain pattern,
1625e.g., on NAND, writing sequentially to erase blocks and discarding
1626before overwriting. The \fBtrimwrite\fR mode works well for this
65fa28ca 1627constraint.
5c4ef02e
JA
1628.TP
1629.B pmemblk
523bad63
TK
1630Read and write using filesystem DAX to a file on a filesystem
1631mounted with DAX on a persistent memory device through the NVML
1632libpmemblk library.
104ee4de 1633.TP
523bad63
TK
1634.B dev\-dax
1635Read and write using device DAX to a persistent memory device (e.g.,
1636/dev/dax0.0) through the NVML libpmem library.
d60e92d1 1637.TP
523bad63
TK
1638.B external
1639Prefix to specify loading an external I/O engine object file. Append
1640the engine filename, e.g. `ioengine=external:/tmp/foo.o' to load
d243fd6d
TK
1641ioengine `foo.o' in `/tmp'. The path can be either
1642absolute or relative. See `engines/skeleton_external.c' in the fio source for
1643details of writing an external I/O engine.
1216cc5a
JB
1644.TP
1645.B filecreate
b71968b1
SW
1646Simply create the files and do no I/O to them. You still need to set
1647\fBfilesize\fR so that all the accounting still occurs, but no actual I/O will be
1648done other than creating the file.
ae0db592
TI
1649.TP
1650.B libpmem
1651Read and write using mmap I/O to a file on a filesystem
1652mounted with DAX on a persistent memory device through the NVML
1653libpmem library.
523bad63
TK
1654.SS "I/O engine specific parameters"
1655In addition, there are some parameters which are only valid when a specific
1656\fBioengine\fR is in use. These are used identically to normal parameters,
1657with the caveat that when used on the command line, they must come after the
1658\fBioengine\fR that defines them is selected.
d60e92d1 1659.TP
523bad63
TK
1660.BI (libaio)userspace_reap
1661Normally, with the libaio engine in use, fio will use the
1662\fBio_getevents\fR\|(3) system call to reap newly returned events. With
1663this flag turned on, the AIO ring will be read directly from user\-space to
1664reap events. The reaping mode is only enabled when polling for a minimum of
16650 events (e.g. when `iodepth_batch_complete=0').
3ce9dcaf 1666.TP
523bad63
TK
1667.BI (pvsync2)hipri
1668Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
1669than normal.
82407585 1670.TP
523bad63
TK
1671.BI (pvsync2)hipri_percentage
1672When hipri is set this determines the probability of a pvsync2 I/O being high
1673priority. The default is 100%.
d60e92d1 1674.TP
523bad63
TK
1675.BI (cpuio)cpuload \fR=\fPint
1676Attempt to use the specified percentage of CPU cycles. This is a mandatory
1677option when using cpuio I/O engine.
997b5680 1678.TP
523bad63
TK
1679.BI (cpuio)cpuchunks \fR=\fPint
1680Split the load into cycles of the given time. In microseconds.
1ad01bd1 1681.TP
523bad63
TK
1682.BI (cpuio)exit_on_io_done \fR=\fPbool
1683Detect when I/O threads are done, then exit.
d60e92d1 1684.TP
523bad63
TK
1685.BI (libhdfs)namenode \fR=\fPstr
1686The hostname or IP address of a HDFS cluster namenode to contact.
d01612f3 1687.TP
523bad63
TK
1688.BI (libhdfs)port
1689The listening port of the HFDS cluster namenode.
d60e92d1 1690.TP
523bad63
TK
1691.BI (netsplice,net)port
1692The TCP or UDP port to bind to or connect to. If this is used with
1693\fBnumjobs\fR to spawn multiple instances of the same job type, then
1694this will be the starting port number since fio will use a range of
1695ports.
d60e92d1 1696.TP
609ac152
SB
1697.BI (rdma)port
1698The port to use for RDMA-CM communication. This should be the same
1699value on the client and the server side.
1700.TP
1701.BI (netsplice,net, rdma)hostname \fR=\fPstr
1702The hostname or IP address to use for TCP, UDP or RDMA-CM based I/O.
1703If the job is a TCP listener or UDP reader, the hostname is not used
1704and must be omitted unless it is a valid UDP multicast address.
591e9e06 1705.TP
523bad63
TK
1706.BI (netsplice,net)interface \fR=\fPstr
1707The IP address of the network interface used to send or receive UDP
1708multicast.
ddf24e42 1709.TP
523bad63
TK
1710.BI (netsplice,net)ttl \fR=\fPint
1711Time\-to\-live value for outgoing UDP multicast packets. Default: 1.
d60e92d1 1712.TP
523bad63
TK
1713.BI (netsplice,net)nodelay \fR=\fPbool
1714Set TCP_NODELAY on TCP connections.
fa769d44 1715.TP
523bad63
TK
1716.BI (netsplice,net)protocol \fR=\fPstr "\fR,\fP proto" \fR=\fPstr
1717The network protocol to use. Accepted values are:
1718.RS
e76b1da4
JA
1719.RS
1720.TP
523bad63
TK
1721.B tcp
1722Transmission control protocol.
e76b1da4 1723.TP
523bad63
TK
1724.B tcpv6
1725Transmission control protocol V6.
e76b1da4 1726.TP
523bad63
TK
1727.B udp
1728User datagram protocol.
1729.TP
1730.B udpv6
1731User datagram protocol V6.
e76b1da4 1732.TP
523bad63
TK
1733.B unix
1734UNIX domain socket.
e76b1da4
JA
1735.RE
1736.P
523bad63
TK
1737When the protocol is TCP or UDP, the port must also be given, as well as the
1738hostname if the job is a TCP listener or UDP reader. For unix sockets, the
1739normal \fBfilename\fR option should be used and the port is invalid.
1740.RE
1741.TP
1742.BI (netsplice,net)listen
1743For TCP network connections, tell fio to listen for incoming connections
1744rather than initiating an outgoing connection. The \fBhostname\fR must
1745be omitted if this option is used.
1746.TP
1747.BI (netsplice,net)pingpong
1748Normally a network writer will just continue writing data, and a network
1749reader will just consume packages. If `pingpong=1' is set, a writer will
1750send its normal payload to the reader, then wait for the reader to send the
1751same payload back. This allows fio to measure network latencies. The
1752submission and completion latencies then measure local time spent sending or
1753receiving, and the completion latency measures how long it took for the
1754other end to receive and send back. For UDP multicast traffic
1755`pingpong=1' should only be set for a single reader when multiple readers
1756are listening to the same address.
1757.TP
1758.BI (netsplice,net)window_size \fR=\fPint
1759Set the desired socket buffer size for the connection.
e76b1da4 1760.TP
523bad63
TK
1761.BI (netsplice,net)mss \fR=\fPint
1762Set the TCP maximum segment size (TCP_MAXSEG).
d60e92d1 1763.TP
523bad63
TK
1764.BI (e4defrag)donorname \fR=\fPstr
1765File will be used as a block donor (swap extents between files).
d60e92d1 1766.TP
523bad63
TK
1767.BI (e4defrag)inplace \fR=\fPint
1768Configure donor file blocks allocation strategy:
1769.RS
1770.RS
d60e92d1 1771.TP
523bad63
TK
1772.B 0
1773Default. Preallocate donor's file on init.
d60e92d1 1774.TP
523bad63
TK
1775.B 1
1776Allocate space immediately inside defragment event, and free right
1777after event.
1778.RE
1779.RE
d60e92d1 1780.TP
d5f9b0ea 1781.BI (rbd,rados)clustername \fR=\fPstr
523bad63 1782Specifies the name of the Ceph cluster.
92d42d69 1783.TP
523bad63
TK
1784.BI (rbd)rbdname \fR=\fPstr
1785Specifies the name of the RBD.
92d42d69 1786.TP
d5f9b0ea
IF
1787.BI (rbd,rados)pool \fR=\fPstr
1788Specifies the name of the Ceph pool containing RBD or RADOS data.
92d42d69 1789.TP
d5f9b0ea 1790.BI (rbd,rados)clientname \fR=\fPstr
523bad63
TK
1791Specifies the username (without the 'client.' prefix) used to access the
1792Ceph cluster. If the \fBclustername\fR is specified, the \fBclientname\fR shall be
1793the full *type.id* string. If no type. prefix is given, fio will add 'client.'
1794by default.
92d42d69 1795.TP
d5f9b0ea
IF
1796.BI (rbd,rados)busy_poll \fR=\fPbool
1797Poll store instead of waiting for completion. Usually this provides better
1798throughput at cost of higher(up to 100%) CPU utilization.
1799.TP
523bad63
TK
1800.BI (mtd)skip_bad \fR=\fPbool
1801Skip operations against known bad blocks.
8116fd24 1802.TP
523bad63
TK
1803.BI (libhdfs)hdfsdirectory
1804libhdfs will create chunk in this HDFS directory.
e0a04ac1 1805.TP
523bad63
TK
1806.BI (libhdfs)chunk_size
1807The size of the chunk to use for each file.
609ac152
SB
1808.TP
1809.BI (rdma)verb \fR=\fPstr
1810The RDMA verb to use on this side of the RDMA ioengine
1811connection. Valid values are write, read, send and recv. These
1812correspond to the equivalent RDMA verbs (e.g. write = rdma_write
1813etc.). Note that this only needs to be specified on the client side of
1814the connection. See the examples folder.
1815.TP
1816.BI (rdma)bindname \fR=\fPstr
1817The name to use to bind the local RDMA-CM connection to a local RDMA
1818device. This could be a hostname or an IPv4 or IPv6 address. On the
1819server side this will be passed into the rdma_bind_addr() function and
1820on the client site it will be used in the rdma_resolve_add()
1821function. This can be useful when multiple paths exist between the
1822client and the server or in certain loopback configurations.
52b81b7c
KD
1823.TP
1824.BI (sg)readfua \fR=\fPbool
1825With readfua option set to 1, read operations include the force
1826unit access (fua) flag. Default: 0.
1827.TP
1828.BI (sg)writefua \fR=\fPbool
1829With writefua option set to 1, write operations include the force
1830unit access (fua) flag. Default: 0.
523bad63
TK
1831.SS "I/O depth"
1832.TP
1833.BI iodepth \fR=\fPint
1834Number of I/O units to keep in flight against the file. Note that
1835increasing \fBiodepth\fR beyond 1 will not affect synchronous ioengines (except
1836for small degrees when \fBverify_async\fR is in use). Even async
1837engines may impose OS restrictions causing the desired depth not to be
1838achieved. This may happen on Linux when using libaio and not setting
1839`direct=1', since buffered I/O is not async on that OS. Keep an
1840eye on the I/O depth distribution in the fio output to verify that the
1841achieved depth is as expected. Default: 1.
1842.TP
1843.BI iodepth_batch_submit \fR=\fPint "\fR,\fP iodepth_batch" \fR=\fPint
1844This defines how many pieces of I/O to submit at once. It defaults to 1
1845which means that we submit each I/O as soon as it is available, but can be
1846raised to submit bigger batches of I/O at the time. If it is set to 0 the
1847\fBiodepth\fR value will be used.
1848.TP
1849.BI iodepth_batch_complete_min \fR=\fPint "\fR,\fP iodepth_batch_complete" \fR=\fPint
1850This defines how many pieces of I/O to retrieve at once. It defaults to 1
1851which means that we'll ask for a minimum of 1 I/O in the retrieval process
1852from the kernel. The I/O retrieval will go on until we hit the limit set by
1853\fBiodepth_low\fR. If this variable is set to 0, then fio will always
1854check for completed events before queuing more I/O. This helps reduce I/O
1855latency, at the cost of more retrieval system calls.
1856.TP
1857.BI iodepth_batch_complete_max \fR=\fPint
1858This defines maximum pieces of I/O to retrieve at once. This variable should
1859be used along with \fBiodepth_batch_complete_min\fR=\fIint\fR variable,
1860specifying the range of min and max amount of I/O which should be
1861retrieved. By default it is equal to \fBiodepth_batch_complete_min\fR
1862value. Example #1:
e0a04ac1 1863.RS
e0a04ac1 1864.RS
e0a04ac1 1865.P
523bad63
TK
1866.PD 0
1867iodepth_batch_complete_min=1
e0a04ac1 1868.P
523bad63
TK
1869iodepth_batch_complete_max=<iodepth>
1870.PD
e0a04ac1
JA
1871.RE
1872.P
523bad63
TK
1873which means that we will retrieve at least 1 I/O and up to the whole
1874submitted queue depth. If none of I/O has been completed yet, we will wait.
1875Example #2:
e8b1961d 1876.RS
523bad63
TK
1877.P
1878.PD 0
1879iodepth_batch_complete_min=0
1880.P
1881iodepth_batch_complete_max=<iodepth>
1882.PD
e8b1961d
JA
1883.RE
1884.P
523bad63
TK
1885which means that we can retrieve up to the whole submitted queue depth, but
1886if none of I/O has been completed yet, we will NOT wait and immediately exit
1887the system call. In this example we simply do polling.
1888.RE
e8b1961d 1889.TP
523bad63
TK
1890.BI iodepth_low \fR=\fPint
1891The low water mark indicating when to start filling the queue
1892again. Defaults to the same as \fBiodepth\fR, meaning that fio will
1893attempt to keep the queue full at all times. If \fBiodepth\fR is set to
1894e.g. 16 and \fBiodepth_low\fR is set to 4, then after fio has filled the queue of
189516 requests, it will let the depth drain down to 4 before starting to fill
1896it again.
d60e92d1 1897.TP
523bad63
TK
1898.BI serialize_overlap \fR=\fPbool
1899Serialize in-flight I/Os that might otherwise cause or suffer from data races.
1900When two or more I/Os are submitted simultaneously, there is no guarantee that
1901the I/Os will be processed or completed in the submitted order. Further, if
1902two or more of those I/Os are writes, any overlapping region between them can
1903become indeterminate/undefined on certain storage. These issues can cause
1904verification to fail erratically when at least one of the racing I/Os is
1905changing data and the overlapping region has a non-zero size. Setting
1906\fBserialize_overlap\fR tells fio to avoid provoking this behavior by explicitly
1907serializing in-flight I/Os that have a non-zero overlap. Note that setting
1908this option can reduce both performance and the \fBiodepth\fR achieved.
1909Additionally this option does not work when \fBio_submit_mode\fR is set to
1910offload. Default: false.
d60e92d1 1911.TP
523bad63
TK
1912.BI io_submit_mode \fR=\fPstr
1913This option controls how fio submits the I/O to the I/O engine. The default
1914is `inline', which means that the fio job threads submit and reap I/O
1915directly. If set to `offload', the job threads will offload I/O submission
1916to a dedicated pool of I/O threads. This requires some coordination and thus
1917has a bit of extra overhead, especially for lower queue depth I/O where it
1918can increase latencies. The benefit is that fio can manage submission rates
1919independently of the device completion rates. This avoids skewed latency
1920reporting if I/O gets backed up on the device side (the coordinated omission
1921problem).
1922.SS "I/O rate"
d60e92d1 1923.TP
523bad63
TK
1924.BI thinktime \fR=\fPtime
1925Stall the job for the specified period of time after an I/O has completed before issuing the
1926next. May be used to simulate processing being done by an application.
1927When the unit is omitted, the value is interpreted in microseconds. See
1928\fBthinktime_blocks\fR and \fBthinktime_spin\fR.
d60e92d1 1929.TP
523bad63
TK
1930.BI thinktime_spin \fR=\fPtime
1931Only valid if \fBthinktime\fR is set \- pretend to spend CPU time doing
1932something with the data received, before falling back to sleeping for the
1933rest of the period specified by \fBthinktime\fR. When the unit is
1934omitted, the value is interpreted in microseconds.
d60e92d1
AC
1935.TP
1936.BI thinktime_blocks \fR=\fPint
523bad63
TK
1937Only valid if \fBthinktime\fR is set \- control how many blocks to issue,
1938before waiting \fBthinktime\fR usecs. If not set, defaults to 1 which will make
1939fio wait \fBthinktime\fR usecs after every block. This effectively makes any
1940queue depth setting redundant, since no more than 1 I/O will be queued
1941before we have to complete it and do our \fBthinktime\fR. In other words, this
1942setting effectively caps the queue depth if the latter is larger.
d60e92d1 1943.TP
6d500c2e 1944.BI rate \fR=\fPint[,int][,int]
523bad63
TK
1945Cap the bandwidth used by this job. The number is in bytes/sec, the normal
1946suffix rules apply. Comma\-separated values may be specified for reads,
1947writes, and trims as described in \fBblocksize\fR.
1948.RS
1949.P
1950For example, using `rate=1m,500k' would limit reads to 1MiB/sec and writes to
1951500KiB/sec. Capping only reads or writes can be done with `rate=,500k' or
1952`rate=500k,' where the former will only limit writes (to 500KiB/sec) and the
1953latter will only limit reads.
1954.RE
d60e92d1 1955.TP
6d500c2e 1956.BI rate_min \fR=\fPint[,int][,int]
523bad63
TK
1957Tell fio to do whatever it can to maintain at least this bandwidth. Failing
1958to meet this requirement will cause the job to exit. Comma\-separated values
1959may be specified for reads, writes, and trims as described in
1960\fBblocksize\fR.
d60e92d1 1961.TP
6d500c2e 1962.BI rate_iops \fR=\fPint[,int][,int]
523bad63
TK
1963Cap the bandwidth to this number of IOPS. Basically the same as
1964\fBrate\fR, just specified independently of bandwidth. If the job is
1965given a block size range instead of a fixed value, the smallest block size
1966is used as the metric. Comma\-separated values may be specified for reads,
1967writes, and trims as described in \fBblocksize\fR.
d60e92d1 1968.TP
6d500c2e 1969.BI rate_iops_min \fR=\fPint[,int][,int]
523bad63
TK
1970If fio doesn't meet this rate of I/O, it will cause the job to exit.
1971Comma\-separated values may be specified for reads, writes, and trims as
1972described in \fBblocksize\fR.
d60e92d1 1973.TP
6de65959 1974.BI rate_process \fR=\fPstr
523bad63
TK
1975This option controls how fio manages rated I/O submissions. The default is
1976`linear', which submits I/O in a linear fashion with fixed delays between
1977I/Os that gets adjusted based on I/O completion rates. If this is set to
1978`poisson', fio will submit I/O based on a more real world random request
6de65959 1979flow, known as the Poisson process
523bad63 1980(\fIhttps://en.wikipedia.org/wiki/Poisson_point_process\fR). The lambda will be
5d02b083 198110^6 / IOPS for the given workload.
1a9bf814
JA
1982.TP
1983.BI rate_ignore_thinktime \fR=\fPbool
1984By default, fio will attempt to catch up to the specified rate setting, if any
1985kind of thinktime setting was used. If this option is set, then fio will
1986ignore the thinktime and continue doing IO at the specified rate, instead of
1987entering a catch-up mode after thinktime is done.
523bad63 1988.SS "I/O latency"
ff6bb260 1989.TP
523bad63 1990.BI latency_target \fR=\fPtime
3e260a46 1991If set, fio will attempt to find the max performance point that the given
523bad63
TK
1992workload will run at while maintaining a latency below this target. When
1993the unit is omitted, the value is interpreted in microseconds. See
1994\fBlatency_window\fR and \fBlatency_percentile\fR.
3e260a46 1995.TP
523bad63 1996.BI latency_window \fR=\fPtime
3e260a46 1997Used with \fBlatency_target\fR to specify the sample window that the job
523bad63
TK
1998is run at varying queue depths to test the performance. When the unit is
1999omitted, the value is interpreted in microseconds.
3e260a46
JA
2000.TP
2001.BI latency_percentile \fR=\fPfloat
523bad63
TK
2002The percentage of I/Os that must fall within the criteria specified by
2003\fBlatency_target\fR and \fBlatency_window\fR. If not set, this
2004defaults to 100.0, meaning that all I/Os must be equal or below to the value
2005set by \fBlatency_target\fR.
2006.TP
2007.BI max_latency \fR=\fPtime
2008If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
2009maximum latency. When the unit is omitted, the value is interpreted in
2010microseconds.
2011.TP
2012.BI rate_cycle \fR=\fPint
2013Average bandwidth for \fBrate\fR and \fBrate_min\fR over this number
2014of milliseconds. Defaults to 1000.
2015.SS "I/O replay"
2016.TP
2017.BI write_iolog \fR=\fPstr
2018Write the issued I/O patterns to the specified file. See
2019\fBread_iolog\fR. Specify a separate file for each job, otherwise the
2020iologs will be interspersed and the file may be corrupt.
2021.TP
2022.BI read_iolog \fR=\fPstr
2023Open an iolog with the specified filename and replay the I/O patterns it
2024contains. This can be used to store a workload and replay it sometime
2025later. The iolog given may also be a blktrace binary file, which allows fio
2026to replay a workload captured by blktrace. See
2027\fBblktrace\fR\|(8) for how to capture such logging data. For blktrace
2028replay, the file needs to be turned into a blkparse binary data file first
2029(`blkparse <device> \-o /dev/null \-d file_for_fio.bin').
3e260a46 2030.TP
523bad63
TK
2031.BI replay_no_stall \fR=\fPbool
2032When replaying I/O with \fBread_iolog\fR the default behavior is to
2033attempt to respect the timestamps within the log and replay them with the
2034appropriate delay between IOPS. By setting this variable fio will not
2035respect the timestamps and attempt to replay them as fast as possible while
2036still respecting ordering. The result is the same I/O pattern to a given
2037device, but different timings.
2038.TP
2039.BI replay_redirect \fR=\fPstr
2040While replaying I/O patterns using \fBread_iolog\fR the default behavior
2041is to replay the IOPS onto the major/minor device that each IOP was recorded
2042from. This is sometimes undesirable because on a different machine those
2043major/minor numbers can map to a different device. Changing hardware on the
2044same system can also result in a different major/minor mapping.
2045\fBreplay_redirect\fR causes all I/Os to be replayed onto the single specified
2046device regardless of the device it was recorded
2047from. i.e. `replay_redirect=/dev/sdc' would cause all I/O
2048in the blktrace or iolog to be replayed onto `/dev/sdc'. This means
2049multiple devices will be replayed onto a single device, if the trace
2050contains multiple devices. If you want multiple devices to be replayed
2051concurrently to multiple redirected devices you must blkparse your trace
2052into separate traces and replay them with independent fio invocations.
2053Unfortunately this also breaks the strict time ordering between multiple
2054device accesses.
2055.TP
2056.BI replay_align \fR=\fPint
2057Force alignment of I/O offsets and lengths in a trace to this power of 2
2058value.
2059.TP
2060.BI replay_scale \fR=\fPint
2061Scale sector offsets down by this factor when replaying traces.
2062.SS "Threads, processes and job synchronization"
2063.TP
2064.BI thread
2065Fio defaults to creating jobs by using fork, however if this option is
2066given, fio will create jobs by using POSIX Threads' function
2067\fBpthread_create\fR\|(3) to create threads instead.
2068.TP
2069.BI wait_for \fR=\fPstr
2070If set, the current job won't be started until all workers of the specified
2071waitee job are done.
2072.\" ignore blank line here from HOWTO as it looks normal without it
2073\fBwait_for\fR operates on the job name basis, so there are a few
2074limitations. First, the waitee must be defined prior to the waiter job
2075(meaning no forward references). Second, if a job is being referenced as a
2076waitee, it must have a unique name (no duplicate waitees).
2077.TP
2078.BI nice \fR=\fPint
2079Run the job with the given nice value. See man \fBnice\fR\|(2).
2080.\" ignore blank line here from HOWTO as it looks normal without it
2081On Windows, values less than \-15 set the process class to "High"; \-1 through
2082\-15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2083priority class.
2084.TP
2085.BI prio \fR=\fPint
2086Set the I/O priority value of this job. Linux limits us to a positive value
2087between 0 and 7, with 0 being the highest. See man
2088\fBionice\fR\|(1). Refer to an appropriate manpage for other operating
2089systems since meaning of priority may differ.
2090.TP
2091.BI prioclass \fR=\fPint
2092Set the I/O priority class. See man \fBionice\fR\|(1).
15501535 2093.TP
d60e92d1 2094.BI cpumask \fR=\fPint
523bad63
TK
2095Set the CPU affinity of this job. The parameter given is a bit mask of
2096allowed CPUs the job may run on. So if you want the allowed CPUs to be 1
2097and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2098\fBsched_setaffinity\fR\|(2). This may not work on all supported
2099operating systems or kernel versions. This option doesn't work well for a
2100higher CPU count than what you can store in an integer mask, so it can only
2101control cpus 1\-32. For boxes with larger CPU counts, use
2102\fBcpus_allowed\fR.
d60e92d1
AC
2103.TP
2104.BI cpus_allowed \fR=\fPstr
523bad63
TK
2105Controls the same options as \fBcpumask\fR, but accepts a textual
2106specification of the permitted CPUs instead. So to use CPUs 1 and 5 you
2107would specify `cpus_allowed=1,5'. This option also allows a range of CPUs
2108to be specified \-\- say you wanted a binding to CPUs 1, 5, and 8 to 15, you
2109would set `cpus_allowed=1,5,8\-15'.
d60e92d1 2110.TP
c2acfbac 2111.BI cpus_allowed_policy \fR=\fPstr
523bad63
TK
2112Set the policy of how fio distributes the CPUs specified by
2113\fBcpus_allowed\fR or \fBcpumask\fR. Two policies are supported:
c2acfbac
JA
2114.RS
2115.RS
2116.TP
2117.B shared
2118All jobs will share the CPU set specified.
2119.TP
2120.B split
2121Each job will get a unique CPU from the CPU set.
2122.RE
2123.P
523bad63
TK
2124\fBshared\fR is the default behavior, if the option isn't specified. If
2125\fBsplit\fR is specified, then fio will will assign one cpu per job. If not
2126enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2127in the set.
c2acfbac 2128.RE
c2acfbac 2129.TP
d0b937ed 2130.BI numa_cpu_nodes \fR=\fPstr
cecbfd47 2131Set this job running on specified NUMA nodes' CPUs. The arguments allow
523bad63
TK
2132comma delimited list of cpu numbers, A\-B ranges, or `all'. Note, to enable
2133NUMA options support, fio must be built on a system with libnuma\-dev(el)
2134installed.
d0b937ed
YR
2135.TP
2136.BI numa_mem_policy \fR=\fPstr
523bad63
TK
2137Set this job's memory policy and corresponding NUMA nodes. Format of the
2138arguments:
39c7a2ca
VF
2139.RS
2140.RS
523bad63
TK
2141.P
2142<mode>[:<nodelist>]
39c7a2ca 2143.RE
523bad63 2144.P
f1dd3fb1 2145`mode' is one of the following memory policies: `default', `prefer',
523bad63
TK
2146`bind', `interleave' or `local'. For `default' and `local' memory
2147policies, no node needs to be specified. For `prefer', only one node is
2148allowed. For `bind' and `interleave' the `nodelist' may be as
2149follows: a comma delimited list of numbers, A\-B ranges, or `all'.
39c7a2ca
VF
2150.RE
2151.TP
523bad63
TK
2152.BI cgroup \fR=\fPstr
2153Add job to this control group. If it doesn't exist, it will be created. The
2154system must have a mounted cgroup blkio mount point for this to work. If
2155your system doesn't have it mounted, you can do so with:
d60e92d1
AC
2156.RS
2157.RS
d60e92d1 2158.P
523bad63
TK
2159# mount \-t cgroup \-o blkio none /cgroup
2160.RE
d60e92d1
AC
2161.RE
2162.TP
523bad63
TK
2163.BI cgroup_weight \fR=\fPint
2164Set the weight of the cgroup to this value. See the documentation that comes
2165with the kernel, allowed values are in the range of 100..1000.
d60e92d1 2166.TP
523bad63
TK
2167.BI cgroup_nodelete \fR=\fPbool
2168Normally fio will delete the cgroups it has created after the job
2169completion. To override this behavior and to leave cgroups around after the
2170job completion, set `cgroup_nodelete=1'. This can be useful if one wants
2171to inspect various cgroup files after job completion. Default: false.
c8eeb9df 2172.TP
523bad63
TK
2173.BI flow_id \fR=\fPint
2174The ID of the flow. If not specified, it defaults to being a global
2175flow. See \fBflow\fR.
d60e92d1 2176.TP
523bad63
TK
2177.BI flow \fR=\fPint
2178Weight in token\-based flow control. If this value is used, then there is
2179a 'flow counter' which is used to regulate the proportion of activity between
2180two or more jobs. Fio attempts to keep this flow counter near zero. The
2181\fBflow\fR parameter stands for how much should be added or subtracted to the
2182flow counter on each iteration of the main I/O loop. That is, if one job has
2183`flow=8' and another job has `flow=\-1', then there will be a roughly 1:8
2184ratio in how much one runs vs the other.
d60e92d1 2185.TP
523bad63
TK
2186.BI flow_watermark \fR=\fPint
2187The maximum value that the absolute value of the flow counter is allowed to
2188reach before the job must wait for a lower value of the counter.
6b7f6851 2189.TP
523bad63
TK
2190.BI flow_sleep \fR=\fPint
2191The period of time, in microseconds, to wait after the flow watermark has
2192been exceeded before retrying operations.
25460cf6 2193.TP
523bad63
TK
2194.BI stonewall "\fR,\fB wait_for_previous"
2195Wait for preceding jobs in the job file to exit, before starting this
2196one. Can be used to insert serialization points in the job file. A stone
2197wall also implies starting a new reporting group, see
2198\fBgroup_reporting\fR.
2378826d 2199.TP
523bad63
TK
2200.BI exitall
2201By default, fio will continue running all other jobs when one job finishes
2202but sometimes this is not the desired action. Setting \fBexitall\fR will
2203instead make fio terminate all other jobs when one job finishes.
e81ecca3 2204.TP
523bad63
TK
2205.BI exec_prerun \fR=\fPstr
2206Before running this job, issue the command specified through
2207\fBsystem\fR\|(3). Output is redirected in a file called `jobname.prerun.txt'.
e9f48479 2208.TP
523bad63
TK
2209.BI exec_postrun \fR=\fPstr
2210After the job completes, issue the command specified though
2211\fBsystem\fR\|(3). Output is redirected in a file called `jobname.postrun.txt'.
d60e92d1 2212.TP
523bad63
TK
2213.BI uid \fR=\fPint
2214Instead of running as the invoking user, set the user ID to this value
2215before the thread/process does any work.
39c1c323 2216.TP
523bad63
TK
2217.BI gid \fR=\fPint
2218Set group ID, see \fBuid\fR.
2219.SS "Verification"
d60e92d1 2220.TP
589e88b7 2221.BI verify_only
523bad63 2222Do not perform specified workload, only verify data still matches previous
5e4c7118 2223invocation of this workload. This option allows one to check data multiple
523bad63
TK
2224times at a later date without overwriting it. This option makes sense only
2225for workloads that write data, and does not support workloads with the
5e4c7118
JA
2226\fBtime_based\fR option set.
2227.TP
d60e92d1 2228.BI do_verify \fR=\fPbool
523bad63
TK
2229Run the verify phase after a write phase. Only valid if \fBverify\fR is
2230set. Default: true.
d60e92d1
AC
2231.TP
2232.BI verify \fR=\fPstr
523bad63
TK
2233If writing to a file, fio can verify the file contents after each iteration
2234of the job. Each verification method also implies verification of special
2235header, which is written to the beginning of each block. This header also
2236includes meta information, like offset of the block, block number, timestamp
2237when block was written, etc. \fBverify\fR can be combined with
2238\fBverify_pattern\fR option. The allowed values are:
d60e92d1
AC
2239.RS
2240.RS
2241.TP
523bad63
TK
2242.B md5
2243Use an md5 sum of the data area and store it in the header of
2244each block.
2245.TP
2246.B crc64
2247Use an experimental crc64 sum of the data area and store it in the
2248header of each block.
2249.TP
2250.B crc32c
2251Use a crc32c sum of the data area and store it in the header of
2252each block. This will automatically use hardware acceleration
2253(e.g. SSE4.2 on an x86 or CRC crypto extensions on ARM64) but will
2254fall back to software crc32c if none is found. Generally the
f1dd3fb1 2255fastest checksum fio supports when hardware accelerated.
523bad63
TK
2256.TP
2257.B crc32c\-intel
2258Synonym for crc32c.
2259.TP
2260.B crc32
2261Use a crc32 sum of the data area and store it in the header of each
2262block.
2263.TP
2264.B crc16
2265Use a crc16 sum of the data area and store it in the header of each
2266block.
2267.TP
2268.B crc7
2269Use a crc7 sum of the data area and store it in the header of each
2270block.
2271.TP
2272.B xxhash
2273Use xxhash as the checksum function. Generally the fastest software
2274checksum that fio supports.
2275.TP
2276.B sha512
2277Use sha512 as the checksum function.
2278.TP
2279.B sha256
2280Use sha256 as the checksum function.
2281.TP
2282.B sha1
2283Use optimized sha1 as the checksum function.
2284.TP
2285.B sha3\-224
2286Use optimized sha3\-224 as the checksum function.
2287.TP
2288.B sha3\-256
2289Use optimized sha3\-256 as the checksum function.
2290.TP
2291.B sha3\-384
2292Use optimized sha3\-384 as the checksum function.
2293.TP
2294.B sha3\-512
2295Use optimized sha3\-512 as the checksum function.
d60e92d1
AC
2296.TP
2297.B meta
523bad63
TK
2298This option is deprecated, since now meta information is included in
2299generic verification header and meta verification happens by
2300default. For detailed information see the description of the
2301\fBverify\fR setting. This option is kept because of
2302compatibility's sake with old configurations. Do not use it.
d60e92d1 2303.TP
59245381 2304.B pattern
523bad63
TK
2305Verify a strict pattern. Normally fio includes a header with some
2306basic information and checksumming, but if this option is set, only
2307the specific pattern set with \fBverify_pattern\fR is verified.
59245381 2308.TP
d60e92d1 2309.B null
523bad63
TK
2310Only pretend to verify. Useful for testing internals with
2311`ioengine=null', not for much else.
d60e92d1 2312.RE
523bad63
TK
2313.P
2314This option can be used for repeated burn\-in tests of a system to make sure
2315that the written data is also correctly read back. If the data direction
2316given is a read or random read, fio will assume that it should verify a
2317previously written file. If the data direction includes any form of write,
2318the verify will be of the newly written data.
d60e92d1
AC
2319.RE
2320.TP
5c9323fb 2321.BI verifysort \fR=\fPbool
523bad63
TK
2322If true, fio will sort written verify blocks when it deems it faster to read
2323them back in a sorted manner. This is often the case when overwriting an
2324existing file, since the blocks are already laid out in the file system. You
2325can ignore this option unless doing huge amounts of really fast I/O where
2326the red\-black tree sorting CPU time becomes significant. Default: true.
d60e92d1 2327.TP
fa769d44 2328.BI verifysort_nr \fR=\fPint
523bad63 2329Pre\-load and sort verify blocks for a read workload.
fa769d44 2330.TP
f7fa2653 2331.BI verify_offset \fR=\fPint
d60e92d1 2332Swap the verification header with data somewhere else in the block before
523bad63 2333writing. It is swapped back before verifying.
d60e92d1 2334.TP
f7fa2653 2335.BI verify_interval \fR=\fPint
523bad63
TK
2336Write the verification header at a finer granularity than the
2337\fBblocksize\fR. It will be written for chunks the size of
2338\fBverify_interval\fR. \fBblocksize\fR should divide this evenly.
d60e92d1 2339.TP
996093bb 2340.BI verify_pattern \fR=\fPstr
523bad63
TK
2341If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2342filling with totally random bytes, but sometimes it's interesting to fill
2343with a known pattern for I/O verification purposes. Depending on the width
2344of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time (it can
2345be either a decimal or a hex number). The \fBverify_pattern\fR if larger than
2346a 32\-bit quantity has to be a hex number that starts with either "0x" or
2347"0X". Use with \fBverify\fR. Also, \fBverify_pattern\fR supports %o
2348format, which means that for each block offset will be written and then
2349verified back, e.g.:
2fa5a241
RP
2350.RS
2351.RS
523bad63
TK
2352.P
2353verify_pattern=%o
2fa5a241 2354.RE
523bad63 2355.P
2fa5a241 2356Or use combination of everything:
2fa5a241 2357.RS
523bad63
TK
2358.P
2359verify_pattern=0xff%o"abcd"\-12
2fa5a241
RP
2360.RE
2361.RE
996093bb 2362.TP
d60e92d1 2363.BI verify_fatal \fR=\fPbool
523bad63
TK
2364Normally fio will keep checking the entire contents before quitting on a
2365block verification failure. If this option is set, fio will exit the job on
2366the first observed failure. Default: false.
d60e92d1 2367.TP
b463e936 2368.BI verify_dump \fR=\fPbool
523bad63
TK
2369If set, dump the contents of both the original data block and the data block
2370we read off disk to files. This allows later analysis to inspect just what
2371kind of data corruption occurred. Off by default.
b463e936 2372.TP
e8462bd8 2373.BI verify_async \fR=\fPint
523bad63
TK
2374Fio will normally verify I/O inline from the submitting thread. This option
2375takes an integer describing how many async offload threads to create for I/O
2376verification instead, causing fio to offload the duty of verifying I/O
2377contents to one or more separate threads. If using this offload option, even
2378sync I/O engines can benefit from using an \fBiodepth\fR setting higher
2379than 1, as it allows them to have I/O in flight while verifies are running.
2380Defaults to 0 async threads, i.e. verification is not asynchronous.
e8462bd8
JA
2381.TP
2382.BI verify_async_cpus \fR=\fPstr
523bad63
TK
2383Tell fio to set the given CPU affinity on the async I/O verification
2384threads. See \fBcpus_allowed\fR for the format used.
e8462bd8 2385.TP
6f87418f
JA
2386.BI verify_backlog \fR=\fPint
2387Fio will normally verify the written contents of a job that utilizes verify
2388once that job has completed. In other words, everything is written then
2389everything is read back and verified. You may want to verify continually
523bad63
TK
2390instead for a variety of reasons. Fio stores the meta data associated with
2391an I/O block in memory, so for large verify workloads, quite a bit of memory
2392would be used up holding this meta data. If this option is enabled, fio will
2393write only N blocks before verifying these blocks.
6f87418f
JA
2394.TP
2395.BI verify_backlog_batch \fR=\fPint
523bad63
TK
2396Control how many blocks fio will verify if \fBverify_backlog\fR is
2397set. If not set, will default to the value of \fBverify_backlog\fR
2398(meaning the entire queue is read back and verified). If
2399\fBverify_backlog_batch\fR is less than \fBverify_backlog\fR then not all
2400blocks will be verified, if \fBverify_backlog_batch\fR is larger than
2401\fBverify_backlog\fR, some blocks will be verified more than once.
2402.TP
2403.BI verify_state_save \fR=\fPbool
2404When a job exits during the write phase of a verify workload, save its
2405current state. This allows fio to replay up until that point, if the verify
2406state is loaded for the verify read phase. The format of the filename is,
2407roughly:
2408.RS
2409.RS
2410.P
2411<type>\-<jobname>\-<jobindex>\-verify.state.
2412.RE
2413.P
2414<type> is "local" for a local run, "sock" for a client/server socket
2415connection, and "ip" (192.168.0.1, for instance) for a networked
2416client/server connection. Defaults to true.
2417.RE
2418.TP
2419.BI verify_state_load \fR=\fPbool
2420If a verify termination trigger was used, fio stores the current write state
2421of each thread. This can be used at verification time so that fio knows how
2422far it should verify. Without this information, fio will run a full
2423verification pass, according to the settings in the job file used. Default
2424false.
6f87418f 2425.TP
fa769d44
SW
2426.BI trim_percentage \fR=\fPint
2427Number of verify blocks to discard/trim.
2428.TP
2429.BI trim_verify_zero \fR=\fPbool
523bad63 2430Verify that trim/discarded blocks are returned as zeros.
fa769d44
SW
2431.TP
2432.BI trim_backlog \fR=\fPint
523bad63 2433Verify that trim/discarded blocks are returned as zeros.
fa769d44
SW
2434.TP
2435.BI trim_backlog_batch \fR=\fPint
523bad63 2436Trim this number of I/O blocks.
fa769d44
SW
2437.TP
2438.BI experimental_verify \fR=\fPbool
2439Enable experimental verification.
523bad63 2440.SS "Steady state"
fa769d44 2441.TP
523bad63
TK
2442.BI steadystate \fR=\fPstr:float "\fR,\fP ss" \fR=\fPstr:float
2443Define the criterion and limit for assessing steady state performance. The
2444first parameter designates the criterion whereas the second parameter sets
2445the threshold. When the criterion falls below the threshold for the
2446specified duration, the job will stop. For example, `iops_slope:0.1%' will
2447direct fio to terminate the job when the least squares regression slope
2448falls below 0.1% of the mean IOPS. If \fBgroup_reporting\fR is enabled
2449this will apply to all jobs in the group. Below is the list of available
2450steady state assessment criteria. All assessments are carried out using only
2451data from the rolling collection window. Threshold limits can be expressed
2452as a fixed value or as a percentage of the mean in the collection window.
2453.RS
2454.RS
d60e92d1 2455.TP
523bad63
TK
2456.B iops
2457Collect IOPS data. Stop the job if all individual IOPS measurements
2458are within the specified limit of the mean IOPS (e.g., `iops:2'
2459means that all individual IOPS values must be within 2 of the mean,
2460whereas `iops:0.2%' means that all individual IOPS values must be
2461within 0.2% of the mean IOPS to terminate the job).
d60e92d1 2462.TP
523bad63
TK
2463.B iops_slope
2464Collect IOPS data and calculate the least squares regression
2465slope. Stop the job if the slope falls below the specified limit.
d60e92d1 2466.TP
523bad63
TK
2467.B bw
2468Collect bandwidth data. Stop the job if all individual bandwidth
2469measurements are within the specified limit of the mean bandwidth.
64bbb865 2470.TP
523bad63
TK
2471.B bw_slope
2472Collect bandwidth data and calculate the least squares regression
2473slope. Stop the job if the slope falls below the specified limit.
2474.RE
2475.RE
d1c46c04 2476.TP
523bad63
TK
2477.BI steadystate_duration \fR=\fPtime "\fR,\fP ss_dur" \fR=\fPtime
2478A rolling window of this duration will be used to judge whether steady state
2479has been reached. Data will be collected once per second. The default is 0
2480which disables steady state detection. When the unit is omitted, the
2481value is interpreted in seconds.
0c63576e 2482.TP
523bad63
TK
2483.BI steadystate_ramp_time \fR=\fPtime "\fR,\fP ss_ramp" \fR=\fPtime
2484Allow the job to run for the specified duration before beginning data
2485collection for checking the steady state job termination criterion. The
2486default is 0. When the unit is omitted, the value is interpreted in seconds.
2487.SS "Measurements and reporting"
0c63576e 2488.TP
3a5db920
JA
2489.BI per_job_logs \fR=\fPbool
2490If set, this generates bw/clat/iops log with per file private filenames. If
523bad63
TK
2491not set, jobs with identical names will share the log filename. Default:
2492true.
2493.TP
2494.BI group_reporting
2495It may sometimes be interesting to display statistics for groups of jobs as
2496a whole instead of for each individual job. This is especially true if
2497\fBnumjobs\fR is used; looking at individual thread/process output
2498quickly becomes unwieldy. To see the final report per\-group instead of
2499per\-job, use \fBgroup_reporting\fR. Jobs in a file will be part of the
2500same reporting group, unless if separated by a \fBstonewall\fR, or by
2501using \fBnew_group\fR.
2502.TP
2503.BI new_group
2504Start a new reporting group. See: \fBgroup_reporting\fR. If not given,
2505all jobs in a file will be part of the same reporting group, unless
2506separated by a \fBstonewall\fR.
2507.TP
2508.BI stats \fR=\fPbool
2509By default, fio collects and shows final output results for all jobs
2510that run. If this option is set to 0, then fio will ignore it in
2511the final stat output.
3a5db920 2512.TP
836bad52 2513.BI write_bw_log \fR=\fPstr
523bad63 2514If given, write a bandwidth log for this job. Can be used to store data of
074f0817 2515the bandwidth of the jobs in their lifetime.
523bad63 2516.RS
074f0817
SW
2517.P
2518If no str argument is given, the default filename of
2519`jobname_type.x.log' is used. Even when the argument is given, fio
2520will still append the type of log. So if one specifies:
523bad63
TK
2521.RS
2522.P
074f0817 2523write_bw_log=foo
523bad63
TK
2524.RE
2525.P
074f0817
SW
2526The actual log name will be `foo_bw.x.log' where `x' is the index
2527of the job (1..N, where N is the number of jobs). If
2528\fBper_job_logs\fR is false, then the filename will not include the
2529`.x` job index.
2530.P
2531The included \fBfio_generate_plots\fR script uses gnuplot to turn these
2532text files into nice graphs. See the \fBLOG FILE FORMATS\fR section for how data is
2533structured within the file.
523bad63 2534.RE
901bb994 2535.TP
074f0817
SW
2536.BI write_lat_log \fR=\fPstr
2537Same as \fBwrite_bw_log\fR, except this option creates I/O
2538submission (e.g., `name_slat.x.log'), completion (e.g.,
2539`name_clat.x.log'), and total (e.g., `name_lat.x.log') latency
2540files instead. See \fBwrite_bw_log\fR for details about the
2541filename format and the \fBLOG FILE FORMATS\fR section for how data is structured
2542within the files.
2543.TP
1e613c9c 2544.BI write_hist_log \fR=\fPstr
074f0817
SW
2545Same as \fBwrite_bw_log\fR but writes an I/O completion latency
2546histogram file (e.g., `name_hist.x.log') instead. Note that this
2547file will be empty unless \fBlog_hist_msec\fR has also been set.
2548See \fBwrite_bw_log\fR for details about the filename format and
2549the \fBLOG FILE FORMATS\fR section for how data is structured
2550within the file.
1e613c9c 2551.TP
c8eeb9df 2552.BI write_iops_log \fR=\fPstr
074f0817
SW
2553Same as \fBwrite_bw_log\fR, but writes an IOPS file (e.g.
2554`name_iops.x.log') instead. See \fBwrite_bw_log\fR for
2555details about the filename format and the \fBLOG FILE FORMATS\fR section for how data
2556is structured within the file.
c8eeb9df 2557.TP
b8bc8cba
JA
2558.BI log_avg_msec \fR=\fPint
2559By default, fio will log an entry in the iops, latency, or bw log for every
523bad63 2560I/O that completes. When writing to the disk log, that can quickly grow to a
b8bc8cba 2561very large size. Setting this option makes fio average the each log entry
e6989e10 2562over the specified period of time, reducing the resolution of the log. See
523bad63
TK
2563\fBlog_max_value\fR as well. Defaults to 0, logging all entries.
2564Also see \fBLOG FILE FORMATS\fR section.
b8bc8cba 2565.TP
1e613c9c 2566.BI log_hist_msec \fR=\fPint
523bad63
TK
2567Same as \fBlog_avg_msec\fR, but logs entries for completion latency
2568histograms. Computing latency percentiles from averages of intervals using
2569\fBlog_avg_msec\fR is inaccurate. Setting this option makes fio log
2570histogram entries over the specified period of time, reducing log sizes for
2571high IOPS devices while retaining percentile accuracy. See
074f0817
SW
2572\fBlog_hist_coarseness\fR and \fBwrite_hist_log\fR as well.
2573Defaults to 0, meaning histogram logging is disabled.
1e613c9c
KC
2574.TP
2575.BI log_hist_coarseness \fR=\fPint
523bad63
TK
2576Integer ranging from 0 to 6, defining the coarseness of the resolution of
2577the histogram logs enabled with \fBlog_hist_msec\fR. For each increment
2578in coarseness, fio outputs half as many bins. Defaults to 0, for which
2579histogram logs contain 1216 latency bins. See \fBLOG FILE FORMATS\fR section.
2580.TP
2581.BI log_max_value \fR=\fPbool
2582If \fBlog_avg_msec\fR is set, fio logs the average over that window. If
2583you instead want to log the maximum value, set this option to 1. Defaults to
25840, meaning that averaged values are logged.
1e613c9c 2585.TP
ae588852 2586.BI log_offset \fR=\fPbool
523bad63
TK
2587If this is set, the iolog options will include the byte offset for the I/O
2588entry as well as the other data values. Defaults to 0 meaning that
2589offsets are not present in logs. Also see \fBLOG FILE FORMATS\fR section.
ae588852 2590.TP
aee2ab67 2591.BI log_compression \fR=\fPint
523bad63
TK
2592If this is set, fio will compress the I/O logs as it goes, to keep the
2593memory footprint lower. When a log reaches the specified size, that chunk is
2594removed and compressed in the background. Given that I/O logs are fairly
2595highly compressible, this yields a nice memory savings for longer runs. The
2596downside is that the compression will consume some background CPU cycles, so
2597it may impact the run. This, however, is also true if the logging ends up
2598consuming most of the system memory. So pick your poison. The I/O logs are
2599saved normally at the end of a run, by decompressing the chunks and storing
2600them in the specified log file. This feature depends on the availability of
2601zlib.
aee2ab67 2602.TP
c08f9fe2 2603.BI log_compression_cpus \fR=\fPstr
523bad63
TK
2604Define the set of CPUs that are allowed to handle online log compression for
2605the I/O jobs. This can provide better isolation between performance
c08f9fe2
JA
2606sensitive jobs, and background compression work.
2607.TP
b26317c9 2608.BI log_store_compressed \fR=\fPbool
c08f9fe2 2609If set, fio will store the log files in a compressed format. They can be
523bad63
TK
2610decompressed with fio, using the \fB\-\-inflate\-log\fR command line
2611parameter. The files will be stored with a `.fz' suffix.
b26317c9 2612.TP
3aea75b1
KC
2613.BI log_unix_epoch \fR=\fPbool
2614If set, fio will log Unix timestamps to the log files produced by enabling
523bad63 2615write_type_log for each log type, instead of the default zero\-based
3aea75b1
KC
2616timestamps.
2617.TP
66347cfa 2618.BI block_error_percentiles \fR=\fPbool
523bad63
TK
2619If set, record errors in trim block\-sized units from writes and trims and
2620output a histogram of how many trims it took to get to errors, and what kind
2621of error was encountered.
d60e92d1 2622.TP
523bad63
TK
2623.BI bwavgtime \fR=\fPint
2624Average the calculated bandwidth over the given time. Value is specified in
2625milliseconds. If the job also does bandwidth logging through
2626\fBwrite_bw_log\fR, then the minimum of this option and
2627\fBlog_avg_msec\fR will be used. Default: 500ms.
d60e92d1 2628.TP
523bad63
TK
2629.BI iopsavgtime \fR=\fPint
2630Average the calculated IOPS over the given time. Value is specified in
2631milliseconds. If the job also does IOPS logging through
2632\fBwrite_iops_log\fR, then the minimum of this option and
2633\fBlog_avg_msec\fR will be used. Default: 500ms.
d60e92d1 2634.TP
d60e92d1 2635.BI disk_util \fR=\fPbool
523bad63
TK
2636Generate disk utilization statistics, if the platform supports it.
2637Default: true.
fa769d44 2638.TP
523bad63
TK
2639.BI disable_lat \fR=\fPbool
2640Disable measurements of total latency numbers. Useful only for cutting back
2641the number of calls to \fBgettimeofday\fR\|(2), as that does impact
2642performance at really high IOPS rates. Note that to really get rid of a
2643large amount of these calls, this option must be used with
2644\fBdisable_slat\fR and \fBdisable_bw_measurement\fR as well.
9e684a49 2645.TP
523bad63
TK
2646.BI disable_clat \fR=\fPbool
2647Disable measurements of completion latency numbers. See
2648\fBdisable_lat\fR.
9e684a49 2649.TP
523bad63
TK
2650.BI disable_slat \fR=\fPbool
2651Disable measurements of submission latency numbers. See
2652\fBdisable_lat\fR.
9e684a49 2653.TP
523bad63
TK
2654.BI disable_bw_measurement \fR=\fPbool "\fR,\fP disable_bw" \fR=\fPbool
2655Disable measurements of throughput/bandwidth numbers. See
2656\fBdisable_lat\fR.
9e684a49 2657.TP
83349190 2658.BI clat_percentiles \fR=\fPbool
b599759b
JA
2659Enable the reporting of percentiles of completion latencies. This option is
2660mutually exclusive with \fBlat_percentiles\fR.
2661.TP
2662.BI lat_percentiles \fR=\fPbool
b71968b1 2663Enable the reporting of percentiles of I/O latencies. This is similar to
b599759b
JA
2664\fBclat_percentiles\fR, except that this includes the submission latency.
2665This option is mutually exclusive with \fBclat_percentiles\fR.
83349190
YH
2666.TP
2667.BI percentile_list \fR=\fPfloat_list
66347cfa 2668Overwrite the default list of percentiles for completion latencies and the
523bad63
TK
2669block error histogram. Each number is a floating number in the range
2670(0,100], and the maximum length of the list is 20. Use ':' to separate the
2671numbers, and list the numbers in ascending order. For example,
2672`\-\-percentile_list=99.5:99.9' will cause fio to report the values of
2673completion latency below which 99.5% and 99.9% of the observed latencies
2674fell, respectively.
e883cb35
JF
2675.TP
2676.BI significant_figures \fR=\fPint
c32ba107
JA
2677If using \fB\-\-output\-format\fR of `normal', set the significant figures
2678to this value. Higher values will yield more precise IOPS and throughput
2679units, while lower values will round. Requires a minimum value of 1 and a
e883cb35 2680maximum value of 10. Defaults to 4.
523bad63 2681.SS "Error handling"
e4585935 2682.TP
523bad63
TK
2683.BI exitall_on_error
2684When one job finishes in error, terminate the rest. The default is to wait
2685for each job to finish.
e4585935 2686.TP
523bad63
TK
2687.BI continue_on_error \fR=\fPstr
2688Normally fio will exit the job on the first observed failure. If this option
2689is set, fio will continue the job when there is a 'non\-fatal error' (EIO or
2690EILSEQ) until the runtime is exceeded or the I/O size specified is
2691completed. If this option is used, there are two more stats that are
2692appended, the total error count and the first error. The error field given
2693in the stats is the first error that was hit during the run.
2694The allowed values are:
2695.RS
2696.RS
046395d7 2697.TP
523bad63
TK
2698.B none
2699Exit on any I/O or verify errors.
de890a1e 2700.TP
523bad63
TK
2701.B read
2702Continue on read errors, exit on all others.
2cafffbe 2703.TP
523bad63
TK
2704.B write
2705Continue on write errors, exit on all others.
a0679ce5 2706.TP
523bad63
TK
2707.B io
2708Continue on any I/O error, exit on all others.
de890a1e 2709.TP
523bad63
TK
2710.B verify
2711Continue on verify errors, exit on all others.
de890a1e 2712.TP
523bad63
TK
2713.B all
2714Continue on all errors.
b93b6a2e 2715.TP
523bad63
TK
2716.B 0
2717Backward\-compatible alias for 'none'.
d3a623de 2718.TP
523bad63
TK
2719.B 1
2720Backward\-compatible alias for 'all'.
2721.RE
2722.RE
1d360ffb 2723.TP
523bad63
TK
2724.BI ignore_error \fR=\fPstr
2725Sometimes you want to ignore some errors during test in that case you can
2726specify error list for each error type, instead of only being able to
2727ignore the default 'non\-fatal error' using \fBcontinue_on_error\fR.
2728`ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST' errors for
2729given error type is separated with ':'. Error may be symbol ('ENOSPC', 'ENOMEM')
2730or integer. Example:
de890a1e
SL
2731.RS
2732.RS
523bad63
TK
2733.P
2734ignore_error=EAGAIN,ENOSPC:122
2735.RE
2736.P
2737This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
2738WRITE. This option works by overriding \fBcontinue_on_error\fR with
2739the list of errors for each error type if any.
2740.RE
de890a1e 2741.TP
523bad63
TK
2742.BI error_dump \fR=\fPbool
2743If set dump every error even if it is non fatal, true by default. If
2744disabled only fatal error will be dumped.
2745.SS "Running predefined workloads"
2746Fio includes predefined profiles that mimic the I/O workloads generated by
2747other tools.
49ccb8c1 2748.TP
523bad63
TK
2749.BI profile \fR=\fPstr
2750The predefined workload to run. Current profiles are:
2751.RS
2752.RS
de890a1e 2753.TP
523bad63
TK
2754.B tiobench
2755Threaded I/O bench (tiotest/tiobench) like workload.
49ccb8c1 2756.TP
523bad63
TK
2757.B act
2758Aerospike Certification Tool (ACT) like workload.
2759.RE
de890a1e
SL
2760.RE
2761.P
523bad63
TK
2762To view a profile's additional options use \fB\-\-cmdhelp\fR after specifying
2763the profile. For example:
2764.RS
2765.TP
2766$ fio \-\-profile=act \-\-cmdhelp
de890a1e 2767.RE
523bad63 2768.SS "Act profile options"
de890a1e 2769.TP
523bad63
TK
2770.BI device\-names \fR=\fPstr
2771Devices to use.
d54fce84 2772.TP
523bad63
TK
2773.BI load \fR=\fPint
2774ACT load multiplier. Default: 1.
7aeb1e94 2775.TP
523bad63
TK
2776.BI test\-duration\fR=\fPtime
2777How long the entire test takes to run. When the unit is omitted, the value
2778is given in seconds. Default: 24h.
1008602c 2779.TP
523bad63
TK
2780.BI threads\-per\-queue\fR=\fPint
2781Number of read I/O threads per device. Default: 8.
e5f34d95 2782.TP
523bad63
TK
2783.BI read\-req\-num\-512\-blocks\fR=\fPint
2784Number of 512B blocks to read at the time. Default: 3.
d54fce84 2785.TP
523bad63
TK
2786.BI large\-block\-op\-kbytes\fR=\fPint
2787Size of large block ops in KiB (writes). Default: 131072.
d54fce84 2788.TP
523bad63
TK
2789.BI prep
2790Set to run ACT prep phase.
2791.SS "Tiobench profile options"
6d500c2e 2792.TP
523bad63
TK
2793.BI size\fR=\fPstr
2794Size in MiB.
0d978694 2795.TP
523bad63
TK
2796.BI block\fR=\fPint
2797Block size in bytes. Default: 4096.
0d978694 2798.TP
523bad63
TK
2799.BI numruns\fR=\fPint
2800Number of runs.
0d978694 2801.TP
523bad63
TK
2802.BI dir\fR=\fPstr
2803Test directory.
65fa28ca 2804.TP
523bad63
TK
2805.BI threads\fR=\fPint
2806Number of threads.
d60e92d1 2807.SH OUTPUT
40943b9a
TK
2808Fio spits out a lot of output. While running, fio will display the status of the
2809jobs created. An example of that would be:
d60e92d1 2810.P
40943b9a
TK
2811.nf
2812 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]
2813.fi
d1429b5c 2814.P
40943b9a
TK
2815The characters inside the first set of square brackets denote the current status of
2816each thread. The first character is the first job defined in the job file, and so
2817forth. The possible values (in typical life cycle order) are:
d60e92d1
AC
2818.RS
2819.TP
40943b9a 2820.PD 0
d60e92d1 2821.B P
40943b9a 2822Thread setup, but not started.
d60e92d1
AC
2823.TP
2824.B C
2825Thread created.
2826.TP
2827.B I
40943b9a
TK
2828Thread initialized, waiting or generating necessary data.
2829.TP
522c29f6 2830.B p
40943b9a
TK
2831Thread running pre\-reading file(s).
2832.TP
2833.B /
2834Thread is in ramp period.
d60e92d1
AC
2835.TP
2836.B R
2837Running, doing sequential reads.
2838.TP
2839.B r
2840Running, doing random reads.
2841.TP
2842.B W
2843Running, doing sequential writes.
2844.TP
2845.B w
2846Running, doing random writes.
2847.TP
2848.B M
2849Running, doing mixed sequential reads/writes.
2850.TP
2851.B m
2852Running, doing mixed random reads/writes.
2853.TP
40943b9a
TK
2854.B D
2855Running, doing sequential trims.
2856.TP
2857.B d
2858Running, doing random trims.
2859.TP
d60e92d1
AC
2860.B F
2861Running, currently waiting for \fBfsync\fR\|(2).
2862.TP
2863.B V
40943b9a
TK
2864Running, doing verification of written data.
2865.TP
2866.B f
2867Thread finishing.
d60e92d1
AC
2868.TP
2869.B E
40943b9a 2870Thread exited, not reaped by main thread yet.
d60e92d1
AC
2871.TP
2872.B \-
40943b9a
TK
2873Thread reaped.
2874.TP
2875.B X
2876Thread reaped, exited with an error.
2877.TP
2878.B K
2879Thread reaped, exited due to signal.
d1429b5c 2880.PD
40943b9a
TK
2881.RE
2882.P
2883Fio will condense the thread string as not to take up more space on the command
2884line than needed. For instance, if you have 10 readers and 10 writers running,
2885the output would look like this:
2886.P
2887.nf
2888 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]
2889.fi
d60e92d1 2890.P
40943b9a
TK
2891Note that the status string is displayed in order, so it's possible to tell which of
2892the jobs are currently doing what. In the example above this means that jobs 1\-\-10
2893are readers and 11\-\-20 are writers.
d60e92d1 2894.P
40943b9a
TK
2895The other values are fairly self explanatory \-\- number of threads currently
2896running and doing I/O, the number of currently open files (f=), the estimated
2897completion percentage, the rate of I/O since last check (read speed listed first,
2898then write speed and optionally trim speed) in terms of bandwidth and IOPS,
2899and time to completion for the current running group. It's impossible to estimate
2900runtime of the following groups (if any).
d60e92d1 2901.P
40943b9a
TK
2902When fio is done (or interrupted by Ctrl\-C), it will show the data for
2903each thread, group of threads, and disks in that order. For each overall thread (or
2904group) the output looks like:
2905.P
2906.nf
2907 Client1: (groupid=0, jobs=1): err= 0: pid=16109: Sat Jun 24 12:07:54 2017
2908 write: IOPS=88, BW=623KiB/s (638kB/s)(30.4MiB/50032msec)
2909 slat (nsec): min=500, max=145500, avg=8318.00, stdev=4781.50
2910 clat (usec): min=170, max=78367, avg=4019.02, stdev=8293.31
2911 lat (usec): min=174, max=78375, avg=4027.34, stdev=8291.79
2912 clat percentiles (usec):
2913 | 1.00th=[ 302], 5.00th=[ 326], 10.00th=[ 343], 20.00th=[ 363],
2914 | 30.00th=[ 392], 40.00th=[ 404], 50.00th=[ 416], 60.00th=[ 445],
2915 | 70.00th=[ 816], 80.00th=[ 6718], 90.00th=[12911], 95.00th=[21627],
2916 | 99.00th=[43779], 99.50th=[51643], 99.90th=[68682], 99.95th=[72877],
2917 | 99.99th=[78119]
2918 bw ( KiB/s): min= 532, max= 686, per=0.10%, avg=622.87, stdev=24.82, samples= 100
2919 iops : min= 76, max= 98, avg=88.98, stdev= 3.54, samples= 100
d3b9694d
VF
2920 lat (usec) : 250=0.04%, 500=64.11%, 750=4.81%, 1000=2.79%
2921 lat (msec) : 2=4.16%, 4=1.84%, 10=4.90%, 20=11.33%, 50=5.37%
2922 lat (msec) : 100=0.65%
40943b9a
TK
2923 cpu : usr=0.27%, sys=0.18%, ctx=12072, majf=0, minf=21
2924 IO depths : 1=85.0%, 2=13.1%, 4=1.8%, 8=0.1%, 16=0.0%, 32=0.0%, >=64=0.0%
2925 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2926 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2927 issued rwt: total=0,4450,0, short=0,0,0, dropped=0,0,0
2928 latency : target=0, window=0, percentile=100.00%, depth=8
2929.fi
2930.P
2931The job name (or first job's name when using \fBgroup_reporting\fR) is printed,
2932along with the group id, count of jobs being aggregated, last error id seen (which
2933is 0 when there are no errors), pid/tid of that thread and the time the job/group
2934completed. Below are the I/O statistics for each data direction performed (showing
2935writes in the example above). In the order listed, they denote:
d60e92d1 2936.RS
d60e92d1 2937.TP
40943b9a
TK
2938.B read/write/trim
2939The string before the colon shows the I/O direction the statistics
2940are for. \fIIOPS\fR is the average I/Os performed per second. \fIBW\fR
2941is the average bandwidth rate shown as: value in power of 2 format
2942(value in power of 10 format). The last two values show: (total
2943I/O performed in power of 2 format / \fIruntime\fR of that thread).
d60e92d1
AC
2944.TP
2945.B slat
40943b9a
TK
2946Submission latency (\fImin\fR being the minimum, \fImax\fR being the
2947maximum, \fIavg\fR being the average, \fIstdev\fR being the standard
2948deviation). This is the time it took to submit the I/O. For
2949sync I/O this row is not displayed as the slat is really the
2950completion latency (since queue/complete is one operation there).
2951This value can be in nanoseconds, microseconds or milliseconds \-\-\-
2952fio will choose the most appropriate base and print that (in the
2953example above nanoseconds was the best scale). Note: in \fB\-\-minimal\fR mode
2954latencies are always expressed in microseconds.
d60e92d1
AC
2955.TP
2956.B clat
40943b9a
TK
2957Completion latency. Same names as slat, this denotes the time from
2958submission to completion of the I/O pieces. For sync I/O, clat will
2959usually be equal (or very close) to 0, as the time from submit to
2960complete is basically just CPU time (I/O has already been done, see slat
2961explanation).
d60e92d1 2962.TP
d3b9694d
VF
2963.B lat
2964Total latency. Same names as slat and clat, this denotes the time from
2965when fio created the I/O unit to completion of the I/O operation.
2966.TP
d60e92d1 2967.B bw
40943b9a
TK
2968Bandwidth statistics based on samples. Same names as the xlat stats,
2969but also includes the number of samples taken (\fIsamples\fR) and an
2970approximate percentage of total aggregate bandwidth this thread
2971received in its group (\fIper\fR). This last value is only really
2972useful if the threads in this group are on the same disk, since they
2973are then competing for disk access.
2974.TP
2975.B iops
2976IOPS statistics based on samples. Same names as \fBbw\fR.
d60e92d1 2977.TP
d3b9694d
VF
2978.B lat (nsec/usec/msec)
2979The distribution of I/O completion latencies. This is the time from when
2980I/O leaves fio and when it gets completed. Unlike the separate
2981read/write/trim sections above, the data here and in the remaining
2982sections apply to all I/Os for the reporting group. 250=0.04% means that
29830.04% of the I/Os completed in under 250us. 500=64.11% means that 64.11%
2984of the I/Os required 250 to 499us for completion.
2985.TP
d60e92d1 2986.B cpu
40943b9a
TK
2987CPU usage. User and system time, along with the number of context
2988switches this thread went through, usage of system and user time, and
2989finally the number of major and minor page faults. The CPU utilization
2990numbers are averages for the jobs in that reporting group, while the
2991context and fault counters are summed.
d60e92d1
AC
2992.TP
2993.B IO depths
40943b9a
TK
2994The distribution of I/O depths over the job lifetime. The numbers are
2995divided into powers of 2 and each entry covers depths from that value
2996up to those that are lower than the next entry \-\- e.g., 16= covers
2997depths from 16 to 31. Note that the range covered by a depth
2998distribution entry can be different to the range covered by the
2999equivalent \fBsubmit\fR/\fBcomplete\fR distribution entry.
3000.TP
3001.B IO submit
3002How many pieces of I/O were submitting in a single submit call. Each
3003entry denotes that amount and below, until the previous entry \-\- e.g.,
300416=100% means that we submitted anywhere between 9 to 16 I/Os per submit
3005call. Note that the range covered by a \fBsubmit\fR distribution entry can
3006be different to the range covered by the equivalent depth distribution
3007entry.
3008.TP
3009.B IO complete
3010Like the above \fBsubmit\fR number, but for completions instead.
3011.TP
3012.B IO issued rwt
3013The number of \fBread/write/trim\fR requests issued, and how many of them were
3014short or dropped.
d60e92d1 3015.TP
d3b9694d 3016.B IO latency
ee21ebee 3017These values are for \fBlatency_target\fR and related options. When
d3b9694d
VF
3018these options are engaged, this section describes the I/O depth required
3019to meet the specified latency target.
d60e92d1 3020.RE
d60e92d1 3021.P
40943b9a
TK
3022After each client has been listed, the group statistics are printed. They
3023will look like this:
3024.P
3025.nf
3026 Run status group 0 (all jobs):
3027 READ: bw=20.9MiB/s (21.9MB/s), 10.4MiB/s\-10.8MiB/s (10.9MB/s\-11.3MB/s), io=64.0MiB (67.1MB), run=2973\-3069msec
3028 WRITE: bw=1231KiB/s (1261kB/s), 616KiB/s\-621KiB/s (630kB/s\-636kB/s), io=64.0MiB (67.1MB), run=52747\-53223msec
3029.fi
3030.P
3031For each data direction it prints:
d60e92d1
AC
3032.RS
3033.TP
40943b9a
TK
3034.B bw
3035Aggregate bandwidth of threads in this group followed by the
3036minimum and maximum bandwidth of all the threads in this group.
3037Values outside of brackets are power\-of\-2 format and those
3038within are the equivalent value in a power\-of\-10 format.
d60e92d1 3039.TP
40943b9a
TK
3040.B io
3041Aggregate I/O performed of all threads in this group. The
3042format is the same as \fBbw\fR.
d60e92d1 3043.TP
40943b9a
TK
3044.B run
3045The smallest and longest runtimes of the threads in this group.
d60e92d1 3046.RE
d60e92d1 3047.P
40943b9a
TK
3048And finally, the disk statistics are printed. This is Linux specific.
3049They will look like this:
3050.P
3051.nf
3052 Disk stats (read/write):
3053 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3054.fi
3055.P
3056Each value is printed for both reads and writes, with reads first. The
3057numbers denote:
d60e92d1
AC
3058.RS
3059.TP
3060.B ios
3061Number of I/Os performed by all groups.
3062.TP
3063.B merge
007c7be9 3064Number of merges performed by the I/O scheduler.
d60e92d1
AC
3065.TP
3066.B ticks
3067Number of ticks we kept the disk busy.
3068.TP
40943b9a 3069.B in_queue
d60e92d1
AC
3070Total time spent in the disk queue.
3071.TP
3072.B util
40943b9a
TK
3073The disk utilization. A value of 100% means we kept the disk
3074busy constantly, 50% would be a disk idling half of the time.
d60e92d1 3075.RE
8423bd11 3076.P
40943b9a
TK
3077It is also possible to get fio to dump the current output while it is running,
3078without terminating the job. To do that, send fio the USR1 signal. You can
3079also get regularly timed dumps by using the \fB\-\-status\-interval\fR
3080parameter, or by creating a file in `/tmp' named
3081`fio\-dump\-status'. If fio sees this file, it will unlink it and dump the
3082current output status.
d60e92d1 3083.SH TERSE OUTPUT
40943b9a
TK
3084For scripted usage where you typically want to generate tables or graphs of the
3085results, fio can output the results in a semicolon separated format. The format
3086is one long line of values, such as:
d60e92d1 3087.P
40943b9a
TK
3088.nf
3089 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%
3090 A description of this job goes here.
3091.fi
d60e92d1 3092.P
40943b9a 3093The job description (if provided) follows on a second line.
d60e92d1 3094.P
40943b9a
TK
3095To enable terse output, use the \fB\-\-minimal\fR or
3096`\-\-output\-format=terse' command line options. The
3097first value is the version of the terse output format. If the output has to be
3098changed for some reason, this number will be incremented by 1 to signify that
3099change.
d60e92d1 3100.P
40943b9a
TK
3101Split up, the format is as follows (comments in brackets denote when a
3102field was introduced or whether it's specific to some terse version):
d60e92d1 3103.P
40943b9a
TK
3104.nf
3105 terse version, fio version [v3], jobname, groupid, error
3106.fi
525c2bfa 3107.RS
40943b9a
TK
3108.P
3109.B
3110READ status:
525c2bfa 3111.RE
40943b9a
TK
3112.P
3113.nf
3114 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3115 Submission latency: min, max, mean, stdev (usec)
3116 Completion latency: min, max, mean, stdev (usec)
3117 Completion latency percentiles: 20 fields (see below)
3118 Total latency: min, max, mean, stdev (usec)
3119 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3120 IOPS [v5]: min, max, mean, stdev, number of samples
3121.fi
d60e92d1 3122.RS
40943b9a
TK
3123.P
3124.B
3125WRITE status:
a2c95580 3126.RE
40943b9a
TK
3127.P
3128.nf
3129 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3130 Submission latency: min, max, mean, stdev (usec)
3131 Completion latency: min, max, mean, stdev (usec)
3132 Completion latency percentiles: 20 fields (see below)
3133 Total latency: min, max, mean, stdev (usec)
3134 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3135 IOPS [v5]: min, max, mean, stdev, number of samples
3136.fi
a2c95580 3137.RS
40943b9a
TK
3138.P
3139.B
3140TRIM status [all but version 3]:
d60e92d1
AC
3141.RE
3142.P
40943b9a
TK
3143.nf
3144 Fields are similar to \fBREAD/WRITE\fR status.
3145.fi
a2c95580 3146.RS
a2c95580 3147.P
40943b9a 3148.B
d1429b5c 3149CPU usage:
d60e92d1
AC
3150.RE
3151.P
40943b9a
TK
3152.nf
3153 user, system, context switches, major faults, minor faults
3154.fi
d60e92d1 3155.RS
40943b9a
TK
3156.P
3157.B
3158I/O depths:
d60e92d1
AC
3159.RE
3160.P
40943b9a
TK
3161.nf
3162 <=1, 2, 4, 8, 16, 32, >=64
3163.fi
562c2d2f 3164.RS
40943b9a
TK
3165.P
3166.B
3167I/O latencies microseconds:
562c2d2f 3168.RE
40943b9a
TK
3169.P
3170.nf
3171 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3172.fi
562c2d2f 3173.RS
40943b9a
TK
3174.P
3175.B
3176I/O latencies milliseconds:
562c2d2f
DN
3177.RE
3178.P
40943b9a
TK
3179.nf
3180 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3181.fi
f2f788dd 3182.RS
40943b9a
TK
3183.P
3184.B
3185Disk utilization [v3]:
f2f788dd
JA
3186.RE
3187.P
40943b9a
TK
3188.nf
3189 disk name, read ios, write ios, read merges, write merges, read ticks, write ticks, time spent in queue, disk utilization percentage
3190.fi
562c2d2f 3191.RS
d60e92d1 3192.P
40943b9a
TK
3193.B
3194Additional Info (dependent on continue_on_error, default off):
d60e92d1 3195.RE
2fc26c3d 3196.P
40943b9a
TK
3197.nf
3198 total # errors, first error code
3199.fi
2fc26c3d
IC
3200.RS
3201.P
40943b9a
TK
3202.B
3203Additional Info (dependent on description being set):
3204.RE
3205.P
2fc26c3d 3206.nf
40943b9a
TK
3207 Text description
3208.fi
3209.P
3210Completion latency percentiles can be a grouping of up to 20 sets, so for the
3211terse output fio writes all of them. Each field will look like this:
3212.P
3213.nf
3214 1.00%=6112
3215.fi
3216.P
3217which is the Xth percentile, and the `usec' latency associated with it.
3218.P
3219For \fBDisk utilization\fR, all disks used by fio are shown. So for each disk there
3220will be a disk utilization section.
3221.P
3222Below is a single line containing short names for each of the fields in the
3223minimal output v3, separated by semicolons:
3224.P
3225.nf
3226 terse_version_3;fio_version;jobname;groupid;error;read_kb;read_bandwidth;read_iops;read_runtime_ms;read_slat_min;read_slat_max;read_slat_mean;read_slat_dev;read_clat_min;read_clat_max;read_clat_mean;read_clat_dev;read_clat_pct01;read_clat_pct02;read_clat_pct03;read_clat_pct04;read_clat_pct05;read_clat_pct06;read_clat_pct07;read_clat_pct08;read_clat_pct09;read_clat_pct10;read_clat_pct11;read_clat_pct12;read_clat_pct13;read_clat_pct14;read_clat_pct15;read_clat_pct16;read_clat_pct17;read_clat_pct18;read_clat_pct19;read_clat_pct20;read_tlat_min;read_lat_max;read_lat_mean;read_lat_dev;read_bw_min;read_bw_max;read_bw_agg_pct;read_bw_mean;read_bw_dev;write_kb;write_bandwidth;write_iops;write_runtime_ms;write_slat_min;write_slat_max;write_slat_mean;write_slat_dev;write_clat_min;write_clat_max;write_clat_mean;write_clat_dev;write_clat_pct01;write_clat_pct02;write_clat_pct03;write_clat_pct04;write_clat_pct05;write_clat_pct06;write_clat_pct07;write_clat_pct08;write_clat_pct09;write_clat_pct10;write_clat_pct11;write_clat_pct12;write_clat_pct13;write_clat_pct14;write_clat_pct15;write_clat_pct16;write_clat_pct17;write_clat_pct18;write_clat_pct19;write_clat_pct20;write_tlat_min;write_lat_max;write_lat_mean;write_lat_dev;write_bw_min;write_bw_max;write_bw_agg_pct;write_bw_mean;write_bw_dev;cpu_user;cpu_sys;cpu_csw;cpu_mjf;cpu_minf;iodepth_1;iodepth_2;iodepth_4;iodepth_8;iodepth_16;iodepth_32;iodepth_64;lat_2us;lat_4us;lat_10us;lat_20us;lat_50us;lat_100us;lat_250us;lat_500us;lat_750us;lat_1000us;lat_2ms;lat_4ms;lat_10ms;lat_20ms;lat_50ms;lat_100ms;lat_250ms;lat_500ms;lat_750ms;lat_1000ms;lat_2000ms;lat_over_2000ms;disk_name;disk_read_iops;disk_write_iops;disk_read_merges;disk_write_merges;disk_read_ticks;write_ticks;disk_queue_time;disk_util
2fc26c3d 3227.fi
44c82dba
VF
3228.SH JSON OUTPUT
3229The \fBjson\fR output format is intended to be both human readable and convenient
3230for automated parsing. For the most part its sections mirror those of the
3231\fBnormal\fR output. The \fBruntime\fR value is reported in msec and the \fBbw\fR value is
3232reported in 1024 bytes per second units.
3233.fi
d9e557ab
VF
3234.SH JSON+ OUTPUT
3235The \fBjson+\fR output format is identical to the \fBjson\fR output format except that it
3236adds a full dump of the completion latency bins. Each \fBbins\fR object contains a
3237set of (key, value) pairs where keys are latency durations and values count how
3238many I/Os had completion latencies of the corresponding duration. For example,
3239consider:
d9e557ab 3240.RS
40943b9a 3241.P
d9e557ab
VF
3242"bins" : { "87552" : 1, "89600" : 1, "94720" : 1, "96768" : 1, "97792" : 1, "99840" : 1, "100864" : 2, "103936" : 6, "104960" : 534, "105984" : 5995, "107008" : 7529, ... }
3243.RE
40943b9a 3244.P
d9e557ab
VF
3245This data indicates that one I/O required 87,552ns to complete, two I/Os required
3246100,864ns to complete, and 7529 I/Os required 107,008ns to complete.
40943b9a 3247.P
d9e557ab 3248Also included with fio is a Python script \fBfio_jsonplus_clat2csv\fR that takes
40943b9a
TK
3249json+ output and generates CSV\-formatted latency data suitable for plotting.
3250.P
d9e557ab 3251The latency durations actually represent the midpoints of latency intervals.
40943b9a 3252For details refer to `stat.h' in the fio source.
29dbd1e5 3253.SH TRACE FILE FORMAT
40943b9a
TK
3254There are two trace file format that you can encounter. The older (v1) format is
3255unsupported since version 1.20\-rc3 (March 2008). It will still be described
29dbd1e5 3256below in case that you get an old trace and want to understand it.
29dbd1e5 3257.P
40943b9a
TK
3258In any case the trace is a simple text file with a single action per line.
3259.TP
29dbd1e5 3260.B Trace file format v1
40943b9a 3261Each line represents a single I/O action in the following format:
29dbd1e5 3262.RS
40943b9a
TK
3263.RS
3264.P
29dbd1e5 3265rw, offset, length
29dbd1e5
JA
3266.RE
3267.P
40943b9a
TK
3268where `rw=0/1' for read/write, and the `offset' and `length' entries being in bytes.
3269.P
3270This format is not supported in fio versions >= 1.20\-rc3.
3271.RE
3272.TP
29dbd1e5 3273.B Trace file format v2
40943b9a
TK
3274The second version of the trace file format was added in fio version 1.17. It
3275allows to access more then one file per trace and has a bigger set of possible
3276file actions.
29dbd1e5 3277.RS
40943b9a 3278.P
29dbd1e5 3279The first line of the trace file has to be:
40943b9a
TK
3280.RS
3281.P
3282"fio version 2 iolog"
3283.RE
3284.P
29dbd1e5 3285Following this can be lines in two different formats, which are described below.
40943b9a
TK
3286.P
3287.B
29dbd1e5 3288The file management format:
40943b9a
TK
3289.RS
3290filename action
29dbd1e5 3291.P
40943b9a 3292The `filename' is given as an absolute path. The `action' can be one of these:
29dbd1e5
JA
3293.RS
3294.TP
3295.B add
40943b9a 3296Add the given `filename' to the trace.
29dbd1e5
JA
3297.TP
3298.B open
40943b9a
TK
3299Open the file with the given `filename'. The `filename' has to have
3300been added with the \fBadd\fR action before.
29dbd1e5
JA
3301.TP
3302.B close
40943b9a
TK
3303Close the file with the given `filename'. The file has to have been
3304\fBopen\fRed before.
3305.RE
29dbd1e5 3306.RE
29dbd1e5 3307.P
40943b9a
TK
3308.B
3309The file I/O action format:
3310.RS
3311filename action offset length
29dbd1e5 3312.P
40943b9a
TK
3313The `filename' is given as an absolute path, and has to have been \fBadd\fRed and
3314\fBopen\fRed before it can be used with this format. The `offset' and `length' are
3315given in bytes. The `action' can be one of these:
29dbd1e5
JA
3316.RS
3317.TP
3318.B wait
40943b9a
TK
3319Wait for `offset' microseconds. Everything below 100 is discarded.
3320The time is relative to the previous `wait' statement.
29dbd1e5
JA
3321.TP
3322.B read
40943b9a 3323Read `length' bytes beginning from `offset'.
29dbd1e5
JA
3324.TP
3325.B write
40943b9a 3326Write `length' bytes beginning from `offset'.
29dbd1e5
JA
3327.TP
3328.B sync
40943b9a 3329\fBfsync\fR\|(2) the file.
29dbd1e5
JA
3330.TP
3331.B datasync
40943b9a 3332\fBfdatasync\fR\|(2) the file.
29dbd1e5
JA
3333.TP
3334.B trim
40943b9a
TK
3335Trim the given file from the given `offset' for `length' bytes.
3336.RE
29dbd1e5 3337.RE
29dbd1e5 3338.SH CPU IDLENESS PROFILING
40943b9a
TK
3339In some cases, we want to understand CPU overhead in a test. For example, we
3340test patches for the specific goodness of whether they reduce CPU usage.
3341Fio implements a balloon approach to create a thread per CPU that runs at idle
3342priority, meaning that it only runs when nobody else needs the cpu.
3343By measuring the amount of work completed by the thread, idleness of each CPU
3344can be derived accordingly.
3345.P
3346An unit work is defined as touching a full page of unsigned characters. Mean and
3347standard deviation of time to complete an unit work is reported in "unit work"
3348section. Options can be chosen to report detailed percpu idleness or overall
3349system idleness by aggregating percpu stats.
29dbd1e5 3350.SH VERIFICATION AND TRIGGERS
40943b9a
TK
3351Fio is usually run in one of two ways, when data verification is done. The first
3352is a normal write job of some sort with verify enabled. When the write phase has
3353completed, fio switches to reads and verifies everything it wrote. The second
3354model is running just the write phase, and then later on running the same job
3355(but with reads instead of writes) to repeat the same I/O patterns and verify
3356the contents. Both of these methods depend on the write phase being completed,
3357as fio otherwise has no idea how much data was written.
3358.P
3359With verification triggers, fio supports dumping the current write state to
3360local files. Then a subsequent read verify workload can load this state and know
3361exactly where to stop. This is useful for testing cases where power is cut to a
3362server in a managed fashion, for instance.
3363.P
29dbd1e5 3364A verification trigger consists of two things:
29dbd1e5 3365.RS
40943b9a
TK
3366.P
33671) Storing the write state of each job.
3368.P
33692) Executing a trigger command.
29dbd1e5 3370.RE
40943b9a
TK
3371.P
3372The write state is relatively small, on the order of hundreds of bytes to single
3373kilobytes. It contains information on the number of completions done, the last X
3374completions, etc.
3375.P
3376A trigger is invoked either through creation ('touch') of a specified file in
3377the system, or through a timeout setting. If fio is run with
3378`\-\-trigger\-file=/tmp/trigger\-file', then it will continually
3379check for the existence of `/tmp/trigger\-file'. When it sees this file, it
3380will fire off the trigger (thus saving state, and executing the trigger
29dbd1e5 3381command).
40943b9a
TK
3382.P
3383For client/server runs, there's both a local and remote trigger. If fio is
3384running as a server backend, it will send the job states back to the client for
3385safe storage, then execute the remote trigger, if specified. If a local trigger
3386is specified, the server will still send back the write state, but the client
3387will then execute the trigger.
29dbd1e5
JA
3388.RE
3389.P
3390.B Verification trigger example
3391.RS
40943b9a
TK
3392Let's say we want to run a powercut test on the remote Linux machine 'server'.
3393Our write workload is in `write\-test.fio'. We want to cut power to 'server' at
3394some point during the run, and we'll run this test from the safety or our local
3395machine, 'localbox'. On the server, we'll start the fio backend normally:
3396.RS
3397.P
3398server# fio \-\-server
3399.RE
3400.P
29dbd1e5 3401and on the client, we'll fire off the workload:
40943b9a
TK
3402.RS
3403.P
3404localbox$ fio \-\-client=server \-\-trigger\-file=/tmp/my\-trigger \-\-trigger\-remote="bash \-c "echo b > /proc/sysrq\-triger""
3405.RE
3406.P
3407We set `/tmp/my\-trigger' as the trigger file, and we tell fio to execute:
3408.RS
3409.P
3410echo b > /proc/sysrq\-trigger
3411.RE
3412.P
3413on the server once it has received the trigger and sent us the write state. This
3414will work, but it's not really cutting power to the server, it's merely
3415abruptly rebooting it. If we have a remote way of cutting power to the server
3416through IPMI or similar, we could do that through a local trigger command
3417instead. Let's assume we have a script that does IPMI reboot of a given hostname,
3418ipmi\-reboot. On localbox, we could then have run fio with a local trigger
3419instead:
3420.RS
3421.P
3422localbox$ fio \-\-client=server \-\-trigger\-file=/tmp/my\-trigger \-\-trigger="ipmi\-reboot server"
3423.RE
3424.P
3425For this case, fio would wait for the server to send us the write state, then
3426execute `ipmi\-reboot server' when that happened.
29dbd1e5
JA
3427.RE
3428.P
3429.B Loading verify state
3430.RS
40943b9a
TK
3431To load stored write state, a read verification job file must contain the
3432\fBverify_state_load\fR option. If that is set, fio will load the previously
29dbd1e5 3433stored state. For a local fio run this is done by loading the files directly,
40943b9a
TK
3434and on a client/server run, the server backend will ask the client to send the
3435files over and load them from there.
29dbd1e5 3436.RE
a3ae5b05 3437.SH LOG FILE FORMATS
a3ae5b05
JA
3438Fio supports a variety of log file formats, for logging latencies, bandwidth,
3439and IOPS. The logs share a common format, which looks like this:
40943b9a 3440.RS
a3ae5b05 3441.P
40943b9a
TK
3442time (msec), value, data direction, block size (bytes), offset (bytes)
3443.RE
3444.P
3445`Time' for the log entry is always in milliseconds. The `value' logged depends
3446on the type of log, it will be one of the following:
3447.RS
a3ae5b05
JA
3448.TP
3449.B Latency log
168bb587 3450Value is latency in nsecs
a3ae5b05
JA
3451.TP
3452.B Bandwidth log
6d500c2e 3453Value is in KiB/sec
a3ae5b05
JA
3454.TP
3455.B IOPS log
40943b9a
TK
3456Value is IOPS
3457.RE
a3ae5b05 3458.P
40943b9a
TK
3459`Data direction' is one of the following:
3460.RS
a3ae5b05
JA
3461.TP
3462.B 0
40943b9a 3463I/O is a READ
a3ae5b05
JA
3464.TP
3465.B 1
40943b9a 3466I/O is a WRITE
a3ae5b05
JA
3467.TP
3468.B 2
40943b9a 3469I/O is a TRIM
a3ae5b05 3470.RE
40943b9a
TK
3471.P
3472The entry's `block size' is always in bytes. The `offset' is the offset, in bytes,
3473from the start of the file, for that particular I/O. The logging of the offset can be
3474toggled with \fBlog_offset\fR.
3475.P
3476Fio defaults to logging every individual I/O. When IOPS are logged for individual
3477I/Os the `value' entry will always be 1. If windowed logging is enabled through
3478\fBlog_avg_msec\fR, fio logs the average values over the specified period of time.
3479If windowed logging is enabled and \fBlog_max_value\fR is set, then fio logs
3480maximum values in that window instead of averages. Since `data direction', `block size'
3481and `offset' are per\-I/O values, if windowed logging is enabled they
3482aren't applicable and will be 0.
49da1240 3483.SH CLIENT / SERVER
40943b9a
TK
3484Normally fio is invoked as a stand\-alone application on the machine where the
3485I/O workload should be generated. However, the backend and frontend of fio can
3486be run separately i.e., the fio server can generate an I/O workload on the "Device
3487Under Test" while being controlled by a client on another machine.
3488.P
3489Start the server on the machine which has access to the storage DUT:
3490.RS
3491.P
3492$ fio \-\-server=args
3493.RE
3494.P
3495where `args' defines what fio listens to. The arguments are of the form
3496`type,hostname' or `IP,port'. `type' is either `ip' (or ip4) for TCP/IP
3497v4, `ip6' for TCP/IP v6, or `sock' for a local unix domain socket.
3498`hostname' is either a hostname or IP address, and `port' is the port to listen
3499to (only valid for TCP/IP, not a local socket). Some examples:
3500.RS
3501.TP
e0ee7a8b 35021) \fBfio \-\-server\fR
40943b9a
TK
3503Start a fio server, listening on all interfaces on the default port (8765).
3504.TP
e0ee7a8b 35052) \fBfio \-\-server=ip:hostname,4444\fR
40943b9a
TK
3506Start a fio server, listening on IP belonging to hostname and on port 4444.
3507.TP
e0ee7a8b 35083) \fBfio \-\-server=ip6:::1,4444\fR
40943b9a
TK
3509Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
3510.TP
e0ee7a8b 35114) \fBfio \-\-server=,4444\fR
40943b9a
TK
3512Start a fio server, listening on all interfaces on port 4444.
3513.TP
e0ee7a8b 35145) \fBfio \-\-server=1.2.3.4\fR
40943b9a
TK
3515Start a fio server, listening on IP 1.2.3.4 on the default port.
3516.TP
e0ee7a8b 35176) \fBfio \-\-server=sock:/tmp/fio.sock\fR
40943b9a
TK
3518Start a fio server, listening on the local socket `/tmp/fio.sock'.
3519.RE
3520.P
3521Once a server is running, a "client" can connect to the fio server with:
3522.RS
3523.P
3524$ fio <local\-args> \-\-client=<server> <remote\-args> <job file(s)>
3525.RE
3526.P
3527where `local\-args' are arguments for the client where it is running, `server'
3528is the connect string, and `remote\-args' and `job file(s)' are sent to the
3529server. The `server' string follows the same format as it does on the server
3530side, to allow IP/hostname/socket and port strings.
3531.P
3532Fio can connect to multiple servers this way:
3533.RS
3534.P
3535$ fio \-\-client=<server1> <job file(s)> \-\-client=<server2> <job file(s)>
3536.RE
3537.P
3538If the job file is located on the fio server, then you can tell the server to
3539load a local file as well. This is done by using \fB\-\-remote\-config\fR:
3540.RS
3541.P
3542$ fio \-\-client=server \-\-remote\-config /path/to/file.fio
3543.RE
3544.P
3545Then fio will open this local (to the server) job file instead of being passed
3546one from the client.
3547.P
ff6bb260 3548If you have many servers (example: 100 VMs/containers), you can input a pathname
40943b9a
TK
3549of a file containing host IPs/names as the parameter value for the
3550\fB\-\-client\fR option. For example, here is an example `host.list'
3551file containing 2 hostnames:
3552.RS
3553.P
3554.PD 0
39b5f61e 3555host1.your.dns.domain
40943b9a 3556.P
39b5f61e 3557host2.your.dns.domain
40943b9a
TK
3558.PD
3559.RE
3560.P
39b5f61e 3561The fio command would then be:
40943b9a
TK
3562.RS
3563.P
3564$ fio \-\-client=host.list <job file(s)>
3565.RE
3566.P
3567In this mode, you cannot input server\-specific parameters or job files \-\- all
39b5f61e 3568servers receive the same job file.
40943b9a
TK
3569.P
3570In order to let `fio \-\-client' runs use a shared filesystem from multiple
3571hosts, `fio \-\-client' now prepends the IP address of the server to the
3572filename. For example, if fio is using the directory `/mnt/nfs/fio' and is
3573writing filename `fileio.tmp', with a \fB\-\-client\fR `hostfile'
3574containing two hostnames `h1' and `h2' with IP addresses 192.168.10.120 and
3575192.168.10.121, then fio will create two files:
3576.RS
3577.P
3578.PD 0
39b5f61e 3579/mnt/nfs/fio/192.168.10.120.fileio.tmp
40943b9a 3580.P
39b5f61e 3581/mnt/nfs/fio/192.168.10.121.fileio.tmp
40943b9a
TK
3582.PD
3583.RE
d60e92d1
AC
3584.SH AUTHORS
3585.B fio
d292596c 3586was written by Jens Axboe <axboe@kernel.dk>.
d1429b5c
AC
3587.br
3588This man page was written by Aaron Carroll <aaronc@cse.unsw.edu.au> based
d60e92d1 3589on documentation by Jens Axboe.
40943b9a
TK
3590.br
3591This man page was rewritten by Tomohiro Kusumi <tkusumi@tuxera.com> based
3592on documentation by Jens Axboe.
d60e92d1 3593.SH "REPORTING BUGS"
482900c9 3594Report bugs to the \fBfio\fR mailing list <fio@vger.kernel.org>.
6468020d 3595.br
40943b9a
TK
3596See \fBREPORTING\-BUGS\fR.
3597.P
3598\fBREPORTING\-BUGS\fR: \fIhttp://git.kernel.dk/cgit/fio/plain/REPORTING\-BUGS\fR
d60e92d1 3599.SH "SEE ALSO"
d1429b5c
AC
3600For further documentation see \fBHOWTO\fR and \fBREADME\fR.
3601.br
40943b9a 3602Sample jobfiles are available in the `examples/' directory.
9040e236 3603.br
40943b9a
TK
3604These are typically located under `/usr/share/doc/fio'.
3605.P
3606\fBHOWTO\fR: \fIhttp://git.kernel.dk/cgit/fio/plain/HOWTO\fR
9040e236 3607.br
40943b9a 3608\fBREADME\fR: \fIhttp://git.kernel.dk/cgit/fio/plain/README\fR