Document io_uring feature
[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
b9921d1a
DZ
23.BI \-\-merge\-blktrace\-only
24Merge blktraces only, don't start any I/O.
25.TP
d60e92d1
AC
26.BI \-\-output \fR=\fPfilename
27Write output to \fIfilename\fR.
28.TP
7db7a5a0
TK
29.BI \-\-output\-format \fR=\fPformat
30Set the reporting \fIformat\fR to `normal', `terse', `json', or
31`json+'. Multiple formats can be selected, separate by a comma. `terse'
32is a CSV based format. `json+' is like `json', except it adds a full
513e37ee 33dump of the latency buckets.
e28ee21d 34.TP
7db7a5a0 35.BI \-\-bandwidth\-log
d23ae827 36Generate aggregate bandwidth logs.
d60e92d1 37.TP
7db7a5a0
TK
38.BI \-\-minimal
39Print statistics in a terse, semicolon\-delimited format.
d60e92d1 40.TP
7db7a5a0
TK
41.BI \-\-append\-terse
42Print statistics in selected mode AND terse, semicolon\-delimited format.
43\fBDeprecated\fR, use \fB\-\-output\-format\fR instead to select multiple formats.
f6a7df53 44.TP
065248bf 45.BI \-\-terse\-version \fR=\fPversion
7db7a5a0 46Set terse \fIversion\fR output format (default `3', or `2', `4', `5').
49da1240 47.TP
7db7a5a0 48.BI \-\-version
bdd88be3
TK
49Print version information and exit.
50.TP
7db7a5a0 51.BI \-\-help
bdd88be3 52Print a summary of the command line options and exit.
49da1240 53.TP
7db7a5a0 54.BI \-\-cpuclock\-test
bdd88be3 55Perform test and validation of internal CPU clock.
fec0f21c 56.TP
bdd88be3 57.BI \-\-crctest \fR=\fP[test]
7db7a5a0 58Test the speed of the built\-in checksumming functions. If no argument is given,
bdd88be3 59all of them are tested. Alternatively, a comma separated list can be passed, in which
fec0f21c
JA
60case the given ones are tested.
61.TP
49da1240 62.BI \-\-cmdhelp \fR=\fPcommand
bdd88be3 63Print help information for \fIcommand\fR. May be `all' for all commands.
49da1240 64.TP
7db7a5a0
TK
65.BI \-\-enghelp \fR=\fP[ioengine[,command]]
66List all commands defined by \fIioengine\fR, or print help for \fIcommand\fR
67defined by \fIioengine\fR. If no \fIioengine\fR is given, list all
68available ioengines.
de890a1e 69.TP
d60e92d1 70.BI \-\-showcmd \fR=\fPjobfile
7db7a5a0 71Convert \fIjobfile\fR to a set of command\-line options.
d60e92d1 72.TP
bdd88be3 73.BI \-\-readonly
4027b2a1 74Turn on safety read\-only checks, preventing writes and trims. The \fB\-\-readonly\fR
bdd88be3 75option is an extra safety guard to prevent users from accidentally starting
4027b2a1
VF
76a write or trim workload when that is not desired. Fio will only modify the
77device under test if `rw=write/randwrite/rw/randrw/trim/randtrim/trimwrite'
78is given. This safety net can be used as an extra precaution.
bdd88be3 79.TP
d60e92d1 80.BI \-\-eta \fR=\fPwhen
7db7a5a0 81Specifies when real\-time ETA estimate should be printed. \fIwhen\fR may
db37d890
JA
82be `always', `never' or `auto'. `auto' is the default, it prints ETA when
83requested if the output is a TTY. `always' disregards the output type, and
84prints ETA when requested. `never' never prints ETA.
85.TP
86.BI \-\-eta\-interval \fR=\fPtime
87By default, fio requests client ETA status roughly every second. With this
88option, the interval is configurable. Fio imposes a minimum allowed time to
89avoid flooding the console, less than 250 msec is not supported.
d60e92d1 90.TP
30b5d57f 91.BI \-\-eta\-newline \fR=\fPtime
bdd88be3
TK
92Force a new line for every \fItime\fR period passed. When the unit is omitted,
93the value is interpreted in seconds.
30b5d57f
JA
94.TP
95.BI \-\-status\-interval \fR=\fPtime
aa6cb459
VF
96Force a full status dump of cumulative (from job start) values at \fItime\fR
97intervals. This option does *not* provide per-period measurements. So
98values such as bandwidth are running averages. When the time unit is omitted,
c1f4de8a
JA
99\fItime\fR is interpreted in seconds. Note that using this option with
100`\-\-output-format=json' will yield output that technically isn't valid json,
101since the output will be collated sets of valid json. It will need to be split
102into valid sets of json after the run.
bdd88be3
TK
103.TP
104.BI \-\-section \fR=\fPname
105Only run specified section \fIname\fR in job file. Multiple sections can be specified.
7db7a5a0 106The \fB\-\-section\fR option allows one to combine related jobs into one file.
bdd88be3 107E.g. one job file could define light, moderate, and heavy sections. Tell
7db7a5a0 108fio to run only the "heavy" section by giving `\-\-section=heavy'
bdd88be3 109command line option. One can also specify the "write" operations in one
7db7a5a0 110section and "verify" operation in another section. The \fB\-\-section\fR option
bdd88be3
TK
111only applies to job sections. The reserved *global* section is always
112parsed and used.
c0a5d35e 113.TP
49da1240 114.BI \-\-alloc\-size \fR=\fPkb
7db7a5a0
TK
115Set the internal smalloc pool size to \fIkb\fR in KiB. The
116\fB\-\-alloc\-size\fR switch allows one to use a larger pool size for smalloc.
bdd88be3
TK
117If running large jobs with randommap enabled, fio can run out of memory.
118Smalloc is an internal allocator for shared structures from a fixed size
119memory pool and can grow to 16 pools. The pool size defaults to 16MiB.
7db7a5a0
TK
120NOTE: While running `.fio_smalloc.*' backing store files are visible
121in `/tmp'.
d60e92d1 122.TP
49da1240
JA
123.BI \-\-warnings\-fatal
124All fio parser warnings are fatal, causing fio to exit with an error.
9183788d 125.TP
49da1240 126.BI \-\-max\-jobs \fR=\fPnr
7db7a5a0 127Set the maximum number of threads/processes to support to \fInr\fR.
7f4811bb
RNS
128NOTE: On Linux, it may be necessary to increase the shared-memory limit
129(`/proc/sys/kernel/shmmax') if fio runs into errors while creating jobs.
d60e92d1 130.TP
49da1240 131.BI \-\-server \fR=\fPargs
7db7a5a0
TK
132Start a backend server, with \fIargs\fR specifying what to listen to.
133See \fBCLIENT/SERVER\fR section.
f57a9c59 134.TP
49da1240 135.BI \-\-daemonize \fR=\fPpidfile
7db7a5a0 136Background a fio server, writing the pid to the given \fIpidfile\fR file.
49da1240 137.TP
bdd88be3 138.BI \-\-client \fR=\fPhostname
7db7a5a0
TK
139Instead of running the jobs locally, send and run them on the given \fIhostname\fR
140or set of \fIhostname\fRs. See \fBCLIENT/SERVER\fR section.
bdd88be3 141.TP
7db7a5a0
TK
142.BI \-\-remote\-config \fR=\fPfile
143Tell fio server to load this local \fIfile\fR.
f2a2ce0e
HL
144.TP
145.BI \-\-idle\-prof \fR=\fPoption
7db7a5a0 146Report CPU idleness. \fIoption\fR is one of the following:
bdd88be3
TK
147.RS
148.RS
149.TP
150.B calibrate
151Run unit work calibration only and exit.
152.TP
153.B system
154Show aggregate system idleness and unit work.
155.TP
156.B percpu
7db7a5a0 157As \fBsystem\fR but also show per CPU idleness.
bdd88be3
TK
158.RE
159.RE
160.TP
7db7a5a0
TK
161.BI \-\-inflate\-log \fR=\fPlog
162Inflate and output compressed \fIlog\fR.
bdd88be3 163.TP
7db7a5a0
TK
164.BI \-\-trigger\-file \fR=\fPfile
165Execute trigger command when \fIfile\fR exists.
bdd88be3 166.TP
7db7a5a0
TK
167.BI \-\-trigger\-timeout \fR=\fPtime
168Execute trigger at this \fItime\fR.
bdd88be3 169.TP
7db7a5a0
TK
170.BI \-\-trigger \fR=\fPcommand
171Set this \fIcommand\fR as local trigger.
bdd88be3 172.TP
7db7a5a0
TK
173.BI \-\-trigger\-remote \fR=\fPcommand
174Set this \fIcommand\fR as remote trigger.
bdd88be3 175.TP
7db7a5a0 176.BI \-\-aux\-path \fR=\fPpath
f4401bf8
SW
177Use the directory specified by \fIpath\fP for generated state files instead
178of the current working directory.
d60e92d1 179.SH "JOB FILE FORMAT"
7a14cf18
TK
180Any parameters following the options will be assumed to be job files, unless
181they match a job file parameter. Multiple job files can be listed and each job
7db7a5a0 182file will be regarded as a separate group. Fio will \fBstonewall\fR execution
7a14cf18
TK
183between each group.
184
185Fio accepts one or more job files describing what it is
186supposed to do. The job file format is the classic ini file, where the names
187enclosed in [] brackets define the job name. You are free to use any ASCII name
188you want, except *global* which has special meaning. Following the job name is
189a sequence of zero or more parameters, one per line, that define the behavior of
190the job. If the first character in a line is a ';' or a '#', the entire line is
191discarded as a comment.
192
193A *global* section sets defaults for the jobs described in that file. A job may
194override a *global* section parameter, and a job file may even have several
195*global* sections if so desired. A job is only affected by a *global* section
196residing above it.
197
7db7a5a0
TK
198The \fB\-\-cmdhelp\fR option also lists all options. If used with an \fIcommand\fR
199argument, \fB\-\-cmdhelp\fR will detail the given \fIcommand\fR.
7a14cf18 200
7db7a5a0
TK
201See the `examples/' directory for inspiration on how to write job files. Note
202the copyright and license requirements currently apply to
203`examples/' files.
07a2919d
NP
204
205Note that the maximum length of a line in the job file is 8192 bytes.
54eb4569
TK
206.SH "JOB FILE PARAMETERS"
207Some parameters take an option of a given type, such as an integer or a
208string. Anywhere a numeric value is required, an arithmetic expression may be
209used, provided it is surrounded by parentheses. Supported operators are:
d59aa780 210.RS
7db7a5a0 211.P
d59aa780 212.B addition (+)
7db7a5a0
TK
213.P
214.B subtraction (\-)
215.P
d59aa780 216.B multiplication (*)
7db7a5a0 217.P
d59aa780 218.B division (/)
7db7a5a0 219.P
d59aa780 220.B modulus (%)
7db7a5a0 221.P
d59aa780
JA
222.B exponentiation (^)
223.RE
d59aa780
JA
224.P
225For time values in expressions, units are microseconds by default. This is
226different than for time values not in expressions (not enclosed in
54eb4569
TK
227parentheses).
228.SH "PARAMETER TYPES"
229The following parameter types are used.
d60e92d1
AC
230.TP
231.I str
6b86fc18
TK
232String. A sequence of alphanumeric characters.
233.TP
234.I time
235Integer with possible time suffix. Without a unit value is interpreted as
236seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
237hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and 'us'
238(or 'usec') for microseconds. For example, use 10m for 10 minutes.
d60e92d1
AC
239.TP
240.I int
6d500c2e
RE
241Integer. A whole number value, which may contain an integer prefix
242and an integer suffix.
0b43a833
TK
243.RS
244.RS
245.P
6b86fc18 246[*integer prefix*] **number** [*integer suffix*]
0b43a833
TK
247.RE
248.P
6b86fc18
TK
249The optional *integer prefix* specifies the number's base. The default
250is decimal. *0x* specifies hexadecimal.
0b43a833 251.P
6b86fc18
TK
252The optional *integer suffix* specifies the number's units, and includes an
253optional unit prefix and an optional unit. For quantities of data, the
254default unit is bytes. For quantities of time, the default unit is seconds
255unless otherwise specified.
0b43a833
TK
256.P
257With `kb_base=1000', fio follows international standards for unit
7db7a5a0 258prefixes. To specify power\-of\-10 decimal values defined in the
6b86fc18 259International System of Units (SI):
0b43a833
TK
260.RS
261.P
7db7a5a0 262.PD 0
eccce61a 263K means kilo (K) or 1000
7db7a5a0 264.P
eccce61a 265M means mega (M) or 1000**2
7db7a5a0 266.P
eccce61a 267G means giga (G) or 1000**3
7db7a5a0 268.P
eccce61a 269T means tera (T) or 1000**4
7db7a5a0 270.P
eccce61a 271P means peta (P) or 1000**5
7db7a5a0 272.PD
0b43a833
TK
273.RE
274.P
7db7a5a0 275To specify power\-of\-2 binary values defined in IEC 80000\-13:
0b43a833
TK
276.RS
277.P
7db7a5a0 278.PD 0
eccce61a 279Ki means kibi (Ki) or 1024
7db7a5a0 280.P
eccce61a 281Mi means mebi (Mi) or 1024**2
7db7a5a0 282.P
eccce61a 283Gi means gibi (Gi) or 1024**3
7db7a5a0 284.P
eccce61a 285Ti means tebi (Ti) or 1024**4
7db7a5a0 286.P
eccce61a 287Pi means pebi (Pi) or 1024**5
7db7a5a0 288.PD
0b43a833
TK
289.RE
290.P
291With `kb_base=1024' (the default), the unit prefixes are opposite
7db7a5a0 292from those specified in the SI and IEC 80000\-13 standards to provide
6b86fc18 293compatibility with old scripts. For example, 4k means 4096.
0b43a833 294.P
6b86fc18
TK
295For quantities of data, an optional unit of 'B' may be included
296(e.g., 'kB' is the same as 'k').
0b43a833 297.P
6b86fc18
TK
298The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
299not milli). 'b' and 'B' both mean byte, not bit.
0b43a833
TK
300.P
301Examples with `kb_base=1000':
302.RS
303.P
7db7a5a0 304.PD 0
6d500c2e 3054 KiB: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
7db7a5a0 306.P
6d500c2e 3071 MiB: 1048576, 1m, 1024k
7db7a5a0 308.P
6d500c2e 3091 MB: 1000000, 1mi, 1000ki
7db7a5a0 310.P
6d500c2e 3111 TiB: 1073741824, 1t, 1024m, 1048576k
7db7a5a0 312.P
6d500c2e 3131 TB: 1000000000, 1ti, 1000mi, 1000000ki
7db7a5a0 314.PD
0b43a833
TK
315.RE
316.P
317Examples with `kb_base=1024' (default):
318.RS
319.P
7db7a5a0 320.PD 0
6d500c2e 3214 KiB: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
7db7a5a0 322.P
6d500c2e 3231 MiB: 1048576, 1m, 1024k
7db7a5a0 324.P
6d500c2e 3251 MB: 1000000, 1mi, 1000ki
7db7a5a0 326.P
6d500c2e 3271 TiB: 1073741824, 1t, 1024m, 1048576k
7db7a5a0 328.P
6d500c2e 3291 TB: 1000000000, 1ti, 1000mi, 1000000ki
7db7a5a0 330.PD
0b43a833
TK
331.RE
332.P
6d500c2e 333To specify times (units are not case sensitive):
0b43a833
TK
334.RS
335.P
7db7a5a0 336.PD 0
6d500c2e 337D means days
7db7a5a0 338.P
6d500c2e 339H means hours
7db7a5a0 340.P
6d500c2e 341M mean minutes
7db7a5a0 342.P
6d500c2e 343s or sec means seconds (default)
7db7a5a0 344.P
6d500c2e 345ms or msec means milliseconds
7db7a5a0 346.P
6d500c2e 347us or usec means microseconds
7db7a5a0 348.PD
0b43a833
TK
349.RE
350.P
6b86fc18 351If the option accepts an upper and lower range, use a colon ':' or
7db7a5a0 352minus '\-' to separate such values. See \fIirange\fR parameter type.
6b86fc18
TK
353If the lower value specified happens to be larger than the upper value
354the two values are swapped.
0b43a833 355.RE
d60e92d1
AC
356.TP
357.I bool
6b86fc18
TK
358Boolean. Usually parsed as an integer, however only defined for
359true and false (1 and 0).
d60e92d1
AC
360.TP
361.I irange
6b86fc18 362Integer range with suffix. Allows value range to be given, such as
7db7a5a0 3631024\-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
6b86fc18 364option allows two sets of ranges, they can be specified with a ',' or '/'
7db7a5a0 365delimiter: 1k\-4k/8k\-32k. Also see \fIint\fR parameter type.
83349190
YH
366.TP
367.I float_list
6b86fc18 368A list of floating point numbers, separated by a ':' character.
523bad63 369.SH "JOB PARAMETERS"
54eb4569 370With the above in mind, here follows the complete list of fio job parameters.
523bad63 371.SS "Units"
d60e92d1 372.TP
523bad63
TK
373.BI kb_base \fR=\fPint
374Select the interpretation of unit prefixes in input parameters.
375.RS
376.RS
d60e92d1 377.TP
523bad63
TK
378.B 1000
379Inputs comply with IEC 80000\-13 and the International
380System of Units (SI). Use:
381.RS
382.P
383.PD 0
384\- power\-of\-2 values with IEC prefixes (e.g., KiB)
385.P
386\- power\-of\-10 values with SI prefixes (e.g., kB)
387.PD
388.RE
389.TP
390.B 1024
391Compatibility mode (default). To avoid breaking old scripts:
392.P
393.RS
394.PD 0
395\- power\-of\-2 values with SI prefixes
396.P
397\- power\-of\-10 values with IEC prefixes
398.PD
399.RE
400.RE
401.P
402See \fBbs\fR for more details on input parameters.
403.P
404Outputs always use correct prefixes. Most outputs include both
405side\-by\-side, like:
406.P
407.RS
408bw=2383.3kB/s (2327.4KiB/s)
409.RE
410.P
411If only one value is reported, then kb_base selects the one to use:
412.P
413.RS
414.PD 0
4151000 \-\- SI prefixes
416.P
4171024 \-\- IEC prefixes
418.PD
419.RE
420.RE
421.TP
422.BI unit_base \fR=\fPint
423Base unit for reporting. Allowed values are:
424.RS
425.RS
426.TP
427.B 0
428Use auto\-detection (default).
429.TP
430.B 8
431Byte based.
432.TP
433.B 1
434Bit based.
435.RE
436.RE
437.SS "Job description"
438.TP
439.BI name \fR=\fPstr
440ASCII name of the job. This may be used to override the name printed by fio
441for this job. Otherwise the job name is used. On the command line this
442parameter has the special purpose of also signaling the start of a new job.
9cc8cb91 443.TP
d60e92d1 444.BI description \fR=\fPstr
523bad63
TK
445Text description of the job. Doesn't do anything except dump this text
446description when this job is run. It's not parsed.
447.TP
448.BI loops \fR=\fPint
449Run the specified number of iterations of this job. Used to repeat the same
450workload a given number of times. Defaults to 1.
451.TP
452.BI numjobs \fR=\fPint
453Create the specified number of clones of this job. Each clone of job
454is spawned as an independent thread or process. May be used to setup a
455larger number of threads/processes doing the same thing. Each thread is
456reported separately; to see statistics for all clones as a whole, use
457\fBgroup_reporting\fR in conjunction with \fBnew_group\fR.
458See \fB\-\-max\-jobs\fR. Default: 1.
459.SS "Time related parameters"
460.TP
461.BI runtime \fR=\fPtime
462Tell fio to terminate processing after the specified period of time. It
463can be quite hard to determine for how long a specified job will run, so
464this parameter is handy to cap the total runtime to a given time. When
f1dd3fb1 465the unit is omitted, the value is interpreted in seconds.
523bad63
TK
466.TP
467.BI time_based
468If set, fio will run for the duration of the \fBruntime\fR specified
469even if the file(s) are completely read or written. It will simply loop over
470the same workload as many times as the \fBruntime\fR allows.
471.TP
472.BI startdelay \fR=\fPirange(int)
473Delay the start of job for the specified amount of time. Can be a single
474value or a range. When given as a range, each thread will choose a value
475randomly from within the range. Value is in seconds if a unit is omitted.
476.TP
477.BI ramp_time \fR=\fPtime
478If set, fio will run the specified workload for this amount of time before
479logging any performance numbers. Useful for letting performance settle
480before logging results, thus minimizing the runtime required for stable
481results. Note that the \fBramp_time\fR is considered lead in time for a job,
482thus it will increase the total runtime if a special timeout or
483\fBruntime\fR is specified. When the unit is omitted, the value is
484given in seconds.
485.TP
486.BI clocksource \fR=\fPstr
487Use the given clocksource as the base of timing. The supported options are:
488.RS
489.RS
490.TP
491.B gettimeofday
492\fBgettimeofday\fR\|(2)
493.TP
494.B clock_gettime
495\fBclock_gettime\fR\|(2)
496.TP
497.B cpu
498Internal CPU clock source
499.RE
500.P
501\fBcpu\fR is the preferred clocksource if it is reliable, as it is very fast (and
502fio is heavy on time calls). Fio will automatically use this clocksource if
503it's supported and considered reliable on the system it is running on,
504unless another clocksource is specifically set. For x86/x86\-64 CPUs, this
505means supporting TSC Invariant.
506.RE
507.TP
508.BI gtod_reduce \fR=\fPbool
509Enable all of the \fBgettimeofday\fR\|(2) reducing options
510(\fBdisable_clat\fR, \fBdisable_slat\fR, \fBdisable_bw_measurement\fR) plus
511reduce precision of the timeout somewhat to really shrink the
512\fBgettimeofday\fR\|(2) call count. With this option enabled, we only do
513about 0.4% of the \fBgettimeofday\fR\|(2) calls we would have done if all
514time keeping was enabled.
515.TP
516.BI gtod_cpu \fR=\fPint
517Sometimes it's cheaper to dedicate a single thread of execution to just
518getting the current time. Fio (and databases, for instance) are very
519intensive on \fBgettimeofday\fR\|(2) calls. With this option, you can set
520one CPU aside for doing nothing but logging current time to a shared memory
521location. Then the other threads/processes that run I/O workloads need only
522copy that segment, instead of entering the kernel with a
523\fBgettimeofday\fR\|(2) call. The CPU set aside for doing these time
524calls will be excluded from other uses. Fio will manually clear it from the
525CPU mask of other jobs.
526.SS "Target file/device"
d60e92d1
AC
527.TP
528.BI directory \fR=\fPstr
523bad63
TK
529Prefix \fBfilename\fRs with this directory. Used to place files in a different
530location than `./'. You can specify a number of directories by
531separating the names with a ':' character. These directories will be
532assigned equally distributed to job clones created by \fBnumjobs\fR as
533long as they are using generated filenames. If specific \fBfilename\fR(s) are
534set fio will use the first listed directory, and thereby matching the
f4401bf8
SW
535\fBfilename\fR semantic (which generates a file for each clone if not
536specified, but lets all clones use the same file if set).
523bad63
TK
537.RS
538.P
ffc90a44 539See the \fBfilename\fR option for information on how to escape ':' and '\\'
523bad63 540characters within the directory path itself.
f4401bf8
SW
541.P
542Note: To control the directory fio will use for internal state files
543use \fB\-\-aux\-path\fR.
523bad63 544.RE
d60e92d1
AC
545.TP
546.BI filename \fR=\fPstr
523bad63
TK
547Fio normally makes up a \fBfilename\fR based on the job name, thread number, and
548file number (see \fBfilename_format\fR). If you want to share files
549between threads in a job or several
550jobs with fixed file paths, specify a \fBfilename\fR for each of them to override
551the default. If the ioengine is file based, you can specify a number of files
552by separating the names with a ':' colon. So if you wanted a job to open
553`/dev/sda' and `/dev/sdb' as the two working files, you would use
554`filename=/dev/sda:/dev/sdb'. This also means that whenever this option is
555specified, \fBnrfiles\fR is ignored. The size of regular files specified
556by this option will be \fBsize\fR divided by number of files unless an
557explicit size is specified by \fBfilesize\fR.
558.RS
559.P
ffc90a44 560Each colon and backslash in the wanted path must be escaped with a '\\'
523bad63
TK
561character. For instance, if the path is `/dev/dsk/foo@3,0:c' then you
562would use `filename=/dev/dsk/foo@3,0\\:c' and if the path is
ffc90a44 563`F:\\filename' then you would use `filename=F\\:\\\\filename'.
523bad63 564.P
ffc90a44
SW
565On Windows, disk devices are accessed as `\\\\.\\PhysicalDrive0' for
566the first device, `\\\\.\\PhysicalDrive1' for the second etc.
523bad63
TK
567Note: Windows and FreeBSD prevent write access to areas
568of the disk containing in\-use data (e.g. filesystems).
569.P
570The filename `\-' is a reserved name, meaning *stdin* or *stdout*. Which
571of the two depends on the read/write direction set.
572.RE
d60e92d1 573.TP
de98bd30 574.BI filename_format \fR=\fPstr
523bad63
TK
575If sharing multiple files between jobs, it is usually necessary to have fio
576generate the exact names that you want. By default, fio will name a file
de98bd30 577based on the default file format specification of
523bad63 578`jobname.jobnumber.filenumber'. With this option, that can be
de98bd30
JA
579customized. Fio will recognize and replace the following keywords in this
580string:
581.RS
582.RS
583.TP
584.B $jobname
585The name of the worker thread or process.
586.TP
587.B $jobnum
588The incremental number of the worker thread or process.
589.TP
590.B $filenum
591The incremental number of the file for that worker thread or process.
592.RE
593.P
523bad63
TK
594To have dependent jobs share a set of files, this option can be set to have
595fio generate filenames that are shared between the two. For instance, if
596`testfiles.$filenum' is specified, file number 4 for any job will be
597named `testfiles.4'. The default of `$jobname.$jobnum.$filenum'
de98bd30 598will be used if no other format specifier is given.
645943c0
JB
599.P
600If you specify a path then the directories will be created up to the main
601directory for the file. So for example if you specify `a/b/c/$jobnum` then the
602directories a/b/c will be created before the file setup part of the job. If you
603specify \fBdirectory\fR then the path will be relative that directory, otherwise
604it is treated as the absolute path.
de98bd30 605.RE
de98bd30 606.TP
922a5be8 607.BI unique_filename \fR=\fPbool
523bad63
TK
608To avoid collisions between networked clients, fio defaults to prefixing any
609generated filenames (with a directory specified) with the source of the
610client connecting. To disable this behavior, set this option to 0.
611.TP
612.BI opendir \fR=\fPstr
613Recursively open any files below directory \fIstr\fR.
922a5be8 614.TP
3ce9dcaf 615.BI lockfile \fR=\fPstr
523bad63
TK
616Fio defaults to not locking any files before it does I/O to them. If a file
617or file descriptor is shared, fio can serialize I/O to that file to make the
618end result consistent. This is usual for emulating real workloads that share
619files. The lock modes are:
3ce9dcaf
JA
620.RS
621.RS
622.TP
623.B none
523bad63 624No locking. The default.
3ce9dcaf
JA
625.TP
626.B exclusive
523bad63 627Only one thread or process may do I/O at a time, excluding all others.
3ce9dcaf
JA
628.TP
629.B readwrite
523bad63
TK
630Read\-write locking on the file. Many readers may
631access the file at the same time, but writes get exclusive access.
3ce9dcaf 632.RE
ce594fbe 633.RE
523bad63
TK
634.TP
635.BI nrfiles \fR=\fPint
636Number of files to use for this job. Defaults to 1. The size of files
637will be \fBsize\fR divided by this unless explicit size is specified by
638\fBfilesize\fR. Files are created for each thread separately, and each
639file will have a file number within its name by default, as explained in
640\fBfilename\fR section.
641.TP
642.BI openfiles \fR=\fPint
643Number of files to keep open at the same time. Defaults to the same as
644\fBnrfiles\fR, can be set smaller to limit the number simultaneous
645opens.
646.TP
647.BI file_service_type \fR=\fPstr
648Defines how fio decides which file from a job to service next. The following
649types are defined:
650.RS
651.RS
652.TP
653.B random
654Choose a file at random.
655.TP
656.B roundrobin
657Round robin over opened files. This is the default.
658.TP
659.B sequential
660Finish one file before moving on to the next. Multiple files can
661still be open depending on \fBopenfiles\fR.
662.TP
663.B zipf
664Use a Zipf distribution to decide what file to access.
665.TP
666.B pareto
667Use a Pareto distribution to decide what file to access.
668.TP
669.B normal
670Use a Gaussian (normal) distribution to decide what file to access.
671.TP
672.B gauss
673Alias for normal.
674.RE
3ce9dcaf 675.P
523bad63
TK
676For \fBrandom\fR, \fBroundrobin\fR, and \fBsequential\fR, a postfix can be appended to
677tell fio how many I/Os to issue before switching to a new file. For example,
678specifying `file_service_type=random:8' would cause fio to issue
6798 I/Os before selecting a new file at random. For the non\-uniform
680distributions, a floating point postfix can be given to influence how the
681distribution is skewed. See \fBrandom_distribution\fR for a description
682of how that would work.
683.RE
684.TP
685.BI ioscheduler \fR=\fPstr
686Attempt to switch the device hosting the file to the specified I/O scheduler
687before running.
688.TP
689.BI create_serialize \fR=\fPbool
690If true, serialize the file creation for the jobs. This may be handy to
691avoid interleaving of data files, which may greatly depend on the filesystem
692used and even the number of processors in the system. Default: true.
693.TP
694.BI create_fsync \fR=\fPbool
695\fBfsync\fR\|(2) the data file after creation. This is the default.
696.TP
697.BI create_on_open \fR=\fPbool
698If true, don't pre\-create files but allow the job's open() to create a file
699when it's time to do I/O. Default: false \-\- pre\-create all necessary files
700when the job starts.
701.TP
702.BI create_only \fR=\fPbool
703If true, fio will only run the setup phase of the job. If files need to be
704laid out or updated on disk, only that will be done \-\- the actual job contents
705are not executed. Default: false.
706.TP
707.BI allow_file_create \fR=\fPbool
708If true, fio is permitted to create files as part of its workload. If this
709option is false, then fio will error out if
710the files it needs to use don't already exist. Default: true.
711.TP
712.BI allow_mounted_write \fR=\fPbool
713If this isn't set, fio will abort jobs that are destructive (e.g. that write)
714to what appears to be a mounted device or partition. This should help catch
715creating inadvertently destructive tests, not realizing that the test will
716destroy data on the mounted file system. Note that some platforms don't allow
717writing against a mounted device regardless of this option. Default: false.
718.TP
719.BI pre_read \fR=\fPbool
720If this is given, files will be pre\-read into memory before starting the
721given I/O operation. This will also clear the \fBinvalidate\fR flag,
722since it is pointless to pre\-read and then drop the cache. This will only
723work for I/O engines that are seek\-able, since they allow you to read the
724same data multiple times. Thus it will not work on non\-seekable I/O engines
725(e.g. network, splice). Default: false.
726.TP
727.BI unlink \fR=\fPbool
728Unlink the job files when done. Not the default, as repeated runs of that
729job would then waste time recreating the file set again and again. Default:
730false.
731.TP
732.BI unlink_each_loop \fR=\fPbool
733Unlink job files after each iteration or loop. Default: false.
734.TP
7b865a2f
BVA
735.BI zonemode \fR=\fPstr
736Accepted values are:
737.RS
738.RS
739.TP
740.B none
741The \fBzonerange\fR, \fBzonesize\fR and \fBzoneskip\fR parameters are ignored.
742.TP
743.B strided
744I/O happens in a single zone until \fBzonesize\fR bytes have been transferred.
745After that number of bytes has been transferred processing of the next zone
746starts.
747.TP
748.B zbd
749Zoned block device mode. I/O happens sequentially in each zone, even if random
750I/O has been selected. Random I/O happens across all zones instead of being
751restricted to a single zone.
752.RE
753.RE
523bad63
TK
754.TP
755.BI zonerange \fR=\fPint
7b865a2f 756Size of a single zone. See also \fBzonesize\fR and \fBzoneskip\fR.
5faddc64
BVA
757.TP
758.BI zonesize \fR=\fPint
7b865a2f
BVA
759For \fBzonemode\fR=strided, this is the number of bytes to transfer before
760skipping \fBzoneskip\fR bytes. If this parameter is smaller than
761\fBzonerange\fR then only a fraction of each zone with \fBzonerange\fR bytes
762will be accessed. If this parameter is larger than \fBzonerange\fR then each
763zone will be accessed multiple times before skipping to the next zone.
764
765For \fBzonemode\fR=zbd, this is the size of a single zone. The \fBzonerange\fR
766parameter is ignored in this mode.
523bad63
TK
767.TP
768.BI zoneskip \fR=\fPint
7b865a2f
BVA
769For \fBzonemode\fR=strided, the number of bytes to skip after \fBzonesize\fR
770bytes of data have been transferred. This parameter must be zero for
771\fBzonemode\fR=zbd.
5faddc64 772
bfbdd35b
BVA
773.TP
774.BI read_beyond_wp \fR=\fPbool
775This parameter applies to \fBzonemode=zbd\fR only.
776
777Zoned block devices are block devices that consist of multiple zones. Each
778zone has a type, e.g. conventional or sequential. A conventional zone can be
779written at any offset that is a multiple of the block size. Sequential zones
780must be written sequentially. The position at which a write must occur is
781called the write pointer. A zoned block device can be either drive
782managed, host managed or host aware. For host managed devices the host must
783ensure that writes happen sequentially. Fio recognizes host managed devices
784and serializes writes to sequential zones for these devices.
785
786If a read occurs in a sequential zone beyond the write pointer then the zoned
787block device will complete the read without reading any data from the storage
788medium. Since such reads lead to unrealistically high bandwidth and IOPS
789numbers fio only reads beyond the write pointer if explicitly told to do
790so. Default: false.
59b07544
BVA
791.TP
792.BI max_open_zones \fR=\fPint
793When running a random write test across an entire drive many more zones will be
794open than in a typical application workload. Hence this command line option
795that allows to limit the number of open zones. The number of open zones is
796defined as the number of zones to which write commands are issued.
a7c2b6fc
BVA
797.TP
798.BI zone_reset_threshold \fR=\fPfloat
799A number between zero and one that indicates the ratio of logical blocks with
800data to the total number of logical blocks in the test above which zones
801should be reset periodically.
802.TP
803.BI zone_reset_frequency \fR=\fPfloat
804A number between zero and one that indicates how often a zone reset should be
805issued if the zone reset threshold has been exceeded. A zone reset is
806submitted after each (1 / zone_reset_frequency) write requests. This and the
807previous parameter can be used to simulate garbage collection activity.
bfbdd35b 808
523bad63
TK
809.SS "I/O type"
810.TP
811.BI direct \fR=\fPbool
812If value is true, use non\-buffered I/O. This is usually O_DIRECT. Note that
8e889110 813OpenBSD and ZFS on Solaris don't support direct I/O. On Windows the synchronous
523bad63
TK
814ioengines don't support direct I/O. Default: false.
815.TP
816.BI atomic \fR=\fPbool
817If value is true, attempt to use atomic direct I/O. Atomic writes are
818guaranteed to be stable once acknowledged by the operating system. Only
819Linux supports O_ATOMIC right now.
820.TP
821.BI buffered \fR=\fPbool
822If value is true, use buffered I/O. This is the opposite of the
823\fBdirect\fR option. Defaults to true.
d60e92d1
AC
824.TP
825.BI readwrite \fR=\fPstr "\fR,\fP rw" \fR=\fPstr
523bad63 826Type of I/O pattern. Accepted values are:
d60e92d1
AC
827.RS
828.RS
829.TP
830.B read
d1429b5c 831Sequential reads.
d60e92d1
AC
832.TP
833.B write
d1429b5c 834Sequential writes.
d60e92d1 835.TP
fa769d44 836.B trim
3740cfc8 837Sequential trims (Linux block devices and SCSI character devices only).
fa769d44 838.TP
d60e92d1 839.B randread
d1429b5c 840Random reads.
d60e92d1
AC
841.TP
842.B randwrite
d1429b5c 843Random writes.
d60e92d1 844.TP
fa769d44 845.B randtrim
3740cfc8 846Random trims (Linux block devices and SCSI character devices only).
fa769d44 847.TP
523bad63
TK
848.B rw,readwrite
849Sequential mixed reads and writes.
d60e92d1 850.TP
ff6bb260 851.B randrw
523bad63 852Random mixed reads and writes.
82a90686
JA
853.TP
854.B trimwrite
523bad63
TK
855Sequential trim+write sequences. Blocks will be trimmed first,
856then the same blocks will be written to.
d60e92d1
AC
857.RE
858.P
523bad63
TK
859Fio defaults to read if the option is not specified. For the mixed I/O
860types, the default is to split them 50/50. For certain types of I/O the
861result may still be skewed a bit, since the speed may be different.
862.P
863It is possible to specify the number of I/Os to do before getting a new
864offset by appending `:<nr>' to the end of the string given. For a
865random read, it would look like `rw=randread:8' for passing in an offset
866modifier with a value of 8. If the suffix is used with a sequential I/O
867pattern, then the `<nr>' value specified will be added to the generated
868offset for each I/O turning sequential I/O into sequential I/O with holes.
869For instance, using `rw=write:4k' will skip 4k for every write. Also see
870the \fBrw_sequencer\fR option.
d60e92d1
AC
871.RE
872.TP
38dad62d 873.BI rw_sequencer \fR=\fPstr
523bad63
TK
874If an offset modifier is given by appending a number to the `rw=\fIstr\fR'
875line, then this option controls how that number modifies the I/O offset
876being generated. Accepted values are:
38dad62d
JA
877.RS
878.RS
879.TP
880.B sequential
523bad63 881Generate sequential offset.
38dad62d
JA
882.TP
883.B identical
523bad63 884Generate the same offset.
38dad62d
JA
885.RE
886.P
523bad63
TK
887\fBsequential\fR is only useful for random I/O, where fio would normally
888generate a new random offset for every I/O. If you append e.g. 8 to randread,
889you would get a new random offset for every 8 I/Os. The result would be a
890seek for only every 8 I/Os, instead of for every I/O. Use `rw=randread:8'
891to specify that. As sequential I/O is already sequential, setting
892\fBsequential\fR for that would not result in any differences. \fBidentical\fR
893behaves in a similar fashion, except it sends the same offset 8 number of
894times before generating a new offset.
38dad62d 895.RE
90fef2d1 896.TP
771e58be
JA
897.BI unified_rw_reporting \fR=\fPbool
898Fio normally reports statistics on a per data direction basis, meaning that
523bad63
TK
899reads, writes, and trims are accounted and reported separately. If this
900option is set fio sums the results and report them as "mixed" instead.
771e58be 901.TP
d60e92d1 902.BI randrepeat \fR=\fPbool
523bad63
TK
903Seed the random number generator used for random I/O patterns in a
904predictable way so the pattern is repeatable across runs. Default: true.
56e2a5fc
CE
905.TP
906.BI allrandrepeat \fR=\fPbool
907Seed all random number generators in a predictable way so results are
523bad63 908repeatable across runs. Default: false.
d60e92d1 909.TP
04778baf
JA
910.BI randseed \fR=\fPint
911Seed the random number generators based on this seed value, to be able to
912control what sequence of output is being generated. If not set, the random
913sequence depends on the \fBrandrepeat\fR setting.
914.TP
a596f047 915.BI fallocate \fR=\fPstr
523bad63
TK
916Whether pre\-allocation is performed when laying down files.
917Accepted values are:
a596f047
EG
918.RS
919.RS
920.TP
921.B none
523bad63 922Do not pre\-allocate space.
a596f047 923.TP
2c3e17be 924.B native
523bad63
TK
925Use a platform's native pre\-allocation call but fall back to
926\fBnone\fR behavior if it fails/is not implemented.
2c3e17be 927.TP
a596f047 928.B posix
523bad63 929Pre\-allocate via \fBposix_fallocate\fR\|(3).
a596f047
EG
930.TP
931.B keep
523bad63
TK
932Pre\-allocate via \fBfallocate\fR\|(2) with
933FALLOC_FL_KEEP_SIZE set.
a596f047
EG
934.TP
935.B 0
523bad63 936Backward\-compatible alias for \fBnone\fR.
a596f047
EG
937.TP
938.B 1
523bad63 939Backward\-compatible alias for \fBposix\fR.
a596f047
EG
940.RE
941.P
523bad63
TK
942May not be available on all supported platforms. \fBkeep\fR is only available
943on Linux. If using ZFS on Solaris this cannot be set to \fBposix\fR
944because ZFS doesn't support pre\-allocation. Default: \fBnative\fR if any
945pre\-allocation methods are available, \fBnone\fR if not.
a596f047 946.RE
7bc8c2cf 947.TP
ecb2083d 948.BI fadvise_hint \fR=\fPstr
c712c97a
JA
949Use \fBposix_fadvise\fR\|(2) or \fBposix_madvise\fR\|(2) to advise the kernel
950what I/O patterns are likely to be issued. Accepted values are:
ecb2083d
JA
951.RS
952.RS
953.TP
954.B 0
955Backwards compatible hint for "no hint".
956.TP
957.B 1
958Backwards compatible hint for "advise with fio workload type". This
523bad63 959uses FADV_RANDOM for a random workload, and FADV_SEQUENTIAL
ecb2083d
JA
960for a sequential workload.
961.TP
962.B sequential
523bad63 963Advise using FADV_SEQUENTIAL.
ecb2083d
JA
964.TP
965.B random
523bad63 966Advise using FADV_RANDOM.
ecb2083d
JA
967.RE
968.RE
d60e92d1 969.TP
8f4b9f24 970.BI write_hint \fR=\fPstr
523bad63
TK
971Use \fBfcntl\fR\|(2) to advise the kernel what life time to expect
972from a write. Only supported on Linux, as of version 4.13. Accepted
8f4b9f24
JA
973values are:
974.RS
975.RS
976.TP
977.B none
978No particular life time associated with this file.
979.TP
980.B short
981Data written to this file has a short life time.
982.TP
983.B medium
984Data written to this file has a medium life time.
985.TP
986.B long
987Data written to this file has a long life time.
988.TP
989.B extreme
990Data written to this file has a very long life time.
991.RE
523bad63
TK
992.P
993The values are all relative to each other, and no absolute meaning
994should be associated with them.
8f4b9f24 995.RE
37659335 996.TP
523bad63
TK
997.BI offset \fR=\fPint
998Start I/O at the provided offset in the file, given as either a fixed size in
83c8b093
JF
999bytes or a percentage. If a percentage is given, the generated offset will be
1000aligned to the minimum \fBblocksize\fR or to the value of \fBoffset_align\fR if
1001provided. Data before the given offset will not be touched. This
523bad63
TK
1002effectively caps the file size at `real_size \- offset'. Can be combined with
1003\fBsize\fR to constrain the start and end range of the I/O workload.
1004A percentage can be specified by a number between 1 and 100 followed by '%',
1005for example, `offset=20%' to specify 20%.
6d500c2e 1006.TP
83c8b093
JF
1007.BI offset_align \fR=\fPint
1008If set to non-zero value, the byte offset generated by a percentage \fBoffset\fR
1009is aligned upwards to this value. Defaults to 0 meaning that a percentage
1010offset is aligned to the minimum block size.
1011.TP
523bad63
TK
1012.BI offset_increment \fR=\fPint
1013If this is provided, then the real offset becomes `\fBoffset\fR + \fBoffset_increment\fR
1014* thread_number', where the thread number is a counter that starts at 0 and
1015is incremented for each sub\-job (i.e. when \fBnumjobs\fR option is
1016specified). This option is useful if there are several jobs which are
1017intended to operate on a file in parallel disjoint segments, with even
1018spacing between the starting points.
6d500c2e 1019.TP
523bad63
TK
1020.BI number_ios \fR=\fPint
1021Fio will normally perform I/Os until it has exhausted the size of the region
1022set by \fBsize\fR, or if it exhaust the allocated time (or hits an error
1023condition). With this setting, the range/size can be set independently of
1024the number of I/Os to perform. When fio reaches this number, it will exit
1025normally and report status. Note that this does not extend the amount of I/O
1026that will be done, it will only stop fio if this condition is met before
1027other end\-of\-job criteria.
d60e92d1 1028.TP
523bad63
TK
1029.BI fsync \fR=\fPint
1030If writing to a file, issue an \fBfsync\fR\|(2) (or its equivalent) of
1031the dirty data for every number of blocks given. For example, if you give 32
1032as a parameter, fio will sync the file after every 32 writes issued. If fio is
1033using non\-buffered I/O, we may not sync the file. The exception is the sg
1034I/O engine, which synchronizes the disk cache anyway. Defaults to 0, which
1035means fio does not periodically issue and wait for a sync to complete. Also
1036see \fBend_fsync\fR and \fBfsync_on_close\fR.
6d500c2e 1037.TP
523bad63
TK
1038.BI fdatasync \fR=\fPint
1039Like \fBfsync\fR but uses \fBfdatasync\fR\|(2) to only sync data and
1040not metadata blocks. In Windows, FreeBSD, and DragonFlyBSD there is no
1041\fBfdatasync\fR\|(2) so this falls back to using \fBfsync\fR\|(2).
1042Defaults to 0, which means fio does not periodically issue and wait for a
1043data\-only sync to complete.
d60e92d1 1044.TP
523bad63
TK
1045.BI write_barrier \fR=\fPint
1046Make every N\-th write a barrier write.
901bb994 1047.TP
523bad63
TK
1048.BI sync_file_range \fR=\fPstr:int
1049Use \fBsync_file_range\fR\|(2) for every \fIint\fR number of write
1050operations. Fio will track range of writes that have happened since the last
1051\fBsync_file_range\fR\|(2) call. \fIstr\fR can currently be one or more of:
1052.RS
1053.RS
fd68418e 1054.TP
523bad63
TK
1055.B wait_before
1056SYNC_FILE_RANGE_WAIT_BEFORE
c5751c62 1057.TP
523bad63
TK
1058.B write
1059SYNC_FILE_RANGE_WRITE
c5751c62 1060.TP
523bad63
TK
1061.B wait_after
1062SYNC_FILE_RANGE_WRITE_AFTER
2fa5a241 1063.RE
523bad63
TK
1064.P
1065So if you do `sync_file_range=wait_before,write:8', fio would use
1066`SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE' for every 8
1067writes. Also see the \fBsync_file_range\fR\|(2) man page. This option is
1068Linux specific.
2fa5a241 1069.RE
ce35b1ec 1070.TP
523bad63
TK
1071.BI overwrite \fR=\fPbool
1072If true, writes to a file will always overwrite existing data. If the file
1073doesn't already exist, it will be created before the write phase begins. If
1074the file exists and is large enough for the specified write phase, nothing
1075will be done. Default: false.
5c94b008 1076.TP
523bad63
TK
1077.BI end_fsync \fR=\fPbool
1078If true, \fBfsync\fR\|(2) file contents when a write stage has completed.
1079Default: false.
d60e92d1 1080.TP
523bad63
TK
1081.BI fsync_on_close \fR=\fPbool
1082If true, fio will \fBfsync\fR\|(2) a dirty file on close. This differs
1083from \fBend_fsync\fR in that it will happen on every file close, not
1084just at the end of the job. Default: false.
d60e92d1 1085.TP
523bad63
TK
1086.BI rwmixread \fR=\fPint
1087Percentage of a mixed workload that should be reads. Default: 50.
1088.TP
1089.BI rwmixwrite \fR=\fPint
1090Percentage of a mixed workload that should be writes. If both
1091\fBrwmixread\fR and \fBrwmixwrite\fR is given and the values do not
1092add up to 100%, the latter of the two will be used to override the
1093first. This may interfere with a given rate setting, if fio is asked to
1094limit reads or writes to a certain rate. If that is the case, then the
1095distribution may be skewed. Default: 50.
1096.TP
1097.BI random_distribution \fR=\fPstr:float[,str:float][,str:float]
1098By default, fio will use a completely uniform random distribution when asked
1099to perform random I/O. Sometimes it is useful to skew the distribution in
1100specific ways, ensuring that some parts of the data is more hot than others.
1101fio includes the following distribution models:
d60e92d1
AC
1102.RS
1103.RS
1104.TP
1105.B random
523bad63 1106Uniform random distribution
8c07860d
JA
1107.TP
1108.B zipf
523bad63 1109Zipf distribution
8c07860d
JA
1110.TP
1111.B pareto
523bad63 1112Pareto distribution
8c07860d 1113.TP
dd3503d3 1114.B normal
523bad63 1115Normal (Gaussian) distribution
dd3503d3 1116.TP
523bad63
TK
1117.B zoned
1118Zoned random distribution
59466396
JA
1119.B zoned_abs
1120Zoned absolute random distribution
d60e92d1
AC
1121.RE
1122.P
523bad63
TK
1123When using a \fBzipf\fR or \fBpareto\fR distribution, an input value is also
1124needed to define the access pattern. For \fBzipf\fR, this is the `Zipf theta'.
1125For \fBpareto\fR, it's the `Pareto power'. Fio includes a test
1126program, \fBfio\-genzipf\fR, that can be used visualize what the given input
1127values will yield in terms of hit rates. If you wanted to use \fBzipf\fR with
1128a `theta' of 1.2, you would use `random_distribution=zipf:1.2' as the
1129option. If a non\-uniform model is used, fio will disable use of the random
1130map. For the \fBnormal\fR distribution, a normal (Gaussian) deviation is
1131supplied as a value between 0 and 100.
1132.P
1133For a \fBzoned\fR distribution, fio supports specifying percentages of I/O
1134access that should fall within what range of the file or device. For
1135example, given a criteria of:
d60e92d1 1136.RS
523bad63
TK
1137.P
1138.PD 0
113960% of accesses should be to the first 10%
1140.P
114130% of accesses should be to the next 20%
1142.P
11438% of accesses should be to the next 30%
1144.P
11452% of accesses should be to the next 40%
1146.PD
1147.RE
1148.P
1149we can define that through zoning of the random accesses. For the above
1150example, the user would do:
1151.RS
1152.P
1153random_distribution=zoned:60/10:30/20:8/30:2/40
1154.RE
1155.P
59466396
JA
1156A \fBzoned_abs\fR distribution works exactly like the\fBzoned\fR, except that
1157it takes absolute sizes. For example, let's say you wanted to define access
1158according to the following criteria:
1159.RS
1160.P
1161.PD 0
116260% of accesses should be to the first 20G
1163.P
116430% of accesses should be to the next 100G
1165.P
116610% of accesses should be to the next 500G
1167.PD
1168.RE
1169.P
1170we can define an absolute zoning distribution with:
1171.RS
1172.P
1173random_distribution=zoned:60/10:30/20:8/30:2/40
1174.RE
1175.P
6a16ece8
JA
1176For both \fBzoned\fR and \fBzoned_abs\fR, fio supports defining up to 256
1177separate zones.
1178.P
59466396 1179Similarly to how \fBbssplit\fR works for setting ranges and percentages
523bad63
TK
1180of block sizes. Like \fBbssplit\fR, it's possible to specify separate
1181zones for reads, writes, and trims. If just one set is given, it'll apply to
1182all of them.
1183.RE
1184.TP
1185.BI percentage_random \fR=\fPint[,int][,int]
1186For a random workload, set how big a percentage should be random. This
1187defaults to 100%, in which case the workload is fully random. It can be set
1188from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1189sequential. Any setting in between will result in a random mix of sequential
1190and random I/O, at the given percentages. Comma\-separated values may be
1191specified for reads, writes, and trims as described in \fBblocksize\fR.
1192.TP
1193.BI norandommap
1194Normally fio will cover every block of the file when doing random I/O. If
1195this option is given, fio will just get a new random offset without looking
1196at past I/O history. This means that some blocks may not be read or written,
1197and that some blocks may be read/written more than once. If this option is
1198used with \fBverify\fR and multiple blocksizes (via \fBbsrange\fR),
1199only intact blocks are verified, i.e., partially\-overwritten blocks are
47e6a6e5
SW
1200ignored. With an async I/O engine and an I/O depth > 1, it is possible for
1201the same block to be overwritten, which can cause verification errors. Either
1202do not use norandommap in this case, or also use the lfsr random generator.
523bad63
TK
1203.TP
1204.BI softrandommap \fR=\fPbool
1205See \fBnorandommap\fR. If fio runs with the random block map enabled and
1206it fails to allocate the map, if this option is set it will continue without
1207a random block map. As coverage will not be as complete as with random maps,
1208this option is disabled by default.
1209.TP
1210.BI random_generator \fR=\fPstr
1211Fio supports the following engines for generating I/O offsets for random I/O:
1212.RS
1213.RS
1214.TP
1215.B tausworthe
1216Strong 2^88 cycle random number generator.
1217.TP
1218.B lfsr
1219Linear feedback shift register generator.
1220.TP
1221.B tausworthe64
1222Strong 64\-bit 2^258 cycle random number generator.
1223.RE
1224.P
1225\fBtausworthe\fR is a strong random number generator, but it requires tracking
1226on the side if we want to ensure that blocks are only read or written
1227once. \fBlfsr\fR guarantees that we never generate the same offset twice, and
1228it's also less computationally expensive. It's not a true random generator,
1229however, though for I/O purposes it's typically good enough. \fBlfsr\fR only
1230works with single block sizes, not with workloads that use multiple block
1231sizes. If used with such a workload, fio may read or write some blocks
1232multiple times. The default value is \fBtausworthe\fR, unless the required
1233space exceeds 2^32 blocks. If it does, then \fBtausworthe64\fR is
1234selected automatically.
1235.RE
1236.SS "Block size"
1237.TP
1238.BI blocksize \fR=\fPint[,int][,int] "\fR,\fB bs" \fR=\fPint[,int][,int]
1239The block size in bytes used for I/O units. Default: 4096. A single value
1240applies to reads, writes, and trims. Comma\-separated values may be
1241specified for reads, writes, and trims. A value not terminated in a comma
1242applies to subsequent types. Examples:
1243.RS
1244.RS
1245.P
1246.PD 0
1247bs=256k means 256k for reads, writes and trims.
1248.P
1249bs=8k,32k means 8k for reads, 32k for writes and trims.
1250.P
1251bs=8k,32k, means 8k for reads, 32k for writes, and default for trims.
1252.P
1253bs=,8k means default for reads, 8k for writes and trims.
1254.P
1255bs=,8k, means default for reads, 8k for writes, and default for trims.
1256.PD
1257.RE
1258.RE
1259.TP
1260.BI blocksize_range \fR=\fPirange[,irange][,irange] "\fR,\fB bsrange" \fR=\fPirange[,irange][,irange]
1261A range of block sizes in bytes for I/O units. The issued I/O unit will
1262always be a multiple of the minimum size, unless
1263\fBblocksize_unaligned\fR is set.
1264Comma\-separated ranges may be specified for reads, writes, and trims as
1265described in \fBblocksize\fR. Example:
1266.RS
1267.RS
1268.P
1269bsrange=1k\-4k,2k\-8k
1270.RE
1271.RE
1272.TP
1273.BI bssplit \fR=\fPstr[,str][,str]
1274Sometimes you want even finer grained control of the block sizes issued, not
1275just an even split between them. This option allows you to weight various
1276block sizes, so that you are able to define a specific amount of block sizes
1277issued. The format for this option is:
1278.RS
1279.RS
1280.P
1281bssplit=blocksize/percentage:blocksize/percentage
1282.RE
1283.P
1284for as many block sizes as needed. So if you want to define a workload that
1285has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would write:
1286.RS
1287.P
1288bssplit=4k/10:64k/50:32k/40
1289.RE
1290.P
1291Ordering does not matter. If the percentage is left blank, fio will fill in
1292the remaining values evenly. So a bssplit option like this one:
1293.RS
1294.P
1295bssplit=4k/50:1k/:32k/
1296.RE
1297.P
1298would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always add up
1299to 100, if bssplit is given a range that adds up to more, it will error out.
1300.P
1301Comma\-separated values may be specified for reads, writes, and trims as
1302described in \fBblocksize\fR.
1303.P
1304If you want a workload that has 50% 2k reads and 50% 4k reads, while having
130590% 4k writes and 10% 8k writes, you would specify:
1306.RS
1307.P
cf04b906 1308bssplit=2k/50:4k/50,4k/90:8k/10
523bad63 1309.RE
6a16ece8
JA
1310.P
1311Fio supports defining up to 64 different weights for each data direction.
523bad63
TK
1312.RE
1313.TP
1314.BI blocksize_unaligned "\fR,\fB bs_unaligned"
1315If set, fio will issue I/O units with any size within
1316\fBblocksize_range\fR, not just multiples of the minimum size. This
1317typically won't work with direct I/O, as that normally requires sector
1318alignment.
1319.TP
1320.BI bs_is_seq_rand \fR=\fPbool
1321If this option is set, fio will use the normal read,write blocksize settings
1322as sequential,random blocksize settings instead. Any random read or write
1323will use the WRITE blocksize settings, and any sequential read or write will
1324use the READ blocksize settings.
1325.TP
1326.BI blockalign \fR=\fPint[,int][,int] "\fR,\fB ba" \fR=\fPint[,int][,int]
1327Boundary to which fio will align random I/O units. Default:
1328\fBblocksize\fR. Minimum alignment is typically 512b for using direct
1329I/O, though it usually depends on the hardware block size. This option is
1330mutually exclusive with using a random map for files, so it will turn off
1331that option. Comma\-separated values may be specified for reads, writes, and
1332trims as described in \fBblocksize\fR.
1333.SS "Buffers and memory"
1334.TP
1335.BI zero_buffers
1336Initialize buffers with all zeros. Default: fill buffers with random data.
1337.TP
1338.BI refill_buffers
1339If this option is given, fio will refill the I/O buffers on every
1340submit. The default is to only fill it at init time and reuse that
1341data. Only makes sense if zero_buffers isn't specified, naturally. If data
1342verification is enabled, \fBrefill_buffers\fR is also automatically enabled.
1343.TP
1344.BI scramble_buffers \fR=\fPbool
1345If \fBrefill_buffers\fR is too costly and the target is using data
1346deduplication, then setting this option will slightly modify the I/O buffer
1347contents to defeat normal de\-dupe attempts. This is not enough to defeat
1348more clever block compression attempts, but it will stop naive dedupe of
1349blocks. Default: true.
1350.TP
1351.BI buffer_compress_percentage \fR=\fPint
72592780
SW
1352If this is set, then fio will attempt to provide I/O buffer content
1353(on WRITEs) that compresses to the specified level. Fio does this by
1354providing a mix of random data followed by fixed pattern data. The
1355fixed pattern is either zeros, or the pattern specified by
1356\fBbuffer_pattern\fR. If the \fBbuffer_pattern\fR option is used, it
1357might skew the compression ratio slightly. Setting
1358\fBbuffer_compress_percentage\fR to a value other than 100 will also
1359enable \fBrefill_buffers\fR in order to reduce the likelihood that
1360adjacent blocks are so similar that they over compress when seen
1361together. See \fBbuffer_compress_chunk\fR for how to set a finer or
1362coarser granularity of the random/fixed data regions. Defaults to unset
1363i.e., buffer data will not adhere to any compression level.
523bad63
TK
1364.TP
1365.BI buffer_compress_chunk \fR=\fPint
72592780
SW
1366This setting allows fio to manage how big the random/fixed data region
1367is when using \fBbuffer_compress_percentage\fR. When
1368\fBbuffer_compress_chunk\fR is set to some non-zero value smaller than the
1369block size, fio can repeat the random/fixed region throughout the I/O
1370buffer at the specified interval (which particularly useful when
1371bigger block sizes are used for a job). When set to 0, fio will use a
1372chunk size that matches the block size resulting in a single
1373random/fixed region within the I/O buffer. Defaults to 512. When the
1374unit is omitted, the value is interpreted in bytes.
523bad63
TK
1375.TP
1376.BI buffer_pattern \fR=\fPstr
1377If set, fio will fill the I/O buffers with this pattern or with the contents
1378of a file. If not set, the contents of I/O buffers are defined by the other
1379options related to buffer contents. The setting can be any pattern of bytes,
1380and can be prefixed with 0x for hex values. It may also be a string, where
1381the string must then be wrapped with "". Or it may also be a filename,
1382where the filename must be wrapped with '' in which case the file is
1383opened and read. Note that not all the file contents will be read if that
1384would cause the buffers to overflow. So, for example:
1385.RS
1386.RS
1387.P
1388.PD 0
1389buffer_pattern='filename'
1390.P
1391or:
1392.P
1393buffer_pattern="abcd"
1394.P
1395or:
1396.P
1397buffer_pattern=\-12
1398.P
1399or:
1400.P
1401buffer_pattern=0xdeadface
1402.PD
1403.RE
1404.P
1405Also you can combine everything together in any order:
1406.RS
1407.P
1408buffer_pattern=0xdeadface"abcd"\-12'filename'
1409.RE
1410.RE
1411.TP
1412.BI dedupe_percentage \fR=\fPint
1413If set, fio will generate this percentage of identical buffers when
1414writing. These buffers will be naturally dedupable. The contents of the
1415buffers depend on what other buffer compression settings have been set. It's
1416possible to have the individual buffers either fully compressible, or not at
72592780
SW
1417all \-\- this option only controls the distribution of unique buffers. Setting
1418this option will also enable \fBrefill_buffers\fR to prevent every buffer
1419being identical.
523bad63
TK
1420.TP
1421.BI invalidate \fR=\fPbool
1422Invalidate the buffer/page cache parts of the files to be used prior to
1423starting I/O if the platform and file type support it. Defaults to true.
1424This will be ignored if \fBpre_read\fR is also specified for the
1425same job.
1426.TP
1427.BI sync \fR=\fPbool
1428Use synchronous I/O for buffered writes. For the majority of I/O engines,
1429this means using O_SYNC. Default: false.
1430.TP
1431.BI iomem \fR=\fPstr "\fR,\fP mem" \fR=\fPstr
1432Fio can use various types of memory as the I/O unit buffer. The allowed
1433values are:
1434.RS
1435.RS
1436.TP
1437.B malloc
1438Use memory from \fBmalloc\fR\|(3) as the buffers. Default memory type.
1439.TP
1440.B shm
1441Use shared memory as the buffers. Allocated through \fBshmget\fR\|(2).
1442.TP
1443.B shmhuge
1444Same as \fBshm\fR, but use huge pages as backing.
1445.TP
1446.B mmap
1447Use \fBmmap\fR\|(2) to allocate buffers. May either be anonymous memory, or can
1448be file backed if a filename is given after the option. The format
1449is `mem=mmap:/path/to/file'.
1450.TP
1451.B mmaphuge
1452Use a memory mapped huge file as the buffer backing. Append filename
1453after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file'.
1454.TP
1455.B mmapshared
1456Same as \fBmmap\fR, but use a MMAP_SHARED mapping.
1457.TP
1458.B cudamalloc
1459Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1460The \fBioengine\fR must be \fBrdma\fR.
1461.RE
1462.P
1463The area allocated is a function of the maximum allowed bs size for the job,
1464multiplied by the I/O depth given. Note that for \fBshmhuge\fR and
1465\fBmmaphuge\fR to work, the system must have free huge pages allocated. This
1466can normally be checked and set by reading/writing
1467`/proc/sys/vm/nr_hugepages' on a Linux system. Fio assumes a huge page
1468is 4MiB in size. So to calculate the number of huge pages you need for a
1469given job file, add up the I/O depth of all jobs (normally one unless
1470\fBiodepth\fR is used) and multiply by the maximum bs set. Then divide
1471that number by the huge page size. You can see the size of the huge pages in
1472`/proc/meminfo'. If no huge pages are allocated by having a non\-zero
1473number in `nr_hugepages', using \fBmmaphuge\fR or \fBshmhuge\fR will fail. Also
1474see \fBhugepage\-size\fR.
1475.P
1476\fBmmaphuge\fR also needs to have hugetlbfs mounted and the file location
1477should point there. So if it's mounted in `/huge', you would use
1478`mem=mmaphuge:/huge/somefile'.
1479.RE
1480.TP
1481.BI iomem_align \fR=\fPint "\fR,\fP mem_align" \fR=\fPint
1482This indicates the memory alignment of the I/O memory buffers. Note that
1483the given alignment is applied to the first I/O unit buffer, if using
1484\fBiodepth\fR the alignment of the following buffers are given by the
1485\fBbs\fR used. In other words, if using a \fBbs\fR that is a
1486multiple of the page sized in the system, all buffers will be aligned to
1487this value. If using a \fBbs\fR that is not page aligned, the alignment
1488of subsequent I/O memory buffers is the sum of the \fBiomem_align\fR and
1489\fBbs\fR used.
1490.TP
1491.BI hugepage\-size \fR=\fPint
1492Defines the size of a huge page. Must at least be equal to the system
1493setting, see `/proc/meminfo'. Defaults to 4MiB. Should probably
1494always be a multiple of megabytes, so using `hugepage\-size=Xm' is the
1495preferred way to set this to avoid setting a non\-pow\-2 bad value.
1496.TP
1497.BI lockmem \fR=\fPint
1498Pin the specified amount of memory with \fBmlock\fR\|(2). Can be used to
1499simulate a smaller amount of memory. The amount specified is per worker.
1500.SS "I/O size"
1501.TP
1502.BI size \fR=\fPint
1503The total size of file I/O for each thread of this job. Fio will run until
1504this many bytes has been transferred, unless runtime is limited by other options
1505(such as \fBruntime\fR, for instance, or increased/decreased by \fBio_size\fR).
1506Fio will divide this size between the available files determined by options
1507such as \fBnrfiles\fR, \fBfilename\fR, unless \fBfilesize\fR is
1508specified by the job. If the result of division happens to be 0, the size is
1509set to the physical size of the given files or devices if they exist.
1510If this option is not specified, fio will use the full size of the given
1511files or devices. If the files do not exist, size must be given. It is also
1512possible to give size as a percentage between 1 and 100. If `size=20%' is
1513given, fio will use 20% of the full size of the given files or devices.
1514Can be combined with \fBoffset\fR to constrain the start and end range
1515that I/O will be done within.
1516.TP
1517.BI io_size \fR=\fPint "\fR,\fB io_limit" \fR=\fPint
1518Normally fio operates within the region set by \fBsize\fR, which means
1519that the \fBsize\fR option sets both the region and size of I/O to be
1520performed. Sometimes that is not what you want. With this option, it is
1521possible to define just the amount of I/O that fio should do. For instance,
1522if \fBsize\fR is set to 20GiB and \fBio_size\fR is set to 5GiB, fio
1523will perform I/O within the first 20GiB but exit when 5GiB have been
1524done. The opposite is also possible \-\- if \fBsize\fR is set to 20GiB,
1525and \fBio_size\fR is set to 40GiB, then fio will do 40GiB of I/O within
1526the 0..20GiB region.
1527.TP
1528.BI filesize \fR=\fPirange(int)
1529Individual file sizes. May be a range, in which case fio will select sizes
1530for files at random within the given range and limited to \fBsize\fR in
1531total (if that is given). If not given, each created file is the same size.
1532This option overrides \fBsize\fR in terms of file size, which means
1533this value is used as a fixed size or possible range of each file.
1534.TP
1535.BI file_append \fR=\fPbool
1536Perform I/O after the end of the file. Normally fio will operate within the
1537size of a file. If this option is set, then fio will append to the file
1538instead. This has identical behavior to setting \fBoffset\fR to the size
1539of a file. This option is ignored on non\-regular files.
1540.TP
1541.BI fill_device \fR=\fPbool "\fR,\fB fill_fs" \fR=\fPbool
1542Sets size to something really large and waits for ENOSPC (no space left on
1543device) as the terminating condition. Only makes sense with sequential
1544write. For a read workload, the mount point will be filled first then I/O
1545started on the result. This option doesn't make sense if operating on a raw
1546device node, since the size of that is already known by the file system.
1547Additionally, writing beyond end\-of\-device will not return ENOSPC there.
1548.SS "I/O engine"
1549.TP
1550.BI ioengine \fR=\fPstr
1551Defines how the job issues I/O to the file. The following types are defined:
1552.RS
1553.RS
1554.TP
1555.B sync
1556Basic \fBread\fR\|(2) or \fBwrite\fR\|(2)
1557I/O. \fBlseek\fR\|(2) is used to position the I/O location.
1558See \fBfsync\fR and \fBfdatasync\fR for syncing write I/Os.
1559.TP
1560.B psync
1561Basic \fBpread\fR\|(2) or \fBpwrite\fR\|(2) I/O. Default on
1562all supported operating systems except for Windows.
1563.TP
1564.B vsync
1565Basic \fBreadv\fR\|(2) or \fBwritev\fR\|(2) I/O. Will emulate
1566queuing by coalescing adjacent I/Os into a single submission.
1567.TP
1568.B pvsync
1569Basic \fBpreadv\fR\|(2) or \fBpwritev\fR\|(2) I/O.
a46c5e01 1570.TP
2cafffbe
JA
1571.B pvsync2
1572Basic \fBpreadv2\fR\|(2) or \fBpwritev2\fR\|(2) I/O.
1573.TP
d60e92d1 1574.B libaio
523bad63
TK
1575Linux native asynchronous I/O. Note that Linux may only support
1576queued behavior with non\-buffered I/O (set `direct=1' or
1577`buffered=0').
1578This engine defines engine specific options.
d60e92d1
AC
1579.TP
1580.B posixaio
523bad63
TK
1581POSIX asynchronous I/O using \fBaio_read\fR\|(3) and
1582\fBaio_write\fR\|(3).
03e20d68
BC
1583.TP
1584.B solarisaio
1585Solaris native asynchronous I/O.
1586.TP
1587.B windowsaio
38f8c318 1588Windows native asynchronous I/O. Default on Windows.
d60e92d1
AC
1589.TP
1590.B mmap
523bad63
TK
1591File is memory mapped with \fBmmap\fR\|(2) and data copied
1592to/from using \fBmemcpy\fR\|(3).
d60e92d1
AC
1593.TP
1594.B splice
523bad63
TK
1595\fBsplice\fR\|(2) is used to transfer the data and
1596\fBvmsplice\fR\|(2) to transfer data from user space to the
1597kernel.
d60e92d1 1598.TP
d60e92d1 1599.B sg
523bad63
TK
1600SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1601ioctl, or if the target is an sg character device we use
1602\fBread\fR\|(2) and \fBwrite\fR\|(2) for asynchronous
1603I/O. Requires \fBfilename\fR option to specify either block or
3740cfc8
VF
1604character devices. This engine supports trim operations. The
1605sg engine includes engine specific options.
d60e92d1
AC
1606.TP
1607.B null
523bad63
TK
1608Doesn't transfer any data, just pretends to. This is mainly used to
1609exercise fio itself and for debugging/testing purposes.
d60e92d1
AC
1610.TP
1611.B net
523bad63
TK
1612Transfer over the network to given `host:port'. Depending on the
1613\fBprotocol\fR used, the \fBhostname\fR, \fBport\fR,
1614\fBlisten\fR and \fBfilename\fR options are used to specify
1615what sort of connection to make, while the \fBprotocol\fR option
1616determines which protocol will be used. This engine defines engine
1617specific options.
d60e92d1
AC
1618.TP
1619.B netsplice
523bad63
TK
1620Like \fBnet\fR, but uses \fBsplice\fR\|(2) and
1621\fBvmsplice\fR\|(2) to map data and send/receive.
1622This engine defines engine specific options.
d60e92d1 1623.TP
53aec0a4 1624.B cpuio
523bad63
TK
1625Doesn't transfer any data, but burns CPU cycles according to the
1626\fBcpuload\fR and \fBcpuchunks\fR options. Setting
1627\fBcpuload\fR\=85 will cause that job to do nothing but burn 85%
1628of the CPU. In case of SMP machines, use `numjobs=<nr_of_cpu>'
1629to get desired CPU usage, as the cpuload only loads a
1630single CPU at the desired rate. A job never finishes unless there is
1631at least one non\-cpuio job.
d60e92d1
AC
1632.TP
1633.B guasi
f1dd3fb1 1634The GUASI I/O engine is the Generic Userspace Asynchronous Syscall
523bad63
TK
1635Interface approach to async I/O. See \fIhttp://www.xmailserver.org/guasi\-lib.html\fR
1636for more info on GUASI.
d60e92d1 1637.TP
21b8aee8 1638.B rdma
523bad63
TK
1639The RDMA I/O engine supports both RDMA memory semantics
1640(RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
609ac152
SB
1641InfiniBand, RoCE and iWARP protocols. This engine defines engine
1642specific options.
d54fce84
DM
1643.TP
1644.B falloc
523bad63
TK
1645I/O engine that does regular fallocate to simulate data transfer as
1646fio ioengine.
1647.RS
1648.P
1649.PD 0
1650DDIR_READ does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1651.P
1652DIR_WRITE does fallocate(,mode = 0).
1653.P
1654DDIR_TRIM does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1655.PD
1656.RE
1657.TP
1658.B ftruncate
1659I/O engine that sends \fBftruncate\fR\|(2) operations in response
1660to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
1661size to the current block offset. \fBblocksize\fR is ignored.
d54fce84
DM
1662.TP
1663.B e4defrag
523bad63
TK
1664I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1665defragment activity in request to DDIR_WRITE event.
0d978694 1666.TP
d5f9b0ea
IF
1667.B rados
1668I/O engine supporting direct access to Ceph Reliable Autonomic Distributed
1669Object Store (RADOS) via librados. This ioengine defines engine specific
1670options.
1671.TP
0d978694 1672.B rbd
523bad63
TK
1673I/O engine supporting direct access to Ceph Rados Block Devices
1674(RBD) via librbd without the need to use the kernel rbd driver. This
1675ioengine defines engine specific options.
a7c386f4 1676.TP
c2f6a13d
LMB
1677.B http
1678I/O engine supporting GET/PUT requests over HTTP(S) with libcurl to
1679a WebDAV or S3 endpoint. This ioengine defines engine specific options.
1680
1681This engine only supports direct IO of iodepth=1; you need to scale this
1682via numjobs. blocksize defines the size of the objects to be created.
1683
1684TRIM is translated to object deletion.
1685.TP
a7c386f4 1686.B gfapi
523bad63
TK
1687Using GlusterFS libgfapi sync interface to direct access to
1688GlusterFS volumes without having to go through FUSE. This ioengine
1689defines engine specific options.
cc47f094 1690.TP
1691.B gfapi_async
523bad63
TK
1692Using GlusterFS libgfapi async interface to direct access to
1693GlusterFS volumes without having to go through FUSE. This ioengine
1694defines engine specific options.
1b10477b 1695.TP
b74e419e 1696.B libhdfs
523bad63
TK
1697Read and write through Hadoop (HDFS). The \fBfilename\fR option
1698is used to specify host,port of the hdfs name\-node to connect. This
1699engine interprets offsets a little differently. In HDFS, files once
1700created cannot be modified so random writes are not possible. To
1701imitate this the libhdfs engine expects a bunch of small files to be
1702created over HDFS and will randomly pick a file from them
1703based on the offset generated by fio backend (see the example
1704job file to create such files, use `rw=write' option). Please
1705note, it may be necessary to set environment variables to work
1706with HDFS/libhdfs properly. Each job uses its own connection to
1707HDFS.
65fa28ca
DE
1708.TP
1709.B mtd
523bad63
TK
1710Read, write and erase an MTD character device (e.g.,
1711`/dev/mtd0'). Discards are treated as erases. Depending on the
1712underlying device type, the I/O may have to go in a certain pattern,
1713e.g., on NAND, writing sequentially to erase blocks and discarding
1714before overwriting. The \fBtrimwrite\fR mode works well for this
65fa28ca 1715constraint.
5c4ef02e
JA
1716.TP
1717.B pmemblk
523bad63 1718Read and write using filesystem DAX to a file on a filesystem
363a5f65 1719mounted with DAX on a persistent memory device through the PMDK
523bad63 1720libpmemblk library.
104ee4de 1721.TP
523bad63
TK
1722.B dev\-dax
1723Read and write using device DAX to a persistent memory device (e.g.,
363a5f65 1724/dev/dax0.0) through the PMDK libpmem library.
d60e92d1 1725.TP
523bad63
TK
1726.B external
1727Prefix to specify loading an external I/O engine object file. Append
1728the engine filename, e.g. `ioengine=external:/tmp/foo.o' to load
d243fd6d
TK
1729ioengine `foo.o' in `/tmp'. The path can be either
1730absolute or relative. See `engines/skeleton_external.c' in the fio source for
1731details of writing an external I/O engine.
1216cc5a
JB
1732.TP
1733.B filecreate
b71968b1
SW
1734Simply create the files and do no I/O to them. You still need to set
1735\fBfilesize\fR so that all the accounting still occurs, but no actual I/O will be
1736done other than creating the file.
ae0db592
TI
1737.TP
1738.B libpmem
1739Read and write using mmap I/O to a file on a filesystem
363a5f65 1740mounted with DAX on a persistent memory device through the PMDK
ae0db592 1741libpmem library.
07751e10
JA
1742.TP
1743.B ime_psync
1744Synchronous read and write using DDN's Infinite Memory Engine (IME). This
1745engine is very basic and issues calls to IME whenever an IO is queued.
1746.TP
1747.B ime_psyncv
1748Synchronous read and write using DDN's Infinite Memory Engine (IME). This
1749engine uses iovecs and will try to stack as much IOs as possible (if the IOs
1750are "contiguous" and the IO depth is not exceeded) before issuing a call to IME.
1751.TP
1752.B ime_aio
1753Asynchronous read and write using DDN's Infinite Memory Engine (IME). This
1754engine will try to stack as much IOs as possible by creating requests for IME.
1755FIO will then decide when to commit these requests.
247ef2aa
KZ
1756.TP
1757.B libiscsi
1758Read and write iscsi lun with libiscsi.
523bad63
TK
1759.SS "I/O engine specific parameters"
1760In addition, there are some parameters which are only valid when a specific
1761\fBioengine\fR is in use. These are used identically to normal parameters,
1762with the caveat that when used on the command line, they must come after the
1763\fBioengine\fR that defines them is selected.
d60e92d1 1764.TP
029b42ac
JA
1765.BI (io_uring)hipri
1766If this option is set, fio will attempt to use polled IO completions. Normal IO
1767completions generate interrupts to signal the completion of IO, polled
1768completions do not. Hence they are require active reaping by the application.
1769The benefits are more efficient IO for high IOPS scenarios, and lower latencies
1770for low queue depth IO.
1771.TP
1772.BI (io_uring)fixedbufs
1773If fio is asked to do direct IO, then Linux will map pages for each IO call, and
1774release them when IO is done. If this option is set, the pages are pre-mapped
1775before IO is started. This eliminates the need to map and release for each IO.
1776This is more efficient, and reduces the IO latency as well.
1777.TP
1778.BI (io_uring)sqthread_poll
1779Normally fio will submit IO by issuing a system call to notify the kernel of
1780available items in the SQ ring. If this option is set, the act of submitting IO
1781will be done by a polling thread in the kernel. This frees up cycles for fio, at
1782the cost of using more CPU in the system.
1783.TP
1784.BI (io_uring)sqthread_poll_cpu
1785When `sqthread_poll` is set, this option provides a way to define which CPU
1786should be used for the polling thread.
1787.TP
523bad63
TK
1788.BI (libaio)userspace_reap
1789Normally, with the libaio engine in use, fio will use the
1790\fBio_getevents\fR\|(3) system call to reap newly returned events. With
1791this flag turned on, the AIO ring will be read directly from user\-space to
1792reap events. The reaping mode is only enabled when polling for a minimum of
17930 events (e.g. when `iodepth_batch_complete=0').
3ce9dcaf 1794.TP
523bad63
TK
1795.BI (pvsync2)hipri
1796Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
1797than normal.
82407585 1798.TP
523bad63
TK
1799.BI (pvsync2)hipri_percentage
1800When hipri is set this determines the probability of a pvsync2 I/O being high
1801priority. The default is 100%.
d60e92d1 1802.TP
523bad63
TK
1803.BI (cpuio)cpuload \fR=\fPint
1804Attempt to use the specified percentage of CPU cycles. This is a mandatory
1805option when using cpuio I/O engine.
997b5680 1806.TP
523bad63
TK
1807.BI (cpuio)cpuchunks \fR=\fPint
1808Split the load into cycles of the given time. In microseconds.
1ad01bd1 1809.TP
523bad63
TK
1810.BI (cpuio)exit_on_io_done \fR=\fPbool
1811Detect when I/O threads are done, then exit.
d60e92d1 1812.TP
523bad63
TK
1813.BI (libhdfs)namenode \fR=\fPstr
1814The hostname or IP address of a HDFS cluster namenode to contact.
d01612f3 1815.TP
523bad63
TK
1816.BI (libhdfs)port
1817The listening port of the HFDS cluster namenode.
d60e92d1 1818.TP
523bad63
TK
1819.BI (netsplice,net)port
1820The TCP or UDP port to bind to or connect to. If this is used with
1821\fBnumjobs\fR to spawn multiple instances of the same job type, then
1822this will be the starting port number since fio will use a range of
1823ports.
d60e92d1 1824.TP
609ac152
SB
1825.BI (rdma)port
1826The port to use for RDMA-CM communication. This should be the same
1827value on the client and the server side.
1828.TP
1829.BI (netsplice,net, rdma)hostname \fR=\fPstr
1830The hostname or IP address to use for TCP, UDP or RDMA-CM based I/O.
1831If the job is a TCP listener or UDP reader, the hostname is not used
1832and must be omitted unless it is a valid UDP multicast address.
591e9e06 1833.TP
523bad63
TK
1834.BI (netsplice,net)interface \fR=\fPstr
1835The IP address of the network interface used to send or receive UDP
1836multicast.
ddf24e42 1837.TP
523bad63
TK
1838.BI (netsplice,net)ttl \fR=\fPint
1839Time\-to\-live value for outgoing UDP multicast packets. Default: 1.
d60e92d1 1840.TP
523bad63
TK
1841.BI (netsplice,net)nodelay \fR=\fPbool
1842Set TCP_NODELAY on TCP connections.
fa769d44 1843.TP
523bad63
TK
1844.BI (netsplice,net)protocol \fR=\fPstr "\fR,\fP proto" \fR=\fPstr
1845The network protocol to use. Accepted values are:
1846.RS
e76b1da4
JA
1847.RS
1848.TP
523bad63
TK
1849.B tcp
1850Transmission control protocol.
e76b1da4 1851.TP
523bad63
TK
1852.B tcpv6
1853Transmission control protocol V6.
e76b1da4 1854.TP
523bad63
TK
1855.B udp
1856User datagram protocol.
1857.TP
1858.B udpv6
1859User datagram protocol V6.
e76b1da4 1860.TP
523bad63
TK
1861.B unix
1862UNIX domain socket.
e76b1da4
JA
1863.RE
1864.P
523bad63
TK
1865When the protocol is TCP or UDP, the port must also be given, as well as the
1866hostname if the job is a TCP listener or UDP reader. For unix sockets, the
1867normal \fBfilename\fR option should be used and the port is invalid.
1868.RE
1869.TP
1870.BI (netsplice,net)listen
1871For TCP network connections, tell fio to listen for incoming connections
1872rather than initiating an outgoing connection. The \fBhostname\fR must
1873be omitted if this option is used.
1874.TP
1875.BI (netsplice,net)pingpong
1876Normally a network writer will just continue writing data, and a network
1877reader will just consume packages. If `pingpong=1' is set, a writer will
1878send its normal payload to the reader, then wait for the reader to send the
1879same payload back. This allows fio to measure network latencies. The
1880submission and completion latencies then measure local time spent sending or
1881receiving, and the completion latency measures how long it took for the
1882other end to receive and send back. For UDP multicast traffic
1883`pingpong=1' should only be set for a single reader when multiple readers
1884are listening to the same address.
1885.TP
1886.BI (netsplice,net)window_size \fR=\fPint
1887Set the desired socket buffer size for the connection.
e76b1da4 1888.TP
523bad63
TK
1889.BI (netsplice,net)mss \fR=\fPint
1890Set the TCP maximum segment size (TCP_MAXSEG).
d60e92d1 1891.TP
523bad63
TK
1892.BI (e4defrag)donorname \fR=\fPstr
1893File will be used as a block donor (swap extents between files).
d60e92d1 1894.TP
523bad63
TK
1895.BI (e4defrag)inplace \fR=\fPint
1896Configure donor file blocks allocation strategy:
1897.RS
1898.RS
d60e92d1 1899.TP
523bad63
TK
1900.B 0
1901Default. Preallocate donor's file on init.
d60e92d1 1902.TP
523bad63
TK
1903.B 1
1904Allocate space immediately inside defragment event, and free right
1905after event.
1906.RE
1907.RE
d60e92d1 1908.TP
d5f9b0ea 1909.BI (rbd,rados)clustername \fR=\fPstr
523bad63 1910Specifies the name of the Ceph cluster.
92d42d69 1911.TP
523bad63
TK
1912.BI (rbd)rbdname \fR=\fPstr
1913Specifies the name of the RBD.
92d42d69 1914.TP
d5f9b0ea
IF
1915.BI (rbd,rados)pool \fR=\fPstr
1916Specifies the name of the Ceph pool containing RBD or RADOS data.
92d42d69 1917.TP
d5f9b0ea 1918.BI (rbd,rados)clientname \fR=\fPstr
523bad63
TK
1919Specifies the username (without the 'client.' prefix) used to access the
1920Ceph cluster. If the \fBclustername\fR is specified, the \fBclientname\fR shall be
1921the full *type.id* string. If no type. prefix is given, fio will add 'client.'
1922by default.
92d42d69 1923.TP
d5f9b0ea
IF
1924.BI (rbd,rados)busy_poll \fR=\fPbool
1925Poll store instead of waiting for completion. Usually this provides better
1926throughput at cost of higher(up to 100%) CPU utilization.
1927.TP
c2f6a13d
LMB
1928.BI (http)http_host \fR=\fPstr
1929Hostname to connect to. For S3, this could be the bucket name. Default
1930is \fBlocalhost\fR
1931.TP
1932.BI (http)http_user \fR=\fPstr
1933Username for HTTP authentication.
1934.TP
1935.BI (http)http_pass \fR=\fPstr
1936Password for HTTP authentication.
1937.TP
09fd2966
LMB
1938.BI (http)https \fR=\fPstr
1939Whether to use HTTPS instead of plain HTTP. \fRon\fP enables HTTPS;
1940\fRinsecure\fP will enable HTTPS, but disable SSL peer verification (use
1941with caution!). Default is \fBoff\fR.
c2f6a13d 1942.TP
09fd2966
LMB
1943.BI (http)http_mode \fR=\fPstr
1944Which HTTP access mode to use: webdav, swift, or s3. Default is
1945\fBwebdav\fR.
c2f6a13d
LMB
1946.TP
1947.BI (http)http_s3_region \fR=\fPstr
1948The S3 region/zone to include in the request. Default is \fBus-east-1\fR.
1949.TP
1950.BI (http)http_s3_key \fR=\fPstr
1951The S3 secret key.
1952.TP
1953.BI (http)http_s3_keyid \fR=\fPstr
1954The S3 key/access id.
1955.TP
09fd2966
LMB
1956.BI (http)http_swift_auth_token \fR=\fPstr
1957The Swift auth token. See the example configuration file on how to
1958retrieve this.
1959.TP
c2f6a13d
LMB
1960.BI (http)http_verbose \fR=\fPint
1961Enable verbose requests from libcurl. Useful for debugging. 1 turns on
1962verbose logging from libcurl, 2 additionally enables HTTP IO tracing.
1963Default is \fB0\fR
1964.TP
523bad63
TK
1965.BI (mtd)skip_bad \fR=\fPbool
1966Skip operations against known bad blocks.
8116fd24 1967.TP
523bad63
TK
1968.BI (libhdfs)hdfsdirectory
1969libhdfs will create chunk in this HDFS directory.
e0a04ac1 1970.TP
523bad63
TK
1971.BI (libhdfs)chunk_size
1972The size of the chunk to use for each file.
609ac152
SB
1973.TP
1974.BI (rdma)verb \fR=\fPstr
1975The RDMA verb to use on this side of the RDMA ioengine
1976connection. Valid values are write, read, send and recv. These
1977correspond to the equivalent RDMA verbs (e.g. write = rdma_write
1978etc.). Note that this only needs to be specified on the client side of
1979the connection. See the examples folder.
1980.TP
1981.BI (rdma)bindname \fR=\fPstr
1982The name to use to bind the local RDMA-CM connection to a local RDMA
1983device. This could be a hostname or an IPv4 or IPv6 address. On the
1984server side this will be passed into the rdma_bind_addr() function and
1985on the client site it will be used in the rdma_resolve_add()
1986function. This can be useful when multiple paths exist between the
1987client and the server or in certain loopback configurations.
52b81b7c
KD
1988.TP
1989.BI (sg)readfua \fR=\fPbool
1990With readfua option set to 1, read operations include the force
1991unit access (fua) flag. Default: 0.
1992.TP
1993.BI (sg)writefua \fR=\fPbool
1994With writefua option set to 1, write operations include the force
1995unit access (fua) flag. Default: 0.
2c3a9150
VF
1996.TP
1997.BI (sg)sg_write_mode \fR=\fPstr
1998Specify the type of write commands to issue. This option can take three
1999values:
2000.RS
2001.RS
2002.TP
2003.B write (default)
2004Write opcodes are issued as usual
2005.TP
2006.B verify
2007Issue WRITE AND VERIFY commands. The BYTCHK bit is set to 0. This
2008directs the device to carry out a medium verification with no data
2009comparison. The writefua option is ignored with this selection.
2010.TP
2011.B same
2012Issue WRITE SAME commands. This transfers a single block to the device
2013and writes this same block of data to a contiguous sequence of LBAs
2014beginning at the specified offset. fio's block size parameter
2015specifies the amount of data written with each command. However, the
2016amount of data actually transferred to the device is equal to the
2017device's block (sector) size. For a device with 512 byte sectors,
2018blocksize=8k will write 16 sectors with each command. fio will still
2019generate 8k of data for each command butonly the first 512 bytes will
2020be used and transferred to the device. The writefua option is ignored
2021with this selection.
2022
523bad63
TK
2023.SS "I/O depth"
2024.TP
2025.BI iodepth \fR=\fPint
2026Number of I/O units to keep in flight against the file. Note that
2027increasing \fBiodepth\fR beyond 1 will not affect synchronous ioengines (except
2028for small degrees when \fBverify_async\fR is in use). Even async
2029engines may impose OS restrictions causing the desired depth not to be
2030achieved. This may happen on Linux when using libaio and not setting
2031`direct=1', since buffered I/O is not async on that OS. Keep an
2032eye on the I/O depth distribution in the fio output to verify that the
2033achieved depth is as expected. Default: 1.
2034.TP
2035.BI iodepth_batch_submit \fR=\fPint "\fR,\fP iodepth_batch" \fR=\fPint
2036This defines how many pieces of I/O to submit at once. It defaults to 1
2037which means that we submit each I/O as soon as it is available, but can be
2038raised to submit bigger batches of I/O at the time. If it is set to 0 the
2039\fBiodepth\fR value will be used.
2040.TP
2041.BI iodepth_batch_complete_min \fR=\fPint "\fR,\fP iodepth_batch_complete" \fR=\fPint
2042This defines how many pieces of I/O to retrieve at once. It defaults to 1
2043which means that we'll ask for a minimum of 1 I/O in the retrieval process
2044from the kernel. The I/O retrieval will go on until we hit the limit set by
2045\fBiodepth_low\fR. If this variable is set to 0, then fio will always
2046check for completed events before queuing more I/O. This helps reduce I/O
2047latency, at the cost of more retrieval system calls.
2048.TP
2049.BI iodepth_batch_complete_max \fR=\fPint
2050This defines maximum pieces of I/O to retrieve at once. This variable should
2051be used along with \fBiodepth_batch_complete_min\fR=\fIint\fR variable,
2052specifying the range of min and max amount of I/O which should be
2053retrieved. By default it is equal to \fBiodepth_batch_complete_min\fR
2054value. Example #1:
e0a04ac1 2055.RS
e0a04ac1 2056.RS
e0a04ac1 2057.P
523bad63
TK
2058.PD 0
2059iodepth_batch_complete_min=1
e0a04ac1 2060.P
523bad63
TK
2061iodepth_batch_complete_max=<iodepth>
2062.PD
e0a04ac1
JA
2063.RE
2064.P
523bad63
TK
2065which means that we will retrieve at least 1 I/O and up to the whole
2066submitted queue depth. If none of I/O has been completed yet, we will wait.
2067Example #2:
e8b1961d 2068.RS
523bad63
TK
2069.P
2070.PD 0
2071iodepth_batch_complete_min=0
2072.P
2073iodepth_batch_complete_max=<iodepth>
2074.PD
e8b1961d
JA
2075.RE
2076.P
523bad63
TK
2077which means that we can retrieve up to the whole submitted queue depth, but
2078if none of I/O has been completed yet, we will NOT wait and immediately exit
2079the system call. In this example we simply do polling.
2080.RE
e8b1961d 2081.TP
523bad63
TK
2082.BI iodepth_low \fR=\fPint
2083The low water mark indicating when to start filling the queue
2084again. Defaults to the same as \fBiodepth\fR, meaning that fio will
2085attempt to keep the queue full at all times. If \fBiodepth\fR is set to
2086e.g. 16 and \fBiodepth_low\fR is set to 4, then after fio has filled the queue of
208716 requests, it will let the depth drain down to 4 before starting to fill
2088it again.
d60e92d1 2089.TP
523bad63
TK
2090.BI serialize_overlap \fR=\fPbool
2091Serialize in-flight I/Os that might otherwise cause or suffer from data races.
2092When two or more I/Os are submitted simultaneously, there is no guarantee that
2093the I/Os will be processed or completed in the submitted order. Further, if
2094two or more of those I/Os are writes, any overlapping region between them can
2095become indeterminate/undefined on certain storage. These issues can cause
2096verification to fail erratically when at least one of the racing I/Os is
2097changing data and the overlapping region has a non-zero size. Setting
2098\fBserialize_overlap\fR tells fio to avoid provoking this behavior by explicitly
2099serializing in-flight I/Os that have a non-zero overlap. Note that setting
2100this option can reduce both performance and the \fBiodepth\fR achieved.
3d6a6f04
VF
2101.RS
2102.P
2103This option only applies to I/Os issued for a single job except when it is
2104enabled along with \fBio_submit_mode\fR=offload. In offload mode, fio
2105will check for overlap among all I/Os submitted by offload jobs with \fBserialize_overlap\fR
307f2246 2106enabled.
3d6a6f04
VF
2107.P
2108Default: false.
2109.RE
d60e92d1 2110.TP
523bad63
TK
2111.BI io_submit_mode \fR=\fPstr
2112This option controls how fio submits the I/O to the I/O engine. The default
2113is `inline', which means that the fio job threads submit and reap I/O
2114directly. If set to `offload', the job threads will offload I/O submission
2115to a dedicated pool of I/O threads. This requires some coordination and thus
2116has a bit of extra overhead, especially for lower queue depth I/O where it
2117can increase latencies. The benefit is that fio can manage submission rates
2118independently of the device completion rates. This avoids skewed latency
2119reporting if I/O gets backed up on the device side (the coordinated omission
2120problem).
2121.SS "I/O rate"
d60e92d1 2122.TP
523bad63
TK
2123.BI thinktime \fR=\fPtime
2124Stall the job for the specified period of time after an I/O has completed before issuing the
2125next. May be used to simulate processing being done by an application.
2126When the unit is omitted, the value is interpreted in microseconds. See
2127\fBthinktime_blocks\fR and \fBthinktime_spin\fR.
d60e92d1 2128.TP
523bad63
TK
2129.BI thinktime_spin \fR=\fPtime
2130Only valid if \fBthinktime\fR is set \- pretend to spend CPU time doing
2131something with the data received, before falling back to sleeping for the
2132rest of the period specified by \fBthinktime\fR. When the unit is
2133omitted, the value is interpreted in microseconds.
d60e92d1
AC
2134.TP
2135.BI thinktime_blocks \fR=\fPint
523bad63
TK
2136Only valid if \fBthinktime\fR is set \- control how many blocks to issue,
2137before waiting \fBthinktime\fR usecs. If not set, defaults to 1 which will make
2138fio wait \fBthinktime\fR usecs after every block. This effectively makes any
2139queue depth setting redundant, since no more than 1 I/O will be queued
2140before we have to complete it and do our \fBthinktime\fR. In other words, this
2141setting effectively caps the queue depth if the latter is larger.
d60e92d1 2142.TP
6d500c2e 2143.BI rate \fR=\fPint[,int][,int]
523bad63
TK
2144Cap the bandwidth used by this job. The number is in bytes/sec, the normal
2145suffix rules apply. Comma\-separated values may be specified for reads,
2146writes, and trims as described in \fBblocksize\fR.
2147.RS
2148.P
2149For example, using `rate=1m,500k' would limit reads to 1MiB/sec and writes to
2150500KiB/sec. Capping only reads or writes can be done with `rate=,500k' or
2151`rate=500k,' where the former will only limit writes (to 500KiB/sec) and the
2152latter will only limit reads.
2153.RE
d60e92d1 2154.TP
6d500c2e 2155.BI rate_min \fR=\fPint[,int][,int]
523bad63
TK
2156Tell fio to do whatever it can to maintain at least this bandwidth. Failing
2157to meet this requirement will cause the job to exit. Comma\-separated values
2158may be specified for reads, writes, and trims as described in
2159\fBblocksize\fR.
d60e92d1 2160.TP
6d500c2e 2161.BI rate_iops \fR=\fPint[,int][,int]
523bad63
TK
2162Cap the bandwidth to this number of IOPS. Basically the same as
2163\fBrate\fR, just specified independently of bandwidth. If the job is
2164given a block size range instead of a fixed value, the smallest block size
2165is used as the metric. Comma\-separated values may be specified for reads,
2166writes, and trims as described in \fBblocksize\fR.
d60e92d1 2167.TP
6d500c2e 2168.BI rate_iops_min \fR=\fPint[,int][,int]
523bad63
TK
2169If fio doesn't meet this rate of I/O, it will cause the job to exit.
2170Comma\-separated values may be specified for reads, writes, and trims as
2171described in \fBblocksize\fR.
d60e92d1 2172.TP
6de65959 2173.BI rate_process \fR=\fPstr
523bad63
TK
2174This option controls how fio manages rated I/O submissions. The default is
2175`linear', which submits I/O in a linear fashion with fixed delays between
2176I/Os that gets adjusted based on I/O completion rates. If this is set to
2177`poisson', fio will submit I/O based on a more real world random request
6de65959 2178flow, known as the Poisson process
523bad63 2179(\fIhttps://en.wikipedia.org/wiki/Poisson_point_process\fR). The lambda will be
5d02b083 218010^6 / IOPS for the given workload.
1a9bf814
JA
2181.TP
2182.BI rate_ignore_thinktime \fR=\fPbool
2183By default, fio will attempt to catch up to the specified rate setting, if any
2184kind of thinktime setting was used. If this option is set, then fio will
2185ignore the thinktime and continue doing IO at the specified rate, instead of
2186entering a catch-up mode after thinktime is done.
523bad63 2187.SS "I/O latency"
ff6bb260 2188.TP
523bad63 2189.BI latency_target \fR=\fPtime
3e260a46 2190If set, fio will attempt to find the max performance point that the given
523bad63
TK
2191workload will run at while maintaining a latency below this target. When
2192the unit is omitted, the value is interpreted in microseconds. See
2193\fBlatency_window\fR and \fBlatency_percentile\fR.
3e260a46 2194.TP
523bad63 2195.BI latency_window \fR=\fPtime
3e260a46 2196Used with \fBlatency_target\fR to specify the sample window that the job
523bad63
TK
2197is run at varying queue depths to test the performance. When the unit is
2198omitted, the value is interpreted in microseconds.
3e260a46
JA
2199.TP
2200.BI latency_percentile \fR=\fPfloat
523bad63
TK
2201The percentage of I/Os that must fall within the criteria specified by
2202\fBlatency_target\fR and \fBlatency_window\fR. If not set, this
2203defaults to 100.0, meaning that all I/Os must be equal or below to the value
2204set by \fBlatency_target\fR.
2205.TP
2206.BI max_latency \fR=\fPtime
2207If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
2208maximum latency. When the unit is omitted, the value is interpreted in
2209microseconds.
2210.TP
2211.BI rate_cycle \fR=\fPint
2212Average bandwidth for \fBrate\fR and \fBrate_min\fR over this number
2213of milliseconds. Defaults to 1000.
2214.SS "I/O replay"
2215.TP
2216.BI write_iolog \fR=\fPstr
2217Write the issued I/O patterns to the specified file. See
2218\fBread_iolog\fR. Specify a separate file for each job, otherwise the
2219iologs will be interspersed and the file may be corrupt.
2220.TP
2221.BI read_iolog \fR=\fPstr
2222Open an iolog with the specified filename and replay the I/O patterns it
2223contains. This can be used to store a workload and replay it sometime
2224later. The iolog given may also be a blktrace binary file, which allows fio
2225to replay a workload captured by blktrace. See
2226\fBblktrace\fR\|(8) for how to capture such logging data. For blktrace
2227replay, the file needs to be turned into a blkparse binary data file first
2228(`blkparse <device> \-o /dev/null \-d file_for_fio.bin').
c70c7f58
AK
2229You can specify a number of files by separating the names with a ':' character.
2230See the \fBfilename\fR option for information on how to escape ':' and '\'
2231characters within the file names. These files will be sequentially assigned to
2232job clones created by \fBnumjobs\fR.
3e260a46 2233.TP
98e7161c
AK
2234.BI read_iolog_chunked \fR=\fPbool
2235Determines how iolog is read. If false (default) entire \fBread_iolog\fR will
2236be read at once. If selected true, input from iolog will be read gradually.
2237Useful when iolog is very large, or it is generated.
2238.TP
b9921d1a
DZ
2239.BI merge_blktrace_file \fR=\fPstr
2240When specified, rather than replaying the logs passed to \fBread_iolog\fR,
2241the logs go through a merge phase which aggregates them into a single blktrace.
2242The resulting file is then passed on as the \fBread_iolog\fR parameter. The
2243intention here is to make the order of events consistent. This limits the
2244influence of the scheduler compared to replaying multiple blktraces via
2245concurrent jobs.
2246.TP
87a48ada
DZ
2247.BI merge_blktrace_scalars \fR=\fPfloat_list
2248This is a percentage based option that is index paired with the list of files
2249passed to \fBread_iolog\fR. When merging is performed, scale the time of each
2250event by the corresponding amount. For example,
2251`\-\-merge_blktrace_scalars="50:100"' runs the first trace in halftime and the
2252second trace in realtime. This knob is separately tunable from
2253\fBreplay_time_scale\fR which scales the trace during runtime and will not
2254change the output of the merge unlike this option.
2255.TP
55bfd8c8
DZ
2256.BI merge_blktrace_iters \fR=\fPfloat_list
2257This is a whole number option that is index paired with the list of files
2258passed to \fBread_iolog\fR. When merging is performed, run each trace for
2259the specified number of iterations. For example,
2260`\-\-merge_blktrace_iters="2:1"' runs the first trace for two iterations
2261and the second trace for one iteration.
2262.TP
523bad63
TK
2263.BI replay_no_stall \fR=\fPbool
2264When replaying I/O with \fBread_iolog\fR the default behavior is to
2265attempt to respect the timestamps within the log and replay them with the
2266appropriate delay between IOPS. By setting this variable fio will not
2267respect the timestamps and attempt to replay them as fast as possible while
2268still respecting ordering. The result is the same I/O pattern to a given
2269device, but different timings.
2270.TP
6dd7fa77
JA
2271.BI replay_time_scale \fR=\fPint
2272When replaying I/O with \fBread_iolog\fR, fio will honor the original timing
2273in the trace. With this option, it's possible to scale the time. It's a
2274percentage option, if set to 50 it means run at 50% the original IO rate in
2275the trace. If set to 200, run at twice the original IO rate. Defaults to 100.
2276.TP
523bad63
TK
2277.BI replay_redirect \fR=\fPstr
2278While replaying I/O patterns using \fBread_iolog\fR the default behavior
2279is to replay the IOPS onto the major/minor device that each IOP was recorded
2280from. This is sometimes undesirable because on a different machine those
2281major/minor numbers can map to a different device. Changing hardware on the
2282same system can also result in a different major/minor mapping.
2283\fBreplay_redirect\fR causes all I/Os to be replayed onto the single specified
2284device regardless of the device it was recorded
2285from. i.e. `replay_redirect=/dev/sdc' would cause all I/O
2286in the blktrace or iolog to be replayed onto `/dev/sdc'. This means
2287multiple devices will be replayed onto a single device, if the trace
2288contains multiple devices. If you want multiple devices to be replayed
2289concurrently to multiple redirected devices you must blkparse your trace
2290into separate traces and replay them with independent fio invocations.
2291Unfortunately this also breaks the strict time ordering between multiple
2292device accesses.
2293.TP
2294.BI replay_align \fR=\fPint
350a535d
DZ
2295Force alignment of the byte offsets in a trace to this value. The value
2296must be a power of 2.
523bad63
TK
2297.TP
2298.BI replay_scale \fR=\fPint
350a535d
DZ
2299Scale bye offsets down by this factor when replaying traces. Should most
2300likely use \fBreplay_align\fR as well.
523bad63
TK
2301.SS "Threads, processes and job synchronization"
2302.TP
38f68906
JA
2303.BI replay_skip \fR=\fPstr
2304Sometimes it's useful to skip certain IO types in a replay trace. This could
2305be, for instance, eliminating the writes in the trace. Or not replaying the
2306trims/discards, if you are redirecting to a device that doesn't support them.
2307This option takes a comma separated list of read, write, trim, sync.
2308.TP
523bad63
TK
2309.BI thread
2310Fio defaults to creating jobs by using fork, however if this option is
2311given, fio will create jobs by using POSIX Threads' function
2312\fBpthread_create\fR\|(3) to create threads instead.
2313.TP
2314.BI wait_for \fR=\fPstr
2315If set, the current job won't be started until all workers of the specified
2316waitee job are done.
2317.\" ignore blank line here from HOWTO as it looks normal without it
2318\fBwait_for\fR operates on the job name basis, so there are a few
2319limitations. First, the waitee must be defined prior to the waiter job
2320(meaning no forward references). Second, if a job is being referenced as a
2321waitee, it must have a unique name (no duplicate waitees).
2322.TP
2323.BI nice \fR=\fPint
2324Run the job with the given nice value. See man \fBnice\fR\|(2).
2325.\" ignore blank line here from HOWTO as it looks normal without it
2326On Windows, values less than \-15 set the process class to "High"; \-1 through
2327\-15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2328priority class.
2329.TP
2330.BI prio \fR=\fPint
2331Set the I/O priority value of this job. Linux limits us to a positive value
2332between 0 and 7, with 0 being the highest. See man
2333\fBionice\fR\|(1). Refer to an appropriate manpage for other operating
2334systems since meaning of priority may differ.
2335.TP
2336.BI prioclass \fR=\fPint
2337Set the I/O priority class. See man \fBionice\fR\|(1).
15501535 2338.TP
d60e92d1 2339.BI cpus_allowed \fR=\fPstr
523bad63 2340Controls the same options as \fBcpumask\fR, but accepts a textual
b570e037
SW
2341specification of the permitted CPUs instead and CPUs are indexed from 0. So
2342to use CPUs 0 and 5 you would specify `cpus_allowed=0,5'. This option also
2343allows a range of CPUs to be specified \-\- say you wanted a binding to CPUs
23440, 5, and 8 to 15, you would set `cpus_allowed=0,5,8\-15'.
2345.RS
2346.P
2347On Windows, when `cpus_allowed' is unset only CPUs from fio's current
2348processor group will be used and affinity settings are inherited from the
2349system. An fio build configured to target Windows 7 makes options that set
2350CPUs processor group aware and values will set both the processor group
2351and a CPU from within that group. For example, on a system where processor
2352group 0 has 40 CPUs and processor group 1 has 32 CPUs, `cpus_allowed'
2353values between 0 and 39 will bind CPUs from processor group 0 and
2354`cpus_allowed' values between 40 and 71 will bind CPUs from processor
2355group 1. When using `cpus_allowed_policy=shared' all CPUs specified by a
2356single `cpus_allowed' option must be from the same processor group. For
2357Windows fio builds not built for Windows 7, CPUs will only be selected from
2358(and be relative to) whatever processor group fio happens to be running in
2359and CPUs from other processor groups cannot be used.
2360.RE
d60e92d1 2361.TP
c2acfbac 2362.BI cpus_allowed_policy \fR=\fPstr
523bad63
TK
2363Set the policy of how fio distributes the CPUs specified by
2364\fBcpus_allowed\fR or \fBcpumask\fR. Two policies are supported:
c2acfbac
JA
2365.RS
2366.RS
2367.TP
2368.B shared
2369All jobs will share the CPU set specified.
2370.TP
2371.B split
2372Each job will get a unique CPU from the CPU set.
2373.RE
2374.P
523bad63
TK
2375\fBshared\fR is the default behavior, if the option isn't specified. If
2376\fBsplit\fR is specified, then fio will will assign one cpu per job. If not
2377enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2378in the set.
c2acfbac 2379.RE
c2acfbac 2380.TP
b570e037
SW
2381.BI cpumask \fR=\fPint
2382Set the CPU affinity of this job. The parameter given is a bit mask of
2383allowed CPUs the job may run on. So if you want the allowed CPUs to be 1
2384and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2385\fBsched_setaffinity\fR\|(2). This may not work on all supported
2386operating systems or kernel versions. This option doesn't work well for a
2387higher CPU count than what you can store in an integer mask, so it can only
2388control cpus 1\-32. For boxes with larger CPU counts, use
2389\fBcpus_allowed\fR.
2390.TP
d0b937ed 2391.BI numa_cpu_nodes \fR=\fPstr
cecbfd47 2392Set this job running on specified NUMA nodes' CPUs. The arguments allow
523bad63
TK
2393comma delimited list of cpu numbers, A\-B ranges, or `all'. Note, to enable
2394NUMA options support, fio must be built on a system with libnuma\-dev(el)
2395installed.
d0b937ed
YR
2396.TP
2397.BI numa_mem_policy \fR=\fPstr
523bad63
TK
2398Set this job's memory policy and corresponding NUMA nodes. Format of the
2399arguments:
39c7a2ca
VF
2400.RS
2401.RS
523bad63
TK
2402.P
2403<mode>[:<nodelist>]
39c7a2ca 2404.RE
523bad63 2405.P
f1dd3fb1 2406`mode' is one of the following memory policies: `default', `prefer',
523bad63
TK
2407`bind', `interleave' or `local'. For `default' and `local' memory
2408policies, no node needs to be specified. For `prefer', only one node is
2409allowed. For `bind' and `interleave' the `nodelist' may be as
2410follows: a comma delimited list of numbers, A\-B ranges, or `all'.
39c7a2ca
VF
2411.RE
2412.TP
523bad63
TK
2413.BI cgroup \fR=\fPstr
2414Add job to this control group. If it doesn't exist, it will be created. The
2415system must have a mounted cgroup blkio mount point for this to work. If
2416your system doesn't have it mounted, you can do so with:
d60e92d1
AC
2417.RS
2418.RS
d60e92d1 2419.P
523bad63
TK
2420# mount \-t cgroup \-o blkio none /cgroup
2421.RE
d60e92d1
AC
2422.RE
2423.TP
523bad63
TK
2424.BI cgroup_weight \fR=\fPint
2425Set the weight of the cgroup to this value. See the documentation that comes
2426with the kernel, allowed values are in the range of 100..1000.
d60e92d1 2427.TP
523bad63
TK
2428.BI cgroup_nodelete \fR=\fPbool
2429Normally fio will delete the cgroups it has created after the job
2430completion. To override this behavior and to leave cgroups around after the
2431job completion, set `cgroup_nodelete=1'. This can be useful if one wants
2432to inspect various cgroup files after job completion. Default: false.
c8eeb9df 2433.TP
523bad63
TK
2434.BI flow_id \fR=\fPint
2435The ID of the flow. If not specified, it defaults to being a global
2436flow. See \fBflow\fR.
d60e92d1 2437.TP
523bad63
TK
2438.BI flow \fR=\fPint
2439Weight in token\-based flow control. If this value is used, then there is
2440a 'flow counter' which is used to regulate the proportion of activity between
2441two or more jobs. Fio attempts to keep this flow counter near zero. The
2442\fBflow\fR parameter stands for how much should be added or subtracted to the
2443flow counter on each iteration of the main I/O loop. That is, if one job has
2444`flow=8' and another job has `flow=\-1', then there will be a roughly 1:8
2445ratio in how much one runs vs the other.
d60e92d1 2446.TP
523bad63
TK
2447.BI flow_watermark \fR=\fPint
2448The maximum value that the absolute value of the flow counter is allowed to
2449reach before the job must wait for a lower value of the counter.
6b7f6851 2450.TP
523bad63
TK
2451.BI flow_sleep \fR=\fPint
2452The period of time, in microseconds, to wait after the flow watermark has
2453been exceeded before retrying operations.
25460cf6 2454.TP
523bad63
TK
2455.BI stonewall "\fR,\fB wait_for_previous"
2456Wait for preceding jobs in the job file to exit, before starting this
2457one. Can be used to insert serialization points in the job file. A stone
2458wall also implies starting a new reporting group, see
2459\fBgroup_reporting\fR.
2378826d 2460.TP
523bad63
TK
2461.BI exitall
2462By default, fio will continue running all other jobs when one job finishes
2463but sometimes this is not the desired action. Setting \fBexitall\fR will
2464instead make fio terminate all other jobs when one job finishes.
e81ecca3 2465.TP
523bad63
TK
2466.BI exec_prerun \fR=\fPstr
2467Before running this job, issue the command specified through
2468\fBsystem\fR\|(3). Output is redirected in a file called `jobname.prerun.txt'.
e9f48479 2469.TP
523bad63
TK
2470.BI exec_postrun \fR=\fPstr
2471After the job completes, issue the command specified though
2472\fBsystem\fR\|(3). Output is redirected in a file called `jobname.postrun.txt'.
d60e92d1 2473.TP
523bad63
TK
2474.BI uid \fR=\fPint
2475Instead of running as the invoking user, set the user ID to this value
2476before the thread/process does any work.
39c1c323 2477.TP
523bad63
TK
2478.BI gid \fR=\fPint
2479Set group ID, see \fBuid\fR.
2480.SS "Verification"
d60e92d1 2481.TP
589e88b7 2482.BI verify_only
523bad63 2483Do not perform specified workload, only verify data still matches previous
5e4c7118 2484invocation of this workload. This option allows one to check data multiple
523bad63
TK
2485times at a later date without overwriting it. This option makes sense only
2486for workloads that write data, and does not support workloads with the
5e4c7118
JA
2487\fBtime_based\fR option set.
2488.TP
d60e92d1 2489.BI do_verify \fR=\fPbool
523bad63
TK
2490Run the verify phase after a write phase. Only valid if \fBverify\fR is
2491set. Default: true.
d60e92d1
AC
2492.TP
2493.BI verify \fR=\fPstr
523bad63
TK
2494If writing to a file, fio can verify the file contents after each iteration
2495of the job. Each verification method also implies verification of special
2496header, which is written to the beginning of each block. This header also
2497includes meta information, like offset of the block, block number, timestamp
2498when block was written, etc. \fBverify\fR can be combined with
2499\fBverify_pattern\fR option. The allowed values are:
d60e92d1
AC
2500.RS
2501.RS
2502.TP
523bad63
TK
2503.B md5
2504Use an md5 sum of the data area and store it in the header of
2505each block.
2506.TP
2507.B crc64
2508Use an experimental crc64 sum of the data area and store it in the
2509header of each block.
2510.TP
2511.B crc32c
2512Use a crc32c sum of the data area and store it in the header of
2513each block. This will automatically use hardware acceleration
2514(e.g. SSE4.2 on an x86 or CRC crypto extensions on ARM64) but will
2515fall back to software crc32c if none is found. Generally the
f1dd3fb1 2516fastest checksum fio supports when hardware accelerated.
523bad63
TK
2517.TP
2518.B crc32c\-intel
2519Synonym for crc32c.
2520.TP
2521.B crc32
2522Use a crc32 sum of the data area and store it in the header of each
2523block.
2524.TP
2525.B crc16
2526Use a crc16 sum of the data area and store it in the header of each
2527block.
2528.TP
2529.B crc7
2530Use a crc7 sum of the data area and store it in the header of each
2531block.
2532.TP
2533.B xxhash
2534Use xxhash as the checksum function. Generally the fastest software
2535checksum that fio supports.
2536.TP
2537.B sha512
2538Use sha512 as the checksum function.
2539.TP
2540.B sha256
2541Use sha256 as the checksum function.
2542.TP
2543.B sha1
2544Use optimized sha1 as the checksum function.
2545.TP
2546.B sha3\-224
2547Use optimized sha3\-224 as the checksum function.
2548.TP
2549.B sha3\-256
2550Use optimized sha3\-256 as the checksum function.
2551.TP
2552.B sha3\-384
2553Use optimized sha3\-384 as the checksum function.
2554.TP
2555.B sha3\-512
2556Use optimized sha3\-512 as the checksum function.
d60e92d1
AC
2557.TP
2558.B meta
523bad63
TK
2559This option is deprecated, since now meta information is included in
2560generic verification header and meta verification happens by
2561default. For detailed information see the description of the
2562\fBverify\fR setting. This option is kept because of
2563compatibility's sake with old configurations. Do not use it.
d60e92d1 2564.TP
59245381 2565.B pattern
523bad63
TK
2566Verify a strict pattern. Normally fio includes a header with some
2567basic information and checksumming, but if this option is set, only
2568the specific pattern set with \fBverify_pattern\fR is verified.
59245381 2569.TP
d60e92d1 2570.B null
523bad63
TK
2571Only pretend to verify. Useful for testing internals with
2572`ioengine=null', not for much else.
d60e92d1 2573.RE
523bad63
TK
2574.P
2575This option can be used for repeated burn\-in tests of a system to make sure
2576that the written data is also correctly read back. If the data direction
2577given is a read or random read, fio will assume that it should verify a
2578previously written file. If the data direction includes any form of write,
2579the verify will be of the newly written data.
47e6a6e5
SW
2580.P
2581To avoid false verification errors, do not use the norandommap option when
2582verifying data with async I/O engines and I/O depths > 1. Or use the
2583norandommap and the lfsr random generator together to avoid writing to the
2584same offset with muliple outstanding I/Os.
d60e92d1
AC
2585.RE
2586.TP
f7fa2653 2587.BI verify_offset \fR=\fPint
d60e92d1 2588Swap the verification header with data somewhere else in the block before
523bad63 2589writing. It is swapped back before verifying.
d60e92d1 2590.TP
f7fa2653 2591.BI verify_interval \fR=\fPint
523bad63
TK
2592Write the verification header at a finer granularity than the
2593\fBblocksize\fR. It will be written for chunks the size of
2594\fBverify_interval\fR. \fBblocksize\fR should divide this evenly.
d60e92d1 2595.TP
996093bb 2596.BI verify_pattern \fR=\fPstr
523bad63
TK
2597If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2598filling with totally random bytes, but sometimes it's interesting to fill
2599with a known pattern for I/O verification purposes. Depending on the width
2600of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time (it can
2601be either a decimal or a hex number). The \fBverify_pattern\fR if larger than
2602a 32\-bit quantity has to be a hex number that starts with either "0x" or
2603"0X". Use with \fBverify\fR. Also, \fBverify_pattern\fR supports %o
2604format, which means that for each block offset will be written and then
2605verified back, e.g.:
2fa5a241
RP
2606.RS
2607.RS
523bad63
TK
2608.P
2609verify_pattern=%o
2fa5a241 2610.RE
523bad63 2611.P
2fa5a241 2612Or use combination of everything:
2fa5a241 2613.RS
523bad63
TK
2614.P
2615verify_pattern=0xff%o"abcd"\-12
2fa5a241
RP
2616.RE
2617.RE
996093bb 2618.TP
d60e92d1 2619.BI verify_fatal \fR=\fPbool
523bad63
TK
2620Normally fio will keep checking the entire contents before quitting on a
2621block verification failure. If this option is set, fio will exit the job on
2622the first observed failure. Default: false.
d60e92d1 2623.TP
b463e936 2624.BI verify_dump \fR=\fPbool
523bad63
TK
2625If set, dump the contents of both the original data block and the data block
2626we read off disk to files. This allows later analysis to inspect just what
2627kind of data corruption occurred. Off by default.
b463e936 2628.TP
e8462bd8 2629.BI verify_async \fR=\fPint
523bad63
TK
2630Fio will normally verify I/O inline from the submitting thread. This option
2631takes an integer describing how many async offload threads to create for I/O
2632verification instead, causing fio to offload the duty of verifying I/O
2633contents to one or more separate threads. If using this offload option, even
2634sync I/O engines can benefit from using an \fBiodepth\fR setting higher
2635than 1, as it allows them to have I/O in flight while verifies are running.
2636Defaults to 0 async threads, i.e. verification is not asynchronous.
e8462bd8
JA
2637.TP
2638.BI verify_async_cpus \fR=\fPstr
523bad63
TK
2639Tell fio to set the given CPU affinity on the async I/O verification
2640threads. See \fBcpus_allowed\fR for the format used.
e8462bd8 2641.TP
6f87418f
JA
2642.BI verify_backlog \fR=\fPint
2643Fio will normally verify the written contents of a job that utilizes verify
2644once that job has completed. In other words, everything is written then
2645everything is read back and verified. You may want to verify continually
523bad63
TK
2646instead for a variety of reasons. Fio stores the meta data associated with
2647an I/O block in memory, so for large verify workloads, quite a bit of memory
2648would be used up holding this meta data. If this option is enabled, fio will
2649write only N blocks before verifying these blocks.
6f87418f
JA
2650.TP
2651.BI verify_backlog_batch \fR=\fPint
523bad63
TK
2652Control how many blocks fio will verify if \fBverify_backlog\fR is
2653set. If not set, will default to the value of \fBverify_backlog\fR
2654(meaning the entire queue is read back and verified). If
2655\fBverify_backlog_batch\fR is less than \fBverify_backlog\fR then not all
2656blocks will be verified, if \fBverify_backlog_batch\fR is larger than
2657\fBverify_backlog\fR, some blocks will be verified more than once.
2658.TP
2659.BI verify_state_save \fR=\fPbool
2660When a job exits during the write phase of a verify workload, save its
2661current state. This allows fio to replay up until that point, if the verify
2662state is loaded for the verify read phase. The format of the filename is,
2663roughly:
2664.RS
2665.RS
2666.P
2667<type>\-<jobname>\-<jobindex>\-verify.state.
2668.RE
2669.P
2670<type> is "local" for a local run, "sock" for a client/server socket
2671connection, and "ip" (192.168.0.1, for instance) for a networked
2672client/server connection. Defaults to true.
2673.RE
2674.TP
2675.BI verify_state_load \fR=\fPbool
2676If a verify termination trigger was used, fio stores the current write state
2677of each thread. This can be used at verification time so that fio knows how
2678far it should verify. Without this information, fio will run a full
2679verification pass, according to the settings in the job file used. Default
2680false.
6f87418f 2681.TP
fa769d44
SW
2682.BI trim_percentage \fR=\fPint
2683Number of verify blocks to discard/trim.
2684.TP
2685.BI trim_verify_zero \fR=\fPbool
523bad63 2686Verify that trim/discarded blocks are returned as zeros.
fa769d44
SW
2687.TP
2688.BI trim_backlog \fR=\fPint
523bad63 2689Verify that trim/discarded blocks are returned as zeros.
fa769d44
SW
2690.TP
2691.BI trim_backlog_batch \fR=\fPint
523bad63 2692Trim this number of I/O blocks.
fa769d44
SW
2693.TP
2694.BI experimental_verify \fR=\fPbool
2695Enable experimental verification.
523bad63 2696.SS "Steady state"
fa769d44 2697.TP
523bad63
TK
2698.BI steadystate \fR=\fPstr:float "\fR,\fP ss" \fR=\fPstr:float
2699Define the criterion and limit for assessing steady state performance. The
2700first parameter designates the criterion whereas the second parameter sets
2701the threshold. When the criterion falls below the threshold for the
2702specified duration, the job will stop. For example, `iops_slope:0.1%' will
2703direct fio to terminate the job when the least squares regression slope
2704falls below 0.1% of the mean IOPS. If \fBgroup_reporting\fR is enabled
2705this will apply to all jobs in the group. Below is the list of available
2706steady state assessment criteria. All assessments are carried out using only
2707data from the rolling collection window. Threshold limits can be expressed
2708as a fixed value or as a percentage of the mean in the collection window.
2709.RS
1cb049d9
VF
2710.P
2711When using this feature, most jobs should include the \fBtime_based\fR
2712and \fBruntime\fR options or the \fBloops\fR option so that fio does not
2713stop running after it has covered the full size of the specified file(s)
2714or device(s).
2715.RS
523bad63 2716.RS
d60e92d1 2717.TP
523bad63
TK
2718.B iops
2719Collect IOPS data. Stop the job if all individual IOPS measurements
2720are within the specified limit of the mean IOPS (e.g., `iops:2'
2721means that all individual IOPS values must be within 2 of the mean,
2722whereas `iops:0.2%' means that all individual IOPS values must be
2723within 0.2% of the mean IOPS to terminate the job).
d60e92d1 2724.TP
523bad63
TK
2725.B iops_slope
2726Collect IOPS data and calculate the least squares regression
2727slope. Stop the job if the slope falls below the specified limit.
d60e92d1 2728.TP
523bad63
TK
2729.B bw
2730Collect bandwidth data. Stop the job if all individual bandwidth
2731measurements are within the specified limit of the mean bandwidth.
64bbb865 2732.TP
523bad63
TK
2733.B bw_slope
2734Collect bandwidth data and calculate the least squares regression
2735slope. Stop the job if the slope falls below the specified limit.
2736.RE
2737.RE
d1c46c04 2738.TP
523bad63
TK
2739.BI steadystate_duration \fR=\fPtime "\fR,\fP ss_dur" \fR=\fPtime
2740A rolling window of this duration will be used to judge whether steady state
2741has been reached. Data will be collected once per second. The default is 0
2742which disables steady state detection. When the unit is omitted, the
2743value is interpreted in seconds.
0c63576e 2744.TP
523bad63
TK
2745.BI steadystate_ramp_time \fR=\fPtime "\fR,\fP ss_ramp" \fR=\fPtime
2746Allow the job to run for the specified duration before beginning data
2747collection for checking the steady state job termination criterion. The
2748default is 0. When the unit is omitted, the value is interpreted in seconds.
2749.SS "Measurements and reporting"
0c63576e 2750.TP
3a5db920
JA
2751.BI per_job_logs \fR=\fPbool
2752If set, this generates bw/clat/iops log with per file private filenames. If
523bad63
TK
2753not set, jobs with identical names will share the log filename. Default:
2754true.
2755.TP
2756.BI group_reporting
2757It may sometimes be interesting to display statistics for groups of jobs as
2758a whole instead of for each individual job. This is especially true if
2759\fBnumjobs\fR is used; looking at individual thread/process output
2760quickly becomes unwieldy. To see the final report per\-group instead of
2761per\-job, use \fBgroup_reporting\fR. Jobs in a file will be part of the
2762same reporting group, unless if separated by a \fBstonewall\fR, or by
2763using \fBnew_group\fR.
2764.TP
2765.BI new_group
2766Start a new reporting group. See: \fBgroup_reporting\fR. If not given,
2767all jobs in a file will be part of the same reporting group, unless
2768separated by a \fBstonewall\fR.
2769.TP
2770.BI stats \fR=\fPbool
2771By default, fio collects and shows final output results for all jobs
2772that run. If this option is set to 0, then fio will ignore it in
2773the final stat output.
3a5db920 2774.TP
836bad52 2775.BI write_bw_log \fR=\fPstr
523bad63 2776If given, write a bandwidth log for this job. Can be used to store data of
074f0817 2777the bandwidth of the jobs in their lifetime.
523bad63 2778.RS
074f0817
SW
2779.P
2780If no str argument is given, the default filename of
2781`jobname_type.x.log' is used. Even when the argument is given, fio
2782will still append the type of log. So if one specifies:
523bad63
TK
2783.RS
2784.P
074f0817 2785write_bw_log=foo
523bad63
TK
2786.RE
2787.P
074f0817
SW
2788The actual log name will be `foo_bw.x.log' where `x' is the index
2789of the job (1..N, where N is the number of jobs). If
2790\fBper_job_logs\fR is false, then the filename will not include the
2791`.x` job index.
2792.P
2793The included \fBfio_generate_plots\fR script uses gnuplot to turn these
2794text files into nice graphs. See the \fBLOG FILE FORMATS\fR section for how data is
2795structured within the file.
523bad63 2796.RE
901bb994 2797.TP
074f0817
SW
2798.BI write_lat_log \fR=\fPstr
2799Same as \fBwrite_bw_log\fR, except this option creates I/O
2800submission (e.g., `name_slat.x.log'), completion (e.g.,
2801`name_clat.x.log'), and total (e.g., `name_lat.x.log') latency
2802files instead. See \fBwrite_bw_log\fR for details about the
2803filename format and the \fBLOG FILE FORMATS\fR section for how data is structured
2804within the files.
2805.TP
1e613c9c 2806.BI write_hist_log \fR=\fPstr
074f0817
SW
2807Same as \fBwrite_bw_log\fR but writes an I/O completion latency
2808histogram file (e.g., `name_hist.x.log') instead. Note that this
2809file will be empty unless \fBlog_hist_msec\fR has also been set.
2810See \fBwrite_bw_log\fR for details about the filename format and
2811the \fBLOG FILE FORMATS\fR section for how data is structured
2812within the file.
1e613c9c 2813.TP
c8eeb9df 2814.BI write_iops_log \fR=\fPstr
074f0817 2815Same as \fBwrite_bw_log\fR, but writes an IOPS file (e.g.
15417073
SW
2816`name_iops.x.log`) instead. Because fio defaults to individual
2817I/O logging, the value entry in the IOPS log will be 1 unless windowed
2818logging (see \fBlog_avg_msec\fR) has been enabled. See
2819\fBwrite_bw_log\fR for details about the filename format and \fBLOG
2820FILE FORMATS\fR for how data is structured within the file.
c8eeb9df 2821.TP
b8bc8cba
JA
2822.BI log_avg_msec \fR=\fPint
2823By default, fio will log an entry in the iops, latency, or bw log for every
523bad63 2824I/O that completes. When writing to the disk log, that can quickly grow to a
b8bc8cba 2825very large size. Setting this option makes fio average the each log entry
e6989e10 2826over the specified period of time, reducing the resolution of the log. See
523bad63
TK
2827\fBlog_max_value\fR as well. Defaults to 0, logging all entries.
2828Also see \fBLOG FILE FORMATS\fR section.
b8bc8cba 2829.TP
1e613c9c 2830.BI log_hist_msec \fR=\fPint
523bad63
TK
2831Same as \fBlog_avg_msec\fR, but logs entries for completion latency
2832histograms. Computing latency percentiles from averages of intervals using
2833\fBlog_avg_msec\fR is inaccurate. Setting this option makes fio log
2834histogram entries over the specified period of time, reducing log sizes for
2835high IOPS devices while retaining percentile accuracy. See
074f0817
SW
2836\fBlog_hist_coarseness\fR and \fBwrite_hist_log\fR as well.
2837Defaults to 0, meaning histogram logging is disabled.
1e613c9c
KC
2838.TP
2839.BI log_hist_coarseness \fR=\fPint
523bad63
TK
2840Integer ranging from 0 to 6, defining the coarseness of the resolution of
2841the histogram logs enabled with \fBlog_hist_msec\fR. For each increment
2842in coarseness, fio outputs half as many bins. Defaults to 0, for which
2843histogram logs contain 1216 latency bins. See \fBLOG FILE FORMATS\fR section.
2844.TP
2845.BI log_max_value \fR=\fPbool
2846If \fBlog_avg_msec\fR is set, fio logs the average over that window. If
2847you instead want to log the maximum value, set this option to 1. Defaults to
28480, meaning that averaged values are logged.
1e613c9c 2849.TP
ae588852 2850.BI log_offset \fR=\fPbool
523bad63
TK
2851If this is set, the iolog options will include the byte offset for the I/O
2852entry as well as the other data values. Defaults to 0 meaning that
2853offsets are not present in logs. Also see \fBLOG FILE FORMATS\fR section.
ae588852 2854.TP
aee2ab67 2855.BI log_compression \fR=\fPint
523bad63
TK
2856If this is set, fio will compress the I/O logs as it goes, to keep the
2857memory footprint lower. When a log reaches the specified size, that chunk is
2858removed and compressed in the background. Given that I/O logs are fairly
2859highly compressible, this yields a nice memory savings for longer runs. The
2860downside is that the compression will consume some background CPU cycles, so
2861it may impact the run. This, however, is also true if the logging ends up
2862consuming most of the system memory. So pick your poison. The I/O logs are
2863saved normally at the end of a run, by decompressing the chunks and storing
2864them in the specified log file. This feature depends on the availability of
2865zlib.
aee2ab67 2866.TP
c08f9fe2 2867.BI log_compression_cpus \fR=\fPstr
523bad63
TK
2868Define the set of CPUs that are allowed to handle online log compression for
2869the I/O jobs. This can provide better isolation between performance
0cf90a62
SW
2870sensitive jobs, and background compression work. See \fBcpus_allowed\fR for
2871the format used.
c08f9fe2 2872.TP
b26317c9 2873.BI log_store_compressed \fR=\fPbool
c08f9fe2 2874If set, fio will store the log files in a compressed format. They can be
523bad63
TK
2875decompressed with fio, using the \fB\-\-inflate\-log\fR command line
2876parameter. The files will be stored with a `.fz' suffix.
b26317c9 2877.TP
3aea75b1
KC
2878.BI log_unix_epoch \fR=\fPbool
2879If set, fio will log Unix timestamps to the log files produced by enabling
523bad63 2880write_type_log for each log type, instead of the default zero\-based
3aea75b1
KC
2881timestamps.
2882.TP
66347cfa 2883.BI block_error_percentiles \fR=\fPbool
523bad63
TK
2884If set, record errors in trim block\-sized units from writes and trims and
2885output a histogram of how many trims it took to get to errors, and what kind
2886of error was encountered.
d60e92d1 2887.TP
523bad63
TK
2888.BI bwavgtime \fR=\fPint
2889Average the calculated bandwidth over the given time. Value is specified in
2890milliseconds. If the job also does bandwidth logging through
2891\fBwrite_bw_log\fR, then the minimum of this option and
2892\fBlog_avg_msec\fR will be used. Default: 500ms.
d60e92d1 2893.TP
523bad63
TK
2894.BI iopsavgtime \fR=\fPint
2895Average the calculated IOPS over the given time. Value is specified in
2896milliseconds. If the job also does IOPS logging through
2897\fBwrite_iops_log\fR, then the minimum of this option and
2898\fBlog_avg_msec\fR will be used. Default: 500ms.
d60e92d1 2899.TP
d60e92d1 2900.BI disk_util \fR=\fPbool
523bad63
TK
2901Generate disk utilization statistics, if the platform supports it.
2902Default: true.
fa769d44 2903.TP
523bad63
TK
2904.BI disable_lat \fR=\fPbool
2905Disable measurements of total latency numbers. Useful only for cutting back
2906the number of calls to \fBgettimeofday\fR\|(2), as that does impact
2907performance at really high IOPS rates. Note that to really get rid of a
2908large amount of these calls, this option must be used with
2909\fBdisable_slat\fR and \fBdisable_bw_measurement\fR as well.
9e684a49 2910.TP
523bad63
TK
2911.BI disable_clat \fR=\fPbool
2912Disable measurements of completion latency numbers. See
2913\fBdisable_lat\fR.
9e684a49 2914.TP
523bad63
TK
2915.BI disable_slat \fR=\fPbool
2916Disable measurements of submission latency numbers. See
2917\fBdisable_lat\fR.
9e684a49 2918.TP
523bad63
TK
2919.BI disable_bw_measurement \fR=\fPbool "\fR,\fP disable_bw" \fR=\fPbool
2920Disable measurements of throughput/bandwidth numbers. See
2921\fBdisable_lat\fR.
9e684a49 2922.TP
83349190 2923.BI clat_percentiles \fR=\fPbool
b599759b
JA
2924Enable the reporting of percentiles of completion latencies. This option is
2925mutually exclusive with \fBlat_percentiles\fR.
2926.TP
2927.BI lat_percentiles \fR=\fPbool
b71968b1 2928Enable the reporting of percentiles of I/O latencies. This is similar to
b599759b
JA
2929\fBclat_percentiles\fR, except that this includes the submission latency.
2930This option is mutually exclusive with \fBclat_percentiles\fR.
83349190
YH
2931.TP
2932.BI percentile_list \fR=\fPfloat_list
66347cfa 2933Overwrite the default list of percentiles for completion latencies and the
523bad63
TK
2934block error histogram. Each number is a floating number in the range
2935(0,100], and the maximum length of the list is 20. Use ':' to separate the
2936numbers, and list the numbers in ascending order. For example,
2937`\-\-percentile_list=99.5:99.9' will cause fio to report the values of
2938completion latency below which 99.5% and 99.9% of the observed latencies
2939fell, respectively.
e883cb35
JF
2940.TP
2941.BI significant_figures \fR=\fPint
c32ba107
JA
2942If using \fB\-\-output\-format\fR of `normal', set the significant figures
2943to this value. Higher values will yield more precise IOPS and throughput
2944units, while lower values will round. Requires a minimum value of 1 and a
e883cb35 2945maximum value of 10. Defaults to 4.
523bad63 2946.SS "Error handling"
e4585935 2947.TP
523bad63
TK
2948.BI exitall_on_error
2949When one job finishes in error, terminate the rest. The default is to wait
2950for each job to finish.
e4585935 2951.TP
523bad63
TK
2952.BI continue_on_error \fR=\fPstr
2953Normally fio will exit the job on the first observed failure. If this option
2954is set, fio will continue the job when there is a 'non\-fatal error' (EIO or
2955EILSEQ) until the runtime is exceeded or the I/O size specified is
2956completed. If this option is used, there are two more stats that are
2957appended, the total error count and the first error. The error field given
2958in the stats is the first error that was hit during the run.
2959The allowed values are:
2960.RS
2961.RS
046395d7 2962.TP
523bad63
TK
2963.B none
2964Exit on any I/O or verify errors.
de890a1e 2965.TP
523bad63
TK
2966.B read
2967Continue on read errors, exit on all others.
2cafffbe 2968.TP
523bad63
TK
2969.B write
2970Continue on write errors, exit on all others.
a0679ce5 2971.TP
523bad63
TK
2972.B io
2973Continue on any I/O error, exit on all others.
de890a1e 2974.TP
523bad63
TK
2975.B verify
2976Continue on verify errors, exit on all others.
de890a1e 2977.TP
523bad63
TK
2978.B all
2979Continue on all errors.
b93b6a2e 2980.TP
523bad63
TK
2981.B 0
2982Backward\-compatible alias for 'none'.
d3a623de 2983.TP
523bad63
TK
2984.B 1
2985Backward\-compatible alias for 'all'.
2986.RE
2987.RE
1d360ffb 2988.TP
523bad63
TK
2989.BI ignore_error \fR=\fPstr
2990Sometimes you want to ignore some errors during test in that case you can
2991specify error list for each error type, instead of only being able to
2992ignore the default 'non\-fatal error' using \fBcontinue_on_error\fR.
2993`ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST' errors for
2994given error type is separated with ':'. Error may be symbol ('ENOSPC', 'ENOMEM')
2995or integer. Example:
de890a1e
SL
2996.RS
2997.RS
523bad63
TK
2998.P
2999ignore_error=EAGAIN,ENOSPC:122
3000.RE
3001.P
3002This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
3003WRITE. This option works by overriding \fBcontinue_on_error\fR with
3004the list of errors for each error type if any.
3005.RE
de890a1e 3006.TP
523bad63
TK
3007.BI error_dump \fR=\fPbool
3008If set dump every error even if it is non fatal, true by default. If
3009disabled only fatal error will be dumped.
3010.SS "Running predefined workloads"
3011Fio includes predefined profiles that mimic the I/O workloads generated by
3012other tools.
49ccb8c1 3013.TP
523bad63
TK
3014.BI profile \fR=\fPstr
3015The predefined workload to run. Current profiles are:
3016.RS
3017.RS
de890a1e 3018.TP
523bad63
TK
3019.B tiobench
3020Threaded I/O bench (tiotest/tiobench) like workload.
49ccb8c1 3021.TP
523bad63
TK
3022.B act
3023Aerospike Certification Tool (ACT) like workload.
3024.RE
de890a1e
SL
3025.RE
3026.P
523bad63
TK
3027To view a profile's additional options use \fB\-\-cmdhelp\fR after specifying
3028the profile. For example:
3029.RS
3030.TP
3031$ fio \-\-profile=act \-\-cmdhelp
de890a1e 3032.RE
523bad63 3033.SS "Act profile options"
de890a1e 3034.TP
523bad63
TK
3035.BI device\-names \fR=\fPstr
3036Devices to use.
d54fce84 3037.TP
523bad63
TK
3038.BI load \fR=\fPint
3039ACT load multiplier. Default: 1.
7aeb1e94 3040.TP
523bad63
TK
3041.BI test\-duration\fR=\fPtime
3042How long the entire test takes to run. When the unit is omitted, the value
3043is given in seconds. Default: 24h.
1008602c 3044.TP
523bad63
TK
3045.BI threads\-per\-queue\fR=\fPint
3046Number of read I/O threads per device. Default: 8.
e5f34d95 3047.TP
523bad63
TK
3048.BI read\-req\-num\-512\-blocks\fR=\fPint
3049Number of 512B blocks to read at the time. Default: 3.
d54fce84 3050.TP
523bad63
TK
3051.BI large\-block\-op\-kbytes\fR=\fPint
3052Size of large block ops in KiB (writes). Default: 131072.
d54fce84 3053.TP
523bad63
TK
3054.BI prep
3055Set to run ACT prep phase.
3056.SS "Tiobench profile options"
6d500c2e 3057.TP
523bad63
TK
3058.BI size\fR=\fPstr
3059Size in MiB.
0d978694 3060.TP
523bad63
TK
3061.BI block\fR=\fPint
3062Block size in bytes. Default: 4096.
0d978694 3063.TP
523bad63
TK
3064.BI numruns\fR=\fPint
3065Number of runs.
0d978694 3066.TP
523bad63
TK
3067.BI dir\fR=\fPstr
3068Test directory.
65fa28ca 3069.TP
523bad63
TK
3070.BI threads\fR=\fPint
3071Number of threads.
d60e92d1 3072.SH OUTPUT
40943b9a
TK
3073Fio spits out a lot of output. While running, fio will display the status of the
3074jobs created. An example of that would be:
d60e92d1 3075.P
40943b9a
TK
3076.nf
3077 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]
3078.fi
d1429b5c 3079.P
40943b9a
TK
3080The characters inside the first set of square brackets denote the current status of
3081each thread. The first character is the first job defined in the job file, and so
3082forth. The possible values (in typical life cycle order) are:
d60e92d1
AC
3083.RS
3084.TP
40943b9a 3085.PD 0
d60e92d1 3086.B P
40943b9a 3087Thread setup, but not started.
d60e92d1
AC
3088.TP
3089.B C
3090Thread created.
3091.TP
3092.B I
40943b9a
TK
3093Thread initialized, waiting or generating necessary data.
3094.TP
522c29f6 3095.B p
40943b9a
TK
3096Thread running pre\-reading file(s).
3097.TP
3098.B /
3099Thread is in ramp period.
d60e92d1
AC
3100.TP
3101.B R
3102Running, doing sequential reads.
3103.TP
3104.B r
3105Running, doing random reads.
3106.TP
3107.B W
3108Running, doing sequential writes.
3109.TP
3110.B w
3111Running, doing random writes.
3112.TP
3113.B M
3114Running, doing mixed sequential reads/writes.
3115.TP
3116.B m
3117Running, doing mixed random reads/writes.
3118.TP
40943b9a
TK
3119.B D
3120Running, doing sequential trims.
3121.TP
3122.B d
3123Running, doing random trims.
3124.TP
d60e92d1
AC
3125.B F
3126Running, currently waiting for \fBfsync\fR\|(2).
3127.TP
3128.B V
40943b9a
TK
3129Running, doing verification of written data.
3130.TP
3131.B f
3132Thread finishing.
d60e92d1
AC
3133.TP
3134.B E
40943b9a 3135Thread exited, not reaped by main thread yet.
d60e92d1
AC
3136.TP
3137.B \-
40943b9a
TK
3138Thread reaped.
3139.TP
3140.B X
3141Thread reaped, exited with an error.
3142.TP
3143.B K
3144Thread reaped, exited due to signal.
d1429b5c 3145.PD
40943b9a
TK
3146.RE
3147.P
3148Fio will condense the thread string as not to take up more space on the command
3149line than needed. For instance, if you have 10 readers and 10 writers running,
3150the output would look like this:
3151.P
3152.nf
3153 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]
3154.fi
d60e92d1 3155.P
40943b9a
TK
3156Note that the status string is displayed in order, so it's possible to tell which of
3157the jobs are currently doing what. In the example above this means that jobs 1\-\-10
3158are readers and 11\-\-20 are writers.
d60e92d1 3159.P
40943b9a
TK
3160The other values are fairly self explanatory \-\- number of threads currently
3161running and doing I/O, the number of currently open files (f=), the estimated
3162completion percentage, the rate of I/O since last check (read speed listed first,
3163then write speed and optionally trim speed) in terms of bandwidth and IOPS,
3164and time to completion for the current running group. It's impossible to estimate
3165runtime of the following groups (if any).
d60e92d1 3166.P
40943b9a
TK
3167When fio is done (or interrupted by Ctrl\-C), it will show the data for
3168each thread, group of threads, and disks in that order. For each overall thread (or
3169group) the output looks like:
3170.P
3171.nf
3172 Client1: (groupid=0, jobs=1): err= 0: pid=16109: Sat Jun 24 12:07:54 2017
3173 write: IOPS=88, BW=623KiB/s (638kB/s)(30.4MiB/50032msec)
3174 slat (nsec): min=500, max=145500, avg=8318.00, stdev=4781.50
3175 clat (usec): min=170, max=78367, avg=4019.02, stdev=8293.31
3176 lat (usec): min=174, max=78375, avg=4027.34, stdev=8291.79
3177 clat percentiles (usec):
3178 | 1.00th=[ 302], 5.00th=[ 326], 10.00th=[ 343], 20.00th=[ 363],
3179 | 30.00th=[ 392], 40.00th=[ 404], 50.00th=[ 416], 60.00th=[ 445],
3180 | 70.00th=[ 816], 80.00th=[ 6718], 90.00th=[12911], 95.00th=[21627],
3181 | 99.00th=[43779], 99.50th=[51643], 99.90th=[68682], 99.95th=[72877],
3182 | 99.99th=[78119]
3183 bw ( KiB/s): min= 532, max= 686, per=0.10%, avg=622.87, stdev=24.82, samples= 100
3184 iops : min= 76, max= 98, avg=88.98, stdev= 3.54, samples= 100
d3b9694d
VF
3185 lat (usec) : 250=0.04%, 500=64.11%, 750=4.81%, 1000=2.79%
3186 lat (msec) : 2=4.16%, 4=1.84%, 10=4.90%, 20=11.33%, 50=5.37%
3187 lat (msec) : 100=0.65%
40943b9a
TK
3188 cpu : usr=0.27%, sys=0.18%, ctx=12072, majf=0, minf=21
3189 IO depths : 1=85.0%, 2=13.1%, 4=1.8%, 8=0.1%, 16=0.0%, 32=0.0%, >=64=0.0%
3190 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3191 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3192 issued rwt: total=0,4450,0, short=0,0,0, dropped=0,0,0
3193 latency : target=0, window=0, percentile=100.00%, depth=8
3194.fi
3195.P
3196The job name (or first job's name when using \fBgroup_reporting\fR) is printed,
3197along with the group id, count of jobs being aggregated, last error id seen (which
3198is 0 when there are no errors), pid/tid of that thread and the time the job/group
3199completed. Below are the I/O statistics for each data direction performed (showing
3200writes in the example above). In the order listed, they denote:
d60e92d1 3201.RS
d60e92d1 3202.TP
40943b9a
TK
3203.B read/write/trim
3204The string before the colon shows the I/O direction the statistics
3205are for. \fIIOPS\fR is the average I/Os performed per second. \fIBW\fR
3206is the average bandwidth rate shown as: value in power of 2 format
3207(value in power of 10 format). The last two values show: (total
3208I/O performed in power of 2 format / \fIruntime\fR of that thread).
d60e92d1
AC
3209.TP
3210.B slat
40943b9a
TK
3211Submission latency (\fImin\fR being the minimum, \fImax\fR being the
3212maximum, \fIavg\fR being the average, \fIstdev\fR being the standard
3213deviation). This is the time it took to submit the I/O. For
3214sync I/O this row is not displayed as the slat is really the
3215completion latency (since queue/complete is one operation there).
3216This value can be in nanoseconds, microseconds or milliseconds \-\-\-
3217fio will choose the most appropriate base and print that (in the
3218example above nanoseconds was the best scale). Note: in \fB\-\-minimal\fR mode
3219latencies are always expressed in microseconds.
d60e92d1
AC
3220.TP
3221.B clat
40943b9a
TK
3222Completion latency. Same names as slat, this denotes the time from
3223submission to completion of the I/O pieces. For sync I/O, clat will
3224usually be equal (or very close) to 0, as the time from submit to
3225complete is basically just CPU time (I/O has already been done, see slat
3226explanation).
d60e92d1 3227.TP
d3b9694d
VF
3228.B lat
3229Total latency. Same names as slat and clat, this denotes the time from
3230when fio created the I/O unit to completion of the I/O operation.
3231.TP
d60e92d1 3232.B bw
40943b9a
TK
3233Bandwidth statistics based on samples. Same names as the xlat stats,
3234but also includes the number of samples taken (\fIsamples\fR) and an
3235approximate percentage of total aggregate bandwidth this thread
3236received in its group (\fIper\fR). This last value is only really
3237useful if the threads in this group are on the same disk, since they
3238are then competing for disk access.
3239.TP
3240.B iops
3241IOPS statistics based on samples. Same names as \fBbw\fR.
d60e92d1 3242.TP
d3b9694d
VF
3243.B lat (nsec/usec/msec)
3244The distribution of I/O completion latencies. This is the time from when
3245I/O leaves fio and when it gets completed. Unlike the separate
3246read/write/trim sections above, the data here and in the remaining
3247sections apply to all I/Os for the reporting group. 250=0.04% means that
32480.04% of the I/Os completed in under 250us. 500=64.11% means that 64.11%
3249of the I/Os required 250 to 499us for completion.
3250.TP
d60e92d1 3251.B cpu
40943b9a
TK
3252CPU usage. User and system time, along with the number of context
3253switches this thread went through, usage of system and user time, and
3254finally the number of major and minor page faults. The CPU utilization
3255numbers are averages for the jobs in that reporting group, while the
3256context and fault counters are summed.
d60e92d1
AC
3257.TP
3258.B IO depths
40943b9a
TK
3259The distribution of I/O depths over the job lifetime. The numbers are
3260divided into powers of 2 and each entry covers depths from that value
3261up to those that are lower than the next entry \-\- e.g., 16= covers
3262depths from 16 to 31. Note that the range covered by a depth
3263distribution entry can be different to the range covered by the
3264equivalent \fBsubmit\fR/\fBcomplete\fR distribution entry.
3265.TP
3266.B IO submit
3267How many pieces of I/O were submitting in a single submit call. Each
3268entry denotes that amount and below, until the previous entry \-\- e.g.,
326916=100% means that we submitted anywhere between 9 to 16 I/Os per submit
3270call. Note that the range covered by a \fBsubmit\fR distribution entry can
3271be different to the range covered by the equivalent depth distribution
3272entry.
3273.TP
3274.B IO complete
3275Like the above \fBsubmit\fR number, but for completions instead.
3276.TP
3277.B IO issued rwt
3278The number of \fBread/write/trim\fR requests issued, and how many of them were
3279short or dropped.
d60e92d1 3280.TP
d3b9694d 3281.B IO latency
ee21ebee 3282These values are for \fBlatency_target\fR and related options. When
d3b9694d
VF
3283these options are engaged, this section describes the I/O depth required
3284to meet the specified latency target.
d60e92d1 3285.RE
d60e92d1 3286.P
40943b9a
TK
3287After each client has been listed, the group statistics are printed. They
3288will look like this:
3289.P
3290.nf
3291 Run status group 0 (all jobs):
3292 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
3293 WRITE: bw=1231KiB/s (1261kB/s), 616KiB/s\-621KiB/s (630kB/s\-636kB/s), io=64.0MiB (67.1MB), run=52747\-53223msec
3294.fi
3295.P
3296For each data direction it prints:
d60e92d1
AC
3297.RS
3298.TP
40943b9a
TK
3299.B bw
3300Aggregate bandwidth of threads in this group followed by the
3301minimum and maximum bandwidth of all the threads in this group.
3302Values outside of brackets are power\-of\-2 format and those
3303within are the equivalent value in a power\-of\-10 format.
d60e92d1 3304.TP
40943b9a
TK
3305.B io
3306Aggregate I/O performed of all threads in this group. The
3307format is the same as \fBbw\fR.
d60e92d1 3308.TP
40943b9a
TK
3309.B run
3310The smallest and longest runtimes of the threads in this group.
d60e92d1 3311.RE
d60e92d1 3312.P
40943b9a
TK
3313And finally, the disk statistics are printed. This is Linux specific.
3314They will look like this:
3315.P
3316.nf
3317 Disk stats (read/write):
3318 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3319.fi
3320.P
3321Each value is printed for both reads and writes, with reads first. The
3322numbers denote:
d60e92d1
AC
3323.RS
3324.TP
3325.B ios
3326Number of I/Os performed by all groups.
3327.TP
3328.B merge
007c7be9 3329Number of merges performed by the I/O scheduler.
d60e92d1
AC
3330.TP
3331.B ticks
3332Number of ticks we kept the disk busy.
3333.TP
40943b9a 3334.B in_queue
d60e92d1
AC
3335Total time spent in the disk queue.
3336.TP
3337.B util
40943b9a
TK
3338The disk utilization. A value of 100% means we kept the disk
3339busy constantly, 50% would be a disk idling half of the time.
d60e92d1 3340.RE
8423bd11 3341.P
40943b9a
TK
3342It is also possible to get fio to dump the current output while it is running,
3343without terminating the job. To do that, send fio the USR1 signal. You can
3344also get regularly timed dumps by using the \fB\-\-status\-interval\fR
3345parameter, or by creating a file in `/tmp' named
3346`fio\-dump\-status'. If fio sees this file, it will unlink it and dump the
3347current output status.
d60e92d1 3348.SH TERSE OUTPUT
40943b9a
TK
3349For scripted usage where you typically want to generate tables or graphs of the
3350results, fio can output the results in a semicolon separated format. The format
3351is one long line of values, such as:
d60e92d1 3352.P
40943b9a
TK
3353.nf
3354 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%
3355 A description of this job goes here.
3356.fi
d60e92d1 3357.P
4e757af1
VF
3358The job description (if provided) follows on a second line for terse v2.
3359It appears on the same line for other terse versions.
d60e92d1 3360.P
40943b9a
TK
3361To enable terse output, use the \fB\-\-minimal\fR or
3362`\-\-output\-format=terse' command line options. The
3363first value is the version of the terse output format. If the output has to be
3364changed for some reason, this number will be incremented by 1 to signify that
3365change.
d60e92d1 3366.P
40943b9a
TK
3367Split up, the format is as follows (comments in brackets denote when a
3368field was introduced or whether it's specific to some terse version):
d60e92d1 3369.P
40943b9a
TK
3370.nf
3371 terse version, fio version [v3], jobname, groupid, error
3372.fi
525c2bfa 3373.RS
40943b9a
TK
3374.P
3375.B
3376READ status:
525c2bfa 3377.RE
40943b9a
TK
3378.P
3379.nf
3380 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3381 Submission latency: min, max, mean, stdev (usec)
3382 Completion latency: min, max, mean, stdev (usec)
3383 Completion latency percentiles: 20 fields (see below)
3384 Total latency: min, max, mean, stdev (usec)
3385 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3386 IOPS [v5]: min, max, mean, stdev, number of samples
3387.fi
d60e92d1 3388.RS
40943b9a
TK
3389.P
3390.B
3391WRITE status:
a2c95580 3392.RE
40943b9a
TK
3393.P
3394.nf
3395 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3396 Submission latency: min, max, mean, stdev (usec)
3397 Completion latency: min, max, mean, stdev (usec)
3398 Completion latency percentiles: 20 fields (see below)
3399 Total latency: min, max, mean, stdev (usec)
3400 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3401 IOPS [v5]: min, max, mean, stdev, number of samples
3402.fi
a2c95580 3403.RS
40943b9a
TK
3404.P
3405.B
3406TRIM status [all but version 3]:
d60e92d1
AC
3407.RE
3408.P
40943b9a
TK
3409.nf
3410 Fields are similar to \fBREAD/WRITE\fR status.
3411.fi
a2c95580 3412.RS
a2c95580 3413.P
40943b9a 3414.B
d1429b5c 3415CPU usage:
d60e92d1
AC
3416.RE
3417.P
40943b9a
TK
3418.nf
3419 user, system, context switches, major faults, minor faults
3420.fi
d60e92d1 3421.RS
40943b9a
TK
3422.P
3423.B
3424I/O depths:
d60e92d1
AC
3425.RE
3426.P
40943b9a
TK
3427.nf
3428 <=1, 2, 4, 8, 16, 32, >=64
3429.fi
562c2d2f 3430.RS
40943b9a
TK
3431.P
3432.B
3433I/O latencies microseconds:
562c2d2f 3434.RE
40943b9a
TK
3435.P
3436.nf
3437 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3438.fi
562c2d2f 3439.RS
40943b9a
TK
3440.P
3441.B
3442I/O latencies milliseconds:
562c2d2f
DN
3443.RE
3444.P
40943b9a
TK
3445.nf
3446 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3447.fi
f2f788dd 3448.RS
40943b9a
TK
3449.P
3450.B
3451Disk utilization [v3]:
f2f788dd
JA
3452.RE
3453.P
40943b9a
TK
3454.nf
3455 disk name, read ios, write ios, read merges, write merges, read ticks, write ticks, time spent in queue, disk utilization percentage
3456.fi
562c2d2f 3457.RS
d60e92d1 3458.P
40943b9a
TK
3459.B
3460Additional Info (dependent on continue_on_error, default off):
d60e92d1 3461.RE
2fc26c3d 3462.P
40943b9a
TK
3463.nf
3464 total # errors, first error code
3465.fi
2fc26c3d
IC
3466.RS
3467.P
40943b9a
TK
3468.B
3469Additional Info (dependent on description being set):
3470.RE
3471.P
2fc26c3d 3472.nf
40943b9a
TK
3473 Text description
3474.fi
3475.P
3476Completion latency percentiles can be a grouping of up to 20 sets, so for the
3477terse output fio writes all of them. Each field will look like this:
3478.P
3479.nf
3480 1.00%=6112
3481.fi
3482.P
3483which is the Xth percentile, and the `usec' latency associated with it.
3484.P
3485For \fBDisk utilization\fR, all disks used by fio are shown. So for each disk there
3486will be a disk utilization section.
3487.P
3488Below is a single line containing short names for each of the fields in the
3489minimal output v3, separated by semicolons:
3490.P
3491.nf
3492 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 3493.fi
4e757af1
VF
3494.P
3495In client/server mode terse output differs from what appears when jobs are run
3496locally. Disk utilization data is omitted from the standard terse output and
3497for v3 and later appears on its own separate line at the end of each terse
3498reporting cycle.
44c82dba
VF
3499.SH JSON OUTPUT
3500The \fBjson\fR output format is intended to be both human readable and convenient
3501for automated parsing. For the most part its sections mirror those of the
3502\fBnormal\fR output. The \fBruntime\fR value is reported in msec and the \fBbw\fR value is
3503reported in 1024 bytes per second units.
3504.fi
d9e557ab
VF
3505.SH JSON+ OUTPUT
3506The \fBjson+\fR output format is identical to the \fBjson\fR output format except that it
3507adds a full dump of the completion latency bins. Each \fBbins\fR object contains a
3508set of (key, value) pairs where keys are latency durations and values count how
3509many I/Os had completion latencies of the corresponding duration. For example,
3510consider:
d9e557ab 3511.RS
40943b9a 3512.P
d9e557ab
VF
3513"bins" : { "87552" : 1, "89600" : 1, "94720" : 1, "96768" : 1, "97792" : 1, "99840" : 1, "100864" : 2, "103936" : 6, "104960" : 534, "105984" : 5995, "107008" : 7529, ... }
3514.RE
40943b9a 3515.P
d9e557ab
VF
3516This data indicates that one I/O required 87,552ns to complete, two I/Os required
3517100,864ns to complete, and 7529 I/Os required 107,008ns to complete.
40943b9a 3518.P
d9e557ab 3519Also included with fio is a Python script \fBfio_jsonplus_clat2csv\fR that takes
40943b9a
TK
3520json+ output and generates CSV\-formatted latency data suitable for plotting.
3521.P
d9e557ab 3522The latency durations actually represent the midpoints of latency intervals.
40943b9a 3523For details refer to `stat.h' in the fio source.
29dbd1e5 3524.SH TRACE FILE FORMAT
40943b9a
TK
3525There are two trace file format that you can encounter. The older (v1) format is
3526unsupported since version 1.20\-rc3 (March 2008). It will still be described
29dbd1e5 3527below in case that you get an old trace and want to understand it.
29dbd1e5 3528.P
40943b9a
TK
3529In any case the trace is a simple text file with a single action per line.
3530.TP
29dbd1e5 3531.B Trace file format v1
40943b9a 3532Each line represents a single I/O action in the following format:
29dbd1e5 3533.RS
40943b9a
TK
3534.RS
3535.P
29dbd1e5 3536rw, offset, length
29dbd1e5
JA
3537.RE
3538.P
40943b9a
TK
3539where `rw=0/1' for read/write, and the `offset' and `length' entries being in bytes.
3540.P
3541This format is not supported in fio versions >= 1.20\-rc3.
3542.RE
3543.TP
29dbd1e5 3544.B Trace file format v2
40943b9a
TK
3545The second version of the trace file format was added in fio version 1.17. It
3546allows to access more then one file per trace and has a bigger set of possible
3547file actions.
29dbd1e5 3548.RS
40943b9a 3549.P
29dbd1e5 3550The first line of the trace file has to be:
40943b9a
TK
3551.RS
3552.P
3553"fio version 2 iolog"
3554.RE
3555.P
29dbd1e5 3556Following this can be lines in two different formats, which are described below.
40943b9a
TK
3557.P
3558.B
29dbd1e5 3559The file management format:
40943b9a
TK
3560.RS
3561filename action
29dbd1e5 3562.P
40943b9a 3563The `filename' is given as an absolute path. The `action' can be one of these:
29dbd1e5
JA
3564.RS
3565.TP
3566.B add
40943b9a 3567Add the given `filename' to the trace.
29dbd1e5
JA
3568.TP
3569.B open
40943b9a
TK
3570Open the file with the given `filename'. The `filename' has to have
3571been added with the \fBadd\fR action before.
29dbd1e5
JA
3572.TP
3573.B close
40943b9a
TK
3574Close the file with the given `filename'. The file has to have been
3575\fBopen\fRed before.
3576.RE
29dbd1e5 3577.RE
29dbd1e5 3578.P
40943b9a
TK
3579.B
3580The file I/O action format:
3581.RS
3582filename action offset length
29dbd1e5 3583.P
40943b9a
TK
3584The `filename' is given as an absolute path, and has to have been \fBadd\fRed and
3585\fBopen\fRed before it can be used with this format. The `offset' and `length' are
3586given in bytes. The `action' can be one of these:
29dbd1e5
JA
3587.RS
3588.TP
3589.B wait
40943b9a
TK
3590Wait for `offset' microseconds. Everything below 100 is discarded.
3591The time is relative to the previous `wait' statement.
29dbd1e5
JA
3592.TP
3593.B read
40943b9a 3594Read `length' bytes beginning from `offset'.
29dbd1e5
JA
3595.TP
3596.B write
40943b9a 3597Write `length' bytes beginning from `offset'.
29dbd1e5
JA
3598.TP
3599.B sync
40943b9a 3600\fBfsync\fR\|(2) the file.
29dbd1e5
JA
3601.TP
3602.B datasync
40943b9a 3603\fBfdatasync\fR\|(2) the file.
29dbd1e5
JA
3604.TP
3605.B trim
40943b9a
TK
3606Trim the given file from the given `offset' for `length' bytes.
3607.RE
29dbd1e5 3608.RE
b9921d1a
DZ
3609.SH I/O REPLAY \- MERGING TRACES
3610Colocation is a common practice used to get the most out of a machine.
3611Knowing which workloads play nicely with each other and which ones don't is
3612a much harder task. While fio can replay workloads concurrently via multiple
3613jobs, it leaves some variability up to the scheduler making results harder to
3614reproduce. Merging is a way to make the order of events consistent.
3615.P
3616Merging is integrated into I/O replay and done when a \fBmerge_blktrace_file\fR
3617is specified. The list of files passed to \fBread_iolog\fR go through the merge
3618process and output a single file stored to the specified file. The output file is
3619passed on as if it were the only file passed to \fBread_iolog\fR. An example would
3620look like:
3621.RS
3622.P
3623$ fio \-\-read_iolog="<file1>:<file2>" \-\-merge_blktrace_file="<output_file>"
3624.RE
3625.P
3626Creating only the merged file can be done by passing the command line argument
3627\fBmerge-blktrace-only\fR.
87a48ada
DZ
3628.P
3629Scaling traces can be done to see the relative impact of any particular trace
3630being slowed down or sped up. \fBmerge_blktrace_scalars\fR takes in a colon
3631separated list of percentage scalars. It is index paired with the files passed
3632to \fBread_iolog\fR.
55bfd8c8
DZ
3633.P
3634With scaling, it may be desirable to match the running time of all traces.
3635This can be done with \fBmerge_blktrace_iters\fR. It is index paired with
3636\fBread_iolog\fR just like \fBmerge_blktrace_scalars\fR.
3637.P
3638In an example, given two traces, A and B, each 60s long. If we want to see
3639the impact of trace A issuing IOs twice as fast and repeat trace A over the
3640runtime of trace B, the following can be done:
3641.RS
3642.P
3643$ fio \-\-read_iolog="<trace_a>:"<trace_b>" \-\-merge_blktrace_file"<output_file>" \-\-merge_blktrace_scalars="50:100" \-\-merge_blktrace_iters="2:1"
3644.RE
3645.P
3646This runs trace A at 2x the speed twice for approximately the same runtime as
3647a single run of trace B.
29dbd1e5 3648.SH CPU IDLENESS PROFILING
40943b9a
TK
3649In some cases, we want to understand CPU overhead in a test. For example, we
3650test patches for the specific goodness of whether they reduce CPU usage.
3651Fio implements a balloon approach to create a thread per CPU that runs at idle
3652priority, meaning that it only runs when nobody else needs the cpu.
3653By measuring the amount of work completed by the thread, idleness of each CPU
3654can be derived accordingly.
3655.P
3656An unit work is defined as touching a full page of unsigned characters. Mean and
3657standard deviation of time to complete an unit work is reported in "unit work"
3658section. Options can be chosen to report detailed percpu idleness or overall
3659system idleness by aggregating percpu stats.
29dbd1e5 3660.SH VERIFICATION AND TRIGGERS
40943b9a
TK
3661Fio is usually run in one of two ways, when data verification is done. The first
3662is a normal write job of some sort with verify enabled. When the write phase has
3663completed, fio switches to reads and verifies everything it wrote. The second
3664model is running just the write phase, and then later on running the same job
3665(but with reads instead of writes) to repeat the same I/O patterns and verify
3666the contents. Both of these methods depend on the write phase being completed,
3667as fio otherwise has no idea how much data was written.
3668.P