fio_time: should include time.h
[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
bdd88be3 80be `always', `never' or `auto'.
d60e92d1 81.TP
30b5d57f 82.BI \-\-eta\-newline \fR=\fPtime
bdd88be3
TK
83Force a new line for every \fItime\fR period passed. When the unit is omitted,
84the value is interpreted in seconds.
30b5d57f
JA
85.TP
86.BI \-\-status\-interval \fR=\fPtime
aa6cb459
VF
87Force a full status dump of cumulative (from job start) values at \fItime\fR
88intervals. This option does *not* provide per-period measurements. So
89values such as bandwidth are running averages. When the time unit is omitted,
90\fItime\fR is interpreted in seconds.
bdd88be3
TK
91.TP
92.BI \-\-section \fR=\fPname
93Only run specified section \fIname\fR in job file. Multiple sections can be specified.
7db7a5a0 94The \fB\-\-section\fR option allows one to combine related jobs into one file.
bdd88be3 95E.g. one job file could define light, moderate, and heavy sections. Tell
7db7a5a0 96fio to run only the "heavy" section by giving `\-\-section=heavy'
bdd88be3 97command line option. One can also specify the "write" operations in one
7db7a5a0 98section and "verify" operation in another section. The \fB\-\-section\fR option
bdd88be3
TK
99only applies to job sections. The reserved *global* section is always
100parsed and used.
c0a5d35e 101.TP
49da1240 102.BI \-\-alloc\-size \fR=\fPkb
7db7a5a0
TK
103Set the internal smalloc pool size to \fIkb\fR in KiB. The
104\fB\-\-alloc\-size\fR switch allows one to use a larger pool size for smalloc.
bdd88be3
TK
105If running large jobs with randommap enabled, fio can run out of memory.
106Smalloc is an internal allocator for shared structures from a fixed size
107memory pool and can grow to 16 pools. The pool size defaults to 16MiB.
7db7a5a0
TK
108NOTE: While running `.fio_smalloc.*' backing store files are visible
109in `/tmp'.
d60e92d1 110.TP
49da1240
JA
111.BI \-\-warnings\-fatal
112All fio parser warnings are fatal, causing fio to exit with an error.
9183788d 113.TP
49da1240 114.BI \-\-max\-jobs \fR=\fPnr
7db7a5a0 115Set the maximum number of threads/processes to support to \fInr\fR.
7f4811bb
RNS
116NOTE: On Linux, it may be necessary to increase the shared-memory limit
117(`/proc/sys/kernel/shmmax') if fio runs into errors while creating jobs.
d60e92d1 118.TP
49da1240 119.BI \-\-server \fR=\fPargs
7db7a5a0
TK
120Start a backend server, with \fIargs\fR specifying what to listen to.
121See \fBCLIENT/SERVER\fR section.
f57a9c59 122.TP
49da1240 123.BI \-\-daemonize \fR=\fPpidfile
7db7a5a0 124Background a fio server, writing the pid to the given \fIpidfile\fR file.
49da1240 125.TP
bdd88be3 126.BI \-\-client \fR=\fPhostname
7db7a5a0
TK
127Instead of running the jobs locally, send and run them on the given \fIhostname\fR
128or set of \fIhostname\fRs. See \fBCLIENT/SERVER\fR section.
bdd88be3 129.TP
7db7a5a0
TK
130.BI \-\-remote\-config \fR=\fPfile
131Tell fio server to load this local \fIfile\fR.
f2a2ce0e
HL
132.TP
133.BI \-\-idle\-prof \fR=\fPoption
7db7a5a0 134Report CPU idleness. \fIoption\fR is one of the following:
bdd88be3
TK
135.RS
136.RS
137.TP
138.B calibrate
139Run unit work calibration only and exit.
140.TP
141.B system
142Show aggregate system idleness and unit work.
143.TP
144.B percpu
7db7a5a0 145As \fBsystem\fR but also show per CPU idleness.
bdd88be3
TK
146.RE
147.RE
148.TP
7db7a5a0
TK
149.BI \-\-inflate\-log \fR=\fPlog
150Inflate and output compressed \fIlog\fR.
bdd88be3 151.TP
7db7a5a0
TK
152.BI \-\-trigger\-file \fR=\fPfile
153Execute trigger command when \fIfile\fR exists.
bdd88be3 154.TP
7db7a5a0
TK
155.BI \-\-trigger\-timeout \fR=\fPtime
156Execute trigger at this \fItime\fR.
bdd88be3 157.TP
7db7a5a0
TK
158.BI \-\-trigger \fR=\fPcommand
159Set this \fIcommand\fR as local trigger.
bdd88be3 160.TP
7db7a5a0
TK
161.BI \-\-trigger\-remote \fR=\fPcommand
162Set this \fIcommand\fR as remote trigger.
bdd88be3 163.TP
7db7a5a0
TK
164.BI \-\-aux\-path \fR=\fPpath
165Use this \fIpath\fR for fio state generated files.
d60e92d1 166.SH "JOB FILE FORMAT"
7a14cf18
TK
167Any parameters following the options will be assumed to be job files, unless
168they match a job file parameter. Multiple job files can be listed and each job
7db7a5a0 169file will be regarded as a separate group. Fio will \fBstonewall\fR execution
7a14cf18
TK
170between each group.
171
172Fio accepts one or more job files describing what it is
173supposed to do. The job file format is the classic ini file, where the names
174enclosed in [] brackets define the job name. You are free to use any ASCII name
175you want, except *global* which has special meaning. Following the job name is
176a sequence of zero or more parameters, one per line, that define the behavior of
177the job. If the first character in a line is a ';' or a '#', the entire line is
178discarded as a comment.
179
180A *global* section sets defaults for the jobs described in that file. A job may
181override a *global* section parameter, and a job file may even have several
182*global* sections if so desired. A job is only affected by a *global* section
183residing above it.
184
7db7a5a0
TK
185The \fB\-\-cmdhelp\fR option also lists all options. If used with an \fIcommand\fR
186argument, \fB\-\-cmdhelp\fR will detail the given \fIcommand\fR.
7a14cf18 187
7db7a5a0
TK
188See the `examples/' directory for inspiration on how to write job files. Note
189the copyright and license requirements currently apply to
190`examples/' files.
54eb4569
TK
191.SH "JOB FILE PARAMETERS"
192Some parameters take an option of a given type, such as an integer or a
193string. Anywhere a numeric value is required, an arithmetic expression may be
194used, provided it is surrounded by parentheses. Supported operators are:
d59aa780 195.RS
7db7a5a0 196.P
d59aa780 197.B addition (+)
7db7a5a0
TK
198.P
199.B subtraction (\-)
200.P
d59aa780 201.B multiplication (*)
7db7a5a0 202.P
d59aa780 203.B division (/)
7db7a5a0 204.P
d59aa780 205.B modulus (%)
7db7a5a0 206.P
d59aa780
JA
207.B exponentiation (^)
208.RE
d59aa780
JA
209.P
210For time values in expressions, units are microseconds by default. This is
211different than for time values not in expressions (not enclosed in
54eb4569
TK
212parentheses).
213.SH "PARAMETER TYPES"
214The following parameter types are used.
d60e92d1
AC
215.TP
216.I str
6b86fc18
TK
217String. A sequence of alphanumeric characters.
218.TP
219.I time
220Integer with possible time suffix. Without a unit value is interpreted as
221seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
222hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and 'us'
223(or 'usec') for microseconds. For example, use 10m for 10 minutes.
d60e92d1
AC
224.TP
225.I int
6d500c2e
RE
226Integer. A whole number value, which may contain an integer prefix
227and an integer suffix.
0b43a833
TK
228.RS
229.RS
230.P
6b86fc18 231[*integer prefix*] **number** [*integer suffix*]
0b43a833
TK
232.RE
233.P
6b86fc18
TK
234The optional *integer prefix* specifies the number's base. The default
235is decimal. *0x* specifies hexadecimal.
0b43a833 236.P
6b86fc18
TK
237The optional *integer suffix* specifies the number's units, and includes an
238optional unit prefix and an optional unit. For quantities of data, the
239default unit is bytes. For quantities of time, the default unit is seconds
240unless otherwise specified.
0b43a833
TK
241.P
242With `kb_base=1000', fio follows international standards for unit
7db7a5a0 243prefixes. To specify power\-of\-10 decimal values defined in the
6b86fc18 244International System of Units (SI):
0b43a833
TK
245.RS
246.P
7db7a5a0 247.PD 0
eccce61a 248K means kilo (K) or 1000
7db7a5a0 249.P
eccce61a 250M means mega (M) or 1000**2
7db7a5a0 251.P
eccce61a 252G means giga (G) or 1000**3
7db7a5a0 253.P
eccce61a 254T means tera (T) or 1000**4
7db7a5a0 255.P
eccce61a 256P means peta (P) or 1000**5
7db7a5a0 257.PD
0b43a833
TK
258.RE
259.P
7db7a5a0 260To specify power\-of\-2 binary values defined in IEC 80000\-13:
0b43a833
TK
261.RS
262.P
7db7a5a0 263.PD 0
eccce61a 264Ki means kibi (Ki) or 1024
7db7a5a0 265.P
eccce61a 266Mi means mebi (Mi) or 1024**2
7db7a5a0 267.P
eccce61a 268Gi means gibi (Gi) or 1024**3
7db7a5a0 269.P
eccce61a 270Ti means tebi (Ti) or 1024**4
7db7a5a0 271.P
eccce61a 272Pi means pebi (Pi) or 1024**5
7db7a5a0 273.PD
0b43a833
TK
274.RE
275.P
276With `kb_base=1024' (the default), the unit prefixes are opposite
7db7a5a0 277from those specified in the SI and IEC 80000\-13 standards to provide
6b86fc18 278compatibility with old scripts. For example, 4k means 4096.
0b43a833 279.P
6b86fc18
TK
280For quantities of data, an optional unit of 'B' may be included
281(e.g., 'kB' is the same as 'k').
0b43a833 282.P
6b86fc18
TK
283The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
284not milli). 'b' and 'B' both mean byte, not bit.
0b43a833
TK
285.P
286Examples with `kb_base=1000':
287.RS
288.P
7db7a5a0 289.PD 0
6d500c2e 2904 KiB: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
7db7a5a0 291.P
6d500c2e 2921 MiB: 1048576, 1m, 1024k
7db7a5a0 293.P
6d500c2e 2941 MB: 1000000, 1mi, 1000ki
7db7a5a0 295.P
6d500c2e 2961 TiB: 1073741824, 1t, 1024m, 1048576k
7db7a5a0 297.P
6d500c2e 2981 TB: 1000000000, 1ti, 1000mi, 1000000ki
7db7a5a0 299.PD
0b43a833
TK
300.RE
301.P
302Examples with `kb_base=1024' (default):
303.RS
304.P
7db7a5a0 305.PD 0
6d500c2e 3064 KiB: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
7db7a5a0 307.P
6d500c2e 3081 MiB: 1048576, 1m, 1024k
7db7a5a0 309.P
6d500c2e 3101 MB: 1000000, 1mi, 1000ki
7db7a5a0 311.P
6d500c2e 3121 TiB: 1073741824, 1t, 1024m, 1048576k
7db7a5a0 313.P
6d500c2e 3141 TB: 1000000000, 1ti, 1000mi, 1000000ki
7db7a5a0 315.PD
0b43a833
TK
316.RE
317.P
6d500c2e 318To specify times (units are not case sensitive):
0b43a833
TK
319.RS
320.P
7db7a5a0 321.PD 0
6d500c2e 322D means days
7db7a5a0 323.P
6d500c2e 324H means hours
7db7a5a0 325.P
6d500c2e 326M mean minutes
7db7a5a0 327.P
6d500c2e 328s or sec means seconds (default)
7db7a5a0 329.P
6d500c2e 330ms or msec means milliseconds
7db7a5a0 331.P
6d500c2e 332us or usec means microseconds
7db7a5a0 333.PD
0b43a833
TK
334.RE
335.P
6b86fc18 336If the option accepts an upper and lower range, use a colon ':' or
7db7a5a0 337minus '\-' to separate such values. See \fIirange\fR parameter type.
6b86fc18
TK
338If the lower value specified happens to be larger than the upper value
339the two values are swapped.
0b43a833 340.RE
d60e92d1
AC
341.TP
342.I bool
6b86fc18
TK
343Boolean. Usually parsed as an integer, however only defined for
344true and false (1 and 0).
d60e92d1
AC
345.TP
346.I irange
6b86fc18 347Integer range with suffix. Allows value range to be given, such as
7db7a5a0 3481024\-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
6b86fc18 349option allows two sets of ranges, they can be specified with a ',' or '/'
7db7a5a0 350delimiter: 1k\-4k/8k\-32k. Also see \fIint\fR parameter type.
83349190
YH
351.TP
352.I float_list
6b86fc18 353A list of floating point numbers, separated by a ':' character.
523bad63 354.SH "JOB PARAMETERS"
54eb4569 355With the above in mind, here follows the complete list of fio job parameters.
523bad63 356.SS "Units"
d60e92d1 357.TP
523bad63
TK
358.BI kb_base \fR=\fPint
359Select the interpretation of unit prefixes in input parameters.
360.RS
361.RS
d60e92d1 362.TP
523bad63
TK
363.B 1000
364Inputs comply with IEC 80000\-13 and the International
365System of Units (SI). Use:
366.RS
367.P
368.PD 0
369\- power\-of\-2 values with IEC prefixes (e.g., KiB)
370.P
371\- power\-of\-10 values with SI prefixes (e.g., kB)
372.PD
373.RE
374.TP
375.B 1024
376Compatibility mode (default). To avoid breaking old scripts:
377.P
378.RS
379.PD 0
380\- power\-of\-2 values with SI prefixes
381.P
382\- power\-of\-10 values with IEC prefixes
383.PD
384.RE
385.RE
386.P
387See \fBbs\fR for more details on input parameters.
388.P
389Outputs always use correct prefixes. Most outputs include both
390side\-by\-side, like:
391.P
392.RS
393bw=2383.3kB/s (2327.4KiB/s)
394.RE
395.P
396If only one value is reported, then kb_base selects the one to use:
397.P
398.RS
399.PD 0
4001000 \-\- SI prefixes
401.P
4021024 \-\- IEC prefixes
403.PD
404.RE
405.RE
406.TP
407.BI unit_base \fR=\fPint
408Base unit for reporting. Allowed values are:
409.RS
410.RS
411.TP
412.B 0
413Use auto\-detection (default).
414.TP
415.B 8
416Byte based.
417.TP
418.B 1
419Bit based.
420.RE
421.RE
422.SS "Job description"
423.TP
424.BI name \fR=\fPstr
425ASCII name of the job. This may be used to override the name printed by fio
426for this job. Otherwise the job name is used. On the command line this
427parameter has the special purpose of also signaling the start of a new job.
9cc8cb91 428.TP
d60e92d1 429.BI description \fR=\fPstr
523bad63
TK
430Text description of the job. Doesn't do anything except dump this text
431description when this job is run. It's not parsed.
432.TP
433.BI loops \fR=\fPint
434Run the specified number of iterations of this job. Used to repeat the same
435workload a given number of times. Defaults to 1.
436.TP
437.BI numjobs \fR=\fPint
438Create the specified number of clones of this job. Each clone of job
439is spawned as an independent thread or process. May be used to setup a
440larger number of threads/processes doing the same thing. Each thread is
441reported separately; to see statistics for all clones as a whole, use
442\fBgroup_reporting\fR in conjunction with \fBnew_group\fR.
443See \fB\-\-max\-jobs\fR. Default: 1.
444.SS "Time related parameters"
445.TP
446.BI runtime \fR=\fPtime
447Tell fio to terminate processing after the specified period of time. It
448can be quite hard to determine for how long a specified job will run, so
449this parameter is handy to cap the total runtime to a given time. When
450the unit is omitted, the value is intepreted in seconds.
451.TP
452.BI time_based
453If set, fio will run for the duration of the \fBruntime\fR specified
454even if the file(s) are completely read or written. It will simply loop over
455the same workload as many times as the \fBruntime\fR allows.
456.TP
457.BI startdelay \fR=\fPirange(int)
458Delay the start of job for the specified amount of time. Can be a single
459value or a range. When given as a range, each thread will choose a value
460randomly from within the range. Value is in seconds if a unit is omitted.
461.TP
462.BI ramp_time \fR=\fPtime
463If set, fio will run the specified workload for this amount of time before
464logging any performance numbers. Useful for letting performance settle
465before logging results, thus minimizing the runtime required for stable
466results. Note that the \fBramp_time\fR is considered lead in time for a job,
467thus it will increase the total runtime if a special timeout or
468\fBruntime\fR is specified. When the unit is omitted, the value is
469given in seconds.
470.TP
471.BI clocksource \fR=\fPstr
472Use the given clocksource as the base of timing. The supported options are:
473.RS
474.RS
475.TP
476.B gettimeofday
477\fBgettimeofday\fR\|(2)
478.TP
479.B clock_gettime
480\fBclock_gettime\fR\|(2)
481.TP
482.B cpu
483Internal CPU clock source
484.RE
485.P
486\fBcpu\fR is the preferred clocksource if it is reliable, as it is very fast (and
487fio is heavy on time calls). Fio will automatically use this clocksource if
488it's supported and considered reliable on the system it is running on,
489unless another clocksource is specifically set. For x86/x86\-64 CPUs, this
490means supporting TSC Invariant.
491.RE
492.TP
493.BI gtod_reduce \fR=\fPbool
494Enable all of the \fBgettimeofday\fR\|(2) reducing options
495(\fBdisable_clat\fR, \fBdisable_slat\fR, \fBdisable_bw_measurement\fR) plus
496reduce precision of the timeout somewhat to really shrink the
497\fBgettimeofday\fR\|(2) call count. With this option enabled, we only do
498about 0.4% of the \fBgettimeofday\fR\|(2) calls we would have done if all
499time keeping was enabled.
500.TP
501.BI gtod_cpu \fR=\fPint
502Sometimes it's cheaper to dedicate a single thread of execution to just
503getting the current time. Fio (and databases, for instance) are very
504intensive on \fBgettimeofday\fR\|(2) calls. With this option, you can set
505one CPU aside for doing nothing but logging current time to a shared memory
506location. Then the other threads/processes that run I/O workloads need only
507copy that segment, instead of entering the kernel with a
508\fBgettimeofday\fR\|(2) call. The CPU set aside for doing these time
509calls will be excluded from other uses. Fio will manually clear it from the
510CPU mask of other jobs.
511.SS "Target file/device"
d60e92d1
AC
512.TP
513.BI directory \fR=\fPstr
523bad63
TK
514Prefix \fBfilename\fRs with this directory. Used to place files in a different
515location than `./'. You can specify a number of directories by
516separating the names with a ':' character. These directories will be
517assigned equally distributed to job clones created by \fBnumjobs\fR as
518long as they are using generated filenames. If specific \fBfilename\fR(s) are
519set fio will use the first listed directory, and thereby matching the
520\fBfilename\fR semantic which generates a file each clone if not specified, but
521let all clones use the same if set.
522.RS
523.P
524See the \fBfilename\fR option for information on how to escape ':' and '\'
525characters within the directory path itself.
526.RE
d60e92d1
AC
527.TP
528.BI filename \fR=\fPstr
523bad63
TK
529Fio normally makes up a \fBfilename\fR based on the job name, thread number, and
530file number (see \fBfilename_format\fR). If you want to share files
531between threads in a job or several
532jobs with fixed file paths, specify a \fBfilename\fR for each of them to override
533the default. If the ioengine is file based, you can specify a number of files
534by separating the names with a ':' colon. So if you wanted a job to open
535`/dev/sda' and `/dev/sdb' as the two working files, you would use
536`filename=/dev/sda:/dev/sdb'. This also means that whenever this option is
537specified, \fBnrfiles\fR is ignored. The size of regular files specified
538by this option will be \fBsize\fR divided by number of files unless an
539explicit size is specified by \fBfilesize\fR.
540.RS
541.P
542Each colon and backslash in the wanted path must be escaped with a '\'
543character. For instance, if the path is `/dev/dsk/foo@3,0:c' then you
544would use `filename=/dev/dsk/foo@3,0\\:c' and if the path is
545`F:\\\\filename' then you would use `filename=F\\:\\\\filename'.
546.P
547On Windows, disk devices are accessed as `\\\\\\\\.\\\\PhysicalDrive0' for
548the first device, `\\\\\\\\.\\\\PhysicalDrive1' for the second etc.
549Note: Windows and FreeBSD prevent write access to areas
550of the disk containing in\-use data (e.g. filesystems).
551.P
552The filename `\-' is a reserved name, meaning *stdin* or *stdout*. Which
553of the two depends on the read/write direction set.
554.RE
d60e92d1 555.TP
de98bd30 556.BI filename_format \fR=\fPstr
523bad63
TK
557If sharing multiple files between jobs, it is usually necessary to have fio
558generate the exact names that you want. By default, fio will name a file
de98bd30 559based on the default file format specification of
523bad63 560`jobname.jobnumber.filenumber'. With this option, that can be
de98bd30
JA
561customized. Fio will recognize and replace the following keywords in this
562string:
563.RS
564.RS
565.TP
566.B $jobname
567The name of the worker thread or process.
568.TP
569.B $jobnum
570The incremental number of the worker thread or process.
571.TP
572.B $filenum
573The incremental number of the file for that worker thread or process.
574.RE
575.P
523bad63
TK
576To have dependent jobs share a set of files, this option can be set to have
577fio generate filenames that are shared between the two. For instance, if
578`testfiles.$filenum' is specified, file number 4 for any job will be
579named `testfiles.4'. The default of `$jobname.$jobnum.$filenum'
de98bd30 580will be used if no other format specifier is given.
645943c0
JB
581.P
582If you specify a path then the directories will be created up to the main
583directory for the file. So for example if you specify `a/b/c/$jobnum` then the
584directories a/b/c will be created before the file setup part of the job. If you
585specify \fBdirectory\fR then the path will be relative that directory, otherwise
586it is treated as the absolute path.
de98bd30 587.RE
de98bd30 588.TP
922a5be8 589.BI unique_filename \fR=\fPbool
523bad63
TK
590To avoid collisions between networked clients, fio defaults to prefixing any
591generated filenames (with a directory specified) with the source of the
592client connecting. To disable this behavior, set this option to 0.
593.TP
594.BI opendir \fR=\fPstr
595Recursively open any files below directory \fIstr\fR.
922a5be8 596.TP
3ce9dcaf 597.BI lockfile \fR=\fPstr
523bad63
TK
598Fio defaults to not locking any files before it does I/O to them. If a file
599or file descriptor is shared, fio can serialize I/O to that file to make the
600end result consistent. This is usual for emulating real workloads that share
601files. The lock modes are:
3ce9dcaf
JA
602.RS
603.RS
604.TP
605.B none
523bad63 606No locking. The default.
3ce9dcaf
JA
607.TP
608.B exclusive
523bad63 609Only one thread or process may do I/O at a time, excluding all others.
3ce9dcaf
JA
610.TP
611.B readwrite
523bad63
TK
612Read\-write locking on the file. Many readers may
613access the file at the same time, but writes get exclusive access.
3ce9dcaf 614.RE
ce594fbe 615.RE
523bad63
TK
616.TP
617.BI nrfiles \fR=\fPint
618Number of files to use for this job. Defaults to 1. The size of files
619will be \fBsize\fR divided by this unless explicit size is specified by
620\fBfilesize\fR. Files are created for each thread separately, and each
621file will have a file number within its name by default, as explained in
622\fBfilename\fR section.
623.TP
624.BI openfiles \fR=\fPint
625Number of files to keep open at the same time. Defaults to the same as
626\fBnrfiles\fR, can be set smaller to limit the number simultaneous
627opens.
628.TP
629.BI file_service_type \fR=\fPstr
630Defines how fio decides which file from a job to service next. The following
631types are defined:
632.RS
633.RS
634.TP
635.B random
636Choose a file at random.
637.TP
638.B roundrobin
639Round robin over opened files. This is the default.
640.TP
641.B sequential
642Finish one file before moving on to the next. Multiple files can
643still be open depending on \fBopenfiles\fR.
644.TP
645.B zipf
646Use a Zipf distribution to decide what file to access.
647.TP
648.B pareto
649Use a Pareto distribution to decide what file to access.
650.TP
651.B normal
652Use a Gaussian (normal) distribution to decide what file to access.
653.TP
654.B gauss
655Alias for normal.
656.RE
3ce9dcaf 657.P
523bad63
TK
658For \fBrandom\fR, \fBroundrobin\fR, and \fBsequential\fR, a postfix can be appended to
659tell fio how many I/Os to issue before switching to a new file. For example,
660specifying `file_service_type=random:8' would cause fio to issue
6618 I/Os before selecting a new file at random. For the non\-uniform
662distributions, a floating point postfix can be given to influence how the
663distribution is skewed. See \fBrandom_distribution\fR for a description
664of how that would work.
665.RE
666.TP
667.BI ioscheduler \fR=\fPstr
668Attempt to switch the device hosting the file to the specified I/O scheduler
669before running.
670.TP
671.BI create_serialize \fR=\fPbool
672If true, serialize the file creation for the jobs. This may be handy to
673avoid interleaving of data files, which may greatly depend on the filesystem
674used and even the number of processors in the system. Default: true.
675.TP
676.BI create_fsync \fR=\fPbool
677\fBfsync\fR\|(2) the data file after creation. This is the default.
678.TP
679.BI create_on_open \fR=\fPbool
680If true, don't pre\-create files but allow the job's open() to create a file
681when it's time to do I/O. Default: false \-\- pre\-create all necessary files
682when the job starts.
683.TP
684.BI create_only \fR=\fPbool
685If true, fio will only run the setup phase of the job. If files need to be
686laid out or updated on disk, only that will be done \-\- the actual job contents
687are not executed. Default: false.
688.TP
689.BI allow_file_create \fR=\fPbool
690If true, fio is permitted to create files as part of its workload. If this
691option is false, then fio will error out if
692the files it needs to use don't already exist. Default: true.
693.TP
694.BI allow_mounted_write \fR=\fPbool
695If this isn't set, fio will abort jobs that are destructive (e.g. that write)
696to what appears to be a mounted device or partition. This should help catch
697creating inadvertently destructive tests, not realizing that the test will
698destroy data on the mounted file system. Note that some platforms don't allow
699writing against a mounted device regardless of this option. Default: false.
700.TP
701.BI pre_read \fR=\fPbool
702If this is given, files will be pre\-read into memory before starting the
703given I/O operation. This will also clear the \fBinvalidate\fR flag,
704since it is pointless to pre\-read and then drop the cache. This will only
705work for I/O engines that are seek\-able, since they allow you to read the
706same data multiple times. Thus it will not work on non\-seekable I/O engines
707(e.g. network, splice). Default: false.
708.TP
709.BI unlink \fR=\fPbool
710Unlink the job files when done. Not the default, as repeated runs of that
711job would then waste time recreating the file set again and again. Default:
712false.
713.TP
714.BI unlink_each_loop \fR=\fPbool
715Unlink job files after each iteration or loop. Default: false.
716.TP
717.BI zonesize \fR=\fPint
718Divide a file into zones of the specified size. See \fBzoneskip\fR.
719.TP
720.BI zonerange \fR=\fPint
721Give size of an I/O zone. See \fBzoneskip\fR.
722.TP
723.BI zoneskip \fR=\fPint
724Skip the specified number of bytes when \fBzonesize\fR data has been
725read. The two zone options can be used to only do I/O on zones of a file.
726.SS "I/O type"
727.TP
728.BI direct \fR=\fPbool
729If value is true, use non\-buffered I/O. This is usually O_DIRECT. Note that
8e889110 730OpenBSD and ZFS on Solaris don't support direct I/O. On Windows the synchronous
523bad63
TK
731ioengines don't support direct I/O. Default: false.
732.TP
733.BI atomic \fR=\fPbool
734If value is true, attempt to use atomic direct I/O. Atomic writes are
735guaranteed to be stable once acknowledged by the operating system. Only
736Linux supports O_ATOMIC right now.
737.TP
738.BI buffered \fR=\fPbool
739If value is true, use buffered I/O. This is the opposite of the
740\fBdirect\fR option. Defaults to true.
d60e92d1
AC
741.TP
742.BI readwrite \fR=\fPstr "\fR,\fP rw" \fR=\fPstr
523bad63 743Type of I/O pattern. Accepted values are:
d60e92d1
AC
744.RS
745.RS
746.TP
747.B read
d1429b5c 748Sequential reads.
d60e92d1
AC
749.TP
750.B write
d1429b5c 751Sequential writes.
d60e92d1 752.TP
fa769d44 753.B trim
169c098d 754Sequential trims (Linux block devices only).
fa769d44 755.TP
d60e92d1 756.B randread
d1429b5c 757Random reads.
d60e92d1
AC
758.TP
759.B randwrite
d1429b5c 760Random writes.
d60e92d1 761.TP
fa769d44 762.B randtrim
169c098d 763Random trims (Linux block devices only).
fa769d44 764.TP
523bad63
TK
765.B rw,readwrite
766Sequential mixed reads and writes.
d60e92d1 767.TP
ff6bb260 768.B randrw
523bad63 769Random mixed reads and writes.
82a90686
JA
770.TP
771.B trimwrite
523bad63
TK
772Sequential trim+write sequences. Blocks will be trimmed first,
773then the same blocks will be written to.
d60e92d1
AC
774.RE
775.P
523bad63
TK
776Fio defaults to read if the option is not specified. For the mixed I/O
777types, the default is to split them 50/50. For certain types of I/O the
778result may still be skewed a bit, since the speed may be different.
779.P
780It is possible to specify the number of I/Os to do before getting a new
781offset by appending `:<nr>' to the end of the string given. For a
782random read, it would look like `rw=randread:8' for passing in an offset
783modifier with a value of 8. If the suffix is used with a sequential I/O
784pattern, then the `<nr>' value specified will be added to the generated
785offset for each I/O turning sequential I/O into sequential I/O with holes.
786For instance, using `rw=write:4k' will skip 4k for every write. Also see
787the \fBrw_sequencer\fR option.
d60e92d1
AC
788.RE
789.TP
38dad62d 790.BI rw_sequencer \fR=\fPstr
523bad63
TK
791If an offset modifier is given by appending a number to the `rw=\fIstr\fR'
792line, then this option controls how that number modifies the I/O offset
793being generated. Accepted values are:
38dad62d
JA
794.RS
795.RS
796.TP
797.B sequential
523bad63 798Generate sequential offset.
38dad62d
JA
799.TP
800.B identical
523bad63 801Generate the same offset.
38dad62d
JA
802.RE
803.P
523bad63
TK
804\fBsequential\fR is only useful for random I/O, where fio would normally
805generate a new random offset for every I/O. If you append e.g. 8 to randread,
806you would get a new random offset for every 8 I/Os. The result would be a
807seek for only every 8 I/Os, instead of for every I/O. Use `rw=randread:8'
808to specify that. As sequential I/O is already sequential, setting
809\fBsequential\fR for that would not result in any differences. \fBidentical\fR
810behaves in a similar fashion, except it sends the same offset 8 number of
811times before generating a new offset.
38dad62d 812.RE
90fef2d1 813.TP
771e58be
JA
814.BI unified_rw_reporting \fR=\fPbool
815Fio normally reports statistics on a per data direction basis, meaning that
523bad63
TK
816reads, writes, and trims are accounted and reported separately. If this
817option is set fio sums the results and report them as "mixed" instead.
771e58be 818.TP
d60e92d1 819.BI randrepeat \fR=\fPbool
523bad63
TK
820Seed the random number generator used for random I/O patterns in a
821predictable way so the pattern is repeatable across runs. Default: true.
56e2a5fc
CE
822.TP
823.BI allrandrepeat \fR=\fPbool
824Seed all random number generators in a predictable way so results are
523bad63 825repeatable across runs. Default: false.
d60e92d1 826.TP
04778baf
JA
827.BI randseed \fR=\fPint
828Seed the random number generators based on this seed value, to be able to
829control what sequence of output is being generated. If not set, the random
830sequence depends on the \fBrandrepeat\fR setting.
831.TP
a596f047 832.BI fallocate \fR=\fPstr
523bad63
TK
833Whether pre\-allocation is performed when laying down files.
834Accepted values are:
a596f047
EG
835.RS
836.RS
837.TP
838.B none
523bad63 839Do not pre\-allocate space.
a596f047 840.TP
2c3e17be 841.B native
523bad63
TK
842Use a platform's native pre\-allocation call but fall back to
843\fBnone\fR behavior if it fails/is not implemented.
2c3e17be 844.TP
a596f047 845.B posix
523bad63 846Pre\-allocate via \fBposix_fallocate\fR\|(3).
a596f047
EG
847.TP
848.B keep
523bad63
TK
849Pre\-allocate via \fBfallocate\fR\|(2) with
850FALLOC_FL_KEEP_SIZE set.
a596f047
EG
851.TP
852.B 0
523bad63 853Backward\-compatible alias for \fBnone\fR.
a596f047
EG
854.TP
855.B 1
523bad63 856Backward\-compatible alias for \fBposix\fR.
a596f047
EG
857.RE
858.P
523bad63
TK
859May not be available on all supported platforms. \fBkeep\fR is only available
860on Linux. If using ZFS on Solaris this cannot be set to \fBposix\fR
861because ZFS doesn't support pre\-allocation. Default: \fBnative\fR if any
862pre\-allocation methods are available, \fBnone\fR if not.
a596f047 863.RE
7bc8c2cf 864.TP
ecb2083d 865.BI fadvise_hint \fR=\fPstr
cf145d90 866Use \fBposix_fadvise\fR\|(2) to advise the kernel what I/O patterns
ecb2083d
JA
867are likely to be issued. Accepted values are:
868.RS
869.RS
870.TP
871.B 0
872Backwards compatible hint for "no hint".
873.TP
874.B 1
875Backwards compatible hint for "advise with fio workload type". This
523bad63 876uses FADV_RANDOM for a random workload, and FADV_SEQUENTIAL
ecb2083d
JA
877for a sequential workload.
878.TP
879.B sequential
523bad63 880Advise using FADV_SEQUENTIAL.
ecb2083d
JA
881.TP
882.B random
523bad63 883Advise using FADV_RANDOM.
ecb2083d
JA
884.RE
885.RE
d60e92d1 886.TP
8f4b9f24 887.BI write_hint \fR=\fPstr
523bad63
TK
888Use \fBfcntl\fR\|(2) to advise the kernel what life time to expect
889from a write. Only supported on Linux, as of version 4.13. Accepted
8f4b9f24
JA
890values are:
891.RS
892.RS
893.TP
894.B none
895No particular life time associated with this file.
896.TP
897.B short
898Data written to this file has a short life time.
899.TP
900.B medium
901Data written to this file has a medium life time.
902.TP
903.B long
904Data written to this file has a long life time.
905.TP
906.B extreme
907Data written to this file has a very long life time.
908.RE
523bad63
TK
909.P
910The values are all relative to each other, and no absolute meaning
911should be associated with them.
8f4b9f24 912.RE
37659335 913.TP
523bad63
TK
914.BI offset \fR=\fPint
915Start I/O at the provided offset in the file, given as either a fixed size in
83c8b093
JF
916bytes or a percentage. If a percentage is given, the generated offset will be
917aligned to the minimum \fBblocksize\fR or to the value of \fBoffset_align\fR if
918provided. Data before the given offset will not be touched. This
523bad63
TK
919effectively caps the file size at `real_size \- offset'. Can be combined with
920\fBsize\fR to constrain the start and end range of the I/O workload.
921A percentage can be specified by a number between 1 and 100 followed by '%',
922for example, `offset=20%' to specify 20%.
6d500c2e 923.TP
83c8b093
JF
924.BI offset_align \fR=\fPint
925If set to non-zero value, the byte offset generated by a percentage \fBoffset\fR
926is aligned upwards to this value. Defaults to 0 meaning that a percentage
927offset is aligned to the minimum block size.
928.TP
523bad63
TK
929.BI offset_increment \fR=\fPint
930If this is provided, then the real offset becomes `\fBoffset\fR + \fBoffset_increment\fR
931* thread_number', where the thread number is a counter that starts at 0 and
932is incremented for each sub\-job (i.e. when \fBnumjobs\fR option is
933specified). This option is useful if there are several jobs which are
934intended to operate on a file in parallel disjoint segments, with even
935spacing between the starting points.
6d500c2e 936.TP
523bad63
TK
937.BI number_ios \fR=\fPint
938Fio will normally perform I/Os until it has exhausted the size of the region
939set by \fBsize\fR, or if it exhaust the allocated time (or hits an error
940condition). With this setting, the range/size can be set independently of
941the number of I/Os to perform. When fio reaches this number, it will exit
942normally and report status. Note that this does not extend the amount of I/O
943that will be done, it will only stop fio if this condition is met before
944other end\-of\-job criteria.
d60e92d1 945.TP
523bad63
TK
946.BI fsync \fR=\fPint
947If writing to a file, issue an \fBfsync\fR\|(2) (or its equivalent) of
948the dirty data for every number of blocks given. For example, if you give 32
949as a parameter, fio will sync the file after every 32 writes issued. If fio is
950using non\-buffered I/O, we may not sync the file. The exception is the sg
951I/O engine, which synchronizes the disk cache anyway. Defaults to 0, which
952means fio does not periodically issue and wait for a sync to complete. Also
953see \fBend_fsync\fR and \fBfsync_on_close\fR.
6d500c2e 954.TP
523bad63
TK
955.BI fdatasync \fR=\fPint
956Like \fBfsync\fR but uses \fBfdatasync\fR\|(2) to only sync data and
957not metadata blocks. In Windows, FreeBSD, and DragonFlyBSD there is no
958\fBfdatasync\fR\|(2) so this falls back to using \fBfsync\fR\|(2).
959Defaults to 0, which means fio does not periodically issue and wait for a
960data\-only sync to complete.
d60e92d1 961.TP
523bad63
TK
962.BI write_barrier \fR=\fPint
963Make every N\-th write a barrier write.
901bb994 964.TP
523bad63
TK
965.BI sync_file_range \fR=\fPstr:int
966Use \fBsync_file_range\fR\|(2) for every \fIint\fR number of write
967operations. Fio will track range of writes that have happened since the last
968\fBsync_file_range\fR\|(2) call. \fIstr\fR can currently be one or more of:
969.RS
970.RS
fd68418e 971.TP
523bad63
TK
972.B wait_before
973SYNC_FILE_RANGE_WAIT_BEFORE
c5751c62 974.TP
523bad63
TK
975.B write
976SYNC_FILE_RANGE_WRITE
c5751c62 977.TP
523bad63
TK
978.B wait_after
979SYNC_FILE_RANGE_WRITE_AFTER
2fa5a241 980.RE
523bad63
TK
981.P
982So if you do `sync_file_range=wait_before,write:8', fio would use
983`SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE' for every 8
984writes. Also see the \fBsync_file_range\fR\|(2) man page. This option is
985Linux specific.
2fa5a241 986.RE
ce35b1ec 987.TP
523bad63
TK
988.BI overwrite \fR=\fPbool
989If true, writes to a file will always overwrite existing data. If the file
990doesn't already exist, it will be created before the write phase begins. If
991the file exists and is large enough for the specified write phase, nothing
992will be done. Default: false.
5c94b008 993.TP
523bad63
TK
994.BI end_fsync \fR=\fPbool
995If true, \fBfsync\fR\|(2) file contents when a write stage has completed.
996Default: false.
d60e92d1 997.TP
523bad63
TK
998.BI fsync_on_close \fR=\fPbool
999If true, fio will \fBfsync\fR\|(2) a dirty file on close. This differs
1000from \fBend_fsync\fR in that it will happen on every file close, not
1001just at the end of the job. Default: false.
d60e92d1 1002.TP
523bad63
TK
1003.BI rwmixread \fR=\fPint
1004Percentage of a mixed workload that should be reads. Default: 50.
1005.TP
1006.BI rwmixwrite \fR=\fPint
1007Percentage of a mixed workload that should be writes. If both
1008\fBrwmixread\fR and \fBrwmixwrite\fR is given and the values do not
1009add up to 100%, the latter of the two will be used to override the
1010first. This may interfere with a given rate setting, if fio is asked to
1011limit reads or writes to a certain rate. If that is the case, then the
1012distribution may be skewed. Default: 50.
1013.TP
1014.BI random_distribution \fR=\fPstr:float[,str:float][,str:float]
1015By default, fio will use a completely uniform random distribution when asked
1016to perform random I/O. Sometimes it is useful to skew the distribution in
1017specific ways, ensuring that some parts of the data is more hot than others.
1018fio includes the following distribution models:
d60e92d1
AC
1019.RS
1020.RS
1021.TP
1022.B random
523bad63 1023Uniform random distribution
8c07860d
JA
1024.TP
1025.B zipf
523bad63 1026Zipf distribution
8c07860d
JA
1027.TP
1028.B pareto
523bad63 1029Pareto distribution
8c07860d 1030.TP
dd3503d3 1031.B normal
523bad63 1032Normal (Gaussian) distribution
dd3503d3 1033.TP
523bad63
TK
1034.B zoned
1035Zoned random distribution
59466396
JA
1036.B zoned_abs
1037Zoned absolute random distribution
d60e92d1
AC
1038.RE
1039.P
523bad63
TK
1040When using a \fBzipf\fR or \fBpareto\fR distribution, an input value is also
1041needed to define the access pattern. For \fBzipf\fR, this is the `Zipf theta'.
1042For \fBpareto\fR, it's the `Pareto power'. Fio includes a test
1043program, \fBfio\-genzipf\fR, that can be used visualize what the given input
1044values will yield in terms of hit rates. If you wanted to use \fBzipf\fR with
1045a `theta' of 1.2, you would use `random_distribution=zipf:1.2' as the
1046option. If a non\-uniform model is used, fio will disable use of the random
1047map. For the \fBnormal\fR distribution, a normal (Gaussian) deviation is
1048supplied as a value between 0 and 100.
1049.P
1050For a \fBzoned\fR distribution, fio supports specifying percentages of I/O
1051access that should fall within what range of the file or device. For
1052example, given a criteria of:
d60e92d1 1053.RS
523bad63
TK
1054.P
1055.PD 0
105660% of accesses should be to the first 10%
1057.P
105830% of accesses should be to the next 20%
1059.P
10608% of accesses should be to the next 30%
1061.P
10622% of accesses should be to the next 40%
1063.PD
1064.RE
1065.P
1066we can define that through zoning of the random accesses. For the above
1067example, the user would do:
1068.RS
1069.P
1070random_distribution=zoned:60/10:30/20:8/30:2/40
1071.RE
1072.P
59466396
JA
1073A \fBzoned_abs\fR distribution works exactly like the\fBzoned\fR, except that
1074it takes absolute sizes. For example, let's say you wanted to define access
1075according to the following criteria:
1076.RS
1077.P
1078.PD 0
107960% of accesses should be to the first 20G
1080.P
108130% of accesses should be to the next 100G
1082.P
108310% of accesses should be to the next 500G
1084.PD
1085.RE
1086.P
1087we can define an absolute zoning distribution with:
1088.RS
1089.P
1090random_distribution=zoned:60/10:30/20:8/30:2/40
1091.RE
1092.P
6a16ece8
JA
1093For both \fBzoned\fR and \fBzoned_abs\fR, fio supports defining up to 256
1094separate zones.
1095.P
59466396 1096Similarly to how \fBbssplit\fR works for setting ranges and percentages
523bad63
TK
1097of block sizes. Like \fBbssplit\fR, it's possible to specify separate
1098zones for reads, writes, and trims. If just one set is given, it'll apply to
1099all of them.
1100.RE
1101.TP
1102.BI percentage_random \fR=\fPint[,int][,int]
1103For a random workload, set how big a percentage should be random. This
1104defaults to 100%, in which case the workload is fully random. It can be set
1105from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1106sequential. Any setting in between will result in a random mix of sequential
1107and random I/O, at the given percentages. Comma\-separated values may be
1108specified for reads, writes, and trims as described in \fBblocksize\fR.
1109.TP
1110.BI norandommap
1111Normally fio will cover every block of the file when doing random I/O. If
1112this option is given, fio will just get a new random offset without looking
1113at past I/O history. This means that some blocks may not be read or written,
1114and that some blocks may be read/written more than once. If this option is
1115used with \fBverify\fR and multiple blocksizes (via \fBbsrange\fR),
1116only intact blocks are verified, i.e., partially\-overwritten blocks are
1117ignored.
1118.TP
1119.BI softrandommap \fR=\fPbool
1120See \fBnorandommap\fR. If fio runs with the random block map enabled and
1121it fails to allocate the map, if this option is set it will continue without
1122a random block map. As coverage will not be as complete as with random maps,
1123this option is disabled by default.
1124.TP
1125.BI random_generator \fR=\fPstr
1126Fio supports the following engines for generating I/O offsets for random I/O:
1127.RS
1128.RS
1129.TP
1130.B tausworthe
1131Strong 2^88 cycle random number generator.
1132.TP
1133.B lfsr
1134Linear feedback shift register generator.
1135.TP
1136.B tausworthe64
1137Strong 64\-bit 2^258 cycle random number generator.
1138.RE
1139.P
1140\fBtausworthe\fR is a strong random number generator, but it requires tracking
1141on the side if we want to ensure that blocks are only read or written
1142once. \fBlfsr\fR guarantees that we never generate the same offset twice, and
1143it's also less computationally expensive. It's not a true random generator,
1144however, though for I/O purposes it's typically good enough. \fBlfsr\fR only
1145works with single block sizes, not with workloads that use multiple block
1146sizes. If used with such a workload, fio may read or write some blocks
1147multiple times. The default value is \fBtausworthe\fR, unless the required
1148space exceeds 2^32 blocks. If it does, then \fBtausworthe64\fR is
1149selected automatically.
1150.RE
1151.SS "Block size"
1152.TP
1153.BI blocksize \fR=\fPint[,int][,int] "\fR,\fB bs" \fR=\fPint[,int][,int]
1154The block size in bytes used for I/O units. Default: 4096. A single value
1155applies to reads, writes, and trims. Comma\-separated values may be
1156specified for reads, writes, and trims. A value not terminated in a comma
1157applies to subsequent types. Examples:
1158.RS
1159.RS
1160.P
1161.PD 0
1162bs=256k means 256k for reads, writes and trims.
1163.P
1164bs=8k,32k means 8k for reads, 32k for writes and trims.
1165.P
1166bs=8k,32k, means 8k for reads, 32k for writes, and default for trims.
1167.P
1168bs=,8k means default for reads, 8k for writes and trims.
1169.P
1170bs=,8k, means default for reads, 8k for writes, and default for trims.
1171.PD
1172.RE
1173.RE
1174.TP
1175.BI blocksize_range \fR=\fPirange[,irange][,irange] "\fR,\fB bsrange" \fR=\fPirange[,irange][,irange]
1176A range of block sizes in bytes for I/O units. The issued I/O unit will
1177always be a multiple of the minimum size, unless
1178\fBblocksize_unaligned\fR is set.
1179Comma\-separated ranges may be specified for reads, writes, and trims as
1180described in \fBblocksize\fR. Example:
1181.RS
1182.RS
1183.P
1184bsrange=1k\-4k,2k\-8k
1185.RE
1186.RE
1187.TP
1188.BI bssplit \fR=\fPstr[,str][,str]
1189Sometimes you want even finer grained control of the block sizes issued, not
1190just an even split between them. This option allows you to weight various
1191block sizes, so that you are able to define a specific amount of block sizes
1192issued. The format for this option is:
1193.RS
1194.RS
1195.P
1196bssplit=blocksize/percentage:blocksize/percentage
1197.RE
1198.P
1199for as many block sizes as needed. So if you want to define a workload that
1200has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would write:
1201.RS
1202.P
1203bssplit=4k/10:64k/50:32k/40
1204.RE
1205.P
1206Ordering does not matter. If the percentage is left blank, fio will fill in
1207the remaining values evenly. So a bssplit option like this one:
1208.RS
1209.P
1210bssplit=4k/50:1k/:32k/
1211.RE
1212.P
1213would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always add up
1214to 100, if bssplit is given a range that adds up to more, it will error out.
1215.P
1216Comma\-separated values may be specified for reads, writes, and trims as
1217described in \fBblocksize\fR.
1218.P
1219If you want a workload that has 50% 2k reads and 50% 4k reads, while having
122090% 4k writes and 10% 8k writes, you would specify:
1221.RS
1222.P
1223bssplit=2k/50:4k/50,4k/90,8k/10
1224.RE
6a16ece8
JA
1225.P
1226Fio supports defining up to 64 different weights for each data direction.
523bad63
TK
1227.RE
1228.TP
1229.BI blocksize_unaligned "\fR,\fB bs_unaligned"
1230If set, fio will issue I/O units with any size within
1231\fBblocksize_range\fR, not just multiples of the minimum size. This
1232typically won't work with direct I/O, as that normally requires sector
1233alignment.
1234.TP
1235.BI bs_is_seq_rand \fR=\fPbool
1236If this option is set, fio will use the normal read,write blocksize settings
1237as sequential,random blocksize settings instead. Any random read or write
1238will use the WRITE blocksize settings, and any sequential read or write will
1239use the READ blocksize settings.
1240.TP
1241.BI blockalign \fR=\fPint[,int][,int] "\fR,\fB ba" \fR=\fPint[,int][,int]
1242Boundary to which fio will align random I/O units. Default:
1243\fBblocksize\fR. Minimum alignment is typically 512b for using direct
1244I/O, though it usually depends on the hardware block size. This option is
1245mutually exclusive with using a random map for files, so it will turn off
1246that option. Comma\-separated values may be specified for reads, writes, and
1247trims as described in \fBblocksize\fR.
1248.SS "Buffers and memory"
1249.TP
1250.BI zero_buffers
1251Initialize buffers with all zeros. Default: fill buffers with random data.
1252.TP
1253.BI refill_buffers
1254If this option is given, fio will refill the I/O buffers on every
1255submit. The default is to only fill it at init time and reuse that
1256data. Only makes sense if zero_buffers isn't specified, naturally. If data
1257verification is enabled, \fBrefill_buffers\fR is also automatically enabled.
1258.TP
1259.BI scramble_buffers \fR=\fPbool
1260If \fBrefill_buffers\fR is too costly and the target is using data
1261deduplication, then setting this option will slightly modify the I/O buffer
1262contents to defeat normal de\-dupe attempts. This is not enough to defeat
1263more clever block compression attempts, but it will stop naive dedupe of
1264blocks. Default: true.
1265.TP
1266.BI buffer_compress_percentage \fR=\fPint
72592780
SW
1267If this is set, then fio will attempt to provide I/O buffer content
1268(on WRITEs) that compresses to the specified level. Fio does this by
1269providing a mix of random data followed by fixed pattern data. The
1270fixed pattern is either zeros, or the pattern specified by
1271\fBbuffer_pattern\fR. If the \fBbuffer_pattern\fR option is used, it
1272might skew the compression ratio slightly. Setting
1273\fBbuffer_compress_percentage\fR to a value other than 100 will also
1274enable \fBrefill_buffers\fR in order to reduce the likelihood that
1275adjacent blocks are so similar that they over compress when seen
1276together. See \fBbuffer_compress_chunk\fR for how to set a finer or
1277coarser granularity of the random/fixed data regions. Defaults to unset
1278i.e., buffer data will not adhere to any compression level.
523bad63
TK
1279.TP
1280.BI buffer_compress_chunk \fR=\fPint
72592780
SW
1281This setting allows fio to manage how big the random/fixed data region
1282is when using \fBbuffer_compress_percentage\fR. When
1283\fBbuffer_compress_chunk\fR is set to some non-zero value smaller than the
1284block size, fio can repeat the random/fixed region throughout the I/O
1285buffer at the specified interval (which particularly useful when
1286bigger block sizes are used for a job). When set to 0, fio will use a
1287chunk size that matches the block size resulting in a single
1288random/fixed region within the I/O buffer. Defaults to 512. When the
1289unit is omitted, the value is interpreted in bytes.
523bad63
TK
1290.TP
1291.BI buffer_pattern \fR=\fPstr
1292If set, fio will fill the I/O buffers with this pattern or with the contents
1293of a file. If not set, the contents of I/O buffers are defined by the other
1294options related to buffer contents. The setting can be any pattern of bytes,
1295and can be prefixed with 0x for hex values. It may also be a string, where
1296the string must then be wrapped with "". Or it may also be a filename,
1297where the filename must be wrapped with '' in which case the file is
1298opened and read. Note that not all the file contents will be read if that
1299would cause the buffers to overflow. So, for example:
1300.RS
1301.RS
1302.P
1303.PD 0
1304buffer_pattern='filename'
1305.P
1306or:
1307.P
1308buffer_pattern="abcd"
1309.P
1310or:
1311.P
1312buffer_pattern=\-12
1313.P
1314or:
1315.P
1316buffer_pattern=0xdeadface
1317.PD
1318.RE
1319.P
1320Also you can combine everything together in any order:
1321.RS
1322.P
1323buffer_pattern=0xdeadface"abcd"\-12'filename'
1324.RE
1325.RE
1326.TP
1327.BI dedupe_percentage \fR=\fPint
1328If set, fio will generate this percentage of identical buffers when
1329writing. These buffers will be naturally dedupable. The contents of the
1330buffers depend on what other buffer compression settings have been set. It's
1331possible to have the individual buffers either fully compressible, or not at
72592780
SW
1332all \-\- this option only controls the distribution of unique buffers. Setting
1333this option will also enable \fBrefill_buffers\fR to prevent every buffer
1334being identical.
523bad63
TK
1335.TP
1336.BI invalidate \fR=\fPbool
1337Invalidate the buffer/page cache parts of the files to be used prior to
1338starting I/O if the platform and file type support it. Defaults to true.
1339This will be ignored if \fBpre_read\fR is also specified for the
1340same job.
1341.TP
1342.BI sync \fR=\fPbool
1343Use synchronous I/O for buffered writes. For the majority of I/O engines,
1344this means using O_SYNC. Default: false.
1345.TP
1346.BI iomem \fR=\fPstr "\fR,\fP mem" \fR=\fPstr
1347Fio can use various types of memory as the I/O unit buffer. The allowed
1348values are:
1349.RS
1350.RS
1351.TP
1352.B malloc
1353Use memory from \fBmalloc\fR\|(3) as the buffers. Default memory type.
1354.TP
1355.B shm
1356Use shared memory as the buffers. Allocated through \fBshmget\fR\|(2).
1357.TP
1358.B shmhuge
1359Same as \fBshm\fR, but use huge pages as backing.
1360.TP
1361.B mmap
1362Use \fBmmap\fR\|(2) to allocate buffers. May either be anonymous memory, or can
1363be file backed if a filename is given after the option. The format
1364is `mem=mmap:/path/to/file'.
1365.TP
1366.B mmaphuge
1367Use a memory mapped huge file as the buffer backing. Append filename
1368after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file'.
1369.TP
1370.B mmapshared
1371Same as \fBmmap\fR, but use a MMAP_SHARED mapping.
1372.TP
1373.B cudamalloc
1374Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1375The \fBioengine\fR must be \fBrdma\fR.
1376.RE
1377.P
1378The area allocated is a function of the maximum allowed bs size for the job,
1379multiplied by the I/O depth given. Note that for \fBshmhuge\fR and
1380\fBmmaphuge\fR to work, the system must have free huge pages allocated. This
1381can normally be checked and set by reading/writing
1382`/proc/sys/vm/nr_hugepages' on a Linux system. Fio assumes a huge page
1383is 4MiB in size. So to calculate the number of huge pages you need for a
1384given job file, add up the I/O depth of all jobs (normally one unless
1385\fBiodepth\fR is used) and multiply by the maximum bs set. Then divide
1386that number by the huge page size. You can see the size of the huge pages in
1387`/proc/meminfo'. If no huge pages are allocated by having a non\-zero
1388number in `nr_hugepages', using \fBmmaphuge\fR or \fBshmhuge\fR will fail. Also
1389see \fBhugepage\-size\fR.
1390.P
1391\fBmmaphuge\fR also needs to have hugetlbfs mounted and the file location
1392should point there. So if it's mounted in `/huge', you would use
1393`mem=mmaphuge:/huge/somefile'.
1394.RE
1395.TP
1396.BI iomem_align \fR=\fPint "\fR,\fP mem_align" \fR=\fPint
1397This indicates the memory alignment of the I/O memory buffers. Note that
1398the given alignment is applied to the first I/O unit buffer, if using
1399\fBiodepth\fR the alignment of the following buffers are given by the
1400\fBbs\fR used. In other words, if using a \fBbs\fR that is a
1401multiple of the page sized in the system, all buffers will be aligned to
1402this value. If using a \fBbs\fR that is not page aligned, the alignment
1403of subsequent I/O memory buffers is the sum of the \fBiomem_align\fR and
1404\fBbs\fR used.
1405.TP
1406.BI hugepage\-size \fR=\fPint
1407Defines the size of a huge page. Must at least be equal to the system
1408setting, see `/proc/meminfo'. Defaults to 4MiB. Should probably
1409always be a multiple of megabytes, so using `hugepage\-size=Xm' is the
1410preferred way to set this to avoid setting a non\-pow\-2 bad value.
1411.TP
1412.BI lockmem \fR=\fPint
1413Pin the specified amount of memory with \fBmlock\fR\|(2). Can be used to
1414simulate a smaller amount of memory. The amount specified is per worker.
1415.SS "I/O size"
1416.TP
1417.BI size \fR=\fPint
1418The total size of file I/O for each thread of this job. Fio will run until
1419this many bytes has been transferred, unless runtime is limited by other options
1420(such as \fBruntime\fR, for instance, or increased/decreased by \fBio_size\fR).
1421Fio will divide this size between the available files determined by options
1422such as \fBnrfiles\fR, \fBfilename\fR, unless \fBfilesize\fR is
1423specified by the job. If the result of division happens to be 0, the size is
1424set to the physical size of the given files or devices if they exist.
1425If this option is not specified, fio will use the full size of the given
1426files or devices. If the files do not exist, size must be given. It is also
1427possible to give size as a percentage between 1 and 100. If `size=20%' is
1428given, fio will use 20% of the full size of the given files or devices.
1429Can be combined with \fBoffset\fR to constrain the start and end range
1430that I/O will be done within.
1431.TP
1432.BI io_size \fR=\fPint "\fR,\fB io_limit" \fR=\fPint
1433Normally fio operates within the region set by \fBsize\fR, which means
1434that the \fBsize\fR option sets both the region and size of I/O to be
1435performed. Sometimes that is not what you want. With this option, it is
1436possible to define just the amount of I/O that fio should do. For instance,
1437if \fBsize\fR is set to 20GiB and \fBio_size\fR is set to 5GiB, fio
1438will perform I/O within the first 20GiB but exit when 5GiB have been
1439done. The opposite is also possible \-\- if \fBsize\fR is set to 20GiB,
1440and \fBio_size\fR is set to 40GiB, then fio will do 40GiB of I/O within
1441the 0..20GiB region.
1442.TP
1443.BI filesize \fR=\fPirange(int)
1444Individual file sizes. May be a range, in which case fio will select sizes
1445for files at random within the given range and limited to \fBsize\fR in
1446total (if that is given). If not given, each created file is the same size.
1447This option overrides \fBsize\fR in terms of file size, which means
1448this value is used as a fixed size or possible range of each file.
1449.TP
1450.BI file_append \fR=\fPbool
1451Perform I/O after the end of the file. Normally fio will operate within the
1452size of a file. If this option is set, then fio will append to the file
1453instead. This has identical behavior to setting \fBoffset\fR to the size
1454of a file. This option is ignored on non\-regular files.
1455.TP
1456.BI fill_device \fR=\fPbool "\fR,\fB fill_fs" \fR=\fPbool
1457Sets size to something really large and waits for ENOSPC (no space left on
1458device) as the terminating condition. Only makes sense with sequential
1459write. For a read workload, the mount point will be filled first then I/O
1460started on the result. This option doesn't make sense if operating on a raw
1461device node, since the size of that is already known by the file system.
1462Additionally, writing beyond end\-of\-device will not return ENOSPC there.
1463.SS "I/O engine"
1464.TP
1465.BI ioengine \fR=\fPstr
1466Defines how the job issues I/O to the file. The following types are defined:
1467.RS
1468.RS
1469.TP
1470.B sync
1471Basic \fBread\fR\|(2) or \fBwrite\fR\|(2)
1472I/O. \fBlseek\fR\|(2) is used to position the I/O location.
1473See \fBfsync\fR and \fBfdatasync\fR for syncing write I/Os.
1474.TP
1475.B psync
1476Basic \fBpread\fR\|(2) or \fBpwrite\fR\|(2) I/O. Default on
1477all supported operating systems except for Windows.
1478.TP
1479.B vsync
1480Basic \fBreadv\fR\|(2) or \fBwritev\fR\|(2) I/O. Will emulate
1481queuing by coalescing adjacent I/Os into a single submission.
1482.TP
1483.B pvsync
1484Basic \fBpreadv\fR\|(2) or \fBpwritev\fR\|(2) I/O.
a46c5e01 1485.TP
2cafffbe
JA
1486.B pvsync2
1487Basic \fBpreadv2\fR\|(2) or \fBpwritev2\fR\|(2) I/O.
1488.TP
d60e92d1 1489.B libaio
523bad63
TK
1490Linux native asynchronous I/O. Note that Linux may only support
1491queued behavior with non\-buffered I/O (set `direct=1' or
1492`buffered=0').
1493This engine defines engine specific options.
d60e92d1
AC
1494.TP
1495.B posixaio
523bad63
TK
1496POSIX asynchronous I/O using \fBaio_read\fR\|(3) and
1497\fBaio_write\fR\|(3).
03e20d68
BC
1498.TP
1499.B solarisaio
1500Solaris native asynchronous I/O.
1501.TP
1502.B windowsaio
38f8c318 1503Windows native asynchronous I/O. Default on Windows.
d60e92d1
AC
1504.TP
1505.B mmap
523bad63
TK
1506File is memory mapped with \fBmmap\fR\|(2) and data copied
1507to/from using \fBmemcpy\fR\|(3).
d60e92d1
AC
1508.TP
1509.B splice
523bad63
TK
1510\fBsplice\fR\|(2) is used to transfer the data and
1511\fBvmsplice\fR\|(2) to transfer data from user space to the
1512kernel.
d60e92d1 1513.TP
d60e92d1 1514.B sg
523bad63
TK
1515SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1516ioctl, or if the target is an sg character device we use
1517\fBread\fR\|(2) and \fBwrite\fR\|(2) for asynchronous
1518I/O. Requires \fBfilename\fR option to specify either block or
1519character devices.
d60e92d1
AC
1520.TP
1521.B null
523bad63
TK
1522Doesn't transfer any data, just pretends to. This is mainly used to
1523exercise fio itself and for debugging/testing purposes.
d60e92d1
AC
1524.TP
1525.B net
523bad63
TK
1526Transfer over the network to given `host:port'. Depending on the
1527\fBprotocol\fR used, the \fBhostname\fR, \fBport\fR,
1528\fBlisten\fR and \fBfilename\fR options are used to specify
1529what sort of connection to make, while the \fBprotocol\fR option
1530determines which protocol will be used. This engine defines engine
1531specific options.
d60e92d1
AC
1532.TP
1533.B netsplice
523bad63
TK
1534Like \fBnet\fR, but uses \fBsplice\fR\|(2) and
1535\fBvmsplice\fR\|(2) to map data and send/receive.
1536This engine defines engine specific options.
d60e92d1 1537.TP
53aec0a4 1538.B cpuio
523bad63
TK
1539Doesn't transfer any data, but burns CPU cycles according to the
1540\fBcpuload\fR and \fBcpuchunks\fR options. Setting
1541\fBcpuload\fR\=85 will cause that job to do nothing but burn 85%
1542of the CPU. In case of SMP machines, use `numjobs=<nr_of_cpu>'
1543to get desired CPU usage, as the cpuload only loads a
1544single CPU at the desired rate. A job never finishes unless there is
1545at least one non\-cpuio job.
d60e92d1
AC
1546.TP
1547.B guasi
523bad63
TK
1548The GUASI I/O engine is the Generic Userspace Asyncronous Syscall
1549Interface approach to async I/O. See \fIhttp://www.xmailserver.org/guasi\-lib.html\fR
1550for more info on GUASI.
d60e92d1 1551.TP
21b8aee8 1552.B rdma
523bad63
TK
1553The RDMA I/O engine supports both RDMA memory semantics
1554(RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
609ac152
SB
1555InfiniBand, RoCE and iWARP protocols. This engine defines engine
1556specific options.
d54fce84
DM
1557.TP
1558.B falloc
523bad63
TK
1559I/O engine that does regular fallocate to simulate data transfer as
1560fio ioengine.
1561.RS
1562.P
1563.PD 0
1564DDIR_READ does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1565.P
1566DIR_WRITE does fallocate(,mode = 0).
1567.P
1568DDIR_TRIM does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1569.PD
1570.RE
1571.TP
1572.B ftruncate
1573I/O engine that sends \fBftruncate\fR\|(2) operations in response
1574to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
1575size to the current block offset. \fBblocksize\fR is ignored.
d54fce84
DM
1576.TP
1577.B e4defrag
523bad63
TK
1578I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1579defragment activity in request to DDIR_WRITE event.
0d978694
DAG
1580.TP
1581.B rbd
523bad63
TK
1582I/O engine supporting direct access to Ceph Rados Block Devices
1583(RBD) via librbd without the need to use the kernel rbd driver. This
1584ioengine defines engine specific options.
a7c386f4 1585.TP
1586.B gfapi
523bad63
TK
1587Using GlusterFS libgfapi sync interface to direct access to
1588GlusterFS volumes without having to go through FUSE. This ioengine
1589defines engine specific options.
cc47f094 1590.TP
1591.B gfapi_async
523bad63
TK
1592Using GlusterFS libgfapi async interface to direct access to
1593GlusterFS volumes without having to go through FUSE. This ioengine
1594defines engine specific options.
1b10477b 1595.TP
b74e419e 1596.B libhdfs
523bad63
TK
1597Read and write through Hadoop (HDFS). The \fBfilename\fR option
1598is used to specify host,port of the hdfs name\-node to connect. This
1599engine interprets offsets a little differently. In HDFS, files once
1600created cannot be modified so random writes are not possible. To
1601imitate this the libhdfs engine expects a bunch of small files to be
1602created over HDFS and will randomly pick a file from them
1603based on the offset generated by fio backend (see the example
1604job file to create such files, use `rw=write' option). Please
1605note, it may be necessary to set environment variables to work
1606with HDFS/libhdfs properly. Each job uses its own connection to
1607HDFS.
65fa28ca
DE
1608.TP
1609.B mtd
523bad63
TK
1610Read, write and erase an MTD character device (e.g.,
1611`/dev/mtd0'). Discards are treated as erases. Depending on the
1612underlying device type, the I/O may have to go in a certain pattern,
1613e.g., on NAND, writing sequentially to erase blocks and discarding
1614before overwriting. The \fBtrimwrite\fR mode works well for this
65fa28ca 1615constraint.
5c4ef02e
JA
1616.TP
1617.B pmemblk
523bad63
TK
1618Read and write using filesystem DAX to a file on a filesystem
1619mounted with DAX on a persistent memory device through the NVML
1620libpmemblk library.
104ee4de 1621.TP
523bad63
TK
1622.B dev\-dax
1623Read and write using device DAX to a persistent memory device (e.g.,
1624/dev/dax0.0) through the NVML libpmem library.
d60e92d1 1625.TP
523bad63
TK
1626.B external
1627Prefix to specify loading an external I/O engine object file. Append
1628the engine filename, e.g. `ioengine=external:/tmp/foo.o' to load
d243fd6d
TK
1629ioengine `foo.o' in `/tmp'. The path can be either
1630absolute or relative. See `engines/skeleton_external.c' in the fio source for
1631details of writing an external I/O engine.
1216cc5a
JB
1632.TP
1633.B filecreate
b71968b1
SW
1634Simply create the files and do no I/O to them. You still need to set
1635\fBfilesize\fR so that all the accounting still occurs, but no actual I/O will be
1636done other than creating the file.
ae0db592
TI
1637.TP
1638.B libpmem
1639Read and write using mmap I/O to a file on a filesystem
1640mounted with DAX on a persistent memory device through the NVML
1641libpmem library.
523bad63
TK
1642.SS "I/O engine specific parameters"
1643In addition, there are some parameters which are only valid when a specific
1644\fBioengine\fR is in use. These are used identically to normal parameters,
1645with the caveat that when used on the command line, they must come after the
1646\fBioengine\fR that defines them is selected.
d60e92d1 1647.TP
523bad63
TK
1648.BI (libaio)userspace_reap
1649Normally, with the libaio engine in use, fio will use the
1650\fBio_getevents\fR\|(3) system call to reap newly returned events. With
1651this flag turned on, the AIO ring will be read directly from user\-space to
1652reap events. The reaping mode is only enabled when polling for a minimum of
16530 events (e.g. when `iodepth_batch_complete=0').
3ce9dcaf 1654.TP
523bad63
TK
1655.BI (pvsync2)hipri
1656Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
1657than normal.
82407585 1658.TP
523bad63
TK
1659.BI (pvsync2)hipri_percentage
1660When hipri is set this determines the probability of a pvsync2 I/O being high
1661priority. The default is 100%.
d60e92d1 1662.TP
523bad63
TK
1663.BI (cpuio)cpuload \fR=\fPint
1664Attempt to use the specified percentage of CPU cycles. This is a mandatory
1665option when using cpuio I/O engine.
997b5680 1666.TP
523bad63
TK
1667.BI (cpuio)cpuchunks \fR=\fPint
1668Split the load into cycles of the given time. In microseconds.
1ad01bd1 1669.TP
523bad63
TK
1670.BI (cpuio)exit_on_io_done \fR=\fPbool
1671Detect when I/O threads are done, then exit.
d60e92d1 1672.TP
523bad63
TK
1673.BI (libhdfs)namenode \fR=\fPstr
1674The hostname or IP address of a HDFS cluster namenode to contact.
d01612f3 1675.TP
523bad63
TK
1676.BI (libhdfs)port
1677The listening port of the HFDS cluster namenode.
d60e92d1 1678.TP
523bad63
TK
1679.BI (netsplice,net)port
1680The TCP or UDP port to bind to or connect to. If this is used with
1681\fBnumjobs\fR to spawn multiple instances of the same job type, then
1682this will be the starting port number since fio will use a range of
1683ports.
d60e92d1 1684.TP
609ac152
SB
1685.BI (rdma)port
1686The port to use for RDMA-CM communication. This should be the same
1687value on the client and the server side.
1688.TP
1689.BI (netsplice,net, rdma)hostname \fR=\fPstr
1690The hostname or IP address to use for TCP, UDP or RDMA-CM based I/O.
1691If the job is a TCP listener or UDP reader, the hostname is not used
1692and must be omitted unless it is a valid UDP multicast address.
591e9e06 1693.TP
523bad63
TK
1694.BI (netsplice,net)interface \fR=\fPstr
1695The IP address of the network interface used to send or receive UDP
1696multicast.
ddf24e42 1697.TP
523bad63
TK
1698.BI (netsplice,net)ttl \fR=\fPint
1699Time\-to\-live value for outgoing UDP multicast packets. Default: 1.
d60e92d1 1700.TP
523bad63
TK
1701.BI (netsplice,net)nodelay \fR=\fPbool
1702Set TCP_NODELAY on TCP connections.
fa769d44 1703.TP
523bad63
TK
1704.BI (netsplice,net)protocol \fR=\fPstr "\fR,\fP proto" \fR=\fPstr
1705The network protocol to use. Accepted values are:
1706.RS
e76b1da4
JA
1707.RS
1708.TP
523bad63
TK
1709.B tcp
1710Transmission control protocol.
e76b1da4 1711.TP
523bad63
TK
1712.B tcpv6
1713Transmission control protocol V6.
e76b1da4 1714.TP
523bad63
TK
1715.B udp
1716User datagram protocol.
1717.TP
1718.B udpv6
1719User datagram protocol V6.
e76b1da4 1720.TP
523bad63
TK
1721.B unix
1722UNIX domain socket.
e76b1da4
JA
1723.RE
1724.P
523bad63
TK
1725When the protocol is TCP or UDP, the port must also be given, as well as the
1726hostname if the job is a TCP listener or UDP reader. For unix sockets, the
1727normal \fBfilename\fR option should be used and the port is invalid.
1728.RE
1729.TP
1730.BI (netsplice,net)listen
1731For TCP network connections, tell fio to listen for incoming connections
1732rather than initiating an outgoing connection. The \fBhostname\fR must
1733be omitted if this option is used.
1734.TP
1735.BI (netsplice,net)pingpong
1736Normally a network writer will just continue writing data, and a network
1737reader will just consume packages. If `pingpong=1' is set, a writer will
1738send its normal payload to the reader, then wait for the reader to send the
1739same payload back. This allows fio to measure network latencies. The
1740submission and completion latencies then measure local time spent sending or
1741receiving, and the completion latency measures how long it took for the
1742other end to receive and send back. For UDP multicast traffic
1743`pingpong=1' should only be set for a single reader when multiple readers
1744are listening to the same address.
1745.TP
1746.BI (netsplice,net)window_size \fR=\fPint
1747Set the desired socket buffer size for the connection.
e76b1da4 1748.TP
523bad63
TK
1749.BI (netsplice,net)mss \fR=\fPint
1750Set the TCP maximum segment size (TCP_MAXSEG).
d60e92d1 1751.TP
523bad63
TK
1752.BI (e4defrag)donorname \fR=\fPstr
1753File will be used as a block donor (swap extents between files).
d60e92d1 1754.TP
523bad63
TK
1755.BI (e4defrag)inplace \fR=\fPint
1756Configure donor file blocks allocation strategy:
1757.RS
1758.RS
d60e92d1 1759.TP
523bad63
TK
1760.B 0
1761Default. Preallocate donor's file on init.
d60e92d1 1762.TP
523bad63
TK
1763.B 1
1764Allocate space immediately inside defragment event, and free right
1765after event.
1766.RE
1767.RE
d60e92d1 1768.TP
523bad63
TK
1769.BI (rbd)clustername \fR=\fPstr
1770Specifies the name of the Ceph cluster.
92d42d69 1771.TP
523bad63
TK
1772.BI (rbd)rbdname \fR=\fPstr
1773Specifies the name of the RBD.
92d42d69 1774.TP
523bad63
TK
1775.BI (rbd)pool \fR=\fPstr
1776Specifies the name of the Ceph pool containing RBD.
92d42d69 1777.TP
523bad63
TK
1778.BI (rbd)clientname \fR=\fPstr
1779Specifies the username (without the 'client.' prefix) used to access the
1780Ceph cluster. If the \fBclustername\fR is specified, the \fBclientname\fR shall be
1781the full *type.id* string. If no type. prefix is given, fio will add 'client.'
1782by default.
92d42d69 1783.TP
523bad63
TK
1784.BI (mtd)skip_bad \fR=\fPbool
1785Skip operations against known bad blocks.
8116fd24 1786.TP
523bad63
TK
1787.BI (libhdfs)hdfsdirectory
1788libhdfs will create chunk in this HDFS directory.
e0a04ac1 1789.TP
523bad63
TK
1790.BI (libhdfs)chunk_size
1791The size of the chunk to use for each file.
609ac152
SB
1792.TP
1793.BI (rdma)verb \fR=\fPstr
1794The RDMA verb to use on this side of the RDMA ioengine
1795connection. Valid values are write, read, send and recv. These
1796correspond to the equivalent RDMA verbs (e.g. write = rdma_write
1797etc.). Note that this only needs to be specified on the client side of
1798the connection. See the examples folder.
1799.TP
1800.BI (rdma)bindname \fR=\fPstr
1801The name to use to bind the local RDMA-CM connection to a local RDMA
1802device. This could be a hostname or an IPv4 or IPv6 address. On the
1803server side this will be passed into the rdma_bind_addr() function and
1804on the client site it will be used in the rdma_resolve_add()
1805function. This can be useful when multiple paths exist between the
1806client and the server or in certain loopback configurations.
523bad63
TK
1807.SS "I/O depth"
1808.TP
1809.BI iodepth \fR=\fPint
1810Number of I/O units to keep in flight against the file. Note that
1811increasing \fBiodepth\fR beyond 1 will not affect synchronous ioengines (except
1812for small degrees when \fBverify_async\fR is in use). Even async
1813engines may impose OS restrictions causing the desired depth not to be
1814achieved. This may happen on Linux when using libaio and not setting
1815`direct=1', since buffered I/O is not async on that OS. Keep an
1816eye on the I/O depth distribution in the fio output to verify that the
1817achieved depth is as expected. Default: 1.
1818.TP
1819.BI iodepth_batch_submit \fR=\fPint "\fR,\fP iodepth_batch" \fR=\fPint
1820This defines how many pieces of I/O to submit at once. It defaults to 1
1821which means that we submit each I/O as soon as it is available, but can be
1822raised to submit bigger batches of I/O at the time. If it is set to 0 the
1823\fBiodepth\fR value will be used.
1824.TP
1825.BI iodepth_batch_complete_min \fR=\fPint "\fR,\fP iodepth_batch_complete" \fR=\fPint
1826This defines how many pieces of I/O to retrieve at once. It defaults to 1
1827which means that we'll ask for a minimum of 1 I/O in the retrieval process
1828from the kernel. The I/O retrieval will go on until we hit the limit set by
1829\fBiodepth_low\fR. If this variable is set to 0, then fio will always
1830check for completed events before queuing more I/O. This helps reduce I/O
1831latency, at the cost of more retrieval system calls.
1832.TP
1833.BI iodepth_batch_complete_max \fR=\fPint
1834This defines maximum pieces of I/O to retrieve at once. This variable should
1835be used along with \fBiodepth_batch_complete_min\fR=\fIint\fR variable,
1836specifying the range of min and max amount of I/O which should be
1837retrieved. By default it is equal to \fBiodepth_batch_complete_min\fR
1838value. Example #1:
e0a04ac1 1839.RS
e0a04ac1 1840.RS
e0a04ac1 1841.P
523bad63
TK
1842.PD 0
1843iodepth_batch_complete_min=1
e0a04ac1 1844.P
523bad63
TK
1845iodepth_batch_complete_max=<iodepth>
1846.PD
e0a04ac1
JA
1847.RE
1848.P
523bad63
TK
1849which means that we will retrieve at least 1 I/O and up to the whole
1850submitted queue depth. If none of I/O has been completed yet, we will wait.
1851Example #2:
e8b1961d 1852.RS
523bad63
TK
1853.P
1854.PD 0
1855iodepth_batch_complete_min=0
1856.P
1857iodepth_batch_complete_max=<iodepth>
1858.PD
e8b1961d
JA
1859.RE
1860.P
523bad63
TK
1861which means that we can retrieve up to the whole submitted queue depth, but
1862if none of I/O has been completed yet, we will NOT wait and immediately exit
1863the system call. In this example we simply do polling.
1864.RE
e8b1961d 1865.TP
523bad63
TK
1866.BI iodepth_low \fR=\fPint
1867The low water mark indicating when to start filling the queue
1868again. Defaults to the same as \fBiodepth\fR, meaning that fio will
1869attempt to keep the queue full at all times. If \fBiodepth\fR is set to
1870e.g. 16 and \fBiodepth_low\fR is set to 4, then after fio has filled the queue of
187116 requests, it will let the depth drain down to 4 before starting to fill
1872it again.
d60e92d1 1873.TP
523bad63
TK
1874.BI serialize_overlap \fR=\fPbool
1875Serialize in-flight I/Os that might otherwise cause or suffer from data races.
1876When two or more I/Os are submitted simultaneously, there is no guarantee that
1877the I/Os will be processed or completed in the submitted order. Further, if
1878two or more of those I/Os are writes, any overlapping region between them can
1879become indeterminate/undefined on certain storage. These issues can cause
1880verification to fail erratically when at least one of the racing I/Os is
1881changing data and the overlapping region has a non-zero size. Setting
1882\fBserialize_overlap\fR tells fio to avoid provoking this behavior by explicitly
1883serializing in-flight I/Os that have a non-zero overlap. Note that setting
1884this option can reduce both performance and the \fBiodepth\fR achieved.
1885Additionally this option does not work when \fBio_submit_mode\fR is set to
1886offload. Default: false.
d60e92d1 1887.TP
523bad63
TK
1888.BI io_submit_mode \fR=\fPstr
1889This option controls how fio submits the I/O to the I/O engine. The default
1890is `inline', which means that the fio job threads submit and reap I/O
1891directly. If set to `offload', the job threads will offload I/O submission
1892to a dedicated pool of I/O threads. This requires some coordination and thus
1893has a bit of extra overhead, especially for lower queue depth I/O where it
1894can increase latencies. The benefit is that fio can manage submission rates
1895independently of the device completion rates. This avoids skewed latency
1896reporting if I/O gets backed up on the device side (the coordinated omission
1897problem).
1898.SS "I/O rate"
d60e92d1 1899.TP
523bad63
TK
1900.BI thinktime \fR=\fPtime
1901Stall the job for the specified period of time after an I/O has completed before issuing the
1902next. May be used to simulate processing being done by an application.
1903When the unit is omitted, the value is interpreted in microseconds. See
1904\fBthinktime_blocks\fR and \fBthinktime_spin\fR.
d60e92d1 1905.TP
523bad63
TK
1906.BI thinktime_spin \fR=\fPtime
1907Only valid if \fBthinktime\fR is set \- pretend to spend CPU time doing
1908something with the data received, before falling back to sleeping for the
1909rest of the period specified by \fBthinktime\fR. When the unit is
1910omitted, the value is interpreted in microseconds.
d60e92d1
AC
1911.TP
1912.BI thinktime_blocks \fR=\fPint
523bad63
TK
1913Only valid if \fBthinktime\fR is set \- control how many blocks to issue,
1914before waiting \fBthinktime\fR usecs. If not set, defaults to 1 which will make
1915fio wait \fBthinktime\fR usecs after every block. This effectively makes any
1916queue depth setting redundant, since no more than 1 I/O will be queued
1917before we have to complete it and do our \fBthinktime\fR. In other words, this
1918setting effectively caps the queue depth if the latter is larger.
d60e92d1 1919.TP
6d500c2e 1920.BI rate \fR=\fPint[,int][,int]
523bad63
TK
1921Cap the bandwidth used by this job. The number is in bytes/sec, the normal
1922suffix rules apply. Comma\-separated values may be specified for reads,
1923writes, and trims as described in \fBblocksize\fR.
1924.RS
1925.P
1926For example, using `rate=1m,500k' would limit reads to 1MiB/sec and writes to
1927500KiB/sec. Capping only reads or writes can be done with `rate=,500k' or
1928`rate=500k,' where the former will only limit writes (to 500KiB/sec) and the
1929latter will only limit reads.
1930.RE
d60e92d1 1931.TP
6d500c2e 1932.BI rate_min \fR=\fPint[,int][,int]
523bad63
TK
1933Tell fio to do whatever it can to maintain at least this bandwidth. Failing
1934to meet this requirement will cause the job to exit. Comma\-separated values
1935may be specified for reads, writes, and trims as described in
1936\fBblocksize\fR.
d60e92d1 1937.TP
6d500c2e 1938.BI rate_iops \fR=\fPint[,int][,int]
523bad63
TK
1939Cap the bandwidth to this number of IOPS. Basically the same as
1940\fBrate\fR, just specified independently of bandwidth. If the job is
1941given a block size range instead of a fixed value, the smallest block size
1942is used as the metric. Comma\-separated values may be specified for reads,
1943writes, and trims as described in \fBblocksize\fR.
d60e92d1 1944.TP
6d500c2e 1945.BI rate_iops_min \fR=\fPint[,int][,int]
523bad63
TK
1946If fio doesn't meet this rate of I/O, it will cause the job to exit.
1947Comma\-separated values may be specified for reads, writes, and trims as
1948described in \fBblocksize\fR.
d60e92d1 1949.TP
6de65959 1950.BI rate_process \fR=\fPstr
523bad63
TK
1951This option controls how fio manages rated I/O submissions. The default is
1952`linear', which submits I/O in a linear fashion with fixed delays between
1953I/Os that gets adjusted based on I/O completion rates. If this is set to
1954`poisson', fio will submit I/O based on a more real world random request
6de65959 1955flow, known as the Poisson process
523bad63 1956(\fIhttps://en.wikipedia.org/wiki/Poisson_point_process\fR). The lambda will be
5d02b083 195710^6 / IOPS for the given workload.
523bad63 1958.SS "I/O latency"
ff6bb260 1959.TP
523bad63 1960.BI latency_target \fR=\fPtime
3e260a46 1961If set, fio will attempt to find the max performance point that the given
523bad63
TK
1962workload will run at while maintaining a latency below this target. When
1963the unit is omitted, the value is interpreted in microseconds. See
1964\fBlatency_window\fR and \fBlatency_percentile\fR.
3e260a46 1965.TP
523bad63 1966.BI latency_window \fR=\fPtime
3e260a46 1967Used with \fBlatency_target\fR to specify the sample window that the job
523bad63
TK
1968is run at varying queue depths to test the performance. When the unit is
1969omitted, the value is interpreted in microseconds.
3e260a46
JA
1970.TP
1971.BI latency_percentile \fR=\fPfloat
523bad63
TK
1972The percentage of I/Os that must fall within the criteria specified by
1973\fBlatency_target\fR and \fBlatency_window\fR. If not set, this
1974defaults to 100.0, meaning that all I/Os must be equal or below to the value
1975set by \fBlatency_target\fR.
1976.TP
1977.BI max_latency \fR=\fPtime
1978If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
1979maximum latency. When the unit is omitted, the value is interpreted in
1980microseconds.
1981.TP
1982.BI rate_cycle \fR=\fPint
1983Average bandwidth for \fBrate\fR and \fBrate_min\fR over this number
1984of milliseconds. Defaults to 1000.
1985.SS "I/O replay"
1986.TP
1987.BI write_iolog \fR=\fPstr
1988Write the issued I/O patterns to the specified file. See
1989\fBread_iolog\fR. Specify a separate file for each job, otherwise the
1990iologs will be interspersed and the file may be corrupt.
1991.TP
1992.BI read_iolog \fR=\fPstr
1993Open an iolog with the specified filename and replay the I/O patterns it
1994contains. This can be used to store a workload and replay it sometime
1995later. The iolog given may also be a blktrace binary file, which allows fio
1996to replay a workload captured by blktrace. See
1997\fBblktrace\fR\|(8) for how to capture such logging data. For blktrace
1998replay, the file needs to be turned into a blkparse binary data file first
1999(`blkparse <device> \-o /dev/null \-d file_for_fio.bin').
3e260a46 2000.TP
523bad63
TK
2001.BI replay_no_stall \fR=\fPbool
2002When replaying I/O with \fBread_iolog\fR the default behavior is to
2003attempt to respect the timestamps within the log and replay them with the
2004appropriate delay between IOPS. By setting this variable fio will not
2005respect the timestamps and attempt to replay them as fast as possible while
2006still respecting ordering. The result is the same I/O pattern to a given
2007device, but different timings.
2008.TP
2009.BI replay_redirect \fR=\fPstr
2010While replaying I/O patterns using \fBread_iolog\fR the default behavior
2011is to replay the IOPS onto the major/minor device that each IOP was recorded
2012from. This is sometimes undesirable because on a different machine those
2013major/minor numbers can map to a different device. Changing hardware on the
2014same system can also result in a different major/minor mapping.
2015\fBreplay_redirect\fR causes all I/Os to be replayed onto the single specified
2016device regardless of the device it was recorded
2017from. i.e. `replay_redirect=/dev/sdc' would cause all I/O
2018in the blktrace or iolog to be replayed onto `/dev/sdc'. This means
2019multiple devices will be replayed onto a single device, if the trace
2020contains multiple devices. If you want multiple devices to be replayed
2021concurrently to multiple redirected devices you must blkparse your trace
2022into separate traces and replay them with independent fio invocations.
2023Unfortunately this also breaks the strict time ordering between multiple
2024device accesses.
2025.TP
2026.BI replay_align \fR=\fPint
2027Force alignment of I/O offsets and lengths in a trace to this power of 2
2028value.
2029.TP
2030.BI replay_scale \fR=\fPint
2031Scale sector offsets down by this factor when replaying traces.
2032.SS "Threads, processes and job synchronization"
2033.TP
2034.BI thread
2035Fio defaults to creating jobs by using fork, however if this option is
2036given, fio will create jobs by using POSIX Threads' function
2037\fBpthread_create\fR\|(3) to create threads instead.
2038.TP
2039.BI wait_for \fR=\fPstr
2040If set, the current job won't be started until all workers of the specified
2041waitee job are done.
2042.\" ignore blank line here from HOWTO as it looks normal without it
2043\fBwait_for\fR operates on the job name basis, so there are a few
2044limitations. First, the waitee must be defined prior to the waiter job
2045(meaning no forward references). Second, if a job is being referenced as a
2046waitee, it must have a unique name (no duplicate waitees).
2047.TP
2048.BI nice \fR=\fPint
2049Run the job with the given nice value. See man \fBnice\fR\|(2).
2050.\" ignore blank line here from HOWTO as it looks normal without it
2051On Windows, values less than \-15 set the process class to "High"; \-1 through
2052\-15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2053priority class.
2054.TP
2055.BI prio \fR=\fPint
2056Set the I/O priority value of this job. Linux limits us to a positive value
2057between 0 and 7, with 0 being the highest. See man
2058\fBionice\fR\|(1). Refer to an appropriate manpage for other operating
2059systems since meaning of priority may differ.
2060.TP
2061.BI prioclass \fR=\fPint
2062Set the I/O priority class. See man \fBionice\fR\|(1).
15501535 2063.TP
d60e92d1 2064.BI cpumask \fR=\fPint
523bad63
TK
2065Set the CPU affinity of this job. The parameter given is a bit mask of
2066allowed CPUs the job may run on. So if you want the allowed CPUs to be 1
2067and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2068\fBsched_setaffinity\fR\|(2). This may not work on all supported
2069operating systems or kernel versions. This option doesn't work well for a
2070higher CPU count than what you can store in an integer mask, so it can only
2071control cpus 1\-32. For boxes with larger CPU counts, use
2072\fBcpus_allowed\fR.
d60e92d1
AC
2073.TP
2074.BI cpus_allowed \fR=\fPstr
523bad63
TK
2075Controls the same options as \fBcpumask\fR, but accepts a textual
2076specification of the permitted CPUs instead. So to use CPUs 1 and 5 you
2077would specify `cpus_allowed=1,5'. This option also allows a range of CPUs
2078to be specified \-\- say you wanted a binding to CPUs 1, 5, and 8 to 15, you
2079would set `cpus_allowed=1,5,8\-15'.
d60e92d1 2080.TP
c2acfbac 2081.BI cpus_allowed_policy \fR=\fPstr
523bad63
TK
2082Set the policy of how fio distributes the CPUs specified by
2083\fBcpus_allowed\fR or \fBcpumask\fR. Two policies are supported:
c2acfbac
JA
2084.RS
2085.RS
2086.TP
2087.B shared
2088All jobs will share the CPU set specified.
2089.TP
2090.B split
2091Each job will get a unique CPU from the CPU set.
2092.RE
2093.P
523bad63
TK
2094\fBshared\fR is the default behavior, if the option isn't specified. If
2095\fBsplit\fR is specified, then fio will will assign one cpu per job. If not
2096enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2097in the set.
c2acfbac 2098.RE
c2acfbac 2099.TP
d0b937ed 2100.BI numa_cpu_nodes \fR=\fPstr
cecbfd47 2101Set this job running on specified NUMA nodes' CPUs. The arguments allow
523bad63
TK
2102comma delimited list of cpu numbers, A\-B ranges, or `all'. Note, to enable
2103NUMA options support, fio must be built on a system with libnuma\-dev(el)
2104installed.
d0b937ed
YR
2105.TP
2106.BI numa_mem_policy \fR=\fPstr
523bad63
TK
2107Set this job's memory policy and corresponding NUMA nodes. Format of the
2108arguments:
39c7a2ca
VF
2109.RS
2110.RS
523bad63
TK
2111.P
2112<mode>[:<nodelist>]
39c7a2ca 2113.RE
523bad63
TK
2114.P
2115`mode' is one of the following memory poicies: `default', `prefer',
2116`bind', `interleave' or `local'. For `default' and `local' memory
2117policies, no node needs to be specified. For `prefer', only one node is
2118allowed. For `bind' and `interleave' the `nodelist' may be as
2119follows: a comma delimited list of numbers, A\-B ranges, or `all'.
39c7a2ca
VF
2120.RE
2121.TP
523bad63
TK
2122.BI cgroup \fR=\fPstr
2123Add job to this control group. If it doesn't exist, it will be created. The
2124system must have a mounted cgroup blkio mount point for this to work. If
2125your system doesn't have it mounted, you can do so with:
d60e92d1
AC
2126.RS
2127.RS
d60e92d1 2128.P
523bad63
TK
2129# mount \-t cgroup \-o blkio none /cgroup
2130.RE
d60e92d1
AC
2131.RE
2132.TP
523bad63
TK
2133.BI cgroup_weight \fR=\fPint
2134Set the weight of the cgroup to this value. See the documentation that comes
2135with the kernel, allowed values are in the range of 100..1000.
d60e92d1 2136.TP
523bad63
TK
2137.BI cgroup_nodelete \fR=\fPbool
2138Normally fio will delete the cgroups it has created after the job
2139completion. To override this behavior and to leave cgroups around after the
2140job completion, set `cgroup_nodelete=1'. This can be useful if one wants
2141to inspect various cgroup files after job completion. Default: false.
c8eeb9df 2142.TP
523bad63
TK
2143.BI flow_id \fR=\fPint
2144The ID of the flow. If not specified, it defaults to being a global
2145flow. See \fBflow\fR.
d60e92d1 2146.TP
523bad63
TK
2147.BI flow \fR=\fPint
2148Weight in token\-based flow control. If this value is used, then there is
2149a 'flow counter' which is used to regulate the proportion of activity between
2150two or more jobs. Fio attempts to keep this flow counter near zero. The
2151\fBflow\fR parameter stands for how much should be added or subtracted to the
2152flow counter on each iteration of the main I/O loop. That is, if one job has
2153`flow=8' and another job has `flow=\-1', then there will be a roughly 1:8
2154ratio in how much one runs vs the other.
d60e92d1 2155.TP
523bad63
TK
2156.BI flow_watermark \fR=\fPint
2157The maximum value that the absolute value of the flow counter is allowed to
2158reach before the job must wait for a lower value of the counter.
6b7f6851 2159.TP
523bad63
TK
2160.BI flow_sleep \fR=\fPint
2161The period of time, in microseconds, to wait after the flow watermark has
2162been exceeded before retrying operations.
25460cf6 2163.TP
523bad63
TK
2164.BI stonewall "\fR,\fB wait_for_previous"
2165Wait for preceding jobs in the job file to exit, before starting this
2166one. Can be used to insert serialization points in the job file. A stone
2167wall also implies starting a new reporting group, see
2168\fBgroup_reporting\fR.
2378826d 2169.TP
523bad63
TK
2170.BI exitall
2171By default, fio will continue running all other jobs when one job finishes
2172but sometimes this is not the desired action. Setting \fBexitall\fR will
2173instead make fio terminate all other jobs when one job finishes.
e81ecca3 2174.TP
523bad63
TK
2175.BI exec_prerun \fR=\fPstr
2176Before running this job, issue the command specified through
2177\fBsystem\fR\|(3). Output is redirected in a file called `jobname.prerun.txt'.
e9f48479 2178.TP
523bad63
TK
2179.BI exec_postrun \fR=\fPstr
2180After the job completes, issue the command specified though
2181\fBsystem\fR\|(3). Output is redirected in a file called `jobname.postrun.txt'.
d60e92d1 2182.TP
523bad63
TK
2183.BI uid \fR=\fPint
2184Instead of running as the invoking user, set the user ID to this value
2185before the thread/process does any work.
39c1c323 2186.TP
523bad63
TK
2187.BI gid \fR=\fPint
2188Set group ID, see \fBuid\fR.
2189.SS "Verification"
d60e92d1 2190.TP
589e88b7 2191.BI verify_only
523bad63 2192Do not perform specified workload, only verify data still matches previous
5e4c7118 2193invocation of this workload. This option allows one to check data multiple
523bad63
TK
2194times at a later date without overwriting it. This option makes sense only
2195for workloads that write data, and does not support workloads with the
5e4c7118
JA
2196\fBtime_based\fR option set.
2197.TP
d60e92d1 2198.BI do_verify \fR=\fPbool
523bad63
TK
2199Run the verify phase after a write phase. Only valid if \fBverify\fR is
2200set. Default: true.
d60e92d1
AC
2201.TP
2202.BI verify \fR=\fPstr
523bad63
TK
2203If writing to a file, fio can verify the file contents after each iteration
2204of the job. Each verification method also implies verification of special
2205header, which is written to the beginning of each block. This header also
2206includes meta information, like offset of the block, block number, timestamp
2207when block was written, etc. \fBverify\fR can be combined with
2208\fBverify_pattern\fR option. The allowed values are:
d60e92d1
AC
2209.RS
2210.RS
2211.TP
523bad63
TK
2212.B md5
2213Use an md5 sum of the data area and store it in the header of
2214each block.
2215.TP
2216.B crc64
2217Use an experimental crc64 sum of the data area and store it in the
2218header of each block.
2219.TP
2220.B crc32c
2221Use a crc32c sum of the data area and store it in the header of
2222each block. This will automatically use hardware acceleration
2223(e.g. SSE4.2 on an x86 or CRC crypto extensions on ARM64) but will
2224fall back to software crc32c if none is found. Generally the
2225fatest checksum fio supports when hardware accelerated.
2226.TP
2227.B crc32c\-intel
2228Synonym for crc32c.
2229.TP
2230.B crc32
2231Use a crc32 sum of the data area and store it in the header of each
2232block.
2233.TP
2234.B crc16
2235Use a crc16 sum of the data area and store it in the header of each
2236block.
2237.TP
2238.B crc7
2239Use a crc7 sum of the data area and store it in the header of each
2240block.
2241.TP
2242.B xxhash
2243Use xxhash as the checksum function. Generally the fastest software
2244checksum that fio supports.
2245.TP
2246.B sha512
2247Use sha512 as the checksum function.
2248.TP
2249.B sha256
2250Use sha256 as the checksum function.
2251.TP
2252.B sha1
2253Use optimized sha1 as the checksum function.
2254.TP
2255.B sha3\-224
2256Use optimized sha3\-224 as the checksum function.
2257.TP
2258.B sha3\-256
2259Use optimized sha3\-256 as the checksum function.
2260.TP
2261.B sha3\-384
2262Use optimized sha3\-384 as the checksum function.
2263.TP
2264.B sha3\-512
2265Use optimized sha3\-512 as the checksum function.
d60e92d1
AC
2266.TP
2267.B meta
523bad63
TK
2268This option is deprecated, since now meta information is included in
2269generic verification header and meta verification happens by
2270default. For detailed information see the description of the
2271\fBverify\fR setting. This option is kept because of
2272compatibility's sake with old configurations. Do not use it.
d60e92d1 2273.TP
59245381 2274.B pattern
523bad63
TK
2275Verify a strict pattern. Normally fio includes a header with some
2276basic information and checksumming, but if this option is set, only
2277the specific pattern set with \fBverify_pattern\fR is verified.
59245381 2278.TP
d60e92d1 2279.B null
523bad63
TK
2280Only pretend to verify. Useful for testing internals with
2281`ioengine=null', not for much else.
d60e92d1 2282.RE
523bad63
TK
2283.P
2284This option can be used for repeated burn\-in tests of a system to make sure
2285that the written data is also correctly read back. If the data direction
2286given is a read or random read, fio will assume that it should verify a
2287previously written file. If the data direction includes any form of write,
2288the verify will be of the newly written data.
d60e92d1
AC
2289.RE
2290.TP
5c9323fb 2291.BI verifysort \fR=\fPbool
523bad63
TK
2292If true, fio will sort written verify blocks when it deems it faster to read
2293them back in a sorted manner. This is often the case when overwriting an
2294existing file, since the blocks are already laid out in the file system. You
2295can ignore this option unless doing huge amounts of really fast I/O where
2296the red\-black tree sorting CPU time becomes significant. Default: true.
d60e92d1 2297.TP
fa769d44 2298.BI verifysort_nr \fR=\fPint
523bad63 2299Pre\-load and sort verify blocks for a read workload.
fa769d44 2300.TP
f7fa2653 2301.BI verify_offset \fR=\fPint
d60e92d1 2302Swap the verification header with data somewhere else in the block before
523bad63 2303writing. It is swapped back before verifying.
d60e92d1 2304.TP
f7fa2653 2305.BI verify_interval \fR=\fPint
523bad63
TK
2306Write the verification header at a finer granularity than the
2307\fBblocksize\fR. It will be written for chunks the size of
2308\fBverify_interval\fR. \fBblocksize\fR should divide this evenly.
d60e92d1 2309.TP
996093bb 2310.BI verify_pattern \fR=\fPstr
523bad63
TK
2311If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2312filling with totally random bytes, but sometimes it's interesting to fill
2313with a known pattern for I/O verification purposes. Depending on the width
2314of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time (it can
2315be either a decimal or a hex number). The \fBverify_pattern\fR if larger than
2316a 32\-bit quantity has to be a hex number that starts with either "0x" or
2317"0X". Use with \fBverify\fR. Also, \fBverify_pattern\fR supports %o
2318format, which means that for each block offset will be written and then
2319verified back, e.g.:
2fa5a241
RP
2320.RS
2321.RS
523bad63
TK
2322.P
2323verify_pattern=%o
2fa5a241 2324.RE
523bad63 2325.P
2fa5a241 2326Or use combination of everything:
2fa5a241 2327.RS
523bad63
TK
2328.P
2329verify_pattern=0xff%o"abcd"\-12
2fa5a241
RP
2330.RE
2331.RE
996093bb 2332.TP
d60e92d1 2333.BI verify_fatal \fR=\fPbool
523bad63
TK
2334Normally fio will keep checking the entire contents before quitting on a
2335block verification failure. If this option is set, fio will exit the job on
2336the first observed failure. Default: false.
d60e92d1 2337.TP
b463e936 2338.BI verify_dump \fR=\fPbool
523bad63
TK
2339If set, dump the contents of both the original data block and the data block
2340we read off disk to files. This allows later analysis to inspect just what
2341kind of data corruption occurred. Off by default.
b463e936 2342.TP
e8462bd8 2343.BI verify_async \fR=\fPint
523bad63
TK
2344Fio will normally verify I/O inline from the submitting thread. This option
2345takes an integer describing how many async offload threads to create for I/O
2346verification instead, causing fio to offload the duty of verifying I/O
2347contents to one or more separate threads. If using this offload option, even
2348sync I/O engines can benefit from using an \fBiodepth\fR setting higher
2349than 1, as it allows them to have I/O in flight while verifies are running.
2350Defaults to 0 async threads, i.e. verification is not asynchronous.
e8462bd8
JA
2351.TP
2352.BI verify_async_cpus \fR=\fPstr
523bad63
TK
2353Tell fio to set the given CPU affinity on the async I/O verification
2354threads. See \fBcpus_allowed\fR for the format used.
e8462bd8 2355.TP
6f87418f
JA
2356.BI verify_backlog \fR=\fPint
2357Fio will normally verify the written contents of a job that utilizes verify
2358once that job has completed. In other words, everything is written then
2359everything is read back and verified. You may want to verify continually
523bad63
TK
2360instead for a variety of reasons. Fio stores the meta data associated with
2361an I/O block in memory, so for large verify workloads, quite a bit of memory
2362would be used up holding this meta data. If this option is enabled, fio will
2363write only N blocks before verifying these blocks.
6f87418f
JA
2364.TP
2365.BI verify_backlog_batch \fR=\fPint
523bad63
TK
2366Control how many blocks fio will verify if \fBverify_backlog\fR is
2367set. If not set, will default to the value of \fBverify_backlog\fR
2368(meaning the entire queue is read back and verified). If
2369\fBverify_backlog_batch\fR is less than \fBverify_backlog\fR then not all
2370blocks will be verified, if \fBverify_backlog_batch\fR is larger than
2371\fBverify_backlog\fR, some blocks will be verified more than once.
2372.TP
2373.BI verify_state_save \fR=\fPbool
2374When a job exits during the write phase of a verify workload, save its
2375current state. This allows fio to replay up until that point, if the verify
2376state is loaded for the verify read phase. The format of the filename is,
2377roughly:
2378.RS
2379.RS
2380.P
2381<type>\-<jobname>\-<jobindex>\-verify.state.
2382.RE
2383.P
2384<type> is "local" for a local run, "sock" for a client/server socket
2385connection, and "ip" (192.168.0.1, for instance) for a networked
2386client/server connection. Defaults to true.
2387.RE
2388.TP
2389.BI verify_state_load \fR=\fPbool
2390If a verify termination trigger was used, fio stores the current write state
2391of each thread. This can be used at verification time so that fio knows how
2392far it should verify. Without this information, fio will run a full
2393verification pass, according to the settings in the job file used. Default
2394false.
6f87418f 2395.TP
fa769d44
SW
2396.BI trim_percentage \fR=\fPint
2397Number of verify blocks to discard/trim.
2398.TP
2399.BI trim_verify_zero \fR=\fPbool
523bad63 2400Verify that trim/discarded blocks are returned as zeros.
fa769d44
SW
2401.TP
2402.BI trim_backlog \fR=\fPint
523bad63 2403Verify that trim/discarded blocks are returned as zeros.
fa769d44
SW
2404.TP
2405.BI trim_backlog_batch \fR=\fPint
523bad63 2406Trim this number of I/O blocks.
fa769d44
SW
2407.TP
2408.BI experimental_verify \fR=\fPbool
2409Enable experimental verification.
523bad63 2410.SS "Steady state"
fa769d44 2411.TP
523bad63
TK
2412.BI steadystate \fR=\fPstr:float "\fR,\fP ss" \fR=\fPstr:float
2413Define the criterion and limit for assessing steady state performance. The
2414first parameter designates the criterion whereas the second parameter sets
2415the threshold. When the criterion falls below the threshold for the
2416specified duration, the job will stop. For example, `iops_slope:0.1%' will
2417direct fio to terminate the job when the least squares regression slope
2418falls below 0.1% of the mean IOPS. If \fBgroup_reporting\fR is enabled
2419this will apply to all jobs in the group. Below is the list of available
2420steady state assessment criteria. All assessments are carried out using only
2421data from the rolling collection window. Threshold limits can be expressed
2422as a fixed value or as a percentage of the mean in the collection window.
2423.RS
2424.RS
d60e92d1 2425.TP
523bad63
TK
2426.B iops
2427Collect IOPS data. Stop the job if all individual IOPS measurements
2428are within the specified limit of the mean IOPS (e.g., `iops:2'
2429means that all individual IOPS values must be within 2 of the mean,
2430whereas `iops:0.2%' means that all individual IOPS values must be
2431within 0.2% of the mean IOPS to terminate the job).
d60e92d1 2432.TP
523bad63
TK
2433.B iops_slope
2434Collect IOPS data and calculate the least squares regression
2435slope. Stop the job if the slope falls below the specified limit.
d60e92d1 2436.TP
523bad63
TK
2437.B bw
2438Collect bandwidth data. Stop the job if all individual bandwidth
2439measurements are within the specified limit of the mean bandwidth.
64bbb865 2440.TP
523bad63
TK
2441.B bw_slope
2442Collect bandwidth data and calculate the least squares regression
2443slope. Stop the job if the slope falls below the specified limit.
2444.RE
2445.RE
d1c46c04 2446.TP
523bad63
TK
2447.BI steadystate_duration \fR=\fPtime "\fR,\fP ss_dur" \fR=\fPtime
2448A rolling window of this duration will be used to judge whether steady state
2449has been reached. Data will be collected once per second. The default is 0
2450which disables steady state detection. When the unit is omitted, the
2451value is interpreted in seconds.
0c63576e 2452.TP
523bad63
TK
2453.BI steadystate_ramp_time \fR=\fPtime "\fR,\fP ss_ramp" \fR=\fPtime
2454Allow the job to run for the specified duration before beginning data
2455collection for checking the steady state job termination criterion. The
2456default is 0. When the unit is omitted, the value is interpreted in seconds.
2457.SS "Measurements and reporting"
0c63576e 2458.TP
3a5db920
JA
2459.BI per_job_logs \fR=\fPbool
2460If set, this generates bw/clat/iops log with per file private filenames. If
523bad63
TK
2461not set, jobs with identical names will share the log filename. Default:
2462true.
2463.TP
2464.BI group_reporting
2465It may sometimes be interesting to display statistics for groups of jobs as
2466a whole instead of for each individual job. This is especially true if
2467\fBnumjobs\fR is used; looking at individual thread/process output
2468quickly becomes unwieldy. To see the final report per\-group instead of
2469per\-job, use \fBgroup_reporting\fR. Jobs in a file will be part of the
2470same reporting group, unless if separated by a \fBstonewall\fR, or by
2471using \fBnew_group\fR.
2472.TP
2473.BI new_group
2474Start a new reporting group. See: \fBgroup_reporting\fR. If not given,
2475all jobs in a file will be part of the same reporting group, unless
2476separated by a \fBstonewall\fR.
2477.TP
2478.BI stats \fR=\fPbool
2479By default, fio collects and shows final output results for all jobs
2480that run. If this option is set to 0, then fio will ignore it in
2481the final stat output.
3a5db920 2482.TP
836bad52 2483.BI write_bw_log \fR=\fPstr
523bad63 2484If given, write a bandwidth log for this job. Can be used to store data of
074f0817 2485the bandwidth of the jobs in their lifetime.
523bad63 2486.RS
074f0817
SW
2487.P
2488If no str argument is given, the default filename of
2489`jobname_type.x.log' is used. Even when the argument is given, fio
2490will still append the type of log. So if one specifies:
523bad63
TK
2491.RS
2492.P
074f0817 2493write_bw_log=foo
523bad63
TK
2494.RE
2495.P
074f0817
SW
2496The actual log name will be `foo_bw.x.log' where `x' is the index
2497of the job (1..N, where N is the number of jobs). If
2498\fBper_job_logs\fR is false, then the filename will not include the
2499`.x` job index.
2500.P
2501The included \fBfio_generate_plots\fR script uses gnuplot to turn these
2502text files into nice graphs. See the \fBLOG FILE FORMATS\fR section for how data is
2503structured within the file.
523bad63 2504.RE
901bb994 2505.TP
074f0817
SW
2506.BI write_lat_log \fR=\fPstr
2507Same as \fBwrite_bw_log\fR, except this option creates I/O
2508submission (e.g., `name_slat.x.log'), completion (e.g.,
2509`name_clat.x.log'), and total (e.g., `name_lat.x.log') latency
2510files instead. See \fBwrite_bw_log\fR for details about the
2511filename format and the \fBLOG FILE FORMATS\fR section for how data is structured
2512within the files.
2513.TP
1e613c9c 2514.BI write_hist_log \fR=\fPstr
074f0817
SW
2515Same as \fBwrite_bw_log\fR but writes an I/O completion latency
2516histogram file (e.g., `name_hist.x.log') instead. Note that this
2517file will be empty unless \fBlog_hist_msec\fR has also been set.
2518See \fBwrite_bw_log\fR for details about the filename format and
2519the \fBLOG FILE FORMATS\fR section for how data is structured
2520within the file.
1e613c9c 2521.TP
c8eeb9df 2522.BI write_iops_log \fR=\fPstr
074f0817
SW
2523Same as \fBwrite_bw_log\fR, but writes an IOPS file (e.g.
2524`name_iops.x.log') instead. See \fBwrite_bw_log\fR for
2525details about the filename format and the \fBLOG FILE FORMATS\fR section for how data
2526is structured within the file.
c8eeb9df 2527.TP
b8bc8cba
JA
2528.BI log_avg_msec \fR=\fPint
2529By default, fio will log an entry in the iops, latency, or bw log for every
523bad63 2530I/O that completes. When writing to the disk log, that can quickly grow to a
b8bc8cba 2531very large size. Setting this option makes fio average the each log entry
e6989e10 2532over the specified period of time, reducing the resolution of the log. See
523bad63
TK
2533\fBlog_max_value\fR as well. Defaults to 0, logging all entries.
2534Also see \fBLOG FILE FORMATS\fR section.
b8bc8cba 2535.TP
1e613c9c 2536.BI log_hist_msec \fR=\fPint
523bad63
TK
2537Same as \fBlog_avg_msec\fR, but logs entries for completion latency
2538histograms. Computing latency percentiles from averages of intervals using
2539\fBlog_avg_msec\fR is inaccurate. Setting this option makes fio log
2540histogram entries over the specified period of time, reducing log sizes for
2541high IOPS devices while retaining percentile accuracy. See
074f0817
SW
2542\fBlog_hist_coarseness\fR and \fBwrite_hist_log\fR as well.
2543Defaults to 0, meaning histogram logging is disabled.
1e613c9c
KC
2544.TP
2545.BI log_hist_coarseness \fR=\fPint
523bad63
TK
2546Integer ranging from 0 to 6, defining the coarseness of the resolution of
2547the histogram logs enabled with \fBlog_hist_msec\fR. For each increment
2548in coarseness, fio outputs half as many bins. Defaults to 0, for which
2549histogram logs contain 1216 latency bins. See \fBLOG FILE FORMATS\fR section.
2550.TP
2551.BI log_max_value \fR=\fPbool
2552If \fBlog_avg_msec\fR is set, fio logs the average over that window. If
2553you instead want to log the maximum value, set this option to 1. Defaults to
25540, meaning that averaged values are logged.
1e613c9c 2555.TP
ae588852 2556.BI log_offset \fR=\fPbool
523bad63
TK
2557If this is set, the iolog options will include the byte offset for the I/O
2558entry as well as the other data values. Defaults to 0 meaning that
2559offsets are not present in logs. Also see \fBLOG FILE FORMATS\fR section.
ae588852 2560.TP
aee2ab67 2561.BI log_compression \fR=\fPint
523bad63
TK
2562If this is set, fio will compress the I/O logs as it goes, to keep the
2563memory footprint lower. When a log reaches the specified size, that chunk is
2564removed and compressed in the background. Given that I/O logs are fairly
2565highly compressible, this yields a nice memory savings for longer runs. The
2566downside is that the compression will consume some background CPU cycles, so
2567it may impact the run. This, however, is also true if the logging ends up
2568consuming most of the system memory. So pick your poison. The I/O logs are
2569saved normally at the end of a run, by decompressing the chunks and storing
2570them in the specified log file. This feature depends on the availability of
2571zlib.
aee2ab67 2572.TP
c08f9fe2 2573.BI log_compression_cpus \fR=\fPstr
523bad63
TK
2574Define the set of CPUs that are allowed to handle online log compression for
2575the I/O jobs. This can provide better isolation between performance
c08f9fe2
JA
2576sensitive jobs, and background compression work.
2577.TP
b26317c9 2578.BI log_store_compressed \fR=\fPbool
c08f9fe2 2579If set, fio will store the log files in a compressed format. They can be
523bad63
TK
2580decompressed with fio, using the \fB\-\-inflate\-log\fR command line
2581parameter. The files will be stored with a `.fz' suffix.
b26317c9 2582.TP
3aea75b1
KC
2583.BI log_unix_epoch \fR=\fPbool
2584If set, fio will log Unix timestamps to the log files produced by enabling
523bad63 2585write_type_log for each log type, instead of the default zero\-based
3aea75b1
KC
2586timestamps.
2587.TP
66347cfa 2588.BI block_error_percentiles \fR=\fPbool
523bad63
TK
2589If set, record errors in trim block\-sized units from writes and trims and
2590output a histogram of how many trims it took to get to errors, and what kind
2591of error was encountered.
d60e92d1 2592.TP
523bad63
TK
2593.BI bwavgtime \fR=\fPint
2594Average the calculated bandwidth over the given time. Value is specified in
2595milliseconds. If the job also does bandwidth logging through
2596\fBwrite_bw_log\fR, then the minimum of this option and
2597\fBlog_avg_msec\fR will be used. Default: 500ms.
d60e92d1 2598.TP
523bad63
TK
2599.BI iopsavgtime \fR=\fPint
2600Average the calculated IOPS over the given time. Value is specified in
2601milliseconds. If the job also does IOPS logging through
2602\fBwrite_iops_log\fR, then the minimum of this option and
2603\fBlog_avg_msec\fR will be used. Default: 500ms.
d60e92d1 2604.TP
d60e92d1 2605.BI disk_util \fR=\fPbool
523bad63
TK
2606Generate disk utilization statistics, if the platform supports it.
2607Default: true.
fa769d44 2608.TP
523bad63
TK
2609.BI disable_lat \fR=\fPbool
2610Disable measurements of total latency numbers. Useful only for cutting back
2611the number of calls to \fBgettimeofday\fR\|(2), as that does impact
2612performance at really high IOPS rates. Note that to really get rid of a
2613large amount of these calls, this option must be used with
2614\fBdisable_slat\fR and \fBdisable_bw_measurement\fR as well.
9e684a49 2615.TP
523bad63
TK
2616.BI disable_clat \fR=\fPbool
2617Disable measurements of completion latency numbers. See
2618\fBdisable_lat\fR.
9e684a49 2619.TP
523bad63
TK
2620.BI disable_slat \fR=\fPbool
2621Disable measurements of submission latency numbers. See
2622\fBdisable_lat\fR.
9e684a49 2623.TP
523bad63
TK
2624.BI disable_bw_measurement \fR=\fPbool "\fR,\fP disable_bw" \fR=\fPbool
2625Disable measurements of throughput/bandwidth numbers. See
2626\fBdisable_lat\fR.
9e684a49 2627.TP
83349190 2628.BI clat_percentiles \fR=\fPbool
b599759b
JA
2629Enable the reporting of percentiles of completion latencies. This option is
2630mutually exclusive with \fBlat_percentiles\fR.
2631.TP
2632.BI lat_percentiles \fR=\fPbool
b71968b1 2633Enable the reporting of percentiles of I/O latencies. This is similar to
b599759b
JA
2634\fBclat_percentiles\fR, except that this includes the submission latency.
2635This option is mutually exclusive with \fBclat_percentiles\fR.
83349190
YH
2636.TP
2637.BI percentile_list \fR=\fPfloat_list
66347cfa 2638Overwrite the default list of percentiles for completion latencies and the
523bad63
TK
2639block error histogram. Each number is a floating number in the range
2640(0,100], and the maximum length of the list is 20. Use ':' to separate the
2641numbers, and list the numbers in ascending order. For example,
2642`\-\-percentile_list=99.5:99.9' will cause fio to report the values of
2643completion latency below which 99.5% and 99.9% of the observed latencies
2644fell, respectively.
e883cb35
JF
2645.TP
2646.BI significant_figures \fR=\fPint
c32ba107
JA
2647If using \fB\-\-output\-format\fR of `normal', set the significant figures
2648to this value. Higher values will yield more precise IOPS and throughput
2649units, while lower values will round. Requires a minimum value of 1 and a
e883cb35 2650maximum value of 10. Defaults to 4.
523bad63 2651.SS "Error handling"
e4585935 2652.TP
523bad63
TK
2653.BI exitall_on_error
2654When one job finishes in error, terminate the rest. The default is to wait
2655for each job to finish.
e4585935 2656.TP
523bad63
TK
2657.BI continue_on_error \fR=\fPstr
2658Normally fio will exit the job on the first observed failure. If this option
2659is set, fio will continue the job when there is a 'non\-fatal error' (EIO or
2660EILSEQ) until the runtime is exceeded or the I/O size specified is
2661completed. If this option is used, there are two more stats that are
2662appended, the total error count and the first error. The error field given
2663in the stats is the first error that was hit during the run.
2664The allowed values are:
2665.RS
2666.RS
046395d7 2667.TP
523bad63
TK
2668.B none
2669Exit on any I/O or verify errors.
de890a1e 2670.TP
523bad63
TK
2671.B read
2672Continue on read errors, exit on all others.
2cafffbe 2673.TP
523bad63
TK
2674.B write
2675Continue on write errors, exit on all others.
a0679ce5 2676.TP
523bad63
TK
2677.B io
2678Continue on any I/O error, exit on all others.
de890a1e 2679.TP
523bad63
TK
2680.B verify
2681Continue on verify errors, exit on all others.
de890a1e 2682.TP
523bad63
TK
2683.B all
2684Continue on all errors.
b93b6a2e 2685.TP
523bad63
TK
2686.B 0
2687Backward\-compatible alias for 'none'.
d3a623de 2688.TP
523bad63
TK
2689.B 1
2690Backward\-compatible alias for 'all'.
2691.RE
2692.RE
1d360ffb 2693.TP
523bad63
TK
2694.BI ignore_error \fR=\fPstr
2695Sometimes you want to ignore some errors during test in that case you can
2696specify error list for each error type, instead of only being able to
2697ignore the default 'non\-fatal error' using \fBcontinue_on_error\fR.
2698`ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST' errors for
2699given error type is separated with ':'. Error may be symbol ('ENOSPC', 'ENOMEM')
2700or integer. Example:
de890a1e
SL
2701.RS
2702.RS
523bad63
TK
2703.P
2704ignore_error=EAGAIN,ENOSPC:122
2705.RE
2706.P
2707This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
2708WRITE. This option works by overriding \fBcontinue_on_error\fR with
2709the list of errors for each error type if any.
2710.RE
de890a1e 2711.TP
523bad63
TK
2712.BI error_dump \fR=\fPbool
2713If set dump every error even if it is non fatal, true by default. If
2714disabled only fatal error will be dumped.
2715.SS "Running predefined workloads"
2716Fio includes predefined profiles that mimic the I/O workloads generated by
2717other tools.
49ccb8c1 2718.TP
523bad63
TK
2719.BI profile \fR=\fPstr
2720The predefined workload to run. Current profiles are:
2721.RS
2722.RS
de890a1e 2723.TP
523bad63
TK
2724.B tiobench
2725Threaded I/O bench (tiotest/tiobench) like workload.
49ccb8c1 2726.TP
523bad63
TK
2727.B act
2728Aerospike Certification Tool (ACT) like workload.
2729.RE
de890a1e
SL
2730.RE
2731.P
523bad63
TK
2732To view a profile's additional options use \fB\-\-cmdhelp\fR after specifying
2733the profile. For example:
2734.RS
2735.TP
2736$ fio \-\-profile=act \-\-cmdhelp
de890a1e 2737.RE
523bad63 2738.SS "Act profile options"
de890a1e 2739.TP
523bad63
TK
2740.BI device\-names \fR=\fPstr
2741Devices to use.
d54fce84 2742.TP
523bad63
TK
2743.BI load \fR=\fPint
2744ACT load multiplier. Default: 1.
7aeb1e94 2745.TP
523bad63
TK
2746.BI test\-duration\fR=\fPtime
2747How long the entire test takes to run. When the unit is omitted, the value
2748is given in seconds. Default: 24h.
1008602c 2749.TP
523bad63
TK
2750.BI threads\-per\-queue\fR=\fPint
2751Number of read I/O threads per device. Default: 8.
e5f34d95 2752.TP
523bad63
TK
2753.BI read\-req\-num\-512\-blocks\fR=\fPint
2754Number of 512B blocks to read at the time. Default: 3.
d54fce84 2755.TP
523bad63
TK
2756.BI large\-block\-op\-kbytes\fR=\fPint
2757Size of large block ops in KiB (writes). Default: 131072.
d54fce84 2758.TP
523bad63
TK
2759.BI prep
2760Set to run ACT prep phase.
2761.SS "Tiobench profile options"
6d500c2e 2762.TP
523bad63
TK
2763.BI size\fR=\fPstr
2764Size in MiB.
0d978694 2765.TP
523bad63
TK
2766.BI block\fR=\fPint
2767Block size in bytes. Default: 4096.
0d978694 2768.TP
523bad63
TK
2769.BI numruns\fR=\fPint
2770Number of runs.
0d978694 2771.TP
523bad63
TK
2772.BI dir\fR=\fPstr
2773Test directory.
65fa28ca 2774.TP
523bad63
TK
2775.BI threads\fR=\fPint
2776Number of threads.
d60e92d1 2777.SH OUTPUT
40943b9a
TK
2778Fio spits out a lot of output. While running, fio will display the status of the
2779jobs created. An example of that would be:
d60e92d1 2780.P
40943b9a
TK
2781.nf
2782 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]
2783.fi
d1429b5c 2784.P
40943b9a
TK
2785The characters inside the first set of square brackets denote the current status of
2786each thread. The first character is the first job defined in the job file, and so
2787forth. The possible values (in typical life cycle order) are:
d60e92d1
AC
2788.RS
2789.TP
40943b9a 2790.PD 0
d60e92d1 2791.B P
40943b9a 2792Thread setup, but not started.
d60e92d1
AC
2793.TP
2794.B C
2795Thread created.
2796.TP
2797.B I
40943b9a
TK
2798Thread initialized, waiting or generating necessary data.
2799.TP
522c29f6 2800.B p
40943b9a
TK
2801Thread running pre\-reading file(s).
2802.TP
2803.B /
2804Thread is in ramp period.
d60e92d1
AC
2805.TP
2806.B R
2807Running, doing sequential reads.
2808.TP
2809.B r
2810Running, doing random reads.
2811.TP
2812.B W
2813Running, doing sequential writes.
2814.TP
2815.B w
2816Running, doing random writes.
2817.TP
2818.B M
2819Running, doing mixed sequential reads/writes.
2820.TP
2821.B m
2822Running, doing mixed random reads/writes.
2823.TP
40943b9a
TK
2824.B D
2825Running, doing sequential trims.
2826.TP
2827.B d
2828Running, doing random trims.
2829.TP
d60e92d1
AC
2830.B F
2831Running, currently waiting for \fBfsync\fR\|(2).
2832.TP
2833.B V
40943b9a
TK
2834Running, doing verification of written data.
2835.TP
2836.B f
2837Thread finishing.
d60e92d1
AC
2838.TP
2839.B E
40943b9a 2840Thread exited, not reaped by main thread yet.
d60e92d1
AC
2841.TP
2842.B \-
40943b9a
TK
2843Thread reaped.
2844.TP
2845.B X
2846Thread reaped, exited with an error.
2847.TP
2848.B K
2849Thread reaped, exited due to signal.
d1429b5c 2850.PD
40943b9a
TK
2851.RE
2852.P
2853Fio will condense the thread string as not to take up more space on the command
2854line than needed. For instance, if you have 10 readers and 10 writers running,
2855the output would look like this:
2856.P
2857.nf
2858 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]
2859.fi
d60e92d1 2860.P
40943b9a
TK
2861Note that the status string is displayed in order, so it's possible to tell which of
2862the jobs are currently doing what. In the example above this means that jobs 1\-\-10
2863are readers and 11\-\-20 are writers.
d60e92d1 2864.P
40943b9a
TK
2865The other values are fairly self explanatory \-\- number of threads currently
2866running and doing I/O, the number of currently open files (f=), the estimated
2867completion percentage, the rate of I/O since last check (read speed listed first,
2868then write speed and optionally trim speed) in terms of bandwidth and IOPS,
2869and time to completion for the current running group. It's impossible to estimate
2870runtime of the following groups (if any).
d60e92d1 2871.P
40943b9a
TK
2872When fio is done (or interrupted by Ctrl\-C), it will show the data for
2873each thread, group of threads, and disks in that order. For each overall thread (or
2874group) the output looks like:
2875.P
2876.nf
2877 Client1: (groupid=0, jobs=1): err= 0: pid=16109: Sat Jun 24 12:07:54 2017
2878 write: IOPS=88, BW=623KiB/s (638kB/s)(30.4MiB/50032msec)
2879 slat (nsec): min=500, max=145500, avg=8318.00, stdev=4781.50
2880 clat (usec): min=170, max=78367, avg=4019.02, stdev=8293.31
2881 lat (usec): min=174, max=78375, avg=4027.34, stdev=8291.79
2882 clat percentiles (usec):
2883 | 1.00th=[ 302], 5.00th=[ 326], 10.00th=[ 343], 20.00th=[ 363],
2884 | 30.00th=[ 392], 40.00th=[ 404], 50.00th=[ 416], 60.00th=[ 445],
2885 | 70.00th=[ 816], 80.00th=[ 6718], 90.00th=[12911], 95.00th=[21627],
2886 | 99.00th=[43779], 99.50th=[51643], 99.90th=[68682], 99.95th=[72877],
2887 | 99.99th=[78119]
2888 bw ( KiB/s): min= 532, max= 686, per=0.10%, avg=622.87, stdev=24.82, samples= 100
2889 iops : min= 76, max= 98, avg=88.98, stdev= 3.54, samples= 100
d3b9694d
VF
2890 lat (usec) : 250=0.04%, 500=64.11%, 750=4.81%, 1000=2.79%
2891 lat (msec) : 2=4.16%, 4=1.84%, 10=4.90%, 20=11.33%, 50=5.37%
2892 lat (msec) : 100=0.65%
40943b9a
TK
2893 cpu : usr=0.27%, sys=0.18%, ctx=12072, majf=0, minf=21
2894 IO depths : 1=85.0%, 2=13.1%, 4=1.8%, 8=0.1%, 16=0.0%, 32=0.0%, >=64=0.0%
2895 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2896 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2897 issued rwt: total=0,4450,0, short=0,0,0, dropped=0,0,0
2898 latency : target=0, window=0, percentile=100.00%, depth=8
2899.fi
2900.P
2901The job name (or first job's name when using \fBgroup_reporting\fR) is printed,
2902along with the group id, count of jobs being aggregated, last error id seen (which
2903is 0 when there are no errors), pid/tid of that thread and the time the job/group
2904completed. Below are the I/O statistics for each data direction performed (showing
2905writes in the example above). In the order listed, they denote:
d60e92d1 2906.RS
d60e92d1 2907.TP
40943b9a
TK
2908.B read/write/trim
2909The string before the colon shows the I/O direction the statistics
2910are for. \fIIOPS\fR is the average I/Os performed per second. \fIBW\fR
2911is the average bandwidth rate shown as: value in power of 2 format
2912(value in power of 10 format). The last two values show: (total
2913I/O performed in power of 2 format / \fIruntime\fR of that thread).
d60e92d1
AC
2914.TP
2915.B slat
40943b9a
TK
2916Submission latency (\fImin\fR being the minimum, \fImax\fR being the
2917maximum, \fIavg\fR being the average, \fIstdev\fR being the standard
2918deviation). This is the time it took to submit the I/O. For
2919sync I/O this row is not displayed as the slat is really the
2920completion latency (since queue/complete is one operation there).
2921This value can be in nanoseconds, microseconds or milliseconds \-\-\-
2922fio will choose the most appropriate base and print that (in the
2923example above nanoseconds was the best scale). Note: in \fB\-\-minimal\fR mode
2924latencies are always expressed in microseconds.
d60e92d1
AC
2925.TP
2926.B clat
40943b9a
TK
2927Completion latency. Same names as slat, this denotes the time from
2928submission to completion of the I/O pieces. For sync I/O, clat will
2929usually be equal (or very close) to 0, as the time from submit to
2930complete is basically just CPU time (I/O has already been done, see slat
2931explanation).
d60e92d1 2932.TP
d3b9694d
VF
2933.B lat
2934Total latency. Same names as slat and clat, this denotes the time from
2935when fio created the I/O unit to completion of the I/O operation.
2936.TP
d60e92d1 2937.B bw
40943b9a
TK
2938Bandwidth statistics based on samples. Same names as the xlat stats,
2939but also includes the number of samples taken (\fIsamples\fR) and an
2940approximate percentage of total aggregate bandwidth this thread
2941received in its group (\fIper\fR). This last value is only really
2942useful if the threads in this group are on the same disk, since they
2943are then competing for disk access.
2944.TP
2945.B iops
2946IOPS statistics based on samples. Same names as \fBbw\fR.
d60e92d1 2947.TP
d3b9694d
VF
2948.B lat (nsec/usec/msec)
2949The distribution of I/O completion latencies. This is the time from when
2950I/O leaves fio and when it gets completed. Unlike the separate
2951read/write/trim sections above, the data here and in the remaining
2952sections apply to all I/Os for the reporting group. 250=0.04% means that
29530.04% of the I/Os completed in under 250us. 500=64.11% means that 64.11%
2954of the I/Os required 250 to 499us for completion.
2955.TP
d60e92d1 2956.B cpu
40943b9a
TK
2957CPU usage. User and system time, along with the number of context
2958switches this thread went through, usage of system and user time, and
2959finally the number of major and minor page faults. The CPU utilization
2960numbers are averages for the jobs in that reporting group, while the
2961context and fault counters are summed.
d60e92d1
AC
2962.TP
2963.B IO depths
40943b9a
TK
2964The distribution of I/O depths over the job lifetime. The numbers are
2965divided into powers of 2 and each entry covers depths from that value
2966up to those that are lower than the next entry \-\- e.g., 16= covers
2967depths from 16 to 31. Note that the range covered by a depth
2968distribution entry can be different to the range covered by the
2969equivalent \fBsubmit\fR/\fBcomplete\fR distribution entry.
2970.TP
2971.B IO submit
2972How many pieces of I/O were submitting in a single submit call. Each
2973entry denotes that amount and below, until the previous entry \-\- e.g.,
297416=100% means that we submitted anywhere between 9 to 16 I/Os per submit
2975call. Note that the range covered by a \fBsubmit\fR distribution entry can
2976be different to the range covered by the equivalent depth distribution
2977entry.
2978.TP
2979.B IO complete
2980Like the above \fBsubmit\fR number, but for completions instead.
2981.TP
2982.B IO issued rwt
2983The number of \fBread/write/trim\fR requests issued, and how many of them were
2984short or dropped.
d60e92d1 2985.TP
d3b9694d 2986.B IO latency
ee21ebee 2987These values are for \fBlatency_target\fR and related options. When
d3b9694d
VF
2988these options are engaged, this section describes the I/O depth required
2989to meet the specified latency target.
d60e92d1 2990.RE
d60e92d1 2991.P
40943b9a
TK
2992After each client has been listed, the group statistics are printed. They
2993will look like this:
2994.P
2995.nf
2996 Run status group 0 (all jobs):
2997 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
2998 WRITE: bw=1231KiB/s (1261kB/s), 616KiB/s\-621KiB/s (630kB/s\-636kB/s), io=64.0MiB (67.1MB), run=52747\-53223msec
2999.fi
3000.P
3001For each data direction it prints:
d60e92d1
AC
3002.RS
3003.TP
40943b9a
TK
3004.B bw
3005Aggregate bandwidth of threads in this group followed by the
3006minimum and maximum bandwidth of all the threads in this group.
3007Values outside of brackets are power\-of\-2 format and those
3008within are the equivalent value in a power\-of\-10 format.
d60e92d1 3009.TP
40943b9a
TK
3010.B io
3011Aggregate I/O performed of all threads in this group. The
3012format is the same as \fBbw\fR.
d60e92d1 3013.TP
40943b9a
TK
3014.B run
3015The smallest and longest runtimes of the threads in this group.
d60e92d1 3016.RE
d60e92d1 3017.P
40943b9a
TK
3018And finally, the disk statistics are printed. This is Linux specific.
3019They will look like this:
3020.P
3021.nf
3022 Disk stats (read/write):
3023 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3024.fi
3025.P
3026Each value is printed for both reads and writes, with reads first. The
3027numbers denote:
d60e92d1
AC
3028.RS
3029.TP
3030.B ios
3031Number of I/Os performed by all groups.
3032.TP
3033.B merge
007c7be9 3034Number of merges performed by the I/O scheduler.
d60e92d1
AC
3035.TP
3036.B ticks
3037Number of ticks we kept the disk busy.
3038.TP
40943b9a 3039.B in_queue
d60e92d1
AC
3040Total time spent in the disk queue.
3041.TP
3042.B util
40943b9a
TK
3043The disk utilization. A value of 100% means we kept the disk
3044busy constantly, 50% would be a disk idling half of the time.
d60e92d1 3045.RE
8423bd11 3046.P
40943b9a
TK
3047It is also possible to get fio to dump the current output while it is running,
3048without terminating the job. To do that, send fio the USR1 signal. You can
3049also get regularly timed dumps by using the \fB\-\-status\-interval\fR
3050parameter, or by creating a file in `/tmp' named
3051`fio\-dump\-status'. If fio sees this file, it will unlink it and dump the
3052current output status.
d60e92d1 3053.SH TERSE OUTPUT
40943b9a
TK
3054For scripted usage where you typically want to generate tables or graphs of the
3055results, fio can output the results in a semicolon separated format. The format
3056is one long line of values, such as:
d60e92d1 3057.P
40943b9a
TK
3058.nf
3059 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%
3060 A description of this job goes here.
3061.fi
d60e92d1 3062.P
40943b9a 3063The job description (if provided) follows on a second line.
d60e92d1 3064.P
40943b9a
TK
3065To enable terse output, use the \fB\-\-minimal\fR or
3066`\-\-output\-format=terse' command line options. The
3067first value is the version of the terse output format. If the output has to be
3068changed for some reason, this number will be incremented by 1 to signify that
3069change.
d60e92d1 3070.P
40943b9a
TK
3071Split up, the format is as follows (comments in brackets denote when a
3072field was introduced or whether it's specific to some terse version):
d60e92d1 3073.P
40943b9a
TK
3074.nf
3075 terse version, fio version [v3], jobname, groupid, error
3076.fi
525c2bfa 3077.RS
40943b9a
TK
3078.P
3079.B
3080READ status:
525c2bfa 3081.RE
40943b9a
TK
3082.P
3083.nf
3084 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3085 Submission latency: min, max, mean, stdev (usec)
3086 Completion latency: min, max, mean, stdev (usec)
3087 Completion latency percentiles: 20 fields (see below)
3088 Total latency: min, max, mean, stdev (usec)
3089 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3090 IOPS [v5]: min, max, mean, stdev, number of samples
3091.fi
d60e92d1 3092.RS
40943b9a
TK
3093.P
3094.B
3095WRITE status:
a2c95580 3096.RE
40943b9a
TK
3097.P
3098.nf
3099 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3100 Submission latency: min, max, mean, stdev (usec)
3101 Completion latency: min, max, mean, stdev (usec)
3102 Completion latency percentiles: 20 fields (see below)
3103 Total latency: min, max, mean, stdev (usec)
3104 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3105 IOPS [v5]: min, max, mean, stdev, number of samples
3106.fi
a2c95580 3107.RS
40943b9a
TK
3108.P
3109.B
3110TRIM status [all but version 3]:
d60e92d1
AC
3111.RE
3112.P
40943b9a
TK
3113.nf
3114 Fields are similar to \fBREAD/WRITE\fR status.
3115.fi
a2c95580 3116.RS
a2c95580 3117.P
40943b9a 3118.B
d1429b5c 3119CPU usage:
d60e92d1
AC
3120.RE
3121.P
40943b9a
TK
3122.nf
3123 user, system, context switches, major faults, minor faults
3124.fi
d60e92d1 3125.RS
40943b9a
TK
3126.P
3127.B
3128I/O depths:
d60e92d1
AC
3129.RE
3130.P
40943b9a
TK
3131.nf
3132 <=1, 2, 4, 8, 16, 32, >=64
3133.fi
562c2d2f 3134.RS
40943b9a
TK
3135.P
3136.B
3137I/O latencies microseconds:
562c2d2f 3138.RE
40943b9a
TK
3139.P
3140.nf
3141 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3142.fi
562c2d2f 3143.RS
40943b9a
TK
3144.P
3145.B
3146I/O latencies milliseconds:
562c2d2f
DN
3147.RE
3148.P
40943b9a
TK
3149.nf
3150 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3151.fi
f2f788dd 3152.RS
40943b9a
TK
3153.P
3154.B
3155Disk utilization [v3]:
f2f788dd
JA
3156.RE
3157.P
40943b9a
TK
3158.nf
3159 disk name, read ios, write ios, read merges, write merges, read ticks, write ticks, time spent in queue, disk utilization percentage
3160.fi
562c2d2f 3161.RS
d60e92d1 3162.P
40943b9a
TK
3163.B
3164Additional Info (dependent on continue_on_error, default off):
d60e92d1 3165.RE
2fc26c3d 3166.P
40943b9a
TK
3167.nf
3168 total # errors, first error code
3169.fi
2fc26c3d
IC
3170.RS
3171.P
40943b9a
TK
3172.B
3173Additional Info (dependent on description being set):
3174.RE
3175.P
2fc26c3d 3176.nf
40943b9a
TK
3177 Text description
3178.fi
3179.P
3180Completion latency percentiles can be a grouping of up to 20 sets, so for the
3181terse output fio writes all of them. Each field will look like this:
3182.P
3183.nf
3184 1.00%=6112
3185.fi
3186.P
3187which is the Xth percentile, and the `usec' latency associated with it.
3188.P
3189For \fBDisk utilization\fR, all disks used by fio are shown. So for each disk there
3190will be a disk utilization section.
3191.P
3192Below is a single line containing short names for each of the fields in the
3193minimal output v3, separated by semicolons:
3194.P
3195.nf
3196 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 3197.fi
44c82dba
VF
3198.SH JSON OUTPUT
3199The \fBjson\fR output format is intended to be both human readable and convenient
3200for automated parsing. For the most part its sections mirror those of the
3201\fBnormal\fR output. The \fBruntime\fR value is reported in msec and the \fBbw\fR value is
3202reported in 1024 bytes per second units.
3203.fi
d9e557ab
VF
3204.SH JSON+ OUTPUT
3205The \fBjson+\fR output format is identical to the \fBjson\fR output format except that it
3206adds a full dump of the completion latency bins. Each \fBbins\fR object contains a
3207set of (key, value) pairs where keys are latency durations and values count how
3208many I/Os had completion latencies of the corresponding duration. For example,
3209consider:
d9e557ab 3210.RS
40943b9a 3211.P
d9e557ab
VF
3212"bins" : { "87552" : 1, "89600" : 1, "94720" : 1, "96768" : 1, "97792" : 1, "99840" : 1, "100864" : 2, "103936" : 6, "104960" : 534, "105984" : 5995, "107008" : 7529, ... }
3213.RE
40943b9a 3214.P
d9e557ab
VF
3215This data indicates that one I/O required 87,552ns to complete, two I/Os required
3216100,864ns to complete, and 7529 I/Os required 107,008ns to complete.
40943b9a 3217.P
d9e557ab 3218Also included with fio is a Python script \fBfio_jsonplus_clat2csv\fR that takes
40943b9a
TK
3219json+ output and generates CSV\-formatted latency data suitable for plotting.
3220.P
d9e557ab 3221The latency durations actually represent the midpoints of latency intervals.
40943b9a 3222For details refer to `stat.h' in the fio source.
29dbd1e5 3223.SH TRACE FILE FORMAT
40943b9a
TK
3224There are two trace file format that you can encounter. The older (v1) format is
3225unsupported since version 1.20\-rc3 (March 2008). It will still be described
29dbd1e5 3226below in case that you get an old trace and want to understand it.
29dbd1e5 3227.P
40943b9a
TK
3228In any case the trace is a simple text file with a single action per line.
3229.TP
29dbd1e5 3230.B Trace file format v1
40943b9a 3231Each line represents a single I/O action in the following format:
29dbd1e5 3232.RS
40943b9a
TK
3233.RS
3234.P
29dbd1e5 3235rw, offset, length
29dbd1e5
JA
3236.RE
3237.P
40943b9a
TK
3238where `rw=0/1' for read/write, and the `offset' and `length' entries being in bytes.
3239.P
3240This format is not supported in fio versions >= 1.20\-rc3.
3241.RE
3242.TP
29dbd1e5 3243.B Trace file format v2
40943b9a
TK
3244The second version of the trace file format was added in fio version 1.17. It
3245allows to access more then one file per trace and has a bigger set of possible
3246file actions.
29dbd1e5 3247.RS
40943b9a 3248.P
29dbd1e5 3249The first line of the trace file has to be:
40943b9a
TK
3250.RS
3251.P
3252"fio version 2 iolog"
3253.RE
3254.P
29dbd1e5 3255Following this can be lines in two different formats, which are described below.
40943b9a
TK
3256.P
3257.B
29dbd1e5 3258The file management format:
40943b9a
TK
3259.RS
3260filename action
29dbd1e5 3261.P
40943b9a 3262The `filename' is given as an absolute path. The `action' can be one of these:
29dbd1e5
JA
3263.RS
3264.TP
3265.B add
40943b9a 3266Add the given `filename' to the trace.
29dbd1e5
JA
3267.TP
3268.B open
40943b9a
TK
3269Open the file with the given `filename'. The `filename' has to have
3270been added with the \fBadd\fR action before.
29dbd1e5
JA
3271.TP
3272.B close
40943b9a
TK
3273Close the file with the given `filename'. The file has to have been
3274\fBopen\fRed before.
3275.RE
29dbd1e5 3276.RE
29dbd1e5 3277.P
40943b9a
TK
3278.B
3279The file I/O action format:
3280.RS
3281filename action offset length
29dbd1e5 3282.P
40943b9a
TK
3283The `filename' is given as an absolute path, and has to have been \fBadd\fRed and
3284\fBopen\fRed before it can be used with this format. The `offset' and `length' are
3285given in bytes. The `action' can be one of these:
29dbd1e5
JA
3286.RS
3287.TP
3288.B wait
40943b9a
TK
3289Wait for `offset' microseconds. Everything below 100 is discarded.
3290The time is relative to the previous `wait' statement.
29dbd1e5
JA
3291.TP
3292.B read
40943b9a 3293Read `length' bytes beginning from `offset'.
29dbd1e5
JA
3294.TP
3295.B write
40943b9a 3296Write `length' bytes beginning from `offset'.
29dbd1e5
JA
3297.TP
3298.B sync
40943b9a 3299\fBfsync\fR\|(2) the file.
29dbd1e5
JA
3300.TP
3301.B datasync
40943b9a 3302\fBfdatasync\fR\|(2) the file.
29dbd1e5
JA
3303.TP
3304.B trim
40943b9a
TK
3305Trim the given file from the given `offset' for `length' bytes.
3306.RE
29dbd1e5 3307.RE
29dbd1e5 3308.SH CPU IDLENESS PROFILING
40943b9a
TK
3309In some cases, we want to understand CPU overhead in a test. For example, we
3310test patches for the specific goodness of whether they reduce CPU usage.
3311Fio implements a balloon approach to create a thread per CPU that runs at idle
3312priority, meaning that it only runs when nobody else needs the cpu.
3313By measuring the amount of work completed by the thread, idleness of each CPU
3314can be derived accordingly.
3315.P
3316An unit work is defined as touching a full page of unsigned characters. Mean and
3317standard deviation of time to complete an unit work is reported in "unit work"
3318section. Options can be chosen to report detailed percpu idleness or overall
3319system idleness by aggregating percpu stats.
29dbd1e5 3320.SH VERIFICATION AND TRIGGERS
40943b9a
TK
3321Fio is usually run in one of two ways, when data verification is done. The first
3322is a normal write job of some sort with verify enabled. When the write phase has
3323completed, fio switches to reads and verifies everything it wrote. The second
3324model is running just the write phase, and then later on running the same job
3325(but with reads instead of writes) to repeat the same I/O patterns and verify
3326the contents. Both of these methods depend on the write phase being completed,
3327as fio otherwise has no idea how much data was written.
3328.P
3329With verification triggers, fio supports dumping the current write state to
3330local files. Then a subsequent read verify workload can load this state and know
3331exactly where to stop. This is useful for testing cases where power is cut to a
3332server in a managed fashion, for instance.
3333.P
29dbd1e5 3334A verification trigger consists of two things:
29dbd1e5 3335.RS
40943b9a
TK
3336.P
33371) Storing the write state of each job.
3338.P
33392) Executing a trigger command.
29dbd1e5 3340.RE
40943b9a
TK
3341.P
3342The write state is relatively small, on the order of hundreds of bytes to single
3343kilobytes. It contains information on the number of completions done, the last X
3344completions, etc.
3345.P
3346A trigger is invoked either through creation ('touch') of a specified file in
3347the system, or through a timeout setting. If fio is run with
3348`\-\-trigger\-file=/tmp/trigger\-file', then it will continually
3349check for the existence of `/tmp/trigger\-file'. When it sees this file, it
3350will fire off the trigger (thus saving state, and executing the trigger
29dbd1e5 3351command).
40943b9a
TK
3352.P
3353For client/server runs, there's both a local and remote trigger. If fio is
3354running as a server backend, it will send the job states back to the client for
3355safe storage, then execute the remote trigger, if specified. If a local trigger
3356is specified, the server will still send back the write state, but the client
3357will then execute the trigger.
29dbd1e5
JA
3358.RE
3359.P
3360.B Verification trigger example
3361.RS
40943b9a
TK
3362Let's say we want to run a powercut test on the remote Linux machine 'server'.
3363Our write workload is in `write\-test.fio'. We want to cut power to 'server' at
3364some point during the run, and we'll run this test from the safety or our local
3365machine, 'localbox'. On the server, we'll start the fio backend normally:
3366.RS
3367.P
3368server# fio \-\-server
3369.RE
3370.P
29dbd1e5 3371and on the client, we'll fire off the workload:
40943b9a
TK
3372.RS
3373.P
3374localbox$ fio \-\-client=server \-\-trigger\-file=/tmp/my\-trigger \-\-trigger\-remote="bash \-c "echo b > /proc/sysrq\-triger""
3375.RE
3376.P
3377We set `/tmp/my\-trigger' as the trigger file, and we tell fio to execute:
3378.RS
3379.P
3380echo b > /proc/sysrq\-trigger
3381.RE
3382.P
3383on the server once it has received the trigger and sent us the write state. This
3384will work, but it's not really cutting power to the server, it's merely
3385abruptly rebooting it. If we have a remote way of cutting power to the server
3386through IPMI or similar, we could do that through a local trigger command
3387instead. Let's assume we have a script that does IPMI reboot of a given hostname,
3388ipmi\-reboot. On localbox, we could then have run fio with a local trigger
3389instead:
3390.RS
3391.P
3392localbox$ fio \-\-client=server \-\-trigger\-file=/tmp/my\-trigger \-\-trigger="ipmi\-reboot server"
3393.RE
3394.P
3395For this case, fio would wait for the server to send us the write state, then
3396execute `ipmi\-reboot server' when that happened.
29dbd1e5
JA
3397.RE
3398.P
3399.B Loading verify state
3400.RS
40943b9a
TK
3401To load stored write state, a read verification job file must contain the
3402\fBverify_state_load\fR option. If that is set, fio will load the previously
29dbd1e5 3403stored state. For a local fio run this is done by loading the files directly,
40943b9a
TK
3404and on a client/server run, the server backend will ask the client to send the
3405files over and load them from there.
29dbd1e5 3406.RE
a3ae5b05 3407.SH LOG FILE FORMATS
a3ae5b05
JA
3408Fio supports a variety of log file formats, for logging latencies, bandwidth,
3409and IOPS. The logs share a common format, which looks like this:
40943b9a 3410.RS
a3ae5b05 3411.P
40943b9a
TK
3412time (msec), value, data direction, block size (bytes), offset (bytes)
3413.RE
3414.P
3415`Time' for the log entry is always in milliseconds. The `value' logged depends
3416on the type of log, it will be one of the following:
3417.RS
a3ae5b05
JA
3418.TP
3419.B Latency log
168bb587 3420Value is latency in nsecs
a3ae5b05
JA
3421.TP
3422.B Bandwidth log
6d500c2e 3423Value is in KiB/sec
a3ae5b05
JA
3424.TP
3425.B IOPS log
40943b9a
TK
3426Value is IOPS
3427.RE
a3ae5b05 3428.P
40943b9a
TK
3429`Data direction' is one of the following:
3430.RS
a3ae5b05
JA
3431.TP
3432.B 0
40943b9a 3433I/O is a READ
a3ae5b05
JA
3434.TP
3435.B 1
40943b9a 3436I/O is a WRITE
a3ae5b05
JA
3437.TP
3438.B 2
40943b9a 3439I/O is a TRIM
a3ae5b05 3440.RE
40943b9a
TK
3441.P
3442The entry's `block size' is always in bytes. The `offset' is the offset, in bytes,
3443from the start of the file, for that particular I/O. The logging of the offset can be
3444toggled with \fBlog_offset\fR.
3445.P
3446Fio defaults to logging every individual I/O. When IOPS are logged for individual
3447I/Os the `value' entry will always be 1. If windowed logging is enabled through
3448\fBlog_avg_msec\fR, fio logs the average values over the specified period of time.
3449If windowed logging is enabled and \fBlog_max_value\fR is set, then fio logs
3450maximum values in that window instead of averages. Since `data direction', `block size'
3451and `offset' are per\-I/O values, if windowed logging is enabled they
3452aren't applicable and will be 0.
49da1240 3453.SH CLIENT / SERVER
40943b9a
TK
3454Normally fio is invoked as a stand\-alone application on the machine where the
3455I/O workload should be generated. However, the backend and frontend of fio can
3456be run separately i.e., the fio server can generate an I/O workload on the "Device
3457Under Test" while being controlled by a client on another machine.
3458.P
3459Start the server on the machine which has access to the storage DUT:
3460.RS
3461.P
3462$ fio \-\-server=args
3463.RE
3464.P
3465where `args' defines what fio listens to. The arguments are of the form
3466`type,hostname' or `IP,port'. `type' is either `ip' (or ip4) for TCP/IP
3467v4, `ip6' for TCP/IP v6, or `sock' for a local unix domain socket.
3468`hostname' is either a hostname or IP address, and `port' is the port to listen
3469to (only valid for TCP/IP, not a local socket). Some examples:
3470.RS
3471.TP
e0ee7a8b 34721) \fBfio \-\-server\fR
40943b9a
TK
3473Start a fio server, listening on all interfaces on the default port (8765).
3474.TP
e0ee7a8b 34752) \fBfio \-\-server=ip:hostname,4444\fR
40943b9a
TK
3476Start a fio server, listening on IP belonging to hostname and on port 4444.
3477.TP
e0ee7a8b 34783) \fBfio \-\-server=ip6:::1,4444\fR
40943b9a
TK
3479Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
3480.TP
e0ee7a8b 34814) \fBfio \-\-server=,4444\fR
40943b9a
TK
3482Start a fio server, listening on all interfaces on port 4444.
3483.TP
e0ee7a8b 34845) \fBfio \-\-server=1.2.3.4\fR
40943b9a
TK
3485Start a fio server, listening on IP 1.2.3.4 on the default port.
3486.TP
e0ee7a8b 34876) \fBfio \-\-server=sock:/tmp/fio.sock\fR
40943b9a
TK
3488Start a fio server, listening on the local socket `/tmp/fio.sock'.
3489.RE
3490.P
3491Once a server is running, a "client" can connect to the fio server with:
3492.RS
3493.P
3494$ fio <local\-args> \-\-client=<server> <remote\-args> <job file(s)>
3495.RE
3496.P
3497where `local\-args' are arguments for the client where it is running, `server'
3498is the connect string, and `remote\-args' and `job file(s)' are sent to the
3499server. The `server' string follows the same format as it does on the server
3500side, to allow IP/hostname/socket and port strings.
3501.P
3502Fio can connect to multiple servers this way:
3503.RS
3504.P
3505$ fio \-\-client=<server1> <job file(s)> \-\-client=<server2> <job file(s)>
3506.RE
3507.P
3508If the job file is located on the fio server, then you can tell the server to
3509load a local file as well. This is done by using \fB\-\-remote\-config\fR:
3510.RS
3511.P
3512$ fio \-\-client=server \-\-remote\-config /path/to/file.fio
3513.RE
3514.P
3515Then fio will open this local (to the server) job file instead of being passed
3516one from the client.
3517.P
ff6bb260 3518If you have many servers (example: 100 VMs/containers), you can input a pathname
40943b9a
TK
3519of a file containing host IPs/names as the parameter value for the
3520\fB\-\-client\fR option. For example, here is an example `host.list'
3521file containing 2 hostnames:
3522.RS
3523.P
3524.PD 0
39b5f61e 3525host1.your.dns.domain
40943b9a 3526.P
39b5f61e 3527host2.your.dns.domain
40943b9a
TK
3528.PD
3529.RE
3530.P
39b5f61e 3531The fio command would then be:
40943b9a
TK
3532.RS
3533.P
3534$ fio \-\-client=host.list <job file(s)>
3535.RE
3536.P
3537In this mode, you cannot input server\-specific parameters or job files \-\- all
39b5f61e 3538servers receive the same job file.
40943b9a
TK
3539.P
3540In order to let `fio \-\-client' runs use a shared filesystem from multiple
3541hosts, `fio \-\-client' now prepends the IP address of the server to the
3542filename. For example, if fio is using the directory `/mnt/nfs/fio' and is
3543writing filename `fileio.tmp', with a \fB\-\-client\fR `hostfile'
3544containing two hostnames `h1' and `h2' with IP addresses 192.168.10.120 and
3545192.168.10.121, then fio will create two files:
3546.RS
3547.P
3548.PD 0
39b5f61e 3549/mnt/nfs/fio/192.168.10.120.fileio.tmp
40943b9a 3550.P
39b5f61e 3551/mnt/nfs/fio/192.168.10.121.fileio.tmp
40943b9a
TK
3552.PD
3553.RE
d60e92d1
AC
3554.SH AUTHORS
3555.B fio
aa58d252 3556was written by Jens Axboe <jens.axboe@oracle.com>,
f8b8f7da 3557now Jens Axboe <axboe@fb.com>.
d1429b5c
AC
3558.br
3559This man page was written by Aaron Carroll <aaronc@cse.unsw.edu.au> based
d60e92d1 3560on documentation by Jens Axboe.
40943b9a
TK
3561.br
3562This man page was rewritten by Tomohiro Kusumi <tkusumi@tuxera.com> based
3563on documentation by Jens Axboe.
d60e92d1 3564.SH "REPORTING BUGS"
482900c9 3565Report bugs to the \fBfio\fR mailing list <fio@vger.kernel.org>.
6468020d 3566.br
40943b9a
TK
3567See \fBREPORTING\-BUGS\fR.
3568.P
3569\fBREPORTING\-BUGS\fR: \fIhttp://git.kernel.dk/cgit/fio/plain/REPORTING\-BUGS\fR
d60e92d1 3570.SH "SEE ALSO"
d1429b5c
AC
3571For further documentation see \fBHOWTO\fR and \fBREADME\fR.
3572.br
40943b9a 3573Sample jobfiles are available in the `examples/' directory.
9040e236 3574.br
40943b9a
TK
3575These are typically located under `/usr/share/doc/fio'.
3576.P
3577\fBHOWTO\fR: \fIhttp://git.kernel.dk/cgit/fio/plain/HOWTO\fR
9040e236 3578.br
40943b9a 3579\fBREADME\fR: \fIhttp://git.kernel.dk/cgit/fio/plain/README\fR