glusterfs: update for new API
[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.
54eb4569
TK
204.SH "JOB FILE PARAMETERS"
205Some parameters take an option of a given type, such as an integer or a
206string. Anywhere a numeric value is required, an arithmetic expression may be
207used, provided it is surrounded by parentheses. Supported operators are:
d59aa780 208.RS
7db7a5a0 209.P
d59aa780 210.B addition (+)
7db7a5a0
TK
211.P
212.B subtraction (\-)
213.P
d59aa780 214.B multiplication (*)
7db7a5a0 215.P
d59aa780 216.B division (/)
7db7a5a0 217.P
d59aa780 218.B modulus (%)
7db7a5a0 219.P
d59aa780
JA
220.B exponentiation (^)
221.RE
d59aa780
JA
222.P
223For time values in expressions, units are microseconds by default. This is
224different than for time values not in expressions (not enclosed in
54eb4569
TK
225parentheses).
226.SH "PARAMETER TYPES"
227The following parameter types are used.
d60e92d1
AC
228.TP
229.I str
6b86fc18
TK
230String. A sequence of alphanumeric characters.
231.TP
232.I time
233Integer with possible time suffix. Without a unit value is interpreted as
234seconds unless otherwise specified. Accepts a suffix of 'd' for days, 'h' for
235hours, 'm' for minutes, 's' for seconds, 'ms' (or 'msec') for milliseconds and 'us'
236(or 'usec') for microseconds. For example, use 10m for 10 minutes.
d60e92d1
AC
237.TP
238.I int
6d500c2e
RE
239Integer. A whole number value, which may contain an integer prefix
240and an integer suffix.
0b43a833
TK
241.RS
242.RS
243.P
6b86fc18 244[*integer prefix*] **number** [*integer suffix*]
0b43a833
TK
245.RE
246.P
6b86fc18
TK
247The optional *integer prefix* specifies the number's base. The default
248is decimal. *0x* specifies hexadecimal.
0b43a833 249.P
6b86fc18
TK
250The optional *integer suffix* specifies the number's units, and includes an
251optional unit prefix and an optional unit. For quantities of data, the
252default unit is bytes. For quantities of time, the default unit is seconds
253unless otherwise specified.
0b43a833
TK
254.P
255With `kb_base=1000', fio follows international standards for unit
7db7a5a0 256prefixes. To specify power\-of\-10 decimal values defined in the
6b86fc18 257International System of Units (SI):
0b43a833
TK
258.RS
259.P
7db7a5a0 260.PD 0
eccce61a 261K means kilo (K) or 1000
7db7a5a0 262.P
eccce61a 263M means mega (M) or 1000**2
7db7a5a0 264.P
eccce61a 265G means giga (G) or 1000**3
7db7a5a0 266.P
eccce61a 267T means tera (T) or 1000**4
7db7a5a0 268.P
eccce61a 269P means peta (P) or 1000**5
7db7a5a0 270.PD
0b43a833
TK
271.RE
272.P
7db7a5a0 273To specify power\-of\-2 binary values defined in IEC 80000\-13:
0b43a833
TK
274.RS
275.P
7db7a5a0 276.PD 0
eccce61a 277Ki means kibi (Ki) or 1024
7db7a5a0 278.P
eccce61a 279Mi means mebi (Mi) or 1024**2
7db7a5a0 280.P
eccce61a 281Gi means gibi (Gi) or 1024**3
7db7a5a0 282.P
eccce61a 283Ti means tebi (Ti) or 1024**4
7db7a5a0 284.P
eccce61a 285Pi means pebi (Pi) or 1024**5
7db7a5a0 286.PD
0b43a833
TK
287.RE
288.P
289With `kb_base=1024' (the default), the unit prefixes are opposite
7db7a5a0 290from those specified in the SI and IEC 80000\-13 standards to provide
6b86fc18 291compatibility with old scripts. For example, 4k means 4096.
0b43a833 292.P
6b86fc18
TK
293For quantities of data, an optional unit of 'B' may be included
294(e.g., 'kB' is the same as 'k').
0b43a833 295.P
6b86fc18
TK
296The *integer suffix* is not case sensitive (e.g., m/mi mean mebi/mega,
297not milli). 'b' and 'B' both mean byte, not bit.
0b43a833
TK
298.P
299Examples with `kb_base=1000':
300.RS
301.P
7db7a5a0 302.PD 0
6d500c2e 3034 KiB: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
7db7a5a0 304.P
6d500c2e 3051 MiB: 1048576, 1m, 1024k
7db7a5a0 306.P
6d500c2e 3071 MB: 1000000, 1mi, 1000ki
7db7a5a0 308.P
6d500c2e 3091 TiB: 1073741824, 1t, 1024m, 1048576k
7db7a5a0 310.P
6d500c2e 3111 TB: 1000000000, 1ti, 1000mi, 1000000ki
7db7a5a0 312.PD
0b43a833
TK
313.RE
314.P
315Examples with `kb_base=1024' (default):
316.RS
317.P
7db7a5a0 318.PD 0
6d500c2e 3194 KiB: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
7db7a5a0 320.P
6d500c2e 3211 MiB: 1048576, 1m, 1024k
7db7a5a0 322.P
6d500c2e 3231 MB: 1000000, 1mi, 1000ki
7db7a5a0 324.P
6d500c2e 3251 TiB: 1073741824, 1t, 1024m, 1048576k
7db7a5a0 326.P
6d500c2e 3271 TB: 1000000000, 1ti, 1000mi, 1000000ki
7db7a5a0 328.PD
0b43a833
TK
329.RE
330.P
6d500c2e 331To specify times (units are not case sensitive):
0b43a833
TK
332.RS
333.P
7db7a5a0 334.PD 0
6d500c2e 335D means days
7db7a5a0 336.P
6d500c2e 337H means hours
7db7a5a0 338.P
6d500c2e 339M mean minutes
7db7a5a0 340.P
6d500c2e 341s or sec means seconds (default)
7db7a5a0 342.P
6d500c2e 343ms or msec means milliseconds
7db7a5a0 344.P
6d500c2e 345us or usec means microseconds
7db7a5a0 346.PD
0b43a833
TK
347.RE
348.P
6b86fc18 349If the option accepts an upper and lower range, use a colon ':' or
7db7a5a0 350minus '\-' to separate such values. See \fIirange\fR parameter type.
6b86fc18
TK
351If the lower value specified happens to be larger than the upper value
352the two values are swapped.
0b43a833 353.RE
d60e92d1
AC
354.TP
355.I bool
6b86fc18
TK
356Boolean. Usually parsed as an integer, however only defined for
357true and false (1 and 0).
d60e92d1
AC
358.TP
359.I irange
6b86fc18 360Integer range with suffix. Allows value range to be given, such as
7db7a5a0 3611024\-4096. A colon may also be used as the separator, e.g. 1k:4k. If the
6b86fc18 362option allows two sets of ranges, they can be specified with a ',' or '/'
7db7a5a0 363delimiter: 1k\-4k/8k\-32k. Also see \fIint\fR parameter type.
83349190
YH
364.TP
365.I float_list
6b86fc18 366A list of floating point numbers, separated by a ':' character.
523bad63 367.SH "JOB PARAMETERS"
54eb4569 368With the above in mind, here follows the complete list of fio job parameters.
523bad63 369.SS "Units"
d60e92d1 370.TP
523bad63
TK
371.BI kb_base \fR=\fPint
372Select the interpretation of unit prefixes in input parameters.
373.RS
374.RS
d60e92d1 375.TP
523bad63
TK
376.B 1000
377Inputs comply with IEC 80000\-13 and the International
378System of Units (SI). Use:
379.RS
380.P
381.PD 0
382\- power\-of\-2 values with IEC prefixes (e.g., KiB)
383.P
384\- power\-of\-10 values with SI prefixes (e.g., kB)
385.PD
386.RE
387.TP
388.B 1024
389Compatibility mode (default). To avoid breaking old scripts:
390.P
391.RS
392.PD 0
393\- power\-of\-2 values with SI prefixes
394.P
395\- power\-of\-10 values with IEC prefixes
396.PD
397.RE
398.RE
399.P
400See \fBbs\fR for more details on input parameters.
401.P
402Outputs always use correct prefixes. Most outputs include both
403side\-by\-side, like:
404.P
405.RS
406bw=2383.3kB/s (2327.4KiB/s)
407.RE
408.P
409If only one value is reported, then kb_base selects the one to use:
410.P
411.RS
412.PD 0
4131000 \-\- SI prefixes
414.P
4151024 \-\- IEC prefixes
416.PD
417.RE
418.RE
419.TP
420.BI unit_base \fR=\fPint
421Base unit for reporting. Allowed values are:
422.RS
423.RS
424.TP
425.B 0
426Use auto\-detection (default).
427.TP
428.B 8
429Byte based.
430.TP
431.B 1
432Bit based.
433.RE
434.RE
435.SS "Job description"
436.TP
437.BI name \fR=\fPstr
438ASCII name of the job. This may be used to override the name printed by fio
439for this job. Otherwise the job name is used. On the command line this
440parameter has the special purpose of also signaling the start of a new job.
9cc8cb91 441.TP
d60e92d1 442.BI description \fR=\fPstr
523bad63
TK
443Text description of the job. Doesn't do anything except dump this text
444description when this job is run. It's not parsed.
445.TP
446.BI loops \fR=\fPint
447Run the specified number of iterations of this job. Used to repeat the same
448workload a given number of times. Defaults to 1.
449.TP
450.BI numjobs \fR=\fPint
451Create the specified number of clones of this job. Each clone of job
452is spawned as an independent thread or process. May be used to setup a
453larger number of threads/processes doing the same thing. Each thread is
454reported separately; to see statistics for all clones as a whole, use
455\fBgroup_reporting\fR in conjunction with \fBnew_group\fR.
456See \fB\-\-max\-jobs\fR. Default: 1.
457.SS "Time related parameters"
458.TP
459.BI runtime \fR=\fPtime
460Tell fio to terminate processing after the specified period of time. It
461can be quite hard to determine for how long a specified job will run, so
462this parameter is handy to cap the total runtime to a given time. When
f1dd3fb1 463the unit is omitted, the value is interpreted in seconds.
523bad63
TK
464.TP
465.BI time_based
466If set, fio will run for the duration of the \fBruntime\fR specified
467even if the file(s) are completely read or written. It will simply loop over
468the same workload as many times as the \fBruntime\fR allows.
469.TP
470.BI startdelay \fR=\fPirange(int)
471Delay the start of job for the specified amount of time. Can be a single
472value or a range. When given as a range, each thread will choose a value
473randomly from within the range. Value is in seconds if a unit is omitted.
474.TP
475.BI ramp_time \fR=\fPtime
476If set, fio will run the specified workload for this amount of time before
477logging any performance numbers. Useful for letting performance settle
478before logging results, thus minimizing the runtime required for stable
479results. Note that the \fBramp_time\fR is considered lead in time for a job,
480thus it will increase the total runtime if a special timeout or
481\fBruntime\fR is specified. When the unit is omitted, the value is
482given in seconds.
483.TP
484.BI clocksource \fR=\fPstr
485Use the given clocksource as the base of timing. The supported options are:
486.RS
487.RS
488.TP
489.B gettimeofday
490\fBgettimeofday\fR\|(2)
491.TP
492.B clock_gettime
493\fBclock_gettime\fR\|(2)
494.TP
495.B cpu
496Internal CPU clock source
497.RE
498.P
499\fBcpu\fR is the preferred clocksource if it is reliable, as it is very fast (and
500fio is heavy on time calls). Fio will automatically use this clocksource if
501it's supported and considered reliable on the system it is running on,
502unless another clocksource is specifically set. For x86/x86\-64 CPUs, this
503means supporting TSC Invariant.
504.RE
505.TP
506.BI gtod_reduce \fR=\fPbool
507Enable all of the \fBgettimeofday\fR\|(2) reducing options
508(\fBdisable_clat\fR, \fBdisable_slat\fR, \fBdisable_bw_measurement\fR) plus
509reduce precision of the timeout somewhat to really shrink the
510\fBgettimeofday\fR\|(2) call count. With this option enabled, we only do
511about 0.4% of the \fBgettimeofday\fR\|(2) calls we would have done if all
512time keeping was enabled.
513.TP
514.BI gtod_cpu \fR=\fPint
515Sometimes it's cheaper to dedicate a single thread of execution to just
516getting the current time. Fio (and databases, for instance) are very
517intensive on \fBgettimeofday\fR\|(2) calls. With this option, you can set
518one CPU aside for doing nothing but logging current time to a shared memory
519location. Then the other threads/processes that run I/O workloads need only
520copy that segment, instead of entering the kernel with a
521\fBgettimeofday\fR\|(2) call. The CPU set aside for doing these time
522calls will be excluded from other uses. Fio will manually clear it from the
523CPU mask of other jobs.
524.SS "Target file/device"
d60e92d1
AC
525.TP
526.BI directory \fR=\fPstr
523bad63
TK
527Prefix \fBfilename\fRs with this directory. Used to place files in a different
528location than `./'. You can specify a number of directories by
529separating the names with a ':' character. These directories will be
530assigned equally distributed to job clones created by \fBnumjobs\fR as
531long as they are using generated filenames. If specific \fBfilename\fR(s) are
532set fio will use the first listed directory, and thereby matching the
f4401bf8
SW
533\fBfilename\fR semantic (which generates a file for each clone if not
534specified, but lets all clones use the same file if set).
523bad63
TK
535.RS
536.P
ffc90a44 537See the \fBfilename\fR option for information on how to escape ':' and '\\'
523bad63 538characters within the directory path itself.
f4401bf8
SW
539.P
540Note: To control the directory fio will use for internal state files
541use \fB\-\-aux\-path\fR.
523bad63 542.RE
d60e92d1
AC
543.TP
544.BI filename \fR=\fPstr
523bad63
TK
545Fio normally makes up a \fBfilename\fR based on the job name, thread number, and
546file number (see \fBfilename_format\fR). If you want to share files
547between threads in a job or several
548jobs with fixed file paths, specify a \fBfilename\fR for each of them to override
549the default. If the ioengine is file based, you can specify a number of files
550by separating the names with a ':' colon. So if you wanted a job to open
551`/dev/sda' and `/dev/sdb' as the two working files, you would use
552`filename=/dev/sda:/dev/sdb'. This also means that whenever this option is
553specified, \fBnrfiles\fR is ignored. The size of regular files specified
554by this option will be \fBsize\fR divided by number of files unless an
555explicit size is specified by \fBfilesize\fR.
556.RS
557.P
ffc90a44 558Each colon and backslash in the wanted path must be escaped with a '\\'
523bad63
TK
559character. For instance, if the path is `/dev/dsk/foo@3,0:c' then you
560would use `filename=/dev/dsk/foo@3,0\\:c' and if the path is
ffc90a44 561`F:\\filename' then you would use `filename=F\\:\\\\filename'.
523bad63 562.P
ffc90a44
SW
563On Windows, disk devices are accessed as `\\\\.\\PhysicalDrive0' for
564the first device, `\\\\.\\PhysicalDrive1' for the second etc.
523bad63
TK
565Note: Windows and FreeBSD prevent write access to areas
566of the disk containing in\-use data (e.g. filesystems).
567.P
568The filename `\-' is a reserved name, meaning *stdin* or *stdout*. Which
569of the two depends on the read/write direction set.
570.RE
d60e92d1 571.TP
de98bd30 572.BI filename_format \fR=\fPstr
523bad63
TK
573If sharing multiple files between jobs, it is usually necessary to have fio
574generate the exact names that you want. By default, fio will name a file
de98bd30 575based on the default file format specification of
523bad63 576`jobname.jobnumber.filenumber'. With this option, that can be
de98bd30
JA
577customized. Fio will recognize and replace the following keywords in this
578string:
579.RS
580.RS
581.TP
582.B $jobname
583The name of the worker thread or process.
584.TP
585.B $jobnum
586The incremental number of the worker thread or process.
587.TP
588.B $filenum
589The incremental number of the file for that worker thread or process.
590.RE
591.P
523bad63
TK
592To have dependent jobs share a set of files, this option can be set to have
593fio generate filenames that are shared between the two. For instance, if
594`testfiles.$filenum' is specified, file number 4 for any job will be
595named `testfiles.4'. The default of `$jobname.$jobnum.$filenum'
de98bd30 596will be used if no other format specifier is given.
645943c0
JB
597.P
598If you specify a path then the directories will be created up to the main
599directory for the file. So for example if you specify `a/b/c/$jobnum` then the
600directories a/b/c will be created before the file setup part of the job. If you
601specify \fBdirectory\fR then the path will be relative that directory, otherwise
602it is treated as the absolute path.
de98bd30 603.RE
de98bd30 604.TP
922a5be8 605.BI unique_filename \fR=\fPbool
523bad63
TK
606To avoid collisions between networked clients, fio defaults to prefixing any
607generated filenames (with a directory specified) with the source of the
608client connecting. To disable this behavior, set this option to 0.
609.TP
610.BI opendir \fR=\fPstr
611Recursively open any files below directory \fIstr\fR.
922a5be8 612.TP
3ce9dcaf 613.BI lockfile \fR=\fPstr
523bad63
TK
614Fio defaults to not locking any files before it does I/O to them. If a file
615or file descriptor is shared, fio can serialize I/O to that file to make the
616end result consistent. This is usual for emulating real workloads that share
617files. The lock modes are:
3ce9dcaf
JA
618.RS
619.RS
620.TP
621.B none
523bad63 622No locking. The default.
3ce9dcaf
JA
623.TP
624.B exclusive
523bad63 625Only one thread or process may do I/O at a time, excluding all others.
3ce9dcaf
JA
626.TP
627.B readwrite
523bad63
TK
628Read\-write locking on the file. Many readers may
629access the file at the same time, but writes get exclusive access.
3ce9dcaf 630.RE
ce594fbe 631.RE
523bad63
TK
632.TP
633.BI nrfiles \fR=\fPint
634Number of files to use for this job. Defaults to 1. The size of files
635will be \fBsize\fR divided by this unless explicit size is specified by
636\fBfilesize\fR. Files are created for each thread separately, and each
637file will have a file number within its name by default, as explained in
638\fBfilename\fR section.
639.TP
640.BI openfiles \fR=\fPint
641Number of files to keep open at the same time. Defaults to the same as
642\fBnrfiles\fR, can be set smaller to limit the number simultaneous
643opens.
644.TP
645.BI file_service_type \fR=\fPstr
646Defines how fio decides which file from a job to service next. The following
647types are defined:
648.RS
649.RS
650.TP
651.B random
652Choose a file at random.
653.TP
654.B roundrobin
655Round robin over opened files. This is the default.
656.TP
657.B sequential
658Finish one file before moving on to the next. Multiple files can
659still be open depending on \fBopenfiles\fR.
660.TP
661.B zipf
662Use a Zipf distribution to decide what file to access.
663.TP
664.B pareto
665Use a Pareto distribution to decide what file to access.
666.TP
667.B normal
668Use a Gaussian (normal) distribution to decide what file to access.
669.TP
670.B gauss
671Alias for normal.
672.RE
3ce9dcaf 673.P
523bad63
TK
674For \fBrandom\fR, \fBroundrobin\fR, and \fBsequential\fR, a postfix can be appended to
675tell fio how many I/Os to issue before switching to a new file. For example,
676specifying `file_service_type=random:8' would cause fio to issue
6778 I/Os before selecting a new file at random. For the non\-uniform
678distributions, a floating point postfix can be given to influence how the
679distribution is skewed. See \fBrandom_distribution\fR for a description
680of how that would work.
681.RE
682.TP
683.BI ioscheduler \fR=\fPstr
684Attempt to switch the device hosting the file to the specified I/O scheduler
685before running.
686.TP
687.BI create_serialize \fR=\fPbool
688If true, serialize the file creation for the jobs. This may be handy to
689avoid interleaving of data files, which may greatly depend on the filesystem
690used and even the number of processors in the system. Default: true.
691.TP
692.BI create_fsync \fR=\fPbool
693\fBfsync\fR\|(2) the data file after creation. This is the default.
694.TP
695.BI create_on_open \fR=\fPbool
696If true, don't pre\-create files but allow the job's open() to create a file
697when it's time to do I/O. Default: false \-\- pre\-create all necessary files
698when the job starts.
699.TP
700.BI create_only \fR=\fPbool
701If true, fio will only run the setup phase of the job. If files need to be
702laid out or updated on disk, only that will be done \-\- the actual job contents
703are not executed. Default: false.
704.TP
705.BI allow_file_create \fR=\fPbool
706If true, fio is permitted to create files as part of its workload. If this
707option is false, then fio will error out if
708the files it needs to use don't already exist. Default: true.
709.TP
710.BI allow_mounted_write \fR=\fPbool
711If this isn't set, fio will abort jobs that are destructive (e.g. that write)
712to what appears to be a mounted device or partition. This should help catch
713creating inadvertently destructive tests, not realizing that the test will
714destroy data on the mounted file system. Note that some platforms don't allow
715writing against a mounted device regardless of this option. Default: false.
716.TP
717.BI pre_read \fR=\fPbool
718If this is given, files will be pre\-read into memory before starting the
719given I/O operation. This will also clear the \fBinvalidate\fR flag,
720since it is pointless to pre\-read and then drop the cache. This will only
721work for I/O engines that are seek\-able, since they allow you to read the
722same data multiple times. Thus it will not work on non\-seekable I/O engines
723(e.g. network, splice). Default: false.
724.TP
725.BI unlink \fR=\fPbool
726Unlink the job files when done. Not the default, as repeated runs of that
727job would then waste time recreating the file set again and again. Default:
728false.
729.TP
730.BI unlink_each_loop \fR=\fPbool
731Unlink job files after each iteration or loop. Default: false.
732.TP
7b865a2f
BVA
733.BI zonemode \fR=\fPstr
734Accepted values are:
735.RS
736.RS
737.TP
738.B none
739The \fBzonerange\fR, \fBzonesize\fR and \fBzoneskip\fR parameters are ignored.
740.TP
741.B strided
742I/O happens in a single zone until \fBzonesize\fR bytes have been transferred.
743After that number of bytes has been transferred processing of the next zone
744starts.
745.TP
746.B zbd
747Zoned block device mode. I/O happens sequentially in each zone, even if random
748I/O has been selected. Random I/O happens across all zones instead of being
749restricted to a single zone.
750.RE
751.RE
523bad63
TK
752.TP
753.BI zonerange \fR=\fPint
7b865a2f 754Size of a single zone. See also \fBzonesize\fR and \fBzoneskip\fR.
5faddc64
BVA
755.TP
756.BI zonesize \fR=\fPint
7b865a2f
BVA
757For \fBzonemode\fR=strided, this is the number of bytes to transfer before
758skipping \fBzoneskip\fR bytes. If this parameter is smaller than
759\fBzonerange\fR then only a fraction of each zone with \fBzonerange\fR bytes
760will be accessed. If this parameter is larger than \fBzonerange\fR then each
761zone will be accessed multiple times before skipping to the next zone.
762
763For \fBzonemode\fR=zbd, this is the size of a single zone. The \fBzonerange\fR
764parameter is ignored in this mode.
523bad63
TK
765.TP
766.BI zoneskip \fR=\fPint
7b865a2f
BVA
767For \fBzonemode\fR=strided, the number of bytes to skip after \fBzonesize\fR
768bytes of data have been transferred. This parameter must be zero for
769\fBzonemode\fR=zbd.
5faddc64 770
bfbdd35b
BVA
771.TP
772.BI read_beyond_wp \fR=\fPbool
773This parameter applies to \fBzonemode=zbd\fR only.
774
775Zoned block devices are block devices that consist of multiple zones. Each
776zone has a type, e.g. conventional or sequential. A conventional zone can be
777written at any offset that is a multiple of the block size. Sequential zones
778must be written sequentially. The position at which a write must occur is
779called the write pointer. A zoned block device can be either drive
780managed, host managed or host aware. For host managed devices the host must
781ensure that writes happen sequentially. Fio recognizes host managed devices
782and serializes writes to sequential zones for these devices.
783
784If a read occurs in a sequential zone beyond the write pointer then the zoned
785block device will complete the read without reading any data from the storage
786medium. Since such reads lead to unrealistically high bandwidth and IOPS
787numbers fio only reads beyond the write pointer if explicitly told to do
788so. Default: false.
59b07544
BVA
789.TP
790.BI max_open_zones \fR=\fPint
791When running a random write test across an entire drive many more zones will be
792open than in a typical application workload. Hence this command line option
793that allows to limit the number of open zones. The number of open zones is
794defined as the number of zones to which write commands are issued.
a7c2b6fc
BVA
795.TP
796.BI zone_reset_threshold \fR=\fPfloat
797A number between zero and one that indicates the ratio of logical blocks with
798data to the total number of logical blocks in the test above which zones
799should be reset periodically.
800.TP
801.BI zone_reset_frequency \fR=\fPfloat
802A number between zero and one that indicates how often a zone reset should be
803issued if the zone reset threshold has been exceeded. A zone reset is
804submitted after each (1 / zone_reset_frequency) write requests. This and the
805previous parameter can be used to simulate garbage collection activity.
bfbdd35b 806
523bad63
TK
807.SS "I/O type"
808.TP
809.BI direct \fR=\fPbool
810If value is true, use non\-buffered I/O. This is usually O_DIRECT. Note that
8e889110 811OpenBSD and ZFS on Solaris don't support direct I/O. On Windows the synchronous
523bad63
TK
812ioengines don't support direct I/O. Default: false.
813.TP
814.BI atomic \fR=\fPbool
815If value is true, attempt to use atomic direct I/O. Atomic writes are
816guaranteed to be stable once acknowledged by the operating system. Only
817Linux supports O_ATOMIC right now.
818.TP
819.BI buffered \fR=\fPbool
820If value is true, use buffered I/O. This is the opposite of the
821\fBdirect\fR option. Defaults to true.
d60e92d1
AC
822.TP
823.BI readwrite \fR=\fPstr "\fR,\fP rw" \fR=\fPstr
523bad63 824Type of I/O pattern. Accepted values are:
d60e92d1
AC
825.RS
826.RS
827.TP
828.B read
d1429b5c 829Sequential reads.
d60e92d1
AC
830.TP
831.B write
d1429b5c 832Sequential writes.
d60e92d1 833.TP
fa769d44 834.B trim
3740cfc8 835Sequential trims (Linux block devices and SCSI character devices only).
fa769d44 836.TP
d60e92d1 837.B randread
d1429b5c 838Random reads.
d60e92d1
AC
839.TP
840.B randwrite
d1429b5c 841Random writes.
d60e92d1 842.TP
fa769d44 843.B randtrim
3740cfc8 844Random trims (Linux block devices and SCSI character devices only).
fa769d44 845.TP
523bad63
TK
846.B rw,readwrite
847Sequential mixed reads and writes.
d60e92d1 848.TP
ff6bb260 849.B randrw
523bad63 850Random mixed reads and writes.
82a90686
JA
851.TP
852.B trimwrite
523bad63
TK
853Sequential trim+write sequences. Blocks will be trimmed first,
854then the same blocks will be written to.
d60e92d1
AC
855.RE
856.P
523bad63
TK
857Fio defaults to read if the option is not specified. For the mixed I/O
858types, the default is to split them 50/50. For certain types of I/O the
859result may still be skewed a bit, since the speed may be different.
860.P
861It is possible to specify the number of I/Os to do before getting a new
862offset by appending `:<nr>' to the end of the string given. For a
863random read, it would look like `rw=randread:8' for passing in an offset
864modifier with a value of 8. If the suffix is used with a sequential I/O
865pattern, then the `<nr>' value specified will be added to the generated
866offset for each I/O turning sequential I/O into sequential I/O with holes.
867For instance, using `rw=write:4k' will skip 4k for every write. Also see
868the \fBrw_sequencer\fR option.
d60e92d1
AC
869.RE
870.TP
38dad62d 871.BI rw_sequencer \fR=\fPstr
523bad63
TK
872If an offset modifier is given by appending a number to the `rw=\fIstr\fR'
873line, then this option controls how that number modifies the I/O offset
874being generated. Accepted values are:
38dad62d
JA
875.RS
876.RS
877.TP
878.B sequential
523bad63 879Generate sequential offset.
38dad62d
JA
880.TP
881.B identical
523bad63 882Generate the same offset.
38dad62d
JA
883.RE
884.P
523bad63
TK
885\fBsequential\fR is only useful for random I/O, where fio would normally
886generate a new random offset for every I/O. If you append e.g. 8 to randread,
887you would get a new random offset for every 8 I/Os. The result would be a
888seek for only every 8 I/Os, instead of for every I/O. Use `rw=randread:8'
889to specify that. As sequential I/O is already sequential, setting
890\fBsequential\fR for that would not result in any differences. \fBidentical\fR
891behaves in a similar fashion, except it sends the same offset 8 number of
892times before generating a new offset.
38dad62d 893.RE
90fef2d1 894.TP
771e58be
JA
895.BI unified_rw_reporting \fR=\fPbool
896Fio normally reports statistics on a per data direction basis, meaning that
523bad63
TK
897reads, writes, and trims are accounted and reported separately. If this
898option is set fio sums the results and report them as "mixed" instead.
771e58be 899.TP
d60e92d1 900.BI randrepeat \fR=\fPbool
523bad63
TK
901Seed the random number generator used for random I/O patterns in a
902predictable way so the pattern is repeatable across runs. Default: true.
56e2a5fc
CE
903.TP
904.BI allrandrepeat \fR=\fPbool
905Seed all random number generators in a predictable way so results are
523bad63 906repeatable across runs. Default: false.
d60e92d1 907.TP
04778baf
JA
908.BI randseed \fR=\fPint
909Seed the random number generators based on this seed value, to be able to
910control what sequence of output is being generated. If not set, the random
911sequence depends on the \fBrandrepeat\fR setting.
912.TP
a596f047 913.BI fallocate \fR=\fPstr
523bad63
TK
914Whether pre\-allocation is performed when laying down files.
915Accepted values are:
a596f047
EG
916.RS
917.RS
918.TP
919.B none
523bad63 920Do not pre\-allocate space.
a596f047 921.TP
2c3e17be 922.B native
523bad63
TK
923Use a platform's native pre\-allocation call but fall back to
924\fBnone\fR behavior if it fails/is not implemented.
2c3e17be 925.TP
a596f047 926.B posix
523bad63 927Pre\-allocate via \fBposix_fallocate\fR\|(3).
a596f047
EG
928.TP
929.B keep
523bad63
TK
930Pre\-allocate via \fBfallocate\fR\|(2) with
931FALLOC_FL_KEEP_SIZE set.
a596f047
EG
932.TP
933.B 0
523bad63 934Backward\-compatible alias for \fBnone\fR.
a596f047
EG
935.TP
936.B 1
523bad63 937Backward\-compatible alias for \fBposix\fR.
a596f047
EG
938.RE
939.P
523bad63
TK
940May not be available on all supported platforms. \fBkeep\fR is only available
941on Linux. If using ZFS on Solaris this cannot be set to \fBposix\fR
942because ZFS doesn't support pre\-allocation. Default: \fBnative\fR if any
943pre\-allocation methods are available, \fBnone\fR if not.
a596f047 944.RE
7bc8c2cf 945.TP
ecb2083d 946.BI fadvise_hint \fR=\fPstr
c712c97a
JA
947Use \fBposix_fadvise\fR\|(2) or \fBposix_madvise\fR\|(2) to advise the kernel
948what I/O patterns are likely to be issued. Accepted values are:
ecb2083d
JA
949.RS
950.RS
951.TP
952.B 0
953Backwards compatible hint for "no hint".
954.TP
955.B 1
956Backwards compatible hint for "advise with fio workload type". This
523bad63 957uses FADV_RANDOM for a random workload, and FADV_SEQUENTIAL
ecb2083d
JA
958for a sequential workload.
959.TP
960.B sequential
523bad63 961Advise using FADV_SEQUENTIAL.
ecb2083d
JA
962.TP
963.B random
523bad63 964Advise using FADV_RANDOM.
ecb2083d
JA
965.RE
966.RE
d60e92d1 967.TP
8f4b9f24 968.BI write_hint \fR=\fPstr
523bad63
TK
969Use \fBfcntl\fR\|(2) to advise the kernel what life time to expect
970from a write. Only supported on Linux, as of version 4.13. Accepted
8f4b9f24
JA
971values are:
972.RS
973.RS
974.TP
975.B none
976No particular life time associated with this file.
977.TP
978.B short
979Data written to this file has a short life time.
980.TP
981.B medium
982Data written to this file has a medium life time.
983.TP
984.B long
985Data written to this file has a long life time.
986.TP
987.B extreme
988Data written to this file has a very long life time.
989.RE
523bad63
TK
990.P
991The values are all relative to each other, and no absolute meaning
992should be associated with them.
8f4b9f24 993.RE
37659335 994.TP
523bad63
TK
995.BI offset \fR=\fPint
996Start I/O at the provided offset in the file, given as either a fixed size in
83c8b093
JF
997bytes or a percentage. If a percentage is given, the generated offset will be
998aligned to the minimum \fBblocksize\fR or to the value of \fBoffset_align\fR if
999provided. Data before the given offset will not be touched. This
523bad63
TK
1000effectively caps the file size at `real_size \- offset'. Can be combined with
1001\fBsize\fR to constrain the start and end range of the I/O workload.
1002A percentage can be specified by a number between 1 and 100 followed by '%',
1003for example, `offset=20%' to specify 20%.
6d500c2e 1004.TP
83c8b093
JF
1005.BI offset_align \fR=\fPint
1006If set to non-zero value, the byte offset generated by a percentage \fBoffset\fR
1007is aligned upwards to this value. Defaults to 0 meaning that a percentage
1008offset is aligned to the minimum block size.
1009.TP
523bad63
TK
1010.BI offset_increment \fR=\fPint
1011If this is provided, then the real offset becomes `\fBoffset\fR + \fBoffset_increment\fR
1012* thread_number', where the thread number is a counter that starts at 0 and
1013is incremented for each sub\-job (i.e. when \fBnumjobs\fR option is
1014specified). This option is useful if there are several jobs which are
1015intended to operate on a file in parallel disjoint segments, with even
1016spacing between the starting points.
6d500c2e 1017.TP
523bad63
TK
1018.BI number_ios \fR=\fPint
1019Fio will normally perform I/Os until it has exhausted the size of the region
1020set by \fBsize\fR, or if it exhaust the allocated time (or hits an error
1021condition). With this setting, the range/size can be set independently of
1022the number of I/Os to perform. When fio reaches this number, it will exit
1023normally and report status. Note that this does not extend the amount of I/O
1024that will be done, it will only stop fio if this condition is met before
1025other end\-of\-job criteria.
d60e92d1 1026.TP
523bad63
TK
1027.BI fsync \fR=\fPint
1028If writing to a file, issue an \fBfsync\fR\|(2) (or its equivalent) of
1029the dirty data for every number of blocks given. For example, if you give 32
1030as a parameter, fio will sync the file after every 32 writes issued. If fio is
1031using non\-buffered I/O, we may not sync the file. The exception is the sg
1032I/O engine, which synchronizes the disk cache anyway. Defaults to 0, which
1033means fio does not periodically issue and wait for a sync to complete. Also
1034see \fBend_fsync\fR and \fBfsync_on_close\fR.
6d500c2e 1035.TP
523bad63
TK
1036.BI fdatasync \fR=\fPint
1037Like \fBfsync\fR but uses \fBfdatasync\fR\|(2) to only sync data and
1038not metadata blocks. In Windows, FreeBSD, and DragonFlyBSD there is no
1039\fBfdatasync\fR\|(2) so this falls back to using \fBfsync\fR\|(2).
1040Defaults to 0, which means fio does not periodically issue and wait for a
1041data\-only sync to complete.
d60e92d1 1042.TP
523bad63
TK
1043.BI write_barrier \fR=\fPint
1044Make every N\-th write a barrier write.
901bb994 1045.TP
523bad63
TK
1046.BI sync_file_range \fR=\fPstr:int
1047Use \fBsync_file_range\fR\|(2) for every \fIint\fR number of write
1048operations. Fio will track range of writes that have happened since the last
1049\fBsync_file_range\fR\|(2) call. \fIstr\fR can currently be one or more of:
1050.RS
1051.RS
fd68418e 1052.TP
523bad63
TK
1053.B wait_before
1054SYNC_FILE_RANGE_WAIT_BEFORE
c5751c62 1055.TP
523bad63
TK
1056.B write
1057SYNC_FILE_RANGE_WRITE
c5751c62 1058.TP
523bad63
TK
1059.B wait_after
1060SYNC_FILE_RANGE_WRITE_AFTER
2fa5a241 1061.RE
523bad63
TK
1062.P
1063So if you do `sync_file_range=wait_before,write:8', fio would use
1064`SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE' for every 8
1065writes. Also see the \fBsync_file_range\fR\|(2) man page. This option is
1066Linux specific.
2fa5a241 1067.RE
ce35b1ec 1068.TP
523bad63
TK
1069.BI overwrite \fR=\fPbool
1070If true, writes to a file will always overwrite existing data. If the file
1071doesn't already exist, it will be created before the write phase begins. If
1072the file exists and is large enough for the specified write phase, nothing
1073will be done. Default: false.
5c94b008 1074.TP
523bad63
TK
1075.BI end_fsync \fR=\fPbool
1076If true, \fBfsync\fR\|(2) file contents when a write stage has completed.
1077Default: false.
d60e92d1 1078.TP
523bad63
TK
1079.BI fsync_on_close \fR=\fPbool
1080If true, fio will \fBfsync\fR\|(2) a dirty file on close. This differs
1081from \fBend_fsync\fR in that it will happen on every file close, not
1082just at the end of the job. Default: false.
d60e92d1 1083.TP
523bad63
TK
1084.BI rwmixread \fR=\fPint
1085Percentage of a mixed workload that should be reads. Default: 50.
1086.TP
1087.BI rwmixwrite \fR=\fPint
1088Percentage of a mixed workload that should be writes. If both
1089\fBrwmixread\fR and \fBrwmixwrite\fR is given and the values do not
1090add up to 100%, the latter of the two will be used to override the
1091first. This may interfere with a given rate setting, if fio is asked to
1092limit reads or writes to a certain rate. If that is the case, then the
1093distribution may be skewed. Default: 50.
1094.TP
1095.BI random_distribution \fR=\fPstr:float[,str:float][,str:float]
1096By default, fio will use a completely uniform random distribution when asked
1097to perform random I/O. Sometimes it is useful to skew the distribution in
1098specific ways, ensuring that some parts of the data is more hot than others.
1099fio includes the following distribution models:
d60e92d1
AC
1100.RS
1101.RS
1102.TP
1103.B random
523bad63 1104Uniform random distribution
8c07860d
JA
1105.TP
1106.B zipf
523bad63 1107Zipf distribution
8c07860d
JA
1108.TP
1109.B pareto
523bad63 1110Pareto distribution
8c07860d 1111.TP
dd3503d3 1112.B normal
523bad63 1113Normal (Gaussian) distribution
dd3503d3 1114.TP
523bad63
TK
1115.B zoned
1116Zoned random distribution
59466396
JA
1117.B zoned_abs
1118Zoned absolute random distribution
d60e92d1
AC
1119.RE
1120.P
523bad63
TK
1121When using a \fBzipf\fR or \fBpareto\fR distribution, an input value is also
1122needed to define the access pattern. For \fBzipf\fR, this is the `Zipf theta'.
1123For \fBpareto\fR, it's the `Pareto power'. Fio includes a test
1124program, \fBfio\-genzipf\fR, that can be used visualize what the given input
1125values will yield in terms of hit rates. If you wanted to use \fBzipf\fR with
1126a `theta' of 1.2, you would use `random_distribution=zipf:1.2' as the
1127option. If a non\-uniform model is used, fio will disable use of the random
1128map. For the \fBnormal\fR distribution, a normal (Gaussian) deviation is
1129supplied as a value between 0 and 100.
1130.P
1131For a \fBzoned\fR distribution, fio supports specifying percentages of I/O
1132access that should fall within what range of the file or device. For
1133example, given a criteria of:
d60e92d1 1134.RS
523bad63
TK
1135.P
1136.PD 0
113760% of accesses should be to the first 10%
1138.P
113930% of accesses should be to the next 20%
1140.P
11418% of accesses should be to the next 30%
1142.P
11432% of accesses should be to the next 40%
1144.PD
1145.RE
1146.P
1147we can define that through zoning of the random accesses. For the above
1148example, the user would do:
1149.RS
1150.P
1151random_distribution=zoned:60/10:30/20:8/30:2/40
1152.RE
1153.P
59466396
JA
1154A \fBzoned_abs\fR distribution works exactly like the\fBzoned\fR, except that
1155it takes absolute sizes. For example, let's say you wanted to define access
1156according to the following criteria:
1157.RS
1158.P
1159.PD 0
116060% of accesses should be to the first 20G
1161.P
116230% of accesses should be to the next 100G
1163.P
116410% of accesses should be to the next 500G
1165.PD
1166.RE
1167.P
1168we can define an absolute zoning distribution with:
1169.RS
1170.P
1171random_distribution=zoned:60/10:30/20:8/30:2/40
1172.RE
1173.P
6a16ece8
JA
1174For both \fBzoned\fR and \fBzoned_abs\fR, fio supports defining up to 256
1175separate zones.
1176.P
59466396 1177Similarly to how \fBbssplit\fR works for setting ranges and percentages
523bad63
TK
1178of block sizes. Like \fBbssplit\fR, it's possible to specify separate
1179zones for reads, writes, and trims. If just one set is given, it'll apply to
1180all of them.
1181.RE
1182.TP
1183.BI percentage_random \fR=\fPint[,int][,int]
1184For a random workload, set how big a percentage should be random. This
1185defaults to 100%, in which case the workload is fully random. It can be set
1186from anywhere from 0 to 100. Setting it to 0 would make the workload fully
1187sequential. Any setting in between will result in a random mix of sequential
1188and random I/O, at the given percentages. Comma\-separated values may be
1189specified for reads, writes, and trims as described in \fBblocksize\fR.
1190.TP
1191.BI norandommap
1192Normally fio will cover every block of the file when doing random I/O. If
1193this option is given, fio will just get a new random offset without looking
1194at past I/O history. This means that some blocks may not be read or written,
1195and that some blocks may be read/written more than once. If this option is
1196used with \fBverify\fR and multiple blocksizes (via \fBbsrange\fR),
1197only intact blocks are verified, i.e., partially\-overwritten blocks are
47e6a6e5
SW
1198ignored. With an async I/O engine and an I/O depth > 1, it is possible for
1199the same block to be overwritten, which can cause verification errors. Either
1200do not use norandommap in this case, or also use the lfsr random generator.
523bad63
TK
1201.TP
1202.BI softrandommap \fR=\fPbool
1203See \fBnorandommap\fR. If fio runs with the random block map enabled and
1204it fails to allocate the map, if this option is set it will continue without
1205a random block map. As coverage will not be as complete as with random maps,
1206this option is disabled by default.
1207.TP
1208.BI random_generator \fR=\fPstr
1209Fio supports the following engines for generating I/O offsets for random I/O:
1210.RS
1211.RS
1212.TP
1213.B tausworthe
1214Strong 2^88 cycle random number generator.
1215.TP
1216.B lfsr
1217Linear feedback shift register generator.
1218.TP
1219.B tausworthe64
1220Strong 64\-bit 2^258 cycle random number generator.
1221.RE
1222.P
1223\fBtausworthe\fR is a strong random number generator, but it requires tracking
1224on the side if we want to ensure that blocks are only read or written
1225once. \fBlfsr\fR guarantees that we never generate the same offset twice, and
1226it's also less computationally expensive. It's not a true random generator,
1227however, though for I/O purposes it's typically good enough. \fBlfsr\fR only
1228works with single block sizes, not with workloads that use multiple block
1229sizes. If used with such a workload, fio may read or write some blocks
1230multiple times. The default value is \fBtausworthe\fR, unless the required
1231space exceeds 2^32 blocks. If it does, then \fBtausworthe64\fR is
1232selected automatically.
1233.RE
1234.SS "Block size"
1235.TP
1236.BI blocksize \fR=\fPint[,int][,int] "\fR,\fB bs" \fR=\fPint[,int][,int]
1237The block size in bytes used for I/O units. Default: 4096. A single value
1238applies to reads, writes, and trims. Comma\-separated values may be
1239specified for reads, writes, and trims. A value not terminated in a comma
1240applies to subsequent types. Examples:
1241.RS
1242.RS
1243.P
1244.PD 0
1245bs=256k means 256k for reads, writes and trims.
1246.P
1247bs=8k,32k means 8k for reads, 32k for writes and trims.
1248.P
1249bs=8k,32k, means 8k for reads, 32k for writes, and default for trims.
1250.P
1251bs=,8k means default for reads, 8k for writes and trims.
1252.P
1253bs=,8k, means default for reads, 8k for writes, and default for trims.
1254.PD
1255.RE
1256.RE
1257.TP
1258.BI blocksize_range \fR=\fPirange[,irange][,irange] "\fR,\fB bsrange" \fR=\fPirange[,irange][,irange]
1259A range of block sizes in bytes for I/O units. The issued I/O unit will
1260always be a multiple of the minimum size, unless
1261\fBblocksize_unaligned\fR is set.
1262Comma\-separated ranges may be specified for reads, writes, and trims as
1263described in \fBblocksize\fR. Example:
1264.RS
1265.RS
1266.P
1267bsrange=1k\-4k,2k\-8k
1268.RE
1269.RE
1270.TP
1271.BI bssplit \fR=\fPstr[,str][,str]
1272Sometimes you want even finer grained control of the block sizes issued, not
1273just an even split between them. This option allows you to weight various
1274block sizes, so that you are able to define a specific amount of block sizes
1275issued. The format for this option is:
1276.RS
1277.RS
1278.P
1279bssplit=blocksize/percentage:blocksize/percentage
1280.RE
1281.P
1282for as many block sizes as needed. So if you want to define a workload that
1283has 50% 64k blocks, 10% 4k blocks, and 40% 32k blocks, you would write:
1284.RS
1285.P
1286bssplit=4k/10:64k/50:32k/40
1287.RE
1288.P
1289Ordering does not matter. If the percentage is left blank, fio will fill in
1290the remaining values evenly. So a bssplit option like this one:
1291.RS
1292.P
1293bssplit=4k/50:1k/:32k/
1294.RE
1295.P
1296would have 50% 4k ios, and 25% 1k and 32k ios. The percentages always add up
1297to 100, if bssplit is given a range that adds up to more, it will error out.
1298.P
1299Comma\-separated values may be specified for reads, writes, and trims as
1300described in \fBblocksize\fR.
1301.P
1302If you want a workload that has 50% 2k reads and 50% 4k reads, while having
130390% 4k writes and 10% 8k writes, you would specify:
1304.RS
1305.P
cf04b906 1306bssplit=2k/50:4k/50,4k/90:8k/10
523bad63 1307.RE
6a16ece8
JA
1308.P
1309Fio supports defining up to 64 different weights for each data direction.
523bad63
TK
1310.RE
1311.TP
1312.BI blocksize_unaligned "\fR,\fB bs_unaligned"
1313If set, fio will issue I/O units with any size within
1314\fBblocksize_range\fR, not just multiples of the minimum size. This
1315typically won't work with direct I/O, as that normally requires sector
1316alignment.
1317.TP
1318.BI bs_is_seq_rand \fR=\fPbool
1319If this option is set, fio will use the normal read,write blocksize settings
1320as sequential,random blocksize settings instead. Any random read or write
1321will use the WRITE blocksize settings, and any sequential read or write will
1322use the READ blocksize settings.
1323.TP
1324.BI blockalign \fR=\fPint[,int][,int] "\fR,\fB ba" \fR=\fPint[,int][,int]
1325Boundary to which fio will align random I/O units. Default:
1326\fBblocksize\fR. Minimum alignment is typically 512b for using direct
1327I/O, though it usually depends on the hardware block size. This option is
1328mutually exclusive with using a random map for files, so it will turn off
1329that option. Comma\-separated values may be specified for reads, writes, and
1330trims as described in \fBblocksize\fR.
1331.SS "Buffers and memory"
1332.TP
1333.BI zero_buffers
1334Initialize buffers with all zeros. Default: fill buffers with random data.
1335.TP
1336.BI refill_buffers
1337If this option is given, fio will refill the I/O buffers on every
1338submit. The default is to only fill it at init time and reuse that
1339data. Only makes sense if zero_buffers isn't specified, naturally. If data
1340verification is enabled, \fBrefill_buffers\fR is also automatically enabled.
1341.TP
1342.BI scramble_buffers \fR=\fPbool
1343If \fBrefill_buffers\fR is too costly and the target is using data
1344deduplication, then setting this option will slightly modify the I/O buffer
1345contents to defeat normal de\-dupe attempts. This is not enough to defeat
1346more clever block compression attempts, but it will stop naive dedupe of
1347blocks. Default: true.
1348.TP
1349.BI buffer_compress_percentage \fR=\fPint
72592780
SW
1350If this is set, then fio will attempt to provide I/O buffer content
1351(on WRITEs) that compresses to the specified level. Fio does this by
1352providing a mix of random data followed by fixed pattern data. The
1353fixed pattern is either zeros, or the pattern specified by
1354\fBbuffer_pattern\fR. If the \fBbuffer_pattern\fR option is used, it
1355might skew the compression ratio slightly. Setting
1356\fBbuffer_compress_percentage\fR to a value other than 100 will also
1357enable \fBrefill_buffers\fR in order to reduce the likelihood that
1358adjacent blocks are so similar that they over compress when seen
1359together. See \fBbuffer_compress_chunk\fR for how to set a finer or
1360coarser granularity of the random/fixed data regions. Defaults to unset
1361i.e., buffer data will not adhere to any compression level.
523bad63
TK
1362.TP
1363.BI buffer_compress_chunk \fR=\fPint
72592780
SW
1364This setting allows fio to manage how big the random/fixed data region
1365is when using \fBbuffer_compress_percentage\fR. When
1366\fBbuffer_compress_chunk\fR is set to some non-zero value smaller than the
1367block size, fio can repeat the random/fixed region throughout the I/O
1368buffer at the specified interval (which particularly useful when
1369bigger block sizes are used for a job). When set to 0, fio will use a
1370chunk size that matches the block size resulting in a single
1371random/fixed region within the I/O buffer. Defaults to 512. When the
1372unit is omitted, the value is interpreted in bytes.
523bad63
TK
1373.TP
1374.BI buffer_pattern \fR=\fPstr
1375If set, fio will fill the I/O buffers with this pattern or with the contents
1376of a file. If not set, the contents of I/O buffers are defined by the other
1377options related to buffer contents. The setting can be any pattern of bytes,
1378and can be prefixed with 0x for hex values. It may also be a string, where
1379the string must then be wrapped with "". Or it may also be a filename,
1380where the filename must be wrapped with '' in which case the file is
1381opened and read. Note that not all the file contents will be read if that
1382would cause the buffers to overflow. So, for example:
1383.RS
1384.RS
1385.P
1386.PD 0
1387buffer_pattern='filename'
1388.P
1389or:
1390.P
1391buffer_pattern="abcd"
1392.P
1393or:
1394.P
1395buffer_pattern=\-12
1396.P
1397or:
1398.P
1399buffer_pattern=0xdeadface
1400.PD
1401.RE
1402.P
1403Also you can combine everything together in any order:
1404.RS
1405.P
1406buffer_pattern=0xdeadface"abcd"\-12'filename'
1407.RE
1408.RE
1409.TP
1410.BI dedupe_percentage \fR=\fPint
1411If set, fio will generate this percentage of identical buffers when
1412writing. These buffers will be naturally dedupable. The contents of the
1413buffers depend on what other buffer compression settings have been set. It's
1414possible to have the individual buffers either fully compressible, or not at
72592780
SW
1415all \-\- this option only controls the distribution of unique buffers. Setting
1416this option will also enable \fBrefill_buffers\fR to prevent every buffer
1417being identical.
523bad63
TK
1418.TP
1419.BI invalidate \fR=\fPbool
1420Invalidate the buffer/page cache parts of the files to be used prior to
1421starting I/O if the platform and file type support it. Defaults to true.
1422This will be ignored if \fBpre_read\fR is also specified for the
1423same job.
1424.TP
1425.BI sync \fR=\fPbool
1426Use synchronous I/O for buffered writes. For the majority of I/O engines,
1427this means using O_SYNC. Default: false.
1428.TP
1429.BI iomem \fR=\fPstr "\fR,\fP mem" \fR=\fPstr
1430Fio can use various types of memory as the I/O unit buffer. The allowed
1431values are:
1432.RS
1433.RS
1434.TP
1435.B malloc
1436Use memory from \fBmalloc\fR\|(3) as the buffers. Default memory type.
1437.TP
1438.B shm
1439Use shared memory as the buffers. Allocated through \fBshmget\fR\|(2).
1440.TP
1441.B shmhuge
1442Same as \fBshm\fR, but use huge pages as backing.
1443.TP
1444.B mmap
1445Use \fBmmap\fR\|(2) to allocate buffers. May either be anonymous memory, or can
1446be file backed if a filename is given after the option. The format
1447is `mem=mmap:/path/to/file'.
1448.TP
1449.B mmaphuge
1450Use a memory mapped huge file as the buffer backing. Append filename
1451after mmaphuge, ala `mem=mmaphuge:/hugetlbfs/file'.
1452.TP
1453.B mmapshared
1454Same as \fBmmap\fR, but use a MMAP_SHARED mapping.
1455.TP
1456.B cudamalloc
1457Use GPU memory as the buffers for GPUDirect RDMA benchmark.
1458The \fBioengine\fR must be \fBrdma\fR.
1459.RE
1460.P
1461The area allocated is a function of the maximum allowed bs size for the job,
1462multiplied by the I/O depth given. Note that for \fBshmhuge\fR and
1463\fBmmaphuge\fR to work, the system must have free huge pages allocated. This
1464can normally be checked and set by reading/writing
1465`/proc/sys/vm/nr_hugepages' on a Linux system. Fio assumes a huge page
1466is 4MiB in size. So to calculate the number of huge pages you need for a
1467given job file, add up the I/O depth of all jobs (normally one unless
1468\fBiodepth\fR is used) and multiply by the maximum bs set. Then divide
1469that number by the huge page size. You can see the size of the huge pages in
1470`/proc/meminfo'. If no huge pages are allocated by having a non\-zero
1471number in `nr_hugepages', using \fBmmaphuge\fR or \fBshmhuge\fR will fail. Also
1472see \fBhugepage\-size\fR.
1473.P
1474\fBmmaphuge\fR also needs to have hugetlbfs mounted and the file location
1475should point there. So if it's mounted in `/huge', you would use
1476`mem=mmaphuge:/huge/somefile'.
1477.RE
1478.TP
1479.BI iomem_align \fR=\fPint "\fR,\fP mem_align" \fR=\fPint
1480This indicates the memory alignment of the I/O memory buffers. Note that
1481the given alignment is applied to the first I/O unit buffer, if using
1482\fBiodepth\fR the alignment of the following buffers are given by the
1483\fBbs\fR used. In other words, if using a \fBbs\fR that is a
1484multiple of the page sized in the system, all buffers will be aligned to
1485this value. If using a \fBbs\fR that is not page aligned, the alignment
1486of subsequent I/O memory buffers is the sum of the \fBiomem_align\fR and
1487\fBbs\fR used.
1488.TP
1489.BI hugepage\-size \fR=\fPint
1490Defines the size of a huge page. Must at least be equal to the system
1491setting, see `/proc/meminfo'. Defaults to 4MiB. Should probably
1492always be a multiple of megabytes, so using `hugepage\-size=Xm' is the
1493preferred way to set this to avoid setting a non\-pow\-2 bad value.
1494.TP
1495.BI lockmem \fR=\fPint
1496Pin the specified amount of memory with \fBmlock\fR\|(2). Can be used to
1497simulate a smaller amount of memory. The amount specified is per worker.
1498.SS "I/O size"
1499.TP
1500.BI size \fR=\fPint
1501The total size of file I/O for each thread of this job. Fio will run until
1502this many bytes has been transferred, unless runtime is limited by other options
1503(such as \fBruntime\fR, for instance, or increased/decreased by \fBio_size\fR).
1504Fio will divide this size between the available files determined by options
1505such as \fBnrfiles\fR, \fBfilename\fR, unless \fBfilesize\fR is
1506specified by the job. If the result of division happens to be 0, the size is
1507set to the physical size of the given files or devices if they exist.
1508If this option is not specified, fio will use the full size of the given
1509files or devices. If the files do not exist, size must be given. It is also
1510possible to give size as a percentage between 1 and 100. If `size=20%' is
1511given, fio will use 20% of the full size of the given files or devices.
1512Can be combined with \fBoffset\fR to constrain the start and end range
1513that I/O will be done within.
1514.TP
1515.BI io_size \fR=\fPint "\fR,\fB io_limit" \fR=\fPint
1516Normally fio operates within the region set by \fBsize\fR, which means
1517that the \fBsize\fR option sets both the region and size of I/O to be
1518performed. Sometimes that is not what you want. With this option, it is
1519possible to define just the amount of I/O that fio should do. For instance,
1520if \fBsize\fR is set to 20GiB and \fBio_size\fR is set to 5GiB, fio
1521will perform I/O within the first 20GiB but exit when 5GiB have been
1522done. The opposite is also possible \-\- if \fBsize\fR is set to 20GiB,
1523and \fBio_size\fR is set to 40GiB, then fio will do 40GiB of I/O within
1524the 0..20GiB region.
1525.TP
1526.BI filesize \fR=\fPirange(int)
1527Individual file sizes. May be a range, in which case fio will select sizes
1528for files at random within the given range and limited to \fBsize\fR in
1529total (if that is given). If not given, each created file is the same size.
1530This option overrides \fBsize\fR in terms of file size, which means
1531this value is used as a fixed size or possible range of each file.
1532.TP
1533.BI file_append \fR=\fPbool
1534Perform I/O after the end of the file. Normally fio will operate within the
1535size of a file. If this option is set, then fio will append to the file
1536instead. This has identical behavior to setting \fBoffset\fR to the size
1537of a file. This option is ignored on non\-regular files.
1538.TP
1539.BI fill_device \fR=\fPbool "\fR,\fB fill_fs" \fR=\fPbool
1540Sets size to something really large and waits for ENOSPC (no space left on
1541device) as the terminating condition. Only makes sense with sequential
1542write. For a read workload, the mount point will be filled first then I/O
1543started on the result. This option doesn't make sense if operating on a raw
1544device node, since the size of that is already known by the file system.
1545Additionally, writing beyond end\-of\-device will not return ENOSPC there.
1546.SS "I/O engine"
1547.TP
1548.BI ioengine \fR=\fPstr
1549Defines how the job issues I/O to the file. The following types are defined:
1550.RS
1551.RS
1552.TP
1553.B sync
1554Basic \fBread\fR\|(2) or \fBwrite\fR\|(2)
1555I/O. \fBlseek\fR\|(2) is used to position the I/O location.
1556See \fBfsync\fR and \fBfdatasync\fR for syncing write I/Os.
1557.TP
1558.B psync
1559Basic \fBpread\fR\|(2) or \fBpwrite\fR\|(2) I/O. Default on
1560all supported operating systems except for Windows.
1561.TP
1562.B vsync
1563Basic \fBreadv\fR\|(2) or \fBwritev\fR\|(2) I/O. Will emulate
1564queuing by coalescing adjacent I/Os into a single submission.
1565.TP
1566.B pvsync
1567Basic \fBpreadv\fR\|(2) or \fBpwritev\fR\|(2) I/O.
a46c5e01 1568.TP
2cafffbe
JA
1569.B pvsync2
1570Basic \fBpreadv2\fR\|(2) or \fBpwritev2\fR\|(2) I/O.
1571.TP
d60e92d1 1572.B libaio
523bad63
TK
1573Linux native asynchronous I/O. Note that Linux may only support
1574queued behavior with non\-buffered I/O (set `direct=1' or
1575`buffered=0').
1576This engine defines engine specific options.
d60e92d1
AC
1577.TP
1578.B posixaio
523bad63
TK
1579POSIX asynchronous I/O using \fBaio_read\fR\|(3) and
1580\fBaio_write\fR\|(3).
03e20d68
BC
1581.TP
1582.B solarisaio
1583Solaris native asynchronous I/O.
1584.TP
1585.B windowsaio
38f8c318 1586Windows native asynchronous I/O. Default on Windows.
d60e92d1
AC
1587.TP
1588.B mmap
523bad63
TK
1589File is memory mapped with \fBmmap\fR\|(2) and data copied
1590to/from using \fBmemcpy\fR\|(3).
d60e92d1
AC
1591.TP
1592.B splice
523bad63
TK
1593\fBsplice\fR\|(2) is used to transfer the data and
1594\fBvmsplice\fR\|(2) to transfer data from user space to the
1595kernel.
d60e92d1 1596.TP
d60e92d1 1597.B sg
523bad63
TK
1598SCSI generic sg v3 I/O. May either be synchronous using the SG_IO
1599ioctl, or if the target is an sg character device we use
1600\fBread\fR\|(2) and \fBwrite\fR\|(2) for asynchronous
1601I/O. Requires \fBfilename\fR option to specify either block or
3740cfc8
VF
1602character devices. This engine supports trim operations. The
1603sg engine includes engine specific options.
d60e92d1
AC
1604.TP
1605.B null
523bad63
TK
1606Doesn't transfer any data, just pretends to. This is mainly used to
1607exercise fio itself and for debugging/testing purposes.
d60e92d1
AC
1608.TP
1609.B net
523bad63
TK
1610Transfer over the network to given `host:port'. Depending on the
1611\fBprotocol\fR used, the \fBhostname\fR, \fBport\fR,
1612\fBlisten\fR and \fBfilename\fR options are used to specify
1613what sort of connection to make, while the \fBprotocol\fR option
1614determines which protocol will be used. This engine defines engine
1615specific options.
d60e92d1
AC
1616.TP
1617.B netsplice
523bad63
TK
1618Like \fBnet\fR, but uses \fBsplice\fR\|(2) and
1619\fBvmsplice\fR\|(2) to map data and send/receive.
1620This engine defines engine specific options.
d60e92d1 1621.TP
53aec0a4 1622.B cpuio
523bad63
TK
1623Doesn't transfer any data, but burns CPU cycles according to the
1624\fBcpuload\fR and \fBcpuchunks\fR options. Setting
1625\fBcpuload\fR\=85 will cause that job to do nothing but burn 85%
1626of the CPU. In case of SMP machines, use `numjobs=<nr_of_cpu>'
1627to get desired CPU usage, as the cpuload only loads a
1628single CPU at the desired rate. A job never finishes unless there is
1629at least one non\-cpuio job.
d60e92d1
AC
1630.TP
1631.B guasi
f1dd3fb1 1632The GUASI I/O engine is the Generic Userspace Asynchronous Syscall
523bad63
TK
1633Interface approach to async I/O. See \fIhttp://www.xmailserver.org/guasi\-lib.html\fR
1634for more info on GUASI.
d60e92d1 1635.TP
21b8aee8 1636.B rdma
523bad63
TK
1637The RDMA I/O engine supports both RDMA memory semantics
1638(RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
609ac152
SB
1639InfiniBand, RoCE and iWARP protocols. This engine defines engine
1640specific options.
d54fce84
DM
1641.TP
1642.B falloc
523bad63
TK
1643I/O engine that does regular fallocate to simulate data transfer as
1644fio ioengine.
1645.RS
1646.P
1647.PD 0
1648DDIR_READ does fallocate(,mode = FALLOC_FL_KEEP_SIZE,).
1649.P
1650DIR_WRITE does fallocate(,mode = 0).
1651.P
1652DDIR_TRIM does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE).
1653.PD
1654.RE
1655.TP
1656.B ftruncate
1657I/O engine that sends \fBftruncate\fR\|(2) operations in response
1658to write (DDIR_WRITE) events. Each ftruncate issued sets the file's
1659size to the current block offset. \fBblocksize\fR is ignored.
d54fce84
DM
1660.TP
1661.B e4defrag
523bad63
TK
1662I/O engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate
1663defragment activity in request to DDIR_WRITE event.
0d978694 1664.TP
d5f9b0ea
IF
1665.B rados
1666I/O engine supporting direct access to Ceph Reliable Autonomic Distributed
1667Object Store (RADOS) via librados. This ioengine defines engine specific
1668options.
1669.TP
0d978694 1670.B rbd
523bad63
TK
1671I/O engine supporting direct access to Ceph Rados Block Devices
1672(RBD) via librbd without the need to use the kernel rbd driver. This
1673ioengine defines engine specific options.
a7c386f4 1674.TP
c2f6a13d
LMB
1675.B http
1676I/O engine supporting GET/PUT requests over HTTP(S) with libcurl to
1677a WebDAV or S3 endpoint. This ioengine defines engine specific options.
1678
1679This engine only supports direct IO of iodepth=1; you need to scale this
1680via numjobs. blocksize defines the size of the objects to be created.
1681
1682TRIM is translated to object deletion.
1683.TP
a7c386f4 1684.B gfapi
523bad63
TK
1685Using GlusterFS libgfapi sync interface to direct access to
1686GlusterFS volumes without having to go through FUSE. This ioengine
1687defines engine specific options.
cc47f094 1688.TP
1689.B gfapi_async
523bad63
TK
1690Using GlusterFS libgfapi async interface to direct access to
1691GlusterFS volumes without having to go through FUSE. This ioengine
1692defines engine specific options.
1b10477b 1693.TP
b74e419e 1694.B libhdfs
523bad63
TK
1695Read and write through Hadoop (HDFS). The \fBfilename\fR option
1696is used to specify host,port of the hdfs name\-node to connect. This
1697engine interprets offsets a little differently. In HDFS, files once
1698created cannot be modified so random writes are not possible. To
1699imitate this the libhdfs engine expects a bunch of small files to be
1700created over HDFS and will randomly pick a file from them
1701based on the offset generated by fio backend (see the example
1702job file to create such files, use `rw=write' option). Please
1703note, it may be necessary to set environment variables to work
1704with HDFS/libhdfs properly. Each job uses its own connection to
1705HDFS.
65fa28ca
DE
1706.TP
1707.B mtd
523bad63
TK
1708Read, write and erase an MTD character device (e.g.,
1709`/dev/mtd0'). Discards are treated as erases. Depending on the
1710underlying device type, the I/O may have to go in a certain pattern,
1711e.g., on NAND, writing sequentially to erase blocks and discarding
1712before overwriting. The \fBtrimwrite\fR mode works well for this
65fa28ca 1713constraint.
5c4ef02e
JA
1714.TP
1715.B pmemblk
523bad63 1716Read and write using filesystem DAX to a file on a filesystem
363a5f65 1717mounted with DAX on a persistent memory device through the PMDK
523bad63 1718libpmemblk library.
104ee4de 1719.TP
523bad63
TK
1720.B dev\-dax
1721Read and write using device DAX to a persistent memory device (e.g.,
363a5f65 1722/dev/dax0.0) through the PMDK libpmem library.
d60e92d1 1723.TP
523bad63
TK
1724.B external
1725Prefix to specify loading an external I/O engine object file. Append
1726the engine filename, e.g. `ioengine=external:/tmp/foo.o' to load
d243fd6d
TK
1727ioengine `foo.o' in `/tmp'. The path can be either
1728absolute or relative. See `engines/skeleton_external.c' in the fio source for
1729details of writing an external I/O engine.
1216cc5a
JB
1730.TP
1731.B filecreate
b71968b1
SW
1732Simply create the files and do no I/O to them. You still need to set
1733\fBfilesize\fR so that all the accounting still occurs, but no actual I/O will be
1734done other than creating the file.
ae0db592
TI
1735.TP
1736.B libpmem
1737Read and write using mmap I/O to a file on a filesystem
363a5f65 1738mounted with DAX on a persistent memory device through the PMDK
ae0db592 1739libpmem library.
07751e10
JA
1740.TP
1741.B ime_psync
1742Synchronous read and write using DDN's Infinite Memory Engine (IME). This
1743engine is very basic and issues calls to IME whenever an IO is queued.
1744.TP
1745.B ime_psyncv
1746Synchronous read and write using DDN's Infinite Memory Engine (IME). This
1747engine uses iovecs and will try to stack as much IOs as possible (if the IOs
1748are "contiguous" and the IO depth is not exceeded) before issuing a call to IME.
1749.TP
1750.B ime_aio
1751Asynchronous read and write using DDN's Infinite Memory Engine (IME). This
1752engine will try to stack as much IOs as possible by creating requests for IME.
1753FIO will then decide when to commit these requests.
247ef2aa
KZ
1754.TP
1755.B libiscsi
1756Read and write iscsi lun with libiscsi.
523bad63
TK
1757.SS "I/O engine specific parameters"
1758In addition, there are some parameters which are only valid when a specific
1759\fBioengine\fR is in use. These are used identically to normal parameters,
1760with the caveat that when used on the command line, they must come after the
1761\fBioengine\fR that defines them is selected.
d60e92d1 1762.TP
523bad63
TK
1763.BI (libaio)userspace_reap
1764Normally, with the libaio engine in use, fio will use the
1765\fBio_getevents\fR\|(3) system call to reap newly returned events. With
1766this flag turned on, the AIO ring will be read directly from user\-space to
1767reap events. The reaping mode is only enabled when polling for a minimum of
17680 events (e.g. when `iodepth_batch_complete=0').
3ce9dcaf 1769.TP
523bad63
TK
1770.BI (pvsync2)hipri
1771Set RWF_HIPRI on I/O, indicating to the kernel that it's of higher priority
1772than normal.
82407585 1773.TP
523bad63
TK
1774.BI (pvsync2)hipri_percentage
1775When hipri is set this determines the probability of a pvsync2 I/O being high
1776priority. The default is 100%.
d60e92d1 1777.TP
523bad63
TK
1778.BI (cpuio)cpuload \fR=\fPint
1779Attempt to use the specified percentage of CPU cycles. This is a mandatory
1780option when using cpuio I/O engine.
997b5680 1781.TP
523bad63
TK
1782.BI (cpuio)cpuchunks \fR=\fPint
1783Split the load into cycles of the given time. In microseconds.
1ad01bd1 1784.TP
523bad63
TK
1785.BI (cpuio)exit_on_io_done \fR=\fPbool
1786Detect when I/O threads are done, then exit.
d60e92d1 1787.TP
523bad63
TK
1788.BI (libhdfs)namenode \fR=\fPstr
1789The hostname or IP address of a HDFS cluster namenode to contact.
d01612f3 1790.TP
523bad63
TK
1791.BI (libhdfs)port
1792The listening port of the HFDS cluster namenode.
d60e92d1 1793.TP
523bad63
TK
1794.BI (netsplice,net)port
1795The TCP or UDP port to bind to or connect to. If this is used with
1796\fBnumjobs\fR to spawn multiple instances of the same job type, then
1797this will be the starting port number since fio will use a range of
1798ports.
d60e92d1 1799.TP
609ac152
SB
1800.BI (rdma)port
1801The port to use for RDMA-CM communication. This should be the same
1802value on the client and the server side.
1803.TP
1804.BI (netsplice,net, rdma)hostname \fR=\fPstr
1805The hostname or IP address to use for TCP, UDP or RDMA-CM based I/O.
1806If the job is a TCP listener or UDP reader, the hostname is not used
1807and must be omitted unless it is a valid UDP multicast address.
591e9e06 1808.TP
523bad63
TK
1809.BI (netsplice,net)interface \fR=\fPstr
1810The IP address of the network interface used to send or receive UDP
1811multicast.
ddf24e42 1812.TP
523bad63
TK
1813.BI (netsplice,net)ttl \fR=\fPint
1814Time\-to\-live value for outgoing UDP multicast packets. Default: 1.
d60e92d1 1815.TP
523bad63
TK
1816.BI (netsplice,net)nodelay \fR=\fPbool
1817Set TCP_NODELAY on TCP connections.
fa769d44 1818.TP
523bad63
TK
1819.BI (netsplice,net)protocol \fR=\fPstr "\fR,\fP proto" \fR=\fPstr
1820The network protocol to use. Accepted values are:
1821.RS
e76b1da4
JA
1822.RS
1823.TP
523bad63
TK
1824.B tcp
1825Transmission control protocol.
e76b1da4 1826.TP
523bad63
TK
1827.B tcpv6
1828Transmission control protocol V6.
e76b1da4 1829.TP
523bad63
TK
1830.B udp
1831User datagram protocol.
1832.TP
1833.B udpv6
1834User datagram protocol V6.
e76b1da4 1835.TP
523bad63
TK
1836.B unix
1837UNIX domain socket.
e76b1da4
JA
1838.RE
1839.P
523bad63
TK
1840When the protocol is TCP or UDP, the port must also be given, as well as the
1841hostname if the job is a TCP listener or UDP reader. For unix sockets, the
1842normal \fBfilename\fR option should be used and the port is invalid.
1843.RE
1844.TP
1845.BI (netsplice,net)listen
1846For TCP network connections, tell fio to listen for incoming connections
1847rather than initiating an outgoing connection. The \fBhostname\fR must
1848be omitted if this option is used.
1849.TP
1850.BI (netsplice,net)pingpong
1851Normally a network writer will just continue writing data, and a network
1852reader will just consume packages. If `pingpong=1' is set, a writer will
1853send its normal payload to the reader, then wait for the reader to send the
1854same payload back. This allows fio to measure network latencies. The
1855submission and completion latencies then measure local time spent sending or
1856receiving, and the completion latency measures how long it took for the
1857other end to receive and send back. For UDP multicast traffic
1858`pingpong=1' should only be set for a single reader when multiple readers
1859are listening to the same address.
1860.TP
1861.BI (netsplice,net)window_size \fR=\fPint
1862Set the desired socket buffer size for the connection.
e76b1da4 1863.TP
523bad63
TK
1864.BI (netsplice,net)mss \fR=\fPint
1865Set the TCP maximum segment size (TCP_MAXSEG).
d60e92d1 1866.TP
523bad63
TK
1867.BI (e4defrag)donorname \fR=\fPstr
1868File will be used as a block donor (swap extents between files).
d60e92d1 1869.TP
523bad63
TK
1870.BI (e4defrag)inplace \fR=\fPint
1871Configure donor file blocks allocation strategy:
1872.RS
1873.RS
d60e92d1 1874.TP
523bad63
TK
1875.B 0
1876Default. Preallocate donor's file on init.
d60e92d1 1877.TP
523bad63
TK
1878.B 1
1879Allocate space immediately inside defragment event, and free right
1880after event.
1881.RE
1882.RE
d60e92d1 1883.TP
d5f9b0ea 1884.BI (rbd,rados)clustername \fR=\fPstr
523bad63 1885Specifies the name of the Ceph cluster.
92d42d69 1886.TP
523bad63
TK
1887.BI (rbd)rbdname \fR=\fPstr
1888Specifies the name of the RBD.
92d42d69 1889.TP
d5f9b0ea
IF
1890.BI (rbd,rados)pool \fR=\fPstr
1891Specifies the name of the Ceph pool containing RBD or RADOS data.
92d42d69 1892.TP
d5f9b0ea 1893.BI (rbd,rados)clientname \fR=\fPstr
523bad63
TK
1894Specifies the username (without the 'client.' prefix) used to access the
1895Ceph cluster. If the \fBclustername\fR is specified, the \fBclientname\fR shall be
1896the full *type.id* string. If no type. prefix is given, fio will add 'client.'
1897by default.
92d42d69 1898.TP
d5f9b0ea
IF
1899.BI (rbd,rados)busy_poll \fR=\fPbool
1900Poll store instead of waiting for completion. Usually this provides better
1901throughput at cost of higher(up to 100%) CPU utilization.
1902.TP
c2f6a13d
LMB
1903.BI (http)http_host \fR=\fPstr
1904Hostname to connect to. For S3, this could be the bucket name. Default
1905is \fBlocalhost\fR
1906.TP
1907.BI (http)http_user \fR=\fPstr
1908Username for HTTP authentication.
1909.TP
1910.BI (http)http_pass \fR=\fPstr
1911Password for HTTP authentication.
1912.TP
09fd2966
LMB
1913.BI (http)https \fR=\fPstr
1914Whether to use HTTPS instead of plain HTTP. \fRon\fP enables HTTPS;
1915\fRinsecure\fP will enable HTTPS, but disable SSL peer verification (use
1916with caution!). Default is \fBoff\fR.
c2f6a13d 1917.TP
09fd2966
LMB
1918.BI (http)http_mode \fR=\fPstr
1919Which HTTP access mode to use: webdav, swift, or s3. Default is
1920\fBwebdav\fR.
c2f6a13d
LMB
1921.TP
1922.BI (http)http_s3_region \fR=\fPstr
1923The S3 region/zone to include in the request. Default is \fBus-east-1\fR.
1924.TP
1925.BI (http)http_s3_key \fR=\fPstr
1926The S3 secret key.
1927.TP
1928.BI (http)http_s3_keyid \fR=\fPstr
1929The S3 key/access id.
1930.TP
09fd2966
LMB
1931.BI (http)http_swift_auth_token \fR=\fPstr
1932The Swift auth token. See the example configuration file on how to
1933retrieve this.
1934.TP
c2f6a13d
LMB
1935.BI (http)http_verbose \fR=\fPint
1936Enable verbose requests from libcurl. Useful for debugging. 1 turns on
1937verbose logging from libcurl, 2 additionally enables HTTP IO tracing.
1938Default is \fB0\fR
1939.TP
523bad63
TK
1940.BI (mtd)skip_bad \fR=\fPbool
1941Skip operations against known bad blocks.
8116fd24 1942.TP
523bad63
TK
1943.BI (libhdfs)hdfsdirectory
1944libhdfs will create chunk in this HDFS directory.
e0a04ac1 1945.TP
523bad63
TK
1946.BI (libhdfs)chunk_size
1947The size of the chunk to use for each file.
609ac152
SB
1948.TP
1949.BI (rdma)verb \fR=\fPstr
1950The RDMA verb to use on this side of the RDMA ioengine
1951connection. Valid values are write, read, send and recv. These
1952correspond to the equivalent RDMA verbs (e.g. write = rdma_write
1953etc.). Note that this only needs to be specified on the client side of
1954the connection. See the examples folder.
1955.TP
1956.BI (rdma)bindname \fR=\fPstr
1957The name to use to bind the local RDMA-CM connection to a local RDMA
1958device. This could be a hostname or an IPv4 or IPv6 address. On the
1959server side this will be passed into the rdma_bind_addr() function and
1960on the client site it will be used in the rdma_resolve_add()
1961function. This can be useful when multiple paths exist between the
1962client and the server or in certain loopback configurations.
52b81b7c
KD
1963.TP
1964.BI (sg)readfua \fR=\fPbool
1965With readfua option set to 1, read operations include the force
1966unit access (fua) flag. Default: 0.
1967.TP
1968.BI (sg)writefua \fR=\fPbool
1969With writefua option set to 1, write operations include the force
1970unit access (fua) flag. Default: 0.
2c3a9150
VF
1971.TP
1972.BI (sg)sg_write_mode \fR=\fPstr
1973Specify the type of write commands to issue. This option can take three
1974values:
1975.RS
1976.RS
1977.TP
1978.B write (default)
1979Write opcodes are issued as usual
1980.TP
1981.B verify
1982Issue WRITE AND VERIFY commands. The BYTCHK bit is set to 0. This
1983directs the device to carry out a medium verification with no data
1984comparison. The writefua option is ignored with this selection.
1985.TP
1986.B same
1987Issue WRITE SAME commands. This transfers a single block to the device
1988and writes this same block of data to a contiguous sequence of LBAs
1989beginning at the specified offset. fio's block size parameter
1990specifies the amount of data written with each command. However, the
1991amount of data actually transferred to the device is equal to the
1992device's block (sector) size. For a device with 512 byte sectors,
1993blocksize=8k will write 16 sectors with each command. fio will still
1994generate 8k of data for each command butonly the first 512 bytes will
1995be used and transferred to the device. The writefua option is ignored
1996with this selection.
1997
523bad63
TK
1998.SS "I/O depth"
1999.TP
2000.BI iodepth \fR=\fPint
2001Number of I/O units to keep in flight against the file. Note that
2002increasing \fBiodepth\fR beyond 1 will not affect synchronous ioengines (except
2003for small degrees when \fBverify_async\fR is in use). Even async
2004engines may impose OS restrictions causing the desired depth not to be
2005achieved. This may happen on Linux when using libaio and not setting
2006`direct=1', since buffered I/O is not async on that OS. Keep an
2007eye on the I/O depth distribution in the fio output to verify that the
2008achieved depth is as expected. Default: 1.
2009.TP
2010.BI iodepth_batch_submit \fR=\fPint "\fR,\fP iodepth_batch" \fR=\fPint
2011This defines how many pieces of I/O to submit at once. It defaults to 1
2012which means that we submit each I/O as soon as it is available, but can be
2013raised to submit bigger batches of I/O at the time. If it is set to 0 the
2014\fBiodepth\fR value will be used.
2015.TP
2016.BI iodepth_batch_complete_min \fR=\fPint "\fR,\fP iodepth_batch_complete" \fR=\fPint
2017This defines how many pieces of I/O to retrieve at once. It defaults to 1
2018which means that we'll ask for a minimum of 1 I/O in the retrieval process
2019from the kernel. The I/O retrieval will go on until we hit the limit set by
2020\fBiodepth_low\fR. If this variable is set to 0, then fio will always
2021check for completed events before queuing more I/O. This helps reduce I/O
2022latency, at the cost of more retrieval system calls.
2023.TP
2024.BI iodepth_batch_complete_max \fR=\fPint
2025This defines maximum pieces of I/O to retrieve at once. This variable should
2026be used along with \fBiodepth_batch_complete_min\fR=\fIint\fR variable,
2027specifying the range of min and max amount of I/O which should be
2028retrieved. By default it is equal to \fBiodepth_batch_complete_min\fR
2029value. Example #1:
e0a04ac1 2030.RS
e0a04ac1 2031.RS
e0a04ac1 2032.P
523bad63
TK
2033.PD 0
2034iodepth_batch_complete_min=1
e0a04ac1 2035.P
523bad63
TK
2036iodepth_batch_complete_max=<iodepth>
2037.PD
e0a04ac1
JA
2038.RE
2039.P
523bad63
TK
2040which means that we will retrieve at least 1 I/O and up to the whole
2041submitted queue depth. If none of I/O has been completed yet, we will wait.
2042Example #2:
e8b1961d 2043.RS
523bad63
TK
2044.P
2045.PD 0
2046iodepth_batch_complete_min=0
2047.P
2048iodepth_batch_complete_max=<iodepth>
2049.PD
e8b1961d
JA
2050.RE
2051.P
523bad63
TK
2052which means that we can retrieve up to the whole submitted queue depth, but
2053if none of I/O has been completed yet, we will NOT wait and immediately exit
2054the system call. In this example we simply do polling.
2055.RE
e8b1961d 2056.TP
523bad63
TK
2057.BI iodepth_low \fR=\fPint
2058The low water mark indicating when to start filling the queue
2059again. Defaults to the same as \fBiodepth\fR, meaning that fio will
2060attempt to keep the queue full at all times. If \fBiodepth\fR is set to
2061e.g. 16 and \fBiodepth_low\fR is set to 4, then after fio has filled the queue of
206216 requests, it will let the depth drain down to 4 before starting to fill
2063it again.
d60e92d1 2064.TP
523bad63
TK
2065.BI serialize_overlap \fR=\fPbool
2066Serialize in-flight I/Os that might otherwise cause or suffer from data races.
2067When two or more I/Os are submitted simultaneously, there is no guarantee that
2068the I/Os will be processed or completed in the submitted order. Further, if
2069two or more of those I/Os are writes, any overlapping region between them can
2070become indeterminate/undefined on certain storage. These issues can cause
2071verification to fail erratically when at least one of the racing I/Os is
2072changing data and the overlapping region has a non-zero size. Setting
2073\fBserialize_overlap\fR tells fio to avoid provoking this behavior by explicitly
2074serializing in-flight I/Os that have a non-zero overlap. Note that setting
2075this option can reduce both performance and the \fBiodepth\fR achieved.
3d6a6f04
VF
2076.RS
2077.P
2078This option only applies to I/Os issued for a single job except when it is
2079enabled along with \fBio_submit_mode\fR=offload. In offload mode, fio
2080will check for overlap among all I/Os submitted by offload jobs with \fBserialize_overlap\fR
307f2246 2081enabled.
3d6a6f04
VF
2082.P
2083Default: false.
2084.RE
d60e92d1 2085.TP
523bad63
TK
2086.BI io_submit_mode \fR=\fPstr
2087This option controls how fio submits the I/O to the I/O engine. The default
2088is `inline', which means that the fio job threads submit and reap I/O
2089directly. If set to `offload', the job threads will offload I/O submission
2090to a dedicated pool of I/O threads. This requires some coordination and thus
2091has a bit of extra overhead, especially for lower queue depth I/O where it
2092can increase latencies. The benefit is that fio can manage submission rates
2093independently of the device completion rates. This avoids skewed latency
2094reporting if I/O gets backed up on the device side (the coordinated omission
2095problem).
2096.SS "I/O rate"
d60e92d1 2097.TP
523bad63
TK
2098.BI thinktime \fR=\fPtime
2099Stall the job for the specified period of time after an I/O has completed before issuing the
2100next. May be used to simulate processing being done by an application.
2101When the unit is omitted, the value is interpreted in microseconds. See
2102\fBthinktime_blocks\fR and \fBthinktime_spin\fR.
d60e92d1 2103.TP
523bad63
TK
2104.BI thinktime_spin \fR=\fPtime
2105Only valid if \fBthinktime\fR is set \- pretend to spend CPU time doing
2106something with the data received, before falling back to sleeping for the
2107rest of the period specified by \fBthinktime\fR. When the unit is
2108omitted, the value is interpreted in microseconds.
d60e92d1
AC
2109.TP
2110.BI thinktime_blocks \fR=\fPint
523bad63
TK
2111Only valid if \fBthinktime\fR is set \- control how many blocks to issue,
2112before waiting \fBthinktime\fR usecs. If not set, defaults to 1 which will make
2113fio wait \fBthinktime\fR usecs after every block. This effectively makes any
2114queue depth setting redundant, since no more than 1 I/O will be queued
2115before we have to complete it and do our \fBthinktime\fR. In other words, this
2116setting effectively caps the queue depth if the latter is larger.
d60e92d1 2117.TP
6d500c2e 2118.BI rate \fR=\fPint[,int][,int]
523bad63
TK
2119Cap the bandwidth used by this job. The number is in bytes/sec, the normal
2120suffix rules apply. Comma\-separated values may be specified for reads,
2121writes, and trims as described in \fBblocksize\fR.
2122.RS
2123.P
2124For example, using `rate=1m,500k' would limit reads to 1MiB/sec and writes to
2125500KiB/sec. Capping only reads or writes can be done with `rate=,500k' or
2126`rate=500k,' where the former will only limit writes (to 500KiB/sec) and the
2127latter will only limit reads.
2128.RE
d60e92d1 2129.TP
6d500c2e 2130.BI rate_min \fR=\fPint[,int][,int]
523bad63
TK
2131Tell fio to do whatever it can to maintain at least this bandwidth. Failing
2132to meet this requirement will cause the job to exit. Comma\-separated values
2133may be specified for reads, writes, and trims as described in
2134\fBblocksize\fR.
d60e92d1 2135.TP
6d500c2e 2136.BI rate_iops \fR=\fPint[,int][,int]
523bad63
TK
2137Cap the bandwidth to this number of IOPS. Basically the same as
2138\fBrate\fR, just specified independently of bandwidth. If the job is
2139given a block size range instead of a fixed value, the smallest block size
2140is used as the metric. Comma\-separated values may be specified for reads,
2141writes, and trims as described in \fBblocksize\fR.
d60e92d1 2142.TP
6d500c2e 2143.BI rate_iops_min \fR=\fPint[,int][,int]
523bad63
TK
2144If fio doesn't meet this rate of I/O, it will cause the job to exit.
2145Comma\-separated values may be specified for reads, writes, and trims as
2146described in \fBblocksize\fR.
d60e92d1 2147.TP
6de65959 2148.BI rate_process \fR=\fPstr
523bad63
TK
2149This option controls how fio manages rated I/O submissions. The default is
2150`linear', which submits I/O in a linear fashion with fixed delays between
2151I/Os that gets adjusted based on I/O completion rates. If this is set to
2152`poisson', fio will submit I/O based on a more real world random request
6de65959 2153flow, known as the Poisson process
523bad63 2154(\fIhttps://en.wikipedia.org/wiki/Poisson_point_process\fR). The lambda will be
5d02b083 215510^6 / IOPS for the given workload.
1a9bf814
JA
2156.TP
2157.BI rate_ignore_thinktime \fR=\fPbool
2158By default, fio will attempt to catch up to the specified rate setting, if any
2159kind of thinktime setting was used. If this option is set, then fio will
2160ignore the thinktime and continue doing IO at the specified rate, instead of
2161entering a catch-up mode after thinktime is done.
523bad63 2162.SS "I/O latency"
ff6bb260 2163.TP
523bad63 2164.BI latency_target \fR=\fPtime
3e260a46 2165If set, fio will attempt to find the max performance point that the given
523bad63
TK
2166workload will run at while maintaining a latency below this target. When
2167the unit is omitted, the value is interpreted in microseconds. See
2168\fBlatency_window\fR and \fBlatency_percentile\fR.
3e260a46 2169.TP
523bad63 2170.BI latency_window \fR=\fPtime
3e260a46 2171Used with \fBlatency_target\fR to specify the sample window that the job
523bad63
TK
2172is run at varying queue depths to test the performance. When the unit is
2173omitted, the value is interpreted in microseconds.
3e260a46
JA
2174.TP
2175.BI latency_percentile \fR=\fPfloat
523bad63
TK
2176The percentage of I/Os that must fall within the criteria specified by
2177\fBlatency_target\fR and \fBlatency_window\fR. If not set, this
2178defaults to 100.0, meaning that all I/Os must be equal or below to the value
2179set by \fBlatency_target\fR.
2180.TP
2181.BI max_latency \fR=\fPtime
2182If set, fio will exit the job with an ETIMEDOUT error if it exceeds this
2183maximum latency. When the unit is omitted, the value is interpreted in
2184microseconds.
2185.TP
2186.BI rate_cycle \fR=\fPint
2187Average bandwidth for \fBrate\fR and \fBrate_min\fR over this number
2188of milliseconds. Defaults to 1000.
2189.SS "I/O replay"
2190.TP
2191.BI write_iolog \fR=\fPstr
2192Write the issued I/O patterns to the specified file. See
2193\fBread_iolog\fR. Specify a separate file for each job, otherwise the
2194iologs will be interspersed and the file may be corrupt.
2195.TP
2196.BI read_iolog \fR=\fPstr
2197Open an iolog with the specified filename and replay the I/O patterns it
2198contains. This can be used to store a workload and replay it sometime
2199later. The iolog given may also be a blktrace binary file, which allows fio
2200to replay a workload captured by blktrace. See
2201\fBblktrace\fR\|(8) for how to capture such logging data. For blktrace
2202replay, the file needs to be turned into a blkparse binary data file first
2203(`blkparse <device> \-o /dev/null \-d file_for_fio.bin').
c70c7f58
AK
2204You can specify a number of files by separating the names with a ':' character.
2205See the \fBfilename\fR option for information on how to escape ':' and '\'
2206characters within the file names. These files will be sequentially assigned to
2207job clones created by \fBnumjobs\fR.
3e260a46 2208.TP
98e7161c
AK
2209.BI read_iolog_chunked \fR=\fPbool
2210Determines how iolog is read. If false (default) entire \fBread_iolog\fR will
2211be read at once. If selected true, input from iolog will be read gradually.
2212Useful when iolog is very large, or it is generated.
2213.TP
b9921d1a
DZ
2214.BI merge_blktrace_file \fR=\fPstr
2215When specified, rather than replaying the logs passed to \fBread_iolog\fR,
2216the logs go through a merge phase which aggregates them into a single blktrace.
2217The resulting file is then passed on as the \fBread_iolog\fR parameter. The
2218intention here is to make the order of events consistent. This limits the
2219influence of the scheduler compared to replaying multiple blktraces via
2220concurrent jobs.
2221.TP
87a48ada
DZ
2222.BI merge_blktrace_scalars \fR=\fPfloat_list
2223This is a percentage based option that is index paired with the list of files
2224passed to \fBread_iolog\fR. When merging is performed, scale the time of each
2225event by the corresponding amount. For example,
2226`\-\-merge_blktrace_scalars="50:100"' runs the first trace in halftime and the
2227second trace in realtime. This knob is separately tunable from
2228\fBreplay_time_scale\fR which scales the trace during runtime and will not
2229change the output of the merge unlike this option.
2230.TP
55bfd8c8
DZ
2231.BI merge_blktrace_iters \fR=\fPfloat_list
2232This is a whole number option that is index paired with the list of files
2233passed to \fBread_iolog\fR. When merging is performed, run each trace for
2234the specified number of iterations. For example,
2235`\-\-merge_blktrace_iters="2:1"' runs the first trace for two iterations
2236and the second trace for one iteration.
2237.TP
523bad63
TK
2238.BI replay_no_stall \fR=\fPbool
2239When replaying I/O with \fBread_iolog\fR the default behavior is to
2240attempt to respect the timestamps within the log and replay them with the
2241appropriate delay between IOPS. By setting this variable fio will not
2242respect the timestamps and attempt to replay them as fast as possible while
2243still respecting ordering. The result is the same I/O pattern to a given
2244device, but different timings.
2245.TP
6dd7fa77
JA
2246.BI replay_time_scale \fR=\fPint
2247When replaying I/O with \fBread_iolog\fR, fio will honor the original timing
2248in the trace. With this option, it's possible to scale the time. It's a
2249percentage option, if set to 50 it means run at 50% the original IO rate in
2250the trace. If set to 200, run at twice the original IO rate. Defaults to 100.
2251.TP
523bad63
TK
2252.BI replay_redirect \fR=\fPstr
2253While replaying I/O patterns using \fBread_iolog\fR the default behavior
2254is to replay the IOPS onto the major/minor device that each IOP was recorded
2255from. This is sometimes undesirable because on a different machine those
2256major/minor numbers can map to a different device. Changing hardware on the
2257same system can also result in a different major/minor mapping.
2258\fBreplay_redirect\fR causes all I/Os to be replayed onto the single specified
2259device regardless of the device it was recorded
2260from. i.e. `replay_redirect=/dev/sdc' would cause all I/O
2261in the blktrace or iolog to be replayed onto `/dev/sdc'. This means
2262multiple devices will be replayed onto a single device, if the trace
2263contains multiple devices. If you want multiple devices to be replayed
2264concurrently to multiple redirected devices you must blkparse your trace
2265into separate traces and replay them with independent fio invocations.
2266Unfortunately this also breaks the strict time ordering between multiple
2267device accesses.
2268.TP
2269.BI replay_align \fR=\fPint
350a535d
DZ
2270Force alignment of the byte offsets in a trace to this value. The value
2271must be a power of 2.
523bad63
TK
2272.TP
2273.BI replay_scale \fR=\fPint
350a535d
DZ
2274Scale bye offsets down by this factor when replaying traces. Should most
2275likely use \fBreplay_align\fR as well.
523bad63
TK
2276.SS "Threads, processes and job synchronization"
2277.TP
38f68906
JA
2278.BI replay_skip \fR=\fPstr
2279Sometimes it's useful to skip certain IO types in a replay trace. This could
2280be, for instance, eliminating the writes in the trace. Or not replaying the
2281trims/discards, if you are redirecting to a device that doesn't support them.
2282This option takes a comma separated list of read, write, trim, sync.
2283.TP
523bad63
TK
2284.BI thread
2285Fio defaults to creating jobs by using fork, however if this option is
2286given, fio will create jobs by using POSIX Threads' function
2287\fBpthread_create\fR\|(3) to create threads instead.
2288.TP
2289.BI wait_for \fR=\fPstr
2290If set, the current job won't be started until all workers of the specified
2291waitee job are done.
2292.\" ignore blank line here from HOWTO as it looks normal without it
2293\fBwait_for\fR operates on the job name basis, so there are a few
2294limitations. First, the waitee must be defined prior to the waiter job
2295(meaning no forward references). Second, if a job is being referenced as a
2296waitee, it must have a unique name (no duplicate waitees).
2297.TP
2298.BI nice \fR=\fPint
2299Run the job with the given nice value. See man \fBnice\fR\|(2).
2300.\" ignore blank line here from HOWTO as it looks normal without it
2301On Windows, values less than \-15 set the process class to "High"; \-1 through
2302\-15 set "Above Normal"; 1 through 15 "Below Normal"; and above 15 "Idle"
2303priority class.
2304.TP
2305.BI prio \fR=\fPint
2306Set the I/O priority value of this job. Linux limits us to a positive value
2307between 0 and 7, with 0 being the highest. See man
2308\fBionice\fR\|(1). Refer to an appropriate manpage for other operating
2309systems since meaning of priority may differ.
2310.TP
2311.BI prioclass \fR=\fPint
2312Set the I/O priority class. See man \fBionice\fR\|(1).
15501535 2313.TP
d60e92d1 2314.BI cpus_allowed \fR=\fPstr
523bad63 2315Controls the same options as \fBcpumask\fR, but accepts a textual
b570e037
SW
2316specification of the permitted CPUs instead and CPUs are indexed from 0. So
2317to use CPUs 0 and 5 you would specify `cpus_allowed=0,5'. This option also
2318allows a range of CPUs to be specified \-\- say you wanted a binding to CPUs
23190, 5, and 8 to 15, you would set `cpus_allowed=0,5,8\-15'.
2320.RS
2321.P
2322On Windows, when `cpus_allowed' is unset only CPUs from fio's current
2323processor group will be used and affinity settings are inherited from the
2324system. An fio build configured to target Windows 7 makes options that set
2325CPUs processor group aware and values will set both the processor group
2326and a CPU from within that group. For example, on a system where processor
2327group 0 has 40 CPUs and processor group 1 has 32 CPUs, `cpus_allowed'
2328values between 0 and 39 will bind CPUs from processor group 0 and
2329`cpus_allowed' values between 40 and 71 will bind CPUs from processor
2330group 1. When using `cpus_allowed_policy=shared' all CPUs specified by a
2331single `cpus_allowed' option must be from the same processor group. For
2332Windows fio builds not built for Windows 7, CPUs will only be selected from
2333(and be relative to) whatever processor group fio happens to be running in
2334and CPUs from other processor groups cannot be used.
2335.RE
d60e92d1 2336.TP
c2acfbac 2337.BI cpus_allowed_policy \fR=\fPstr
523bad63
TK
2338Set the policy of how fio distributes the CPUs specified by
2339\fBcpus_allowed\fR or \fBcpumask\fR. Two policies are supported:
c2acfbac
JA
2340.RS
2341.RS
2342.TP
2343.B shared
2344All jobs will share the CPU set specified.
2345.TP
2346.B split
2347Each job will get a unique CPU from the CPU set.
2348.RE
2349.P
523bad63
TK
2350\fBshared\fR is the default behavior, if the option isn't specified. If
2351\fBsplit\fR is specified, then fio will will assign one cpu per job. If not
2352enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs
2353in the set.
c2acfbac 2354.RE
c2acfbac 2355.TP
b570e037
SW
2356.BI cpumask \fR=\fPint
2357Set the CPU affinity of this job. The parameter given is a bit mask of
2358allowed CPUs the job may run on. So if you want the allowed CPUs to be 1
2359and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man
2360\fBsched_setaffinity\fR\|(2). This may not work on all supported
2361operating systems or kernel versions. This option doesn't work well for a
2362higher CPU count than what you can store in an integer mask, so it can only
2363control cpus 1\-32. For boxes with larger CPU counts, use
2364\fBcpus_allowed\fR.
2365.TP
d0b937ed 2366.BI numa_cpu_nodes \fR=\fPstr
cecbfd47 2367Set this job running on specified NUMA nodes' CPUs. The arguments allow
523bad63
TK
2368comma delimited list of cpu numbers, A\-B ranges, or `all'. Note, to enable
2369NUMA options support, fio must be built on a system with libnuma\-dev(el)
2370installed.
d0b937ed
YR
2371.TP
2372.BI numa_mem_policy \fR=\fPstr
523bad63
TK
2373Set this job's memory policy and corresponding NUMA nodes. Format of the
2374arguments:
39c7a2ca
VF
2375.RS
2376.RS
523bad63
TK
2377.P
2378<mode>[:<nodelist>]
39c7a2ca 2379.RE
523bad63 2380.P
f1dd3fb1 2381`mode' is one of the following memory policies: `default', `prefer',
523bad63
TK
2382`bind', `interleave' or `local'. For `default' and `local' memory
2383policies, no node needs to be specified. For `prefer', only one node is
2384allowed. For `bind' and `interleave' the `nodelist' may be as
2385follows: a comma delimited list of numbers, A\-B ranges, or `all'.
39c7a2ca
VF
2386.RE
2387.TP
523bad63
TK
2388.BI cgroup \fR=\fPstr
2389Add job to this control group. If it doesn't exist, it will be created. The
2390system must have a mounted cgroup blkio mount point for this to work. If
2391your system doesn't have it mounted, you can do so with:
d60e92d1
AC
2392.RS
2393.RS
d60e92d1 2394.P
523bad63
TK
2395# mount \-t cgroup \-o blkio none /cgroup
2396.RE
d60e92d1
AC
2397.RE
2398.TP
523bad63
TK
2399.BI cgroup_weight \fR=\fPint
2400Set the weight of the cgroup to this value. See the documentation that comes
2401with the kernel, allowed values are in the range of 100..1000.
d60e92d1 2402.TP
523bad63
TK
2403.BI cgroup_nodelete \fR=\fPbool
2404Normally fio will delete the cgroups it has created after the job
2405completion. To override this behavior and to leave cgroups around after the
2406job completion, set `cgroup_nodelete=1'. This can be useful if one wants
2407to inspect various cgroup files after job completion. Default: false.
c8eeb9df 2408.TP
523bad63
TK
2409.BI flow_id \fR=\fPint
2410The ID of the flow. If not specified, it defaults to being a global
2411flow. See \fBflow\fR.
d60e92d1 2412.TP
523bad63
TK
2413.BI flow \fR=\fPint
2414Weight in token\-based flow control. If this value is used, then there is
2415a 'flow counter' which is used to regulate the proportion of activity between
2416two or more jobs. Fio attempts to keep this flow counter near zero. The
2417\fBflow\fR parameter stands for how much should be added or subtracted to the
2418flow counter on each iteration of the main I/O loop. That is, if one job has
2419`flow=8' and another job has `flow=\-1', then there will be a roughly 1:8
2420ratio in how much one runs vs the other.
d60e92d1 2421.TP
523bad63
TK
2422.BI flow_watermark \fR=\fPint
2423The maximum value that the absolute value of the flow counter is allowed to
2424reach before the job must wait for a lower value of the counter.
6b7f6851 2425.TP
523bad63
TK
2426.BI flow_sleep \fR=\fPint
2427The period of time, in microseconds, to wait after the flow watermark has
2428been exceeded before retrying operations.
25460cf6 2429.TP
523bad63
TK
2430.BI stonewall "\fR,\fB wait_for_previous"
2431Wait for preceding jobs in the job file to exit, before starting this
2432one. Can be used to insert serialization points in the job file. A stone
2433wall also implies starting a new reporting group, see
2434\fBgroup_reporting\fR.
2378826d 2435.TP
523bad63
TK
2436.BI exitall
2437By default, fio will continue running all other jobs when one job finishes
2438but sometimes this is not the desired action. Setting \fBexitall\fR will
2439instead make fio terminate all other jobs when one job finishes.
e81ecca3 2440.TP
523bad63
TK
2441.BI exec_prerun \fR=\fPstr
2442Before running this job, issue the command specified through
2443\fBsystem\fR\|(3). Output is redirected in a file called `jobname.prerun.txt'.
e9f48479 2444.TP
523bad63
TK
2445.BI exec_postrun \fR=\fPstr
2446After the job completes, issue the command specified though
2447\fBsystem\fR\|(3). Output is redirected in a file called `jobname.postrun.txt'.
d60e92d1 2448.TP
523bad63
TK
2449.BI uid \fR=\fPint
2450Instead of running as the invoking user, set the user ID to this value
2451before the thread/process does any work.
39c1c323 2452.TP
523bad63
TK
2453.BI gid \fR=\fPint
2454Set group ID, see \fBuid\fR.
2455.SS "Verification"
d60e92d1 2456.TP
589e88b7 2457.BI verify_only
523bad63 2458Do not perform specified workload, only verify data still matches previous
5e4c7118 2459invocation of this workload. This option allows one to check data multiple
523bad63
TK
2460times at a later date without overwriting it. This option makes sense only
2461for workloads that write data, and does not support workloads with the
5e4c7118
JA
2462\fBtime_based\fR option set.
2463.TP
d60e92d1 2464.BI do_verify \fR=\fPbool
523bad63
TK
2465Run the verify phase after a write phase. Only valid if \fBverify\fR is
2466set. Default: true.
d60e92d1
AC
2467.TP
2468.BI verify \fR=\fPstr
523bad63
TK
2469If writing to a file, fio can verify the file contents after each iteration
2470of the job. Each verification method also implies verification of special
2471header, which is written to the beginning of each block. This header also
2472includes meta information, like offset of the block, block number, timestamp
2473when block was written, etc. \fBverify\fR can be combined with
2474\fBverify_pattern\fR option. The allowed values are:
d60e92d1
AC
2475.RS
2476.RS
2477.TP
523bad63
TK
2478.B md5
2479Use an md5 sum of the data area and store it in the header of
2480each block.
2481.TP
2482.B crc64
2483Use an experimental crc64 sum of the data area and store it in the
2484header of each block.
2485.TP
2486.B crc32c
2487Use a crc32c sum of the data area and store it in the header of
2488each block. This will automatically use hardware acceleration
2489(e.g. SSE4.2 on an x86 or CRC crypto extensions on ARM64) but will
2490fall back to software crc32c if none is found. Generally the
f1dd3fb1 2491fastest checksum fio supports when hardware accelerated.
523bad63
TK
2492.TP
2493.B crc32c\-intel
2494Synonym for crc32c.
2495.TP
2496.B crc32
2497Use a crc32 sum of the data area and store it in the header of each
2498block.
2499.TP
2500.B crc16
2501Use a crc16 sum of the data area and store it in the header of each
2502block.
2503.TP
2504.B crc7
2505Use a crc7 sum of the data area and store it in the header of each
2506block.
2507.TP
2508.B xxhash
2509Use xxhash as the checksum function. Generally the fastest software
2510checksum that fio supports.
2511.TP
2512.B sha512
2513Use sha512 as the checksum function.
2514.TP
2515.B sha256
2516Use sha256 as the checksum function.
2517.TP
2518.B sha1
2519Use optimized sha1 as the checksum function.
2520.TP
2521.B sha3\-224
2522Use optimized sha3\-224 as the checksum function.
2523.TP
2524.B sha3\-256
2525Use optimized sha3\-256 as the checksum function.
2526.TP
2527.B sha3\-384
2528Use optimized sha3\-384 as the checksum function.
2529.TP
2530.B sha3\-512
2531Use optimized sha3\-512 as the checksum function.
d60e92d1
AC
2532.TP
2533.B meta
523bad63
TK
2534This option is deprecated, since now meta information is included in
2535generic verification header and meta verification happens by
2536default. For detailed information see the description of the
2537\fBverify\fR setting. This option is kept because of
2538compatibility's sake with old configurations. Do not use it.
d60e92d1 2539.TP
59245381 2540.B pattern
523bad63
TK
2541Verify a strict pattern. Normally fio includes a header with some
2542basic information and checksumming, but if this option is set, only
2543the specific pattern set with \fBverify_pattern\fR is verified.
59245381 2544.TP
d60e92d1 2545.B null
523bad63
TK
2546Only pretend to verify. Useful for testing internals with
2547`ioengine=null', not for much else.
d60e92d1 2548.RE
523bad63
TK
2549.P
2550This option can be used for repeated burn\-in tests of a system to make sure
2551that the written data is also correctly read back. If the data direction
2552given is a read or random read, fio will assume that it should verify a
2553previously written file. If the data direction includes any form of write,
2554the verify will be of the newly written data.
47e6a6e5
SW
2555.P
2556To avoid false verification errors, do not use the norandommap option when
2557verifying data with async I/O engines and I/O depths > 1. Or use the
2558norandommap and the lfsr random generator together to avoid writing to the
2559same offset with muliple outstanding I/Os.
d60e92d1
AC
2560.RE
2561.TP
f7fa2653 2562.BI verify_offset \fR=\fPint
d60e92d1 2563Swap the verification header with data somewhere else in the block before
523bad63 2564writing. It is swapped back before verifying.
d60e92d1 2565.TP
f7fa2653 2566.BI verify_interval \fR=\fPint
523bad63
TK
2567Write the verification header at a finer granularity than the
2568\fBblocksize\fR. It will be written for chunks the size of
2569\fBverify_interval\fR. \fBblocksize\fR should divide this evenly.
d60e92d1 2570.TP
996093bb 2571.BI verify_pattern \fR=\fPstr
523bad63
TK
2572If set, fio will fill the I/O buffers with this pattern. Fio defaults to
2573filling with totally random bytes, but sometimes it's interesting to fill
2574with a known pattern for I/O verification purposes. Depending on the width
2575of the pattern, fio will fill 1/2/3/4 bytes of the buffer at the time (it can
2576be either a decimal or a hex number). The \fBverify_pattern\fR if larger than
2577a 32\-bit quantity has to be a hex number that starts with either "0x" or
2578"0X". Use with \fBverify\fR. Also, \fBverify_pattern\fR supports %o
2579format, which means that for each block offset will be written and then
2580verified back, e.g.:
2fa5a241
RP
2581.RS
2582.RS
523bad63
TK
2583.P
2584verify_pattern=%o
2fa5a241 2585.RE
523bad63 2586.P
2fa5a241 2587Or use combination of everything:
2fa5a241 2588.RS
523bad63
TK
2589.P
2590verify_pattern=0xff%o"abcd"\-12
2fa5a241
RP
2591.RE
2592.RE
996093bb 2593.TP
d60e92d1 2594.BI verify_fatal \fR=\fPbool
523bad63
TK
2595Normally fio will keep checking the entire contents before quitting on a
2596block verification failure. If this option is set, fio will exit the job on
2597the first observed failure. Default: false.
d60e92d1 2598.TP
b463e936 2599.BI verify_dump \fR=\fPbool
523bad63
TK
2600If set, dump the contents of both the original data block and the data block
2601we read off disk to files. This allows later analysis to inspect just what
2602kind of data corruption occurred. Off by default.
b463e936 2603.TP
e8462bd8 2604.BI verify_async \fR=\fPint
523bad63
TK
2605Fio will normally verify I/O inline from the submitting thread. This option
2606takes an integer describing how many async offload threads to create for I/O
2607verification instead, causing fio to offload the duty of verifying I/O
2608contents to one or more separate threads. If using this offload option, even
2609sync I/O engines can benefit from using an \fBiodepth\fR setting higher
2610than 1, as it allows them to have I/O in flight while verifies are running.
2611Defaults to 0 async threads, i.e. verification is not asynchronous.
e8462bd8
JA
2612.TP
2613.BI verify_async_cpus \fR=\fPstr
523bad63
TK
2614Tell fio to set the given CPU affinity on the async I/O verification
2615threads. See \fBcpus_allowed\fR for the format used.
e8462bd8 2616.TP
6f87418f
JA
2617.BI verify_backlog \fR=\fPint
2618Fio will normally verify the written contents of a job that utilizes verify
2619once that job has completed. In other words, everything is written then
2620everything is read back and verified. You may want to verify continually
523bad63
TK
2621instead for a variety of reasons. Fio stores the meta data associated with
2622an I/O block in memory, so for large verify workloads, quite a bit of memory
2623would be used up holding this meta data. If this option is enabled, fio will
2624write only N blocks before verifying these blocks.
6f87418f
JA
2625.TP
2626.BI verify_backlog_batch \fR=\fPint
523bad63
TK
2627Control how many blocks fio will verify if \fBverify_backlog\fR is
2628set. If not set, will default to the value of \fBverify_backlog\fR
2629(meaning the entire queue is read back and verified). If
2630\fBverify_backlog_batch\fR is less than \fBverify_backlog\fR then not all
2631blocks will be verified, if \fBverify_backlog_batch\fR is larger than
2632\fBverify_backlog\fR, some blocks will be verified more than once.
2633.TP
2634.BI verify_state_save \fR=\fPbool
2635When a job exits during the write phase of a verify workload, save its
2636current state. This allows fio to replay up until that point, if the verify
2637state is loaded for the verify read phase. The format of the filename is,
2638roughly:
2639.RS
2640.RS
2641.P
2642<type>\-<jobname>\-<jobindex>\-verify.state.
2643.RE
2644.P
2645<type> is "local" for a local run, "sock" for a client/server socket
2646connection, and "ip" (192.168.0.1, for instance) for a networked
2647client/server connection. Defaults to true.
2648.RE
2649.TP
2650.BI verify_state_load \fR=\fPbool
2651If a verify termination trigger was used, fio stores the current write state
2652of each thread. This can be used at verification time so that fio knows how
2653far it should verify. Without this information, fio will run a full
2654verification pass, according to the settings in the job file used. Default
2655false.
6f87418f 2656.TP
fa769d44
SW
2657.BI trim_percentage \fR=\fPint
2658Number of verify blocks to discard/trim.
2659.TP
2660.BI trim_verify_zero \fR=\fPbool
523bad63 2661Verify that trim/discarded blocks are returned as zeros.
fa769d44
SW
2662.TP
2663.BI trim_backlog \fR=\fPint
523bad63 2664Verify that trim/discarded blocks are returned as zeros.
fa769d44
SW
2665.TP
2666.BI trim_backlog_batch \fR=\fPint
523bad63 2667Trim this number of I/O blocks.
fa769d44
SW
2668.TP
2669.BI experimental_verify \fR=\fPbool
2670Enable experimental verification.
523bad63 2671.SS "Steady state"
fa769d44 2672.TP
523bad63
TK
2673.BI steadystate \fR=\fPstr:float "\fR,\fP ss" \fR=\fPstr:float
2674Define the criterion and limit for assessing steady state performance. The
2675first parameter designates the criterion whereas the second parameter sets
2676the threshold. When the criterion falls below the threshold for the
2677specified duration, the job will stop. For example, `iops_slope:0.1%' will
2678direct fio to terminate the job when the least squares regression slope
2679falls below 0.1% of the mean IOPS. If \fBgroup_reporting\fR is enabled
2680this will apply to all jobs in the group. Below is the list of available
2681steady state assessment criteria. All assessments are carried out using only
2682data from the rolling collection window. Threshold limits can be expressed
2683as a fixed value or as a percentage of the mean in the collection window.
2684.RS
1cb049d9
VF
2685.P
2686When using this feature, most jobs should include the \fBtime_based\fR
2687and \fBruntime\fR options or the \fBloops\fR option so that fio does not
2688stop running after it has covered the full size of the specified file(s)
2689or device(s).
2690.RS
523bad63 2691.RS
d60e92d1 2692.TP
523bad63
TK
2693.B iops
2694Collect IOPS data. Stop the job if all individual IOPS measurements
2695are within the specified limit of the mean IOPS (e.g., `iops:2'
2696means that all individual IOPS values must be within 2 of the mean,
2697whereas `iops:0.2%' means that all individual IOPS values must be
2698within 0.2% of the mean IOPS to terminate the job).
d60e92d1 2699.TP
523bad63
TK
2700.B iops_slope
2701Collect IOPS data and calculate the least squares regression
2702slope. Stop the job if the slope falls below the specified limit.
d60e92d1 2703.TP
523bad63
TK
2704.B bw
2705Collect bandwidth data. Stop the job if all individual bandwidth
2706measurements are within the specified limit of the mean bandwidth.
64bbb865 2707.TP
523bad63
TK
2708.B bw_slope
2709Collect bandwidth data and calculate the least squares regression
2710slope. Stop the job if the slope falls below the specified limit.
2711.RE
2712.RE
d1c46c04 2713.TP
523bad63
TK
2714.BI steadystate_duration \fR=\fPtime "\fR,\fP ss_dur" \fR=\fPtime
2715A rolling window of this duration will be used to judge whether steady state
2716has been reached. Data will be collected once per second. The default is 0
2717which disables steady state detection. When the unit is omitted, the
2718value is interpreted in seconds.
0c63576e 2719.TP
523bad63
TK
2720.BI steadystate_ramp_time \fR=\fPtime "\fR,\fP ss_ramp" \fR=\fPtime
2721Allow the job to run for the specified duration before beginning data
2722collection for checking the steady state job termination criterion. The
2723default is 0. When the unit is omitted, the value is interpreted in seconds.
2724.SS "Measurements and reporting"
0c63576e 2725.TP
3a5db920
JA
2726.BI per_job_logs \fR=\fPbool
2727If set, this generates bw/clat/iops log with per file private filenames. If
523bad63
TK
2728not set, jobs with identical names will share the log filename. Default:
2729true.
2730.TP
2731.BI group_reporting
2732It may sometimes be interesting to display statistics for groups of jobs as
2733a whole instead of for each individual job. This is especially true if
2734\fBnumjobs\fR is used; looking at individual thread/process output
2735quickly becomes unwieldy. To see the final report per\-group instead of
2736per\-job, use \fBgroup_reporting\fR. Jobs in a file will be part of the
2737same reporting group, unless if separated by a \fBstonewall\fR, or by
2738using \fBnew_group\fR.
2739.TP
2740.BI new_group
2741Start a new reporting group. See: \fBgroup_reporting\fR. If not given,
2742all jobs in a file will be part of the same reporting group, unless
2743separated by a \fBstonewall\fR.
2744.TP
2745.BI stats \fR=\fPbool
2746By default, fio collects and shows final output results for all jobs
2747that run. If this option is set to 0, then fio will ignore it in
2748the final stat output.
3a5db920 2749.TP
836bad52 2750.BI write_bw_log \fR=\fPstr
523bad63 2751If given, write a bandwidth log for this job. Can be used to store data of
074f0817 2752the bandwidth of the jobs in their lifetime.
523bad63 2753.RS
074f0817
SW
2754.P
2755If no str argument is given, the default filename of
2756`jobname_type.x.log' is used. Even when the argument is given, fio
2757will still append the type of log. So if one specifies:
523bad63
TK
2758.RS
2759.P
074f0817 2760write_bw_log=foo
523bad63
TK
2761.RE
2762.P
074f0817
SW
2763The actual log name will be `foo_bw.x.log' where `x' is the index
2764of the job (1..N, where N is the number of jobs). If
2765\fBper_job_logs\fR is false, then the filename will not include the
2766`.x` job index.
2767.P
2768The included \fBfio_generate_plots\fR script uses gnuplot to turn these
2769text files into nice graphs. See the \fBLOG FILE FORMATS\fR section for how data is
2770structured within the file.
523bad63 2771.RE
901bb994 2772.TP
074f0817
SW
2773.BI write_lat_log \fR=\fPstr
2774Same as \fBwrite_bw_log\fR, except this option creates I/O
2775submission (e.g., `name_slat.x.log'), completion (e.g.,
2776`name_clat.x.log'), and total (e.g., `name_lat.x.log') latency
2777files instead. See \fBwrite_bw_log\fR for details about the
2778filename format and the \fBLOG FILE FORMATS\fR section for how data is structured
2779within the files.
2780.TP
1e613c9c 2781.BI write_hist_log \fR=\fPstr
074f0817
SW
2782Same as \fBwrite_bw_log\fR but writes an I/O completion latency
2783histogram file (e.g., `name_hist.x.log') instead. Note that this
2784file will be empty unless \fBlog_hist_msec\fR has also been set.
2785See \fBwrite_bw_log\fR for details about the filename format and
2786the \fBLOG FILE FORMATS\fR section for how data is structured
2787within the file.
1e613c9c 2788.TP
c8eeb9df 2789.BI write_iops_log \fR=\fPstr
074f0817 2790Same as \fBwrite_bw_log\fR, but writes an IOPS file (e.g.
15417073
SW
2791`name_iops.x.log`) instead. Because fio defaults to individual
2792I/O logging, the value entry in the IOPS log will be 1 unless windowed
2793logging (see \fBlog_avg_msec\fR) has been enabled. See
2794\fBwrite_bw_log\fR for details about the filename format and \fBLOG
2795FILE FORMATS\fR for how data is structured within the file.
c8eeb9df 2796.TP
b8bc8cba
JA
2797.BI log_avg_msec \fR=\fPint
2798By default, fio will log an entry in the iops, latency, or bw log for every
523bad63 2799I/O that completes. When writing to the disk log, that can quickly grow to a
b8bc8cba 2800very large size. Setting this option makes fio average the each log entry
e6989e10 2801over the specified period of time, reducing the resolution of the log. See
523bad63
TK
2802\fBlog_max_value\fR as well. Defaults to 0, logging all entries.
2803Also see \fBLOG FILE FORMATS\fR section.
b8bc8cba 2804.TP
1e613c9c 2805.BI log_hist_msec \fR=\fPint
523bad63
TK
2806Same as \fBlog_avg_msec\fR, but logs entries for completion latency
2807histograms. Computing latency percentiles from averages of intervals using
2808\fBlog_avg_msec\fR is inaccurate. Setting this option makes fio log
2809histogram entries over the specified period of time, reducing log sizes for
2810high IOPS devices while retaining percentile accuracy. See
074f0817
SW
2811\fBlog_hist_coarseness\fR and \fBwrite_hist_log\fR as well.
2812Defaults to 0, meaning histogram logging is disabled.
1e613c9c
KC
2813.TP
2814.BI log_hist_coarseness \fR=\fPint
523bad63
TK
2815Integer ranging from 0 to 6, defining the coarseness of the resolution of
2816the histogram logs enabled with \fBlog_hist_msec\fR. For each increment
2817in coarseness, fio outputs half as many bins. Defaults to 0, for which
2818histogram logs contain 1216 latency bins. See \fBLOG FILE FORMATS\fR section.
2819.TP
2820.BI log_max_value \fR=\fPbool
2821If \fBlog_avg_msec\fR is set, fio logs the average over that window. If
2822you instead want to log the maximum value, set this option to 1. Defaults to
28230, meaning that averaged values are logged.
1e613c9c 2824.TP
ae588852 2825.BI log_offset \fR=\fPbool
523bad63
TK
2826If this is set, the iolog options will include the byte offset for the I/O
2827entry as well as the other data values. Defaults to 0 meaning that
2828offsets are not present in logs. Also see \fBLOG FILE FORMATS\fR section.
ae588852 2829.TP
aee2ab67 2830.BI log_compression \fR=\fPint
523bad63
TK
2831If this is set, fio will compress the I/O logs as it goes, to keep the
2832memory footprint lower. When a log reaches the specified size, that chunk is
2833removed and compressed in the background. Given that I/O logs are fairly
2834highly compressible, this yields a nice memory savings for longer runs. The
2835downside is that the compression will consume some background CPU cycles, so
2836it may impact the run. This, however, is also true if the logging ends up
2837consuming most of the system memory. So pick your poison. The I/O logs are
2838saved normally at the end of a run, by decompressing the chunks and storing
2839them in the specified log file. This feature depends on the availability of
2840zlib.
aee2ab67 2841.TP
c08f9fe2 2842.BI log_compression_cpus \fR=\fPstr
523bad63
TK
2843Define the set of CPUs that are allowed to handle online log compression for
2844the I/O jobs. This can provide better isolation between performance
0cf90a62
SW
2845sensitive jobs, and background compression work. See \fBcpus_allowed\fR for
2846the format used.
c08f9fe2 2847.TP
b26317c9 2848.BI log_store_compressed \fR=\fPbool
c08f9fe2 2849If set, fio will store the log files in a compressed format. They can be
523bad63
TK
2850decompressed with fio, using the \fB\-\-inflate\-log\fR command line
2851parameter. The files will be stored with a `.fz' suffix.
b26317c9 2852.TP
3aea75b1
KC
2853.BI log_unix_epoch \fR=\fPbool
2854If set, fio will log Unix timestamps to the log files produced by enabling
523bad63 2855write_type_log for each log type, instead of the default zero\-based
3aea75b1
KC
2856timestamps.
2857.TP
66347cfa 2858.BI block_error_percentiles \fR=\fPbool
523bad63
TK
2859If set, record errors in trim block\-sized units from writes and trims and
2860output a histogram of how many trims it took to get to errors, and what kind
2861of error was encountered.
d60e92d1 2862.TP
523bad63
TK
2863.BI bwavgtime \fR=\fPint
2864Average the calculated bandwidth over the given time. Value is specified in
2865milliseconds. If the job also does bandwidth logging through
2866\fBwrite_bw_log\fR, then the minimum of this option and
2867\fBlog_avg_msec\fR will be used. Default: 500ms.
d60e92d1 2868.TP
523bad63
TK
2869.BI iopsavgtime \fR=\fPint
2870Average the calculated IOPS over the given time. Value is specified in
2871milliseconds. If the job also does IOPS logging through
2872\fBwrite_iops_log\fR, then the minimum of this option and
2873\fBlog_avg_msec\fR will be used. Default: 500ms.
d60e92d1 2874.TP
d60e92d1 2875.BI disk_util \fR=\fPbool
523bad63
TK
2876Generate disk utilization statistics, if the platform supports it.
2877Default: true.
fa769d44 2878.TP
523bad63
TK
2879.BI disable_lat \fR=\fPbool
2880Disable measurements of total latency numbers. Useful only for cutting back
2881the number of calls to \fBgettimeofday\fR\|(2), as that does impact
2882performance at really high IOPS rates. Note that to really get rid of a
2883large amount of these calls, this option must be used with
2884\fBdisable_slat\fR and \fBdisable_bw_measurement\fR as well.
9e684a49 2885.TP
523bad63
TK
2886.BI disable_clat \fR=\fPbool
2887Disable measurements of completion latency numbers. See
2888\fBdisable_lat\fR.
9e684a49 2889.TP
523bad63
TK
2890.BI disable_slat \fR=\fPbool
2891Disable measurements of submission latency numbers. See
2892\fBdisable_lat\fR.
9e684a49 2893.TP
523bad63
TK
2894.BI disable_bw_measurement \fR=\fPbool "\fR,\fP disable_bw" \fR=\fPbool
2895Disable measurements of throughput/bandwidth numbers. See
2896\fBdisable_lat\fR.
9e684a49 2897.TP
83349190 2898.BI clat_percentiles \fR=\fPbool
b599759b
JA
2899Enable the reporting of percentiles of completion latencies. This option is
2900mutually exclusive with \fBlat_percentiles\fR.
2901.TP
2902.BI lat_percentiles \fR=\fPbool
b71968b1 2903Enable the reporting of percentiles of I/O latencies. This is similar to
b599759b
JA
2904\fBclat_percentiles\fR, except that this includes the submission latency.
2905This option is mutually exclusive with \fBclat_percentiles\fR.
83349190
YH
2906.TP
2907.BI percentile_list \fR=\fPfloat_list
66347cfa 2908Overwrite the default list of percentiles for completion latencies and the
523bad63
TK
2909block error histogram. Each number is a floating number in the range
2910(0,100], and the maximum length of the list is 20. Use ':' to separate the
2911numbers, and list the numbers in ascending order. For example,
2912`\-\-percentile_list=99.5:99.9' will cause fio to report the values of
2913completion latency below which 99.5% and 99.9% of the observed latencies
2914fell, respectively.
e883cb35
JF
2915.TP
2916.BI significant_figures \fR=\fPint
c32ba107
JA
2917If using \fB\-\-output\-format\fR of `normal', set the significant figures
2918to this value. Higher values will yield more precise IOPS and throughput
2919units, while lower values will round. Requires a minimum value of 1 and a
e883cb35 2920maximum value of 10. Defaults to 4.
523bad63 2921.SS "Error handling"
e4585935 2922.TP
523bad63
TK
2923.BI exitall_on_error
2924When one job finishes in error, terminate the rest. The default is to wait
2925for each job to finish.
e4585935 2926.TP
523bad63
TK
2927.BI continue_on_error \fR=\fPstr
2928Normally fio will exit the job on the first observed failure. If this option
2929is set, fio will continue the job when there is a 'non\-fatal error' (EIO or
2930EILSEQ) until the runtime is exceeded or the I/O size specified is
2931completed. If this option is used, there are two more stats that are
2932appended, the total error count and the first error. The error field given
2933in the stats is the first error that was hit during the run.
2934The allowed values are:
2935.RS
2936.RS
046395d7 2937.TP
523bad63
TK
2938.B none
2939Exit on any I/O or verify errors.
de890a1e 2940.TP
523bad63
TK
2941.B read
2942Continue on read errors, exit on all others.
2cafffbe 2943.TP
523bad63
TK
2944.B write
2945Continue on write errors, exit on all others.
a0679ce5 2946.TP
523bad63
TK
2947.B io
2948Continue on any I/O error, exit on all others.
de890a1e 2949.TP
523bad63
TK
2950.B verify
2951Continue on verify errors, exit on all others.
de890a1e 2952.TP
523bad63
TK
2953.B all
2954Continue on all errors.
b93b6a2e 2955.TP
523bad63
TK
2956.B 0
2957Backward\-compatible alias for 'none'.
d3a623de 2958.TP
523bad63
TK
2959.B 1
2960Backward\-compatible alias for 'all'.
2961.RE
2962.RE
1d360ffb 2963.TP
523bad63
TK
2964.BI ignore_error \fR=\fPstr
2965Sometimes you want to ignore some errors during test in that case you can
2966specify error list for each error type, instead of only being able to
2967ignore the default 'non\-fatal error' using \fBcontinue_on_error\fR.
2968`ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST' errors for
2969given error type is separated with ':'. Error may be symbol ('ENOSPC', 'ENOMEM')
2970or integer. Example:
de890a1e
SL
2971.RS
2972.RS
523bad63
TK
2973.P
2974ignore_error=EAGAIN,ENOSPC:122
2975.RE
2976.P
2977This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from
2978WRITE. This option works by overriding \fBcontinue_on_error\fR with
2979the list of errors for each error type if any.
2980.RE
de890a1e 2981.TP
523bad63
TK
2982.BI error_dump \fR=\fPbool
2983If set dump every error even if it is non fatal, true by default. If
2984disabled only fatal error will be dumped.
2985.SS "Running predefined workloads"
2986Fio includes predefined profiles that mimic the I/O workloads generated by
2987other tools.
49ccb8c1 2988.TP
523bad63
TK
2989.BI profile \fR=\fPstr
2990The predefined workload to run. Current profiles are:
2991.RS
2992.RS
de890a1e 2993.TP
523bad63
TK
2994.B tiobench
2995Threaded I/O bench (tiotest/tiobench) like workload.
49ccb8c1 2996.TP
523bad63
TK
2997.B act
2998Aerospike Certification Tool (ACT) like workload.
2999.RE
de890a1e
SL
3000.RE
3001.P
523bad63
TK
3002To view a profile's additional options use \fB\-\-cmdhelp\fR after specifying
3003the profile. For example:
3004.RS
3005.TP
3006$ fio \-\-profile=act \-\-cmdhelp
de890a1e 3007.RE
523bad63 3008.SS "Act profile options"
de890a1e 3009.TP
523bad63
TK
3010.BI device\-names \fR=\fPstr
3011Devices to use.
d54fce84 3012.TP
523bad63
TK
3013.BI load \fR=\fPint
3014ACT load multiplier. Default: 1.
7aeb1e94 3015.TP
523bad63
TK
3016.BI test\-duration\fR=\fPtime
3017How long the entire test takes to run. When the unit is omitted, the value
3018is given in seconds. Default: 24h.
1008602c 3019.TP
523bad63
TK
3020.BI threads\-per\-queue\fR=\fPint
3021Number of read I/O threads per device. Default: 8.
e5f34d95 3022.TP
523bad63
TK
3023.BI read\-req\-num\-512\-blocks\fR=\fPint
3024Number of 512B blocks to read at the time. Default: 3.
d54fce84 3025.TP
523bad63
TK
3026.BI large\-block\-op\-kbytes\fR=\fPint
3027Size of large block ops in KiB (writes). Default: 131072.
d54fce84 3028.TP
523bad63
TK
3029.BI prep
3030Set to run ACT prep phase.
3031.SS "Tiobench profile options"
6d500c2e 3032.TP
523bad63
TK
3033.BI size\fR=\fPstr
3034Size in MiB.
0d978694 3035.TP
523bad63
TK
3036.BI block\fR=\fPint
3037Block size in bytes. Default: 4096.
0d978694 3038.TP
523bad63
TK
3039.BI numruns\fR=\fPint
3040Number of runs.
0d978694 3041.TP
523bad63
TK
3042.BI dir\fR=\fPstr
3043Test directory.
65fa28ca 3044.TP
523bad63
TK
3045.BI threads\fR=\fPint
3046Number of threads.
d60e92d1 3047.SH OUTPUT
40943b9a
TK
3048Fio spits out a lot of output. While running, fio will display the status of the
3049jobs created. An example of that would be:
d60e92d1 3050.P
40943b9a
TK
3051.nf
3052 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]
3053.fi
d1429b5c 3054.P
40943b9a
TK
3055The characters inside the first set of square brackets denote the current status of
3056each thread. The first character is the first job defined in the job file, and so
3057forth. The possible values (in typical life cycle order) are:
d60e92d1
AC
3058.RS
3059.TP
40943b9a 3060.PD 0
d60e92d1 3061.B P
40943b9a 3062Thread setup, but not started.
d60e92d1
AC
3063.TP
3064.B C
3065Thread created.
3066.TP
3067.B I
40943b9a
TK
3068Thread initialized, waiting or generating necessary data.
3069.TP
522c29f6 3070.B p
40943b9a
TK
3071Thread running pre\-reading file(s).
3072.TP
3073.B /
3074Thread is in ramp period.
d60e92d1
AC
3075.TP
3076.B R
3077Running, doing sequential reads.
3078.TP
3079.B r
3080Running, doing random reads.
3081.TP
3082.B W
3083Running, doing sequential writes.
3084.TP
3085.B w
3086Running, doing random writes.
3087.TP
3088.B M
3089Running, doing mixed sequential reads/writes.
3090.TP
3091.B m
3092Running, doing mixed random reads/writes.
3093.TP
40943b9a
TK
3094.B D
3095Running, doing sequential trims.
3096.TP
3097.B d
3098Running, doing random trims.
3099.TP
d60e92d1
AC
3100.B F
3101Running, currently waiting for \fBfsync\fR\|(2).
3102.TP
3103.B V
40943b9a
TK
3104Running, doing verification of written data.
3105.TP
3106.B f
3107Thread finishing.
d60e92d1
AC
3108.TP
3109.B E
40943b9a 3110Thread exited, not reaped by main thread yet.
d60e92d1
AC
3111.TP
3112.B \-
40943b9a
TK
3113Thread reaped.
3114.TP
3115.B X
3116Thread reaped, exited with an error.
3117.TP
3118.B K
3119Thread reaped, exited due to signal.
d1429b5c 3120.PD
40943b9a
TK
3121.RE
3122.P
3123Fio will condense the thread string as not to take up more space on the command
3124line than needed. For instance, if you have 10 readers and 10 writers running,
3125the output would look like this:
3126.P
3127.nf
3128 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]
3129.fi
d60e92d1 3130.P
40943b9a
TK
3131Note that the status string is displayed in order, so it's possible to tell which of
3132the jobs are currently doing what. In the example above this means that jobs 1\-\-10
3133are readers and 11\-\-20 are writers.
d60e92d1 3134.P
40943b9a
TK
3135The other values are fairly self explanatory \-\- number of threads currently
3136running and doing I/O, the number of currently open files (f=), the estimated
3137completion percentage, the rate of I/O since last check (read speed listed first,
3138then write speed and optionally trim speed) in terms of bandwidth and IOPS,
3139and time to completion for the current running group. It's impossible to estimate
3140runtime of the following groups (if any).
d60e92d1 3141.P
40943b9a
TK
3142When fio is done (or interrupted by Ctrl\-C), it will show the data for
3143each thread, group of threads, and disks in that order. For each overall thread (or
3144group) the output looks like:
3145.P
3146.nf
3147 Client1: (groupid=0, jobs=1): err= 0: pid=16109: Sat Jun 24 12:07:54 2017
3148 write: IOPS=88, BW=623KiB/s (638kB/s)(30.4MiB/50032msec)
3149 slat (nsec): min=500, max=145500, avg=8318.00, stdev=4781.50
3150 clat (usec): min=170, max=78367, avg=4019.02, stdev=8293.31
3151 lat (usec): min=174, max=78375, avg=4027.34, stdev=8291.79
3152 clat percentiles (usec):
3153 | 1.00th=[ 302], 5.00th=[ 326], 10.00th=[ 343], 20.00th=[ 363],
3154 | 30.00th=[ 392], 40.00th=[ 404], 50.00th=[ 416], 60.00th=[ 445],
3155 | 70.00th=[ 816], 80.00th=[ 6718], 90.00th=[12911], 95.00th=[21627],
3156 | 99.00th=[43779], 99.50th=[51643], 99.90th=[68682], 99.95th=[72877],
3157 | 99.99th=[78119]
3158 bw ( KiB/s): min= 532, max= 686, per=0.10%, avg=622.87, stdev=24.82, samples= 100
3159 iops : min= 76, max= 98, avg=88.98, stdev= 3.54, samples= 100
d3b9694d
VF
3160 lat (usec) : 250=0.04%, 500=64.11%, 750=4.81%, 1000=2.79%
3161 lat (msec) : 2=4.16%, 4=1.84%, 10=4.90%, 20=11.33%, 50=5.37%
3162 lat (msec) : 100=0.65%
40943b9a
TK
3163 cpu : usr=0.27%, sys=0.18%, ctx=12072, majf=0, minf=21
3164 IO depths : 1=85.0%, 2=13.1%, 4=1.8%, 8=0.1%, 16=0.0%, 32=0.0%, >=64=0.0%
3165 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3166 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
3167 issued rwt: total=0,4450,0, short=0,0,0, dropped=0,0,0
3168 latency : target=0, window=0, percentile=100.00%, depth=8
3169.fi
3170.P
3171The job name (or first job's name when using \fBgroup_reporting\fR) is printed,
3172along with the group id, count of jobs being aggregated, last error id seen (which
3173is 0 when there are no errors), pid/tid of that thread and the time the job/group
3174completed. Below are the I/O statistics for each data direction performed (showing
3175writes in the example above). In the order listed, they denote:
d60e92d1 3176.RS
d60e92d1 3177.TP
40943b9a
TK
3178.B read/write/trim
3179The string before the colon shows the I/O direction the statistics
3180are for. \fIIOPS\fR is the average I/Os performed per second. \fIBW\fR
3181is the average bandwidth rate shown as: value in power of 2 format
3182(value in power of 10 format). The last two values show: (total
3183I/O performed in power of 2 format / \fIruntime\fR of that thread).
d60e92d1
AC
3184.TP
3185.B slat
40943b9a
TK
3186Submission latency (\fImin\fR being the minimum, \fImax\fR being the
3187maximum, \fIavg\fR being the average, \fIstdev\fR being the standard
3188deviation). This is the time it took to submit the I/O. For
3189sync I/O this row is not displayed as the slat is really the
3190completion latency (since queue/complete is one operation there).
3191This value can be in nanoseconds, microseconds or milliseconds \-\-\-
3192fio will choose the most appropriate base and print that (in the
3193example above nanoseconds was the best scale). Note: in \fB\-\-minimal\fR mode
3194latencies are always expressed in microseconds.
d60e92d1
AC
3195.TP
3196.B clat
40943b9a
TK
3197Completion latency. Same names as slat, this denotes the time from
3198submission to completion of the I/O pieces. For sync I/O, clat will
3199usually be equal (or very close) to 0, as the time from submit to
3200complete is basically just CPU time (I/O has already been done, see slat
3201explanation).
d60e92d1 3202.TP
d3b9694d
VF
3203.B lat
3204Total latency. Same names as slat and clat, this denotes the time from
3205when fio created the I/O unit to completion of the I/O operation.
3206.TP
d60e92d1 3207.B bw
40943b9a
TK
3208Bandwidth statistics based on samples. Same names as the xlat stats,
3209but also includes the number of samples taken (\fIsamples\fR) and an
3210approximate percentage of total aggregate bandwidth this thread
3211received in its group (\fIper\fR). This last value is only really
3212useful if the threads in this group are on the same disk, since they
3213are then competing for disk access.
3214.TP
3215.B iops
3216IOPS statistics based on samples. Same names as \fBbw\fR.
d60e92d1 3217.TP
d3b9694d
VF
3218.B lat (nsec/usec/msec)
3219The distribution of I/O completion latencies. This is the time from when
3220I/O leaves fio and when it gets completed. Unlike the separate
3221read/write/trim sections above, the data here and in the remaining
3222sections apply to all I/Os for the reporting group. 250=0.04% means that
32230.04% of the I/Os completed in under 250us. 500=64.11% means that 64.11%
3224of the I/Os required 250 to 499us for completion.
3225.TP
d60e92d1 3226.B cpu
40943b9a
TK
3227CPU usage. User and system time, along with the number of context
3228switches this thread went through, usage of system and user time, and
3229finally the number of major and minor page faults. The CPU utilization
3230numbers are averages for the jobs in that reporting group, while the
3231context and fault counters are summed.
d60e92d1
AC
3232.TP
3233.B IO depths
40943b9a
TK
3234The distribution of I/O depths over the job lifetime. The numbers are
3235divided into powers of 2 and each entry covers depths from that value
3236up to those that are lower than the next entry \-\- e.g., 16= covers
3237depths from 16 to 31. Note that the range covered by a depth
3238distribution entry can be different to the range covered by the
3239equivalent \fBsubmit\fR/\fBcomplete\fR distribution entry.
3240.TP
3241.B IO submit
3242How many pieces of I/O were submitting in a single submit call. Each
3243entry denotes that amount and below, until the previous entry \-\- e.g.,
324416=100% means that we submitted anywhere between 9 to 16 I/Os per submit
3245call. Note that the range covered by a \fBsubmit\fR distribution entry can
3246be different to the range covered by the equivalent depth distribution
3247entry.
3248.TP
3249.B IO complete
3250Like the above \fBsubmit\fR number, but for completions instead.
3251.TP
3252.B IO issued rwt
3253The number of \fBread/write/trim\fR requests issued, and how many of them were
3254short or dropped.
d60e92d1 3255.TP
d3b9694d 3256.B IO latency
ee21ebee 3257These values are for \fBlatency_target\fR and related options. When
d3b9694d
VF
3258these options are engaged, this section describes the I/O depth required
3259to meet the specified latency target.
d60e92d1 3260.RE
d60e92d1 3261.P
40943b9a
TK
3262After each client has been listed, the group statistics are printed. They
3263will look like this:
3264.P
3265.nf
3266 Run status group 0 (all jobs):
3267 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
3268 WRITE: bw=1231KiB/s (1261kB/s), 616KiB/s\-621KiB/s (630kB/s\-636kB/s), io=64.0MiB (67.1MB), run=52747\-53223msec
3269.fi
3270.P
3271For each data direction it prints:
d60e92d1
AC
3272.RS
3273.TP
40943b9a
TK
3274.B bw
3275Aggregate bandwidth of threads in this group followed by the
3276minimum and maximum bandwidth of all the threads in this group.
3277Values outside of brackets are power\-of\-2 format and those
3278within are the equivalent value in a power\-of\-10 format.
d60e92d1 3279.TP
40943b9a
TK
3280.B io
3281Aggregate I/O performed of all threads in this group. The
3282format is the same as \fBbw\fR.
d60e92d1 3283.TP
40943b9a
TK
3284.B run
3285The smallest and longest runtimes of the threads in this group.
d60e92d1 3286.RE
d60e92d1 3287.P
40943b9a
TK
3288And finally, the disk statistics are printed. This is Linux specific.
3289They will look like this:
3290.P
3291.nf
3292 Disk stats (read/write):
3293 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
3294.fi
3295.P
3296Each value is printed for both reads and writes, with reads first. The
3297numbers denote:
d60e92d1
AC
3298.RS
3299.TP
3300.B ios
3301Number of I/Os performed by all groups.
3302.TP
3303.B merge
007c7be9 3304Number of merges performed by the I/O scheduler.
d60e92d1
AC
3305.TP
3306.B ticks
3307Number of ticks we kept the disk busy.
3308.TP
40943b9a 3309.B in_queue
d60e92d1
AC
3310Total time spent in the disk queue.
3311.TP
3312.B util
40943b9a
TK
3313The disk utilization. A value of 100% means we kept the disk
3314busy constantly, 50% would be a disk idling half of the time.
d60e92d1 3315.RE
8423bd11 3316.P
40943b9a
TK
3317It is also possible to get fio to dump the current output while it is running,
3318without terminating the job. To do that, send fio the USR1 signal. You can
3319also get regularly timed dumps by using the \fB\-\-status\-interval\fR
3320parameter, or by creating a file in `/tmp' named
3321`fio\-dump\-status'. If fio sees this file, it will unlink it and dump the
3322current output status.
d60e92d1 3323.SH TERSE OUTPUT
40943b9a
TK
3324For scripted usage where you typically want to generate tables or graphs of the
3325results, fio can output the results in a semicolon separated format. The format
3326is one long line of values, such as:
d60e92d1 3327.P
40943b9a
TK
3328.nf
3329 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%
3330 A description of this job goes here.
3331.fi
d60e92d1 3332.P
4e757af1
VF
3333The job description (if provided) follows on a second line for terse v2.
3334It appears on the same line for other terse versions.
d60e92d1 3335.P
40943b9a
TK
3336To enable terse output, use the \fB\-\-minimal\fR or
3337`\-\-output\-format=terse' command line options. The
3338first value is the version of the terse output format. If the output has to be
3339changed for some reason, this number will be incremented by 1 to signify that
3340change.
d60e92d1 3341.P
40943b9a
TK
3342Split up, the format is as follows (comments in brackets denote when a
3343field was introduced or whether it's specific to some terse version):
d60e92d1 3344.P
40943b9a
TK
3345.nf
3346 terse version, fio version [v3], jobname, groupid, error
3347.fi
525c2bfa 3348.RS
40943b9a
TK
3349.P
3350.B
3351READ status:
525c2bfa 3352.RE
40943b9a
TK
3353.P
3354.nf
3355 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3356 Submission latency: min, max, mean, stdev (usec)
3357 Completion latency: min, max, mean, stdev (usec)
3358 Completion latency percentiles: 20 fields (see below)
3359 Total latency: min, max, mean, stdev (usec)
3360 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3361 IOPS [v5]: min, max, mean, stdev, number of samples
3362.fi
d60e92d1 3363.RS
40943b9a
TK
3364.P
3365.B
3366WRITE status:
a2c95580 3367.RE
40943b9a
TK
3368.P
3369.nf
3370 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
3371 Submission latency: min, max, mean, stdev (usec)
3372 Completion latency: min, max, mean, stdev (usec)
3373 Completion latency percentiles: 20 fields (see below)
3374 Total latency: min, max, mean, stdev (usec)
3375 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev, number of samples [v5]
3376 IOPS [v5]: min, max, mean, stdev, number of samples
3377.fi
a2c95580 3378.RS
40943b9a
TK
3379.P
3380.B
3381TRIM status [all but version 3]:
d60e92d1
AC
3382.RE
3383.P
40943b9a
TK
3384.nf
3385 Fields are similar to \fBREAD/WRITE\fR status.
3386.fi
a2c95580 3387.RS
a2c95580 3388.P
40943b9a 3389.B
d1429b5c 3390CPU usage:
d60e92d1
AC
3391.RE
3392.P
40943b9a
TK
3393.nf
3394 user, system, context switches, major faults, minor faults
3395.fi
d60e92d1 3396.RS
40943b9a
TK
3397.P
3398.B
3399I/O depths:
d60e92d1
AC
3400.RE
3401.P
40943b9a
TK
3402.nf
3403 <=1, 2, 4, 8, 16, 32, >=64
3404.fi
562c2d2f 3405.RS
40943b9a
TK
3406.P
3407.B
3408I/O latencies microseconds:
562c2d2f 3409.RE
40943b9a
TK
3410.P
3411.nf
3412 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
3413.fi
562c2d2f 3414.RS
40943b9a
TK
3415.P
3416.B
3417I/O latencies milliseconds:
562c2d2f
DN
3418.RE
3419.P
40943b9a
TK
3420.nf
3421 <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
3422.fi
f2f788dd 3423.RS
40943b9a
TK
3424.P
3425.B
3426Disk utilization [v3]:
f2f788dd
JA
3427.RE
3428.P
40943b9a
TK
3429.nf
3430 disk name, read ios, write ios, read merges, write merges, read ticks, write ticks, time spent in queue, disk utilization percentage
3431.fi
562c2d2f 3432.RS
d60e92d1 3433.P
40943b9a
TK
3434.B
3435Additional Info (dependent on continue_on_error, default off):
d60e92d1 3436.RE
2fc26c3d 3437.P
40943b9a
TK
3438.nf
3439 total # errors, first error code
3440.fi
2fc26c3d
IC
3441.RS
3442.P
40943b9a
TK
3443.B
3444Additional Info (dependent on description being set):
3445.RE
3446.P
2fc26c3d 3447.nf
40943b9a
TK
3448 Text description
3449.fi
3450.P
3451Completion latency percentiles can be a grouping of up to 20 sets, so for the
3452terse output fio writes all of them. Each field will look like this:
3453.P
3454.nf
3455 1.00%=6112
3456.fi
3457.P
3458which is the Xth percentile, and the `usec' latency associated with it.
3459.P
3460For \fBDisk utilization\fR, all disks used by fio are shown. So for each disk there
3461will be a disk utilization section.
3462.P
3463Below is a single line containing short names for each of the fields in the
3464minimal output v3, separated by semicolons:
3465.P
3466.nf
3467 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 3468.fi
4e757af1
VF
3469.P
3470In client/server mode terse output differs from what appears when jobs are run
3471locally. Disk utilization data is omitted from the standard terse output and
3472for v3 and later appears on its own separate line at the end of each terse
3473reporting cycle.
44c82dba
VF
3474.SH JSON OUTPUT
3475The \fBjson\fR output format is intended to be both human readable and convenient
3476for automated parsing. For the most part its sections mirror those of the
3477\fBnormal\fR output. The \fBruntime\fR value is reported in msec and the \fBbw\fR value is
3478reported in 1024 bytes per second units.
3479.fi
d9e557ab
VF
3480.SH JSON+ OUTPUT
3481The \fBjson+\fR output format is identical to the \fBjson\fR output format except that it
3482adds a full dump of the completion latency bins. Each \fBbins\fR object contains a
3483set of (key, value) pairs where keys are latency durations and values count how
3484many I/Os had completion latencies of the corresponding duration. For example,
3485consider:
d9e557ab 3486.RS
40943b9a 3487.P
d9e557ab
VF
3488"bins" : { "87552" : 1, "89600" : 1, "94720" : 1, "96768" : 1, "97792" : 1, "99840" : 1, "100864" : 2, "103936" : 6, "104960" : 534, "105984" : 5995, "107008" : 7529, ... }
3489.RE
40943b9a 3490.P
d9e557ab
VF
3491This data indicates that one I/O required 87,552ns to complete, two I/Os required
3492100,864ns to complete, and 7529 I/Os required 107,008ns to complete.
40943b9a 3493.P
d9e557ab 3494Also included with fio is a Python script \fBfio_jsonplus_clat2csv\fR that takes
40943b9a
TK
3495json+ output and generates CSV\-formatted latency data suitable for plotting.
3496.P
d9e557ab 3497The latency durations actually represent the midpoints of latency intervals.
40943b9a 3498For details refer to `stat.h' in the fio source.
29dbd1e5 3499.SH TRACE FILE FORMAT
40943b9a
TK
3500There are two trace file format that you can encounter. The older (v1) format is
3501unsupported since version 1.20\-rc3 (March 2008). It will still be described
29dbd1e5 3502below in case that you get an old trace and want to understand it.
29dbd1e5 3503.P
40943b9a
TK
3504In any case the trace is a simple text file with a single action per line.
3505.TP
29dbd1e5 3506.B Trace file format v1
40943b9a 3507Each line represents a single I/O action in the following format:
29dbd1e5 3508.RS
40943b9a
TK
3509.RS
3510.P
29dbd1e5 3511rw, offset, length
29dbd1e5
JA
3512.RE
3513.P
40943b9a
TK
3514where `rw=0/1' for read/write, and the `offset' and `length' entries being in bytes.
3515.P
3516This format is not supported in fio versions >= 1.20\-rc3.
3517.RE
3518.TP
29dbd1e5 3519.B Trace file format v2
40943b9a
TK
3520The second version of the trace file format was added in fio version 1.17. It
3521allows to access more then one file per trace and has a bigger set of possible
3522file actions.
29dbd1e5 3523.RS
40943b9a 3524.P
29dbd1e5 3525The first line of the trace file has to be:
40943b9a
TK
3526.RS
3527.P
3528"fio version 2 iolog"
3529.RE
3530.P
29dbd1e5 3531Following this can be lines in two different formats, which are described below.
40943b9a
TK
3532.P
3533.B
29dbd1e5 3534The file management format:
40943b9a
TK
3535.RS
3536filename action
29dbd1e5 3537.P
40943b9a 3538The `filename' is given as an absolute path. The `action' can be one of these:
29dbd1e5
JA
3539.RS
3540.TP
3541.B add
40943b9a 3542Add the given `filename' to the trace.
29dbd1e5
JA
3543.TP
3544.B open
40943b9a
TK
3545Open the file with the given `filename'. The `filename' has to have
3546been added with the \fBadd\fR action before.
29dbd1e5
JA
3547.TP
3548.B close
40943b9a
TK
3549Close the file with the given `filename'. The file has to have been
3550\fBopen\fRed before.
3551.RE
29dbd1e5 3552.RE
29dbd1e5 3553.P
40943b9a
TK
3554.B
3555The file I/O action format:
3556.RS
3557filename action offset length
29dbd1e5 3558.P
40943b9a
TK
3559The `filename' is given as an absolute path, and has to have been \fBadd\fRed and
3560\fBopen\fRed before it can be used with this format. The `offset' and `length' are
3561given in bytes. The `action' can be one of these:
29dbd1e5
JA
3562.RS
3563.TP
3564.B wait
40943b9a
TK
3565Wait for `offset' microseconds. Everything below 100 is discarded.
3566The time is relative to the previous `wait' statement.
29dbd1e5
JA
3567.TP
3568.B read
40943b9a 3569Read `length' bytes beginning from `offset'.
29dbd1e5
JA
3570.TP
3571.B write
40943b9a 3572Write `length' bytes beginning from `offset'.
29dbd1e5
JA
3573.TP
3574.B sync
40943b9a 3575\fBfsync\fR\|(2) the file.
29dbd1e5
JA
3576.TP
3577.B datasync
40943b9a 3578\fBfdatasync\fR\|(2) the file.
29dbd1e5
JA
3579.TP
3580.B trim
40943b9a
TK
3581Trim the given file from the given `offset' for `length' bytes.
3582.RE
29dbd1e5 3583.RE
b9921d1a
DZ
3584.SH I/O REPLAY \- MERGING TRACES
3585Colocation is a common practice used to get the most out of a machine.
3586Knowing which workloads play nicely with each other and which ones don't is
3587a much harder task. While fio can replay workloads concurrently via multiple
3588jobs, it leaves some variability up to the scheduler making results harder to
3589reproduce. Merging is a way to make the order of events consistent.
3590.P
3591Merging is integrated into I/O replay and done when a \fBmerge_blktrace_file\fR
3592is specified. The list of files passed to \fBread_iolog\fR go through the merge
3593process and output a single file stored to the specified file. The output file is
3594passed on as if it were the only file passed to \fBread_iolog\fR. An example would
3595look like:
3596.RS
3597.P
3598$ fio \-\-read_iolog="<file1>:<file2>" \-\-merge_blktrace_file="<output_file>"
3599.RE
3600.P
3601Creating only the merged file can be done by passing the command line argument
3602\fBmerge-blktrace-only\fR.
87a48ada
DZ
3603.P
3604Scaling traces can be done to see the relative impact of any particular trace
3605being slowed down or sped up. \fBmerge_blktrace_scalars\fR takes in a colon
3606separated list of percentage scalars. It is index paired with the files passed
3607to \fBread_iolog\fR.
55bfd8c8
DZ
3608.P
3609With scaling, it may be desirable to match the running time of all traces.
3610This can be done with \fBmerge_blktrace_iters\fR. It is index paired with
3611\fBread_iolog\fR just like \fBmerge_blktrace_scalars\fR.
3612.P
3613In an example, given two traces, A and B, each 60s long. If we want to see
3614the impact of trace A issuing IOs twice as fast and repeat trace A over the
3615runtime of trace B, the following can be done:
3616.RS
3617.P
3618$ fio \-\-read_iolog="<trace_a>:"<trace_b>" \-\-merge_blktrace_file"<output_file>" \-\-merge_blktrace_scalars="50:100" \-\-merge_blktrace_iters="2:1"
3619.RE
3620.P
3621This runs trace A at 2x the speed twice for approximately the same runtime as
3622a single run of trace B.
29dbd1e5 3623.SH CPU IDLENESS PROFILING
40943b9a
TK
3624In some cases, we want to understand CPU overhead in a test. For example, we
3625test patches for the specific goodness of whether they reduce CPU usage.
3626Fio implements a balloon approach to create a thread per CPU that runs at idle
3627priority, meaning that it only runs when nobody else needs the cpu.
3628By measuring the amount of work completed by the thread, idleness of each CPU
3629can be derived accordingly.
3630.P
3631An unit work is defined as touching a full page of unsigned characters. Mean and
3632standard deviation of time to complete an unit work is reported in "unit work"
3633section. Options can be chosen to report detailed percpu idleness or overall
3634system idleness by aggregating percpu stats.
29dbd1e5 3635.SH VERIFICATION AND TRIGGERS
40943b9a
TK
3636Fio is usually run in one of two ways, when data verification is done. The first
3637is a normal write job of some sort with verify enabled. When the write phase has
3638completed, fio switches to reads and verifies everything it wrote. The second
3639model is running just the write phase, and then later on running the same job
3640(but with reads instead of writes) to repeat the same I/O patterns and verify
3641the contents. Both of these methods depend on the write phase being completed,
3642as fio otherwise has no idea how much data was written.
3643.P
3644With verification triggers, fio supports dumping the current write state to
3645local files. Then a subsequent read verify workload can load this state and know
3646exactly where to stop. This is useful for testing cases where power is cut to a
3647server in a managed fashion, for instance.
3648.P
29dbd1e5 3649A verification trigger consists of two things:
29dbd1e5 3650.RS
40943b9a
TK
3651.P
36521) Storing the write state of each job.
3653.P
36542) Executing a trigger command.
29dbd1e5 3655.RE
40943b9a
TK
3656.P
3657The write state is relatively small, on the order of hundreds of bytes to single
3658kilobytes. It contains information on the number of completions done, the last X
3659completions, etc.
3660.P
3661A trigger is invoked either through creation ('touch') of a specified file in
3662the system, or through a timeout setting. If fio is run with
3663`\-\-trigger\-file=/tmp/trigger\-file', then it will continually
3664check for the existence of `/tmp/trigger\-file'. When it sees this file, it
3665will fire off the trigger (thus saving state, and executing the trigger
29dbd1e5 3666command).
40943b9a
TK
3667.