Add support for waiting for other jobs by name
[fio.git] / fio.1
CommitLineData
f8b8f7da 1.TH fio 1 "December 2014" "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
JA
15.BI \-\-debug \fR=\fPtype
16Enable verbose tracing of various fio actions. May be `all' for all types
17or individual types separated by a comma (eg \-\-debug=io,file). `help' will
18list all available tracing options.
19.TP
d60e92d1
AC
20.BI \-\-output \fR=\fPfilename
21Write output to \fIfilename\fR.
22.TP
e28ee21d 23.BI \-\-output-format \fR=\fPformat
513e37ee
VF
24Set the reporting format to \fInormal\fR, \fIterse\fR, \fIjson\fR, or
25\fIjson+\fR. Multiple formats can be selected, separate by a comma. \fIterse\fR
26is a CSV based format. \fIjson+\fR is like \fIjson\fR, except it adds a full
27dump of the latency buckets.
e28ee21d 28.TP
b2cecdc2 29.BI \-\-runtime \fR=\fPruntime
30Limit run time to \fIruntime\fR seconds.
d60e92d1 31.TP
d60e92d1
AC
32.B \-\-bandwidth\-log
33Generate per-job bandwidth logs.
34.TP
35.B \-\-minimal
d1429b5c 36Print statistics in a terse, semicolon-delimited format.
d60e92d1 37.TP
f6a7df53
JA
38.B \-\-append-terse
39Print statistics in selected mode AND terse, semicolon-delimited format.
40Deprecated, use \-\-output-format instead to select multiple formats.
41.TP
49da1240
JA
42.B \-\-version
43Display version information and exit.
44.TP
065248bf 45.BI \-\-terse\-version \fR=\fPversion
4d658652 46Set terse version output format (Current version 3, or older version 2).
49da1240
JA
47.TP
48.B \-\-help
49Display usage information and exit.
50.TP
fec0f21c
JA
51.B \-\-cpuclock-test
52Perform test and validation of internal CPU clock
53.TP
54.BI \-\-crctest[\fR=\fPtest]
55Test the speed of the builtin checksumming functions. If no argument is given,
56all of them are tested. Or a comma separated list can be passed, in which
57case the given ones are tested.
58.TP
49da1240
JA
59.BI \-\-cmdhelp \fR=\fPcommand
60Print help information for \fIcommand\fR. May be `all' for all commands.
61.TP
de890a1e
SL
62.BI \-\-enghelp \fR=\fPioengine[,command]
63List all commands defined by \fIioengine\fR, or print help for \fIcommand\fR defined by \fIioengine\fR.
64.TP
d60e92d1
AC
65.BI \-\-showcmd \fR=\fPjobfile
66Convert \fIjobfile\fR to a set of command-line options.
67.TP
d60e92d1
AC
68.BI \-\-eta \fR=\fPwhen
69Specifies when real-time ETA estimate should be printed. \fIwhen\fR may
70be one of `always', `never' or `auto'.
71.TP
30b5d57f
JA
72.BI \-\-eta\-newline \fR=\fPtime
73Force an ETA newline for every `time` period passed.
74.TP
75.BI \-\-status\-interval \fR=\fPtime
76Report full output status every `time` period passed.
77.TP
49da1240
JA
78.BI \-\-readonly
79Turn on safety read-only checks, preventing any attempted write.
80.TP
c0a5d35e 81.BI \-\-section \fR=\fPsec
cf145d90 82Only run section \fIsec\fR from job file. This option can be used multiple times to add more sections to run.
c0a5d35e 83.TP
49da1240
JA
84.BI \-\-alloc\-size \fR=\fPkb
85Set the internal smalloc pool size to \fIkb\fP kilobytes.
d60e92d1 86.TP
49da1240
JA
87.BI \-\-warnings\-fatal
88All fio parser warnings are fatal, causing fio to exit with an error.
9183788d 89.TP
49da1240 90.BI \-\-max\-jobs \fR=\fPnr
57e118a2 91Set the maximum allowed number of jobs (threads/processes) to support.
d60e92d1 92.TP
49da1240
JA
93.BI \-\-server \fR=\fPargs
94Start a backend server, with \fIargs\fP specifying what to listen to. See client/server section.
f57a9c59 95.TP
49da1240
JA
96.BI \-\-daemonize \fR=\fPpidfile
97Background a fio server, writing the pid to the given pid file.
98.TP
99.BI \-\-client \fR=\fPhost
39b5f61e 100Instead of running the jobs locally, send and run them on the given host or set of hosts. See client/server section.
f2a2ce0e
HL
101.TP
102.BI \-\-idle\-prof \fR=\fPoption
103Report cpu idleness on a system or percpu basis (\fIoption\fP=system,percpu) or run unit work calibration only (\fIoption\fP=calibrate).
d60e92d1
AC
104.SH "JOB FILE FORMAT"
105Job files are in `ini' format. They consist of one or more
106job definitions, which begin with a job name in square brackets and
107extend to the next job name. The job name can be any ASCII string
108except `global', which has a special meaning. Following the job name is
109a sequence of zero or more parameters, one per line, that define the
110behavior of the job. Any line starting with a `;' or `#' character is
d1429b5c 111considered a comment and ignored.
d9956b64
AC
112.P
113If \fIjobfile\fR is specified as `-', the job file will be read from
114standard input.
d60e92d1
AC
115.SS "Global Section"
116The global section contains default parameters for jobs specified in the
117job file. A job is only affected by global sections residing above it,
118and there may be any number of global sections. Specific job definitions
119may override any parameter set in global sections.
120.SH "JOB PARAMETERS"
121.SS Types
b470a02c
SC
122Some parameters may take arguments of a specific type.
123Anywhere a numeric value is required, an arithmetic expression may be used,
d59aa780
JA
124provided it is surrounded by parentheses. Supported operators are:
125.RS
126.RS
127.TP
128.B addition (+)
129.TP
130.B subtraction (-)
131.TP
132.B multiplication (*)
133.TP
134.B division (/)
135.TP
136.B modulus (%)
137.TP
138.B exponentiation (^)
139.RE
140.RE
141.P
142For time values in expressions, units are microseconds by default. This is
143different than for time values not in expressions (not enclosed in
144parentheses). The types used are:
d60e92d1
AC
145.TP
146.I str
147String: a sequence of alphanumeric characters.
148.TP
149.I int
d60e92d1 150SI integer: a whole number, possibly containing a suffix denoting the base unit
b09da8fa
JA
151of the value. Accepted suffixes are `k', 'M', 'G', 'T', and 'P', denoting
152kilo (1024), mega (1024^2), giga (1024^3), tera (1024^4), and peta (1024^5)
74454ce4
CE
153respectively. If prefixed with '0x', the value is assumed to be base 16
154(hexadecimal). A suffix may include a trailing 'b', for instance 'kb' is
155identical to 'k'. You can specify a base 10 value by using 'KiB', 'MiB','GiB',
156etc. This is useful for disk drives where values are often given in base 10
157values. Specifying '30GiB' will get you 30*1000^3 bytes.
158When specifying times the default suffix meaning changes, still denoting the
159base unit of the value, but accepted suffixes are 'D' (days), 'H' (hours), 'M'
0de5b26f
JA
160(minutes), 'S' Seconds, 'ms' (or msec) milli seconds, 'us' (or 'usec') micro
161seconds. Time values without a unit specify seconds.
74454ce4 162The suffixes are not case sensitive.
d60e92d1
AC
163.TP
164.I bool
165Boolean: a true or false value. `0' denotes false, `1' denotes true.
166.TP
167.I irange
168Integer range: a range of integers specified in the format
d1429b5c
AC
169\fIlower\fR:\fIupper\fR or \fIlower\fR\-\fIupper\fR. \fIlower\fR and
170\fIupper\fR may contain a suffix as described above. If an option allows two
171sets of ranges, they are separated with a `,' or `/' character. For example:
172`8\-8k/8M\-4G'.
83349190
YH
173.TP
174.I float_list
175List of floating numbers: A list of floating numbers, separated by
cecbfd47 176a ':' character.
d60e92d1
AC
177.SS "Parameter List"
178.TP
179.BI name \fR=\fPstr
d9956b64 180May be used to override the job name. On the command line, this parameter
d60e92d1
AC
181has the special purpose of signalling the start of a new job.
182.TP
9cc8cb91
AK
183.BI wait_for \fR=\fPstr
184Specifies the name of the already defined job to wait for. Single waitee name
185only may be specified. If set, the job won't be started until all workers of
186the waitee job are done. Wait_for operates on the job name basis, so there are
187a few limitations. First, the waitee must be defined prior to the waiter job
188(meaning no forward references). Second, if a job is being referenced as a
189waitee, it must have a unique name (no duplicate waitees).
190.TP
d60e92d1
AC
191.BI description \fR=\fPstr
192Human-readable description of the job. It is printed when the job is run, but
193otherwise has no special purpose.
194.TP
195.BI directory \fR=\fPstr
196Prefix filenames with this directory. Used to place files in a location other
197than `./'.
bcbfeefa
CE
198You can specify a number of directories by separating the names with a ':'
199character. These directories will be assigned equally distributed to job clones
200creates with \fInumjobs\fR as long as they are using generated filenames.
201If specific \fIfilename(s)\fR are set fio will use the first listed directory,
202and thereby matching the \fIfilename\fR semantic which generates a file each
67445b63
JA
203clone if not specified, but let all clones use the same if set. See
204\fIfilename\fR for considerations regarding escaping certain characters on
205some platforms.
d60e92d1
AC
206.TP
207.BI filename \fR=\fPstr
208.B fio
209normally makes up a file name based on the job name, thread number, and file
d1429b5c 210number. If you want to share files between threads in a job or several jobs,
de890a1e
SL
211specify a \fIfilename\fR for each of them to override the default.
212If the I/O engine is file-based, you can specify
d1429b5c
AC
213a number of files by separating the names with a `:' character. `\-' is a
214reserved name, meaning stdin or stdout, depending on the read/write direction
67445b63
JA
215set. On Windows, disk devices are accessed as \\.\PhysicalDrive0 for the first
216device, \\.\PhysicalDrive1 for the second etc. Note: Windows and FreeBSD
217prevent write access to areas of the disk containing in-use data
218(e.g. filesystems). If the wanted filename does need to include a colon, then
4904acd5
JM
219escape that with a '\\' character. For instance, if the filename is
220"/dev/dsk/foo@3,0:c", then you would use filename="/dev/dsk/foo@3,0\\:c".
d60e92d1 221.TP
de98bd30 222.BI filename_format \fR=\fPstr
ce594fbe 223If sharing multiple files between jobs, it is usually necessary to have
de98bd30
JA
224fio generate the exact names that you want. By default, fio will name a file
225based on the default file format specification of
226\fBjobname.jobnumber.filenumber\fP. With this option, that can be
227customized. Fio will recognize and replace the following keywords in this
228string:
229.RS
230.RS
231.TP
232.B $jobname
233The name of the worker thread or process.
234.TP
235.B $jobnum
236The incremental number of the worker thread or process.
237.TP
238.B $filenum
239The incremental number of the file for that worker thread or process.
240.RE
241.P
242To have dependent jobs share a set of files, this option can be set to
243have fio generate filenames that are shared between the two. For instance,
244if \fBtestfiles.$filenum\fR is specified, file number 4 for any job will
245be named \fBtestfiles.4\fR. The default of \fB$jobname.$jobnum.$filenum\fR
246will be used if no other format specifier is given.
247.RE
248.P
249.TP
3ce9dcaf
JA
250.BI lockfile \fR=\fPstr
251Fio defaults to not locking any files before it does IO to them. If a file or
252file descriptor is shared, fio can serialize IO to that file to make the end
253result consistent. This is usual for emulating real workloads that share files.
254The lock modes are:
255.RS
256.RS
257.TP
258.B none
259No locking. This is the default.
260.TP
261.B exclusive
cf145d90 262Only one thread or process may do IO at a time, excluding all others.
3ce9dcaf
JA
263.TP
264.B readwrite
265Read-write locking on the file. Many readers may access the file at the same
266time, but writes get exclusive access.
267.RE
ce594fbe 268.RE
3ce9dcaf 269.P
d60e92d1
AC
270.BI opendir \fR=\fPstr
271Recursively open any files below directory \fIstr\fR.
272.TP
273.BI readwrite \fR=\fPstr "\fR,\fP rw" \fR=\fPstr
274Type of I/O pattern. Accepted values are:
275.RS
276.RS
277.TP
278.B read
d1429b5c 279Sequential reads.
d60e92d1
AC
280.TP
281.B write
d1429b5c 282Sequential writes.
d60e92d1 283.TP
fa769d44
SW
284.B trim
285Sequential trim (Linux block devices only).
286.TP
d60e92d1 287.B randread
d1429b5c 288Random reads.
d60e92d1
AC
289.TP
290.B randwrite
d1429b5c 291Random writes.
d60e92d1 292.TP
fa769d44
SW
293.B randtrim
294Random trim (Linux block devices only).
295.TP
10b023db 296.B rw, readwrite
d1429b5c 297Mixed sequential reads and writes.
d60e92d1 298.TP
ff6bb260 299.B randrw
d1429b5c 300Mixed random reads and writes.
82a90686
JA
301.TP
302.B trimwrite
303Trim and write mixed workload. Blocks will be trimmed first, then the same
304blocks will be written to.
d60e92d1
AC
305.RE
306.P
38dad62d
JA
307For mixed I/O, the default split is 50/50. For certain types of io the result
308may still be skewed a bit, since the speed may be different. It is possible to
3b7fa9ec 309specify a number of IO's to do before getting a new offset, this is done by
38dad62d
JA
310appending a `:\fI<nr>\fR to the end of the string given. For a random read, it
311would look like \fBrw=randread:8\fR for passing in an offset modifier with a
059b0802
JA
312value of 8. If the postfix is used with a sequential IO pattern, then the value
313specified will be added to the generated offset for each IO. For instance,
314using \fBrw=write:4k\fR will skip 4k for every write. It turns sequential IO
315into sequential IO with holes. See the \fBrw_sequencer\fR option.
d60e92d1
AC
316.RE
317.TP
38dad62d
JA
318.BI rw_sequencer \fR=\fPstr
319If an offset modifier is given by appending a number to the \fBrw=<str>\fR line,
320then this option controls how that number modifies the IO offset being
321generated. Accepted values are:
322.RS
323.RS
324.TP
325.B sequential
326Generate sequential offset
327.TP
328.B identical
329Generate the same offset
330.RE
331.P
332\fBsequential\fR is only useful for random IO, where fio would normally
333generate a new random offset for every IO. If you append eg 8 to randread, you
334would get a new random offset for every 8 IO's. The result would be a seek for
335only every 8 IO's, instead of for every IO. Use \fBrw=randread:8\fR to specify
336that. As sequential IO is already sequential, setting \fBsequential\fR for that
337would not result in any differences. \fBidentical\fR behaves in a similar
338fashion, except it sends the same offset 8 number of times before generating a
339new offset.
340.RE
341.P
342.TP
90fef2d1
JA
343.BI kb_base \fR=\fPint
344The base unit for a kilobyte. The defacto base is 2^10, 1024. Storage
345manufacturers like to use 10^3 or 1000 as a base ten unit instead, for obvious
5c9323fb 346reasons. Allowed values are 1024 or 1000, with 1024 being the default.
90fef2d1 347.TP
771e58be
JA
348.BI unified_rw_reporting \fR=\fPbool
349Fio normally reports statistics on a per data direction basis, meaning that
350read, write, and trim are accounted and reported separately. If this option is
cf145d90 351set fio sums the results and reports them as "mixed" instead.
771e58be 352.TP
d60e92d1 353.BI randrepeat \fR=\fPbool
56e2a5fc
CE
354Seed the random number generator used for random I/O patterns in a predictable
355way so the pattern is repeatable across runs. Default: true.
356.TP
357.BI allrandrepeat \fR=\fPbool
358Seed all random number generators in a predictable way so results are
359repeatable across runs. Default: false.
d60e92d1 360.TP
04778baf
JA
361.BI randseed \fR=\fPint
362Seed the random number generators based on this seed value, to be able to
363control what sequence of output is being generated. If not set, the random
364sequence depends on the \fBrandrepeat\fR setting.
365.TP
a596f047
EG
366.BI fallocate \fR=\fPstr
367Whether pre-allocation is performed when laying down files. Accepted values
368are:
369.RS
370.RS
371.TP
372.B none
373Do not pre-allocate space.
374.TP
375.B posix
ccc2b328 376Pre-allocate via \fBposix_fallocate\fR\|(3).
a596f047
EG
377.TP
378.B keep
ccc2b328 379Pre-allocate via \fBfallocate\fR\|(2) with FALLOC_FL_KEEP_SIZE set.
a596f047
EG
380.TP
381.B 0
382Backward-compatible alias for 'none'.
383.TP
384.B 1
385Backward-compatible alias for 'posix'.
386.RE
387.P
388May not be available on all supported platforms. 'keep' is only
389available on Linux. If using ZFS on Solaris this must be set to 'none'
390because ZFS doesn't support it. Default: 'posix'.
391.RE
7bc8c2cf 392.TP
d60e92d1 393.BI fadvise_hint \fR=\fPbool
cf145d90 394Use \fBposix_fadvise\fR\|(2) to advise the kernel what I/O patterns
d1429b5c 395are likely to be issued. Default: true.
d60e92d1 396.TP
37659335
JA
397.BI fadvise_stream \fR=\fPint
398Use \fBposix_fadvise\fR\|(2) to advise the kernel what stream ID the
399writes issued belong to. Only supported on Linux. Note, this option
400may change going forward.
401.TP
f7fa2653 402.BI size \fR=\fPint
d60e92d1 403Total size of I/O for this job. \fBfio\fR will run until this many bytes have
a4d3b4db
JA
404been transferred, unless limited by other options (\fBruntime\fR, for instance,
405or increased/descreased by \fBio_size\fR). Unless \fBnrfiles\fR and
406\fBfilesize\fR options are given, this amount will be divided between the
407available files for the job. If not set, fio will use the full size of the
408given files or devices. If the files do not exist, size must be given. It is
409also possible to give size as a percentage between 1 and 100. If size=20% is
410given, fio will use 20% of the full size of the given files or devices.
411.TP
412.BI io_size \fR=\fPint "\fR,\fB io_limit \fR=\fPint
77731b29
JA
413Normally fio operates within the region set by \fBsize\fR, which means that
414the \fBsize\fR option sets both the region and size of IO to be performed.
415Sometimes that is not what you want. With this option, it is possible to
416define just the amount of IO that fio should do. For instance, if \fBsize\fR
417is set to 20G and \fBio_limit\fR is set to 5G, fio will perform IO within
a4d3b4db
JA
418the first 20G but exit when 5G have been done. The opposite is also
419possible - if \fBsize\fR is set to 20G, and \fBio_size\fR is set to 40G, then
420fio will do 40G of IO within the 0..20G region.
d60e92d1 421.TP
74586c1e 422.BI fill_device \fR=\fPbool "\fR,\fB fill_fs" \fR=\fPbool
3ce9dcaf
JA
423Sets size to something really large and waits for ENOSPC (no space left on
424device) as the terminating condition. Only makes sense with sequential write.
425For a read workload, the mount point will be filled first then IO started on
4f12432e
JA
426the result. This option doesn't make sense if operating on a raw device node,
427since the size of that is already known by the file system. Additionally,
428writing beyond end-of-device will not return ENOSPC there.
3ce9dcaf 429.TP
d60e92d1
AC
430.BI filesize \fR=\fPirange
431Individual file sizes. May be a range, in which case \fBfio\fR will select sizes
d1429b5c
AC
432for files at random within the given range, limited to \fBsize\fR in total (if
433that is given). If \fBfilesize\fR is not specified, each created file is the
434same size.
d60e92d1 435.TP
bedc9dc2
JA
436.BI file_append \fR=\fPbool
437Perform IO after the end of the file. Normally fio will operate within the
438size of a file. If this option is set, then fio will append to the file
439instead. This has identical behavior to setting \fRoffset\fP to the size
0aae4ce7 440of a file. This option is ignored on non-regular files.
bedc9dc2 441.TP
f7fa2653 442.BI blocksize \fR=\fPint[,int] "\fR,\fB bs" \fR=\fPint[,int]
d9472271
JA
443Block size for I/O units. Default: 4k. Values for reads, writes, and trims
444can be specified separately in the format \fIread\fR,\fIwrite\fR,\fItrim\fR
445either of which may be empty to leave that value at its default. If a trailing
446comma isn't given, the remainder will inherit the last value set.
d60e92d1 447.TP
9183788d 448.BI blocksize_range \fR=\fPirange[,irange] "\fR,\fB bsrange" \fR=\fPirange[,irange]
d1429b5c
AC
449Specify a range of I/O block sizes. The issued I/O unit will always be a
450multiple of the minimum size, unless \fBblocksize_unaligned\fR is set. Applies
9183788d 451to both reads and writes if only one range is given, but can be specified
de8f6de9 452separately with a comma separating the values. Example: bsrange=1k-4k,2k-8k.
9183788d
JA
453Also (see \fBblocksize\fR).
454.TP
455.BI bssplit \fR=\fPstr
456This option allows even finer grained control of the block sizes issued,
457not just even splits between them. With this option, you can weight various
458block sizes for exact control of the issued IO for a job that has mixed
459block sizes. The format of the option is bssplit=blocksize/percentage,
5982a925 460optionally adding as many definitions as needed separated by a colon.
9183788d 461Example: bssplit=4k/10:64k/50:32k/40 would issue 50% 64k blocks, 10% 4k
c83cdd3e
JA
462blocks and 40% 32k blocks. \fBbssplit\fR also supports giving separate
463splits to reads and writes. The format is identical to what the
464\fBbs\fR option accepts, the read and write parts are separated with a
465comma.
d60e92d1
AC
466.TP
467.B blocksize_unaligned\fR,\fP bs_unaligned
d1429b5c
AC
468If set, any size in \fBblocksize_range\fR may be used. This typically won't
469work with direct I/O, as that normally requires sector alignment.
d60e92d1 470.TP
2b7a01d0 471.BI blockalign \fR=\fPint[,int] "\fR,\fB ba" \fR=\fPint[,int]
639ce0f3
MS
472At what boundary to align random IO offsets. Defaults to the same as 'blocksize'
473the minimum blocksize given. Minimum alignment is typically 512b
2b7a01d0
JA
474for using direct IO, though it usually depends on the hardware block size.
475This option is mutually exclusive with using a random map for files, so it
476will turn off that option.
43602667 477.TP
6aca9b3d
JA
478.BI bs_is_seq_rand \fR=\fPbool
479If this option is set, fio will use the normal read,write blocksize settings as
480sequential,random instead. Any random read or write will use the WRITE
481blocksize settings, and any sequential read or write will use the READ
482blocksize setting.
483.TP
d60e92d1 484.B zero_buffers
cf145d90 485Initialize buffers with all zeros. Default: fill buffers with random data.
d60e92d1 486.TP
901bb994
JA
487.B refill_buffers
488If this option is given, fio will refill the IO buffers on every submit. The
489default is to only fill it at init time and reuse that data. Only makes sense
490if zero_buffers isn't specified, naturally. If data verification is enabled,
491refill_buffers is also automatically enabled.
492.TP
fd68418e
JA
493.BI scramble_buffers \fR=\fPbool
494If \fBrefill_buffers\fR is too costly and the target is using data
495deduplication, then setting this option will slightly modify the IO buffer
496contents to defeat normal de-dupe attempts. This is not enough to defeat
497more clever block compression attempts, but it will stop naive dedupe
498of blocks. Default: true.
499.TP
c5751c62
JA
500.BI buffer_compress_percentage \fR=\fPint
501If this is set, then fio will attempt to provide IO buffer content (on WRITEs)
502that compress to the specified level. Fio does this by providing a mix of
d1af2894
JA
503random data and a fixed pattern. The fixed pattern is either zeroes, or the
504pattern specified by \fBbuffer_pattern\fR. If the pattern option is used, it
505might skew the compression ratio slightly. Note that this is per block size
506unit, for file/disk wide compression level that matches this setting. Note
507that this is per block size unit, for file/disk wide compression level that
508matches this setting, you'll also want to set refill_buffers.
c5751c62
JA
509.TP
510.BI buffer_compress_chunk \fR=\fPint
511See \fBbuffer_compress_percentage\fR. This setting allows fio to manage how
512big the ranges of random data and zeroed data is. Without this set, fio will
513provide \fBbuffer_compress_percentage\fR of blocksize random data, followed by
514the remaining zeroed. With this set to some chunk size smaller than the block
515size, fio can alternate random and zeroed data throughout the IO buffer.
516.TP
ce35b1ec 517.BI buffer_pattern \fR=\fPstr
cf145d90
CVB
518If set, fio will fill the IO buffers with this pattern. If not set, the contents
519of IO buffers is defined by the other options related to buffer contents. The
ce35b1ec 520setting can be any pattern of bytes, and can be prefixed with 0x for hex
02975b64 521values. It may also be a string, where the string must then be wrapped with
2fa5a241
RP
522"", e.g.:
523.RS
524.RS
525\fBbuffer_pattern\fR="abcd"
526.RS
527or
528.RE
529\fBbuffer_pattern\fR=-12
530.RS
531or
532.RE
533\fBbuffer_pattern\fR=0xdeadface
534.RE
535.LP
536Also you can combine everything together in any order:
537.LP
538.RS
539\fBbuffer_pattern\fR=0xdeadface"abcd"-12
540.RE
541.RE
ce35b1ec 542.TP
5c94b008
JA
543.BI dedupe_percentage \fR=\fPint
544If set, fio will generate this percentage of identical buffers when writing.
545These buffers will be naturally dedupable. The contents of the buffers depend
546on what other buffer compression settings have been set. It's possible to have
547the individual buffers either fully compressible, or not at all. This option
548only controls the distribution of unique buffers.
549.TP
d60e92d1
AC
550.BI nrfiles \fR=\fPint
551Number of files to use for this job. Default: 1.
552.TP
553.BI openfiles \fR=\fPint
554Number of files to keep open at the same time. Default: \fBnrfiles\fR.
555.TP
556.BI file_service_type \fR=\fPstr
557Defines how files to service are selected. The following types are defined:
558.RS
559.RS
560.TP
561.B random
5c9323fb 562Choose a file at random.
d60e92d1
AC
563.TP
564.B roundrobin
cf145d90 565Round robin over opened files (default).
5c9323fb 566.TP
6b7f6851
JA
567.B sequential
568Do each file in the set sequentially.
d60e92d1
AC
569.RE
570.P
cf145d90 571The number of I/Os to issue before switching to a new file can be specified by
d60e92d1
AC
572appending `:\fIint\fR' to the service type.
573.RE
574.TP
575.BI ioengine \fR=\fPstr
576Defines how the job issues I/O. The following types are defined:
577.RS
578.RS
579.TP
580.B sync
ccc2b328 581Basic \fBread\fR\|(2) or \fBwrite\fR\|(2) I/O. \fBfseek\fR\|(2) is used to
d60e92d1
AC
582position the I/O location.
583.TP
a31041ea 584.B psync
ccc2b328 585Basic \fBpread\fR\|(2) or \fBpwrite\fR\|(2) I/O.
a31041ea 586.TP
9183788d 587.B vsync
ccc2b328 588Basic \fBreadv\fR\|(2) or \fBwritev\fR\|(2) I/O. Will emulate queuing by
cecbfd47 589coalescing adjacent IOs into a single submission.
9183788d 590.TP
a46c5e01 591.B pvsync
ccc2b328 592Basic \fBpreadv\fR\|(2) or \fBpwritev\fR\|(2) I/O.
a46c5e01 593.TP
d60e92d1 594.B libaio
de890a1e 595Linux native asynchronous I/O. This ioengine defines engine specific options.
d60e92d1
AC
596.TP
597.B posixaio
ccc2b328 598POSIX asynchronous I/O using \fBaio_read\fR\|(3) and \fBaio_write\fR\|(3).
03e20d68
BC
599.TP
600.B solarisaio
601Solaris native asynchronous I/O.
602.TP
603.B windowsaio
604Windows native asynchronous I/O.
d60e92d1
AC
605.TP
606.B mmap
ccc2b328
SW
607File is memory mapped with \fBmmap\fR\|(2) and data copied using
608\fBmemcpy\fR\|(3).
d60e92d1
AC
609.TP
610.B splice
ccc2b328 611\fBsplice\fR\|(2) is used to transfer the data and \fBvmsplice\fR\|(2) to
d1429b5c 612transfer data from user-space to the kernel.
d60e92d1
AC
613.TP
614.B syslet-rw
615Use the syslet system calls to make regular read/write asynchronous.
616.TP
617.B sg
618SCSI generic sg v3 I/O. May be either synchronous using the SG_IO ioctl, or if
ccc2b328
SW
619the target is an sg character device, we use \fBread\fR\|(2) and
620\fBwrite\fR\|(2) for asynchronous I/O.
d60e92d1
AC
621.TP
622.B null
623Doesn't transfer any data, just pretends to. Mainly used to exercise \fBfio\fR
624itself and for debugging and testing purposes.
625.TP
626.B net
de890a1e
SL
627Transfer over the network. The protocol to be used can be defined with the
628\fBprotocol\fR parameter. Depending on the protocol, \fBfilename\fR,
629\fBhostname\fR, \fBport\fR, or \fBlisten\fR must be specified.
630This ioengine defines engine specific options.
d60e92d1
AC
631.TP
632.B netsplice
ccc2b328 633Like \fBnet\fR, but uses \fBsplice\fR\|(2) and \fBvmsplice\fR\|(2) to map data
de890a1e 634and send/receive. This ioengine defines engine specific options.
d60e92d1 635.TP
53aec0a4 636.B cpuio
d60e92d1
AC
637Doesn't transfer any data, but burns CPU cycles according to \fBcpuload\fR and
638\fBcpucycles\fR parameters.
639.TP
640.B guasi
641The GUASI I/O engine is the Generic Userspace Asynchronous Syscall Interface
cecbfd47 642approach to asynchronous I/O.
d1429b5c
AC
643.br
644See <http://www.xmailserver.org/guasi\-lib.html>.
d60e92d1 645.TP
21b8aee8 646.B rdma
85286c5c
BVA
647The RDMA I/O engine supports both RDMA memory semantics (RDMA_WRITE/RDMA_READ)
648and channel semantics (Send/Recv) for the InfiniBand, RoCE and iWARP protocols.
21b8aee8 649.TP
d60e92d1
AC
650.B external
651Loads an external I/O engine object file. Append the engine filename as
652`:\fIenginepath\fR'.
d54fce84
DM
653.TP
654.B falloc
cecbfd47 655 IO engine that does regular linux native fallocate call to simulate data
d54fce84
DM
656transfer as fio ioengine
657.br
658 DDIR_READ does fallocate(,mode = FALLOC_FL_KEEP_SIZE,)
659.br
0981fd71 660 DIR_WRITE does fallocate(,mode = 0)
d54fce84
DM
661.br
662 DDIR_TRIM does fallocate(,mode = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE)
663.TP
664.B e4defrag
665IO engine that does regular EXT4_IOC_MOVE_EXT ioctls to simulate defragment activity
666request to DDIR_WRITE event
0d978694
DAG
667.TP
668.B rbd
ff6bb260
SL
669IO engine supporting direct access to Ceph Rados Block Devices (RBD) via librbd
670without the need to use the kernel rbd driver. This ioengine defines engine specific
0d978694 671options.
a7c386f4 672.TP
673.B gfapi
cc47f094 674Using Glusterfs libgfapi sync interface to direct access to Glusterfs volumes without
675having to go through FUSE. This ioengine defines engine specific
676options.
677.TP
678.B gfapi_async
679Using Glusterfs libgfapi async interface to direct access to Glusterfs volumes without
a7c386f4 680having to go through FUSE. This ioengine defines engine specific
681options.
1b10477b 682.TP
b74e419e
MM
683.B libhdfs
684Read and write through Hadoop (HDFS). The \fBfilename\fR option is used to
685specify host,port of the hdfs name-node to connect. This engine interprets
686offsets a little differently. In HDFS, files once created cannot be modified.
687So random writes are not possible. To imitate this, libhdfs engine expects
688bunch of small files to be created over HDFS, and engine will randomly pick a
689file out of those files based on the offset generated by fio backend. (see the
690example job file to create such files, use rw=write option). Please note, you
691might want to set necessary environment variables to work with hdfs/libhdfs
692properly.
65fa28ca
DE
693.TP
694.B mtd
695Read, write and erase an MTD character device (e.g., /dev/mtd0). Discards are
696treated as erases. Depending on the underlying device type, the I/O may have
697to go in a certain pattern, e.g., on NAND, writing sequentially to erase blocks
698and discarding before overwriting. The writetrim mode works well for this
699constraint.
d60e92d1 700.RE
595e1734 701.P
d60e92d1
AC
702.RE
703.TP
704.BI iodepth \fR=\fPint
8489dae4
SK
705Number of I/O units to keep in flight against the file. Note that increasing
706iodepth beyond 1 will not affect synchronous ioengines (except for small
cf145d90 707degress when verify_async is in use). Even async engines may impose OS
ee72ca09
JA
708restrictions causing the desired depth not to be achieved. This may happen on
709Linux when using libaio and not setting \fBdirect\fR=1, since buffered IO is
710not async on that OS. Keep an eye on the IO depth distribution in the
711fio output to verify that the achieved depth is as expected. Default: 1.
d60e92d1 712.TP
e63a0b2f
RP
713.BI iodepth_batch \fR=\fPint "\fR,\fP iodepth_batch_submit" \fR=\fPint
714This defines how many pieces of IO to submit at once. It defaults to 1
715which means that we submit each IO as soon as it is available, but can
716be raised to submit bigger batches of IO at the time. If it is set to 0
717the \fBiodepth\fR value will be used.
d60e92d1 718.TP
82407585 719.BI iodepth_batch_complete_min \fR=\fPint "\fR,\fP iodepth_batch_complete" \fR=\fPint
3ce9dcaf
JA
720This defines how many pieces of IO to retrieve at once. It defaults to 1 which
721 means that we'll ask for a minimum of 1 IO in the retrieval process from the
722kernel. The IO retrieval will go on until we hit the limit set by
723\fBiodepth_low\fR. If this variable is set to 0, then fio will always check for
724completed events before queuing more IO. This helps reduce IO latency, at the
725cost of more retrieval system calls.
726.TP
82407585
RP
727.BI iodepth_batch_complete_max \fR=\fPint
728This defines maximum pieces of IO to
729retrieve at once. This variable should be used along with
730\fBiodepth_batch_complete_min\fR=int variable, specifying the range
731of min and max amount of IO which should be retrieved. By default
732it is equal to \fBiodepth_batch_complete_min\fR value.
733
734Example #1:
735.RS
736.RS
737\fBiodepth_batch_complete_min\fR=1
738.LP
739\fBiodepth_batch_complete_max\fR=<iodepth>
740.RE
741
742which means that we will retrieve at leat 1 IO and up to the
743whole submitted queue depth. If none of IO has been completed
744yet, we will wait.
745
746Example #2:
747.RS
748\fBiodepth_batch_complete_min\fR=0
749.LP
750\fBiodepth_batch_complete_max\fR=<iodepth>
751.RE
752
753which means that we can retrieve up to the whole submitted
754queue depth, but if none of IO has been completed yet, we will
755NOT wait and immediately exit the system call. In this example
756we simply do polling.
757.RE
758.TP
d60e92d1
AC
759.BI iodepth_low \fR=\fPint
760Low watermark indicating when to start filling the queue again. Default:
ff6bb260 761\fBiodepth\fR.
d60e92d1 762.TP
1ad01bd1
JA
763.BI io_submit_mode \fR=\fPstr
764This option controls how fio submits the IO to the IO engine. The default is
765\fBinline\fR, which means that the fio job threads submit and reap IO directly.
766If set to \fBoffload\fR, the job threads will offload IO submission to a
767dedicated pool of IO threads. This requires some coordination and thus has a
768bit of extra overhead, especially for lower queue depth IO where it can
769increase latencies. The benefit is that fio can manage submission rates
770independently of the device completion rates. This avoids skewed latency
771reporting if IO gets back up on the device side (the coordinated omission
772problem).
773.TP
d60e92d1
AC
774.BI direct \fR=\fPbool
775If true, use non-buffered I/O (usually O_DIRECT). Default: false.
776.TP
d01612f3
CM
777.BI atomic \fR=\fPbool
778If value is true, attempt to use atomic direct IO. Atomic writes are guaranteed
779to be stable once acknowledged by the operating system. Only Linux supports
780O_ATOMIC right now.
781.TP
d60e92d1
AC
782.BI buffered \fR=\fPbool
783If true, use buffered I/O. This is the opposite of the \fBdirect\fR parameter.
784Default: true.
785.TP
f7fa2653 786.BI offset \fR=\fPint
d60e92d1
AC
787Offset in the file to start I/O. Data before the offset will not be touched.
788.TP
591e9e06
JA
789.BI offset_increment \fR=\fPint
790If this is provided, then the real offset becomes the
69bdd6ba
JH
791offset + offset_increment * thread_number, where the thread number is a
792counter that starts at 0 and is incremented for each sub-job (i.e. when
793numjobs option is specified). This option is useful if there are several jobs
794which are intended to operate on a file in parallel disjoint segments, with
795even spacing between the starting points.
591e9e06 796.TP
ddf24e42
JA
797.BI number_ios \fR=\fPint
798Fio will normally perform IOs until it has exhausted the size of the region
799set by \fBsize\fR, or if it exhaust the allocated time (or hits an error
800condition). With this setting, the range/size can be set independently of
801the number of IOs to perform. When fio reaches this number, it will exit
be3fec7d
JA
802normally and report status. Note that this does not extend the amount
803of IO that will be done, it will only stop fio if this condition is met
804before other end-of-job criteria.
ddf24e42 805.TP
d60e92d1 806.BI fsync \fR=\fPint
d1429b5c
AC
807How many I/Os to perform before issuing an \fBfsync\fR\|(2) of dirty data. If
8080, don't sync. Default: 0.
d60e92d1 809.TP
5f9099ea
JA
810.BI fdatasync \fR=\fPint
811Like \fBfsync\fR, but uses \fBfdatasync\fR\|(2) instead to only sync the
812data parts of the file. Default: 0.
813.TP
fa769d44
SW
814.BI write_barrier \fR=\fPint
815Make every Nth write a barrier write.
816.TP
e76b1da4 817.BI sync_file_range \fR=\fPstr:int
ccc2b328
SW
818Use \fBsync_file_range\fR\|(2) for every \fRval\fP number of write operations. Fio will
819track range of writes that have happened since the last \fBsync_file_range\fR\|(2) call.
e76b1da4
JA
820\fRstr\fP can currently be one or more of:
821.RS
822.TP
823.B wait_before
824SYNC_FILE_RANGE_WAIT_BEFORE
825.TP
826.B write
827SYNC_FILE_RANGE_WRITE
828.TP
829.B wait_after
830SYNC_FILE_RANGE_WRITE
831.TP
832.RE
833.P
834So if you do sync_file_range=wait_before,write:8, fio would use
835\fBSYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE\fP for every 8 writes.
ccc2b328 836Also see the \fBsync_file_range\fR\|(2) man page. This option is Linux specific.
e76b1da4 837.TP
d60e92d1 838.BI overwrite \fR=\fPbool
d1429b5c 839If writing, setup the file first and do overwrites. Default: false.
d60e92d1
AC
840.TP
841.BI end_fsync \fR=\fPbool
dbd11ead 842Sync file contents when a write stage has completed. Default: false.
d60e92d1
AC
843.TP
844.BI fsync_on_close \fR=\fPbool
845If true, sync file contents on close. This differs from \fBend_fsync\fR in that
d1429b5c 846it will happen on every close, not just at the end of the job. Default: false.
d60e92d1 847.TP
d60e92d1
AC
848.BI rwmixread \fR=\fPint
849Percentage of a mixed workload that should be reads. Default: 50.
850.TP
851.BI rwmixwrite \fR=\fPint
d1429b5c 852Percentage of a mixed workload that should be writes. If \fBrwmixread\fR and
c35dd7a6
JA
853\fBrwmixwrite\fR are given and do not sum to 100%, the latter of the two
854overrides the first. This may interfere with a given rate setting, if fio is
855asked to limit reads or writes to a certain rate. If that is the case, then
856the distribution may be skewed. Default: 50.
d60e92d1 857.TP
92d42d69
JA
858.BI random_distribution \fR=\fPstr:float
859By default, fio will use a completely uniform random distribution when asked
860to perform random IO. Sometimes it is useful to skew the distribution in
861specific ways, ensuring that some parts of the data is more hot than others.
862Fio includes the following distribution models:
863.RS
864.TP
865.B random
866Uniform random distribution
867.TP
868.B zipf
869Zipf distribution
870.TP
871.B pareto
872Pareto distribution
873.TP
874.RE
875.P
876When using a zipf or pareto distribution, an input value is also needed to
877define the access pattern. For zipf, this is the zipf theta. For pareto,
878it's the pareto power. Fio includes a test program, genzipf, that can be
879used visualize what the given input values will yield in terms of hit rates.
880If you wanted to use zipf with a theta of 1.2, you would use
881random_distribution=zipf:1.2 as the option. If a non-uniform model is used,
882fio will disable use of the random map.
883.TP
211c9b89
JA
884.BI percentage_random \fR=\fPint
885For a random workload, set how big a percentage should be random. This defaults
886to 100%, in which case the workload is fully random. It can be set from
887anywhere from 0 to 100. Setting it to 0 would make the workload fully
d9472271
JA
888sequential. It is possible to set different values for reads, writes, and
889trim. To do so, simply use a comma separated list. See \fBblocksize\fR.
211c9b89 890.TP
d60e92d1
AC
891.B norandommap
892Normally \fBfio\fR will cover every block of the file when doing random I/O. If
893this parameter is given, a new offset will be chosen without looking at past
894I/O history. This parameter is mutually exclusive with \fBverify\fR.
895.TP
744492c9 896.BI softrandommap \fR=\fPbool
3ce9dcaf
JA
897See \fBnorandommap\fR. If fio runs with the random block map enabled and it
898fails to allocate the map, if this option is set it will continue without a
899random block map. As coverage will not be as complete as with random maps, this
900option is disabled by default.
901.TP
e8b1961d
JA
902.BI random_generator \fR=\fPstr
903Fio supports the following engines for generating IO offsets for random IO:
904.RS
905.TP
906.B tausworthe
907Strong 2^88 cycle random number generator
908.TP
909.B lfsr
910Linear feedback shift register generator
911.TP
c3546b53
JA
912.B tausworthe64
913Strong 64-bit 2^258 cycle random number generator
914.TP
e8b1961d
JA
915.RE
916.P
917Tausworthe is a strong random number generator, but it requires tracking on the
918side if we want to ensure that blocks are only read or written once. LFSR
919guarantees that we never generate the same offset twice, and it's also less
920computationally expensive. It's not a true random generator, however, though
921for IO purposes it's typically good enough. LFSR only works with single block
922sizes, not with workloads that use multiple block sizes. If used with such a
923workload, fio may read or write some blocks multiple times.
924.TP
d60e92d1 925.BI nice \fR=\fPint
ccc2b328 926Run job with given nice value. See \fBnice\fR\|(2).
d60e92d1
AC
927.TP
928.BI prio \fR=\fPint
929Set I/O priority value of this job between 0 (highest) and 7 (lowest). See
ccc2b328 930\fBionice\fR\|(1).
d60e92d1
AC
931.TP
932.BI prioclass \fR=\fPint
ccc2b328 933Set I/O priority class. See \fBionice\fR\|(1).
d60e92d1
AC
934.TP
935.BI thinktime \fR=\fPint
936Stall job for given number of microseconds between issuing I/Os.
937.TP
938.BI thinktime_spin \fR=\fPint
939Pretend to spend CPU time for given number of microseconds, sleeping the rest
940of the time specified by \fBthinktime\fR. Only valid if \fBthinktime\fR is set.
941.TP
942.BI thinktime_blocks \fR=\fPint
4d01ece6
JA
943Only valid if thinktime is set - control how many blocks to issue, before
944waiting \fBthinktime\fR microseconds. If not set, defaults to 1 which will
945make fio wait \fBthinktime\fR microseconds after every block. This
946effectively makes any queue depth setting redundant, since no more than 1 IO
947will be queued before we have to complete it and do our thinktime. In other
948words, this setting effectively caps the queue depth if the latter is larger.
d60e92d1
AC
949Default: 1.
950.TP
951.BI rate \fR=\fPint
c35dd7a6
JA
952Cap bandwidth used by this job. The number is in bytes/sec, the normal postfix
953rules apply. You can use \fBrate\fR=500k to limit reads and writes to 500k each,
954or you can specify read and writes separately. Using \fBrate\fR=1m,500k would
955limit reads to 1MB/sec and writes to 500KB/sec. Capping only reads or writes
956can be done with \fBrate\fR=,500k or \fBrate\fR=500k,. The former will only
957limit writes (to 500KB/sec), the latter will only limit reads.
d60e92d1 958.TP
6d428bcd 959.BI rate_min \fR=\fPint
d60e92d1 960Tell \fBfio\fR to do whatever it can to maintain at least the given bandwidth.
c35dd7a6
JA
961Failing to meet this requirement will cause the job to exit. The same format
962as \fBrate\fR is used for read vs write separation.
d60e92d1
AC
963.TP
964.BI rate_iops \fR=\fPint
c35dd7a6
JA
965Cap the bandwidth to this number of IOPS. Basically the same as rate, just
966specified independently of bandwidth. The same format as \fBrate\fR is used for
de8f6de9 967read vs write separation. If \fBblocksize\fR is a range, the smallest block
c35dd7a6 968size is used as the metric.
d60e92d1
AC
969.TP
970.BI rate_iops_min \fR=\fPint
c35dd7a6 971If this rate of I/O is not met, the job will exit. The same format as \fBrate\fR
de8f6de9 972is used for read vs write separation.
d60e92d1 973.TP
6de65959
JA
974.BI rate_process \fR=\fPstr
975This option controls how fio manages rated IO submissions. The default is
976\fBlinear\fR, which submits IO in a linear fashion with fixed delays between
977IOs that gets adjusted based on IO completion rates. If this is set to
978\fBpoisson\fR, fio will submit IO based on a more real world random request
979flow, known as the Poisson process
5d02b083
JA
980(https://en.wikipedia.org/wiki/Poisson_process). The lambda will be
98110^6 / IOPS for the given workload.
ff6bb260 982.TP
6d428bcd
JA
983.BI rate_cycle \fR=\fPint
984Average bandwidth for \fBrate\fR and \fBrate_min\fR over this number of
d60e92d1
AC
985milliseconds. Default: 1000ms.
986.TP
3e260a46
JA
987.BI latency_target \fR=\fPint
988If set, fio will attempt to find the max performance point that the given
989workload will run at while maintaining a latency below this target. The
990values is given in microseconds. See \fBlatency_window\fR and
991\fBlatency_percentile\fR.
992.TP
993.BI latency_window \fR=\fPint
994Used with \fBlatency_target\fR to specify the sample window that the job
995is run at varying queue depths to test the performance. The value is given
996in microseconds.
997.TP
998.BI latency_percentile \fR=\fPfloat
999The percentage of IOs that must fall within the criteria specified by
1000\fBlatency_target\fR and \fBlatency_window\fR. If not set, this defaults
1001to 100.0, meaning that all IOs must be equal or below to the value set
1002by \fBlatency_target\fR.
1003.TP
15501535
JA
1004.BI max_latency \fR=\fPint
1005If set, fio will exit the job if it exceeds this maximum latency. It will exit
1006with an ETIME error.
1007.TP
d60e92d1
AC
1008.BI cpumask \fR=\fPint
1009Set CPU affinity for this job. \fIint\fR is a bitmask of allowed CPUs the job
1010may run on. See \fBsched_setaffinity\fR\|(2).
1011.TP
1012.BI cpus_allowed \fR=\fPstr
1013Same as \fBcpumask\fR, but allows a comma-delimited list of CPU numbers.
1014.TP
c2acfbac
JA
1015.BI cpus_allowed_policy \fR=\fPstr
1016Set the policy of how fio distributes the CPUs specified by \fBcpus_allowed\fR
1017or \fBcpumask\fR. Two policies are supported:
1018.RS
1019.RS
1020.TP
1021.B shared
1022All jobs will share the CPU set specified.
1023.TP
1024.B split
1025Each job will get a unique CPU from the CPU set.
1026.RE
1027.P
1028\fBshared\fR is the default behaviour, if the option isn't specified. If
ada083cd
JA
1029\fBsplit\fR is specified, then fio will assign one cpu per job. If not enough
1030CPUs are given for the jobs listed, then fio will roundrobin the CPUs in
1031the set.
c2acfbac
JA
1032.RE
1033.P
1034.TP
d0b937ed 1035.BI numa_cpu_nodes \fR=\fPstr
cecbfd47 1036Set this job running on specified NUMA nodes' CPUs. The arguments allow
d0b937ed
YR
1037comma delimited list of cpu numbers, A-B ranges, or 'all'.
1038.TP
1039.BI numa_mem_policy \fR=\fPstr
1040Set this job's memory policy and corresponding NUMA nodes. Format of
cecbfd47 1041the arguments:
d0b937ed
YR
1042.RS
1043.TP
1044.B <mode>[:<nodelist>]
1045.TP
1046.B mode
1047is one of the following memory policy:
1048.TP
1049.B default, prefer, bind, interleave, local
1050.TP
1051.RE
1052For \fBdefault\fR and \fBlocal\fR memory policy, no \fBnodelist\fR is
1053needed to be specified. For \fBprefer\fR, only one node is
1054allowed. For \fBbind\fR and \fBinterleave\fR, \fBnodelist\fR allows
1055comma delimited list of numbers, A-B ranges, or 'all'.
1056.TP
23ed19b0
CE
1057.BI startdelay \fR=\fPirange
1058Delay start of job for the specified number of seconds. Supports all time
1059suffixes to allow specification of hours, minutes, seconds and
bd66aa2c 1060milliseconds - seconds are the default if a unit is omitted.
23ed19b0
CE
1061Can be given as a range which causes each thread to choose randomly out of the
1062range.
d60e92d1
AC
1063.TP
1064.BI runtime \fR=\fPint
1065Terminate processing after the specified number of seconds.
1066.TP
1067.B time_based
1068If given, run for the specified \fBruntime\fR duration even if the files are
1069completely read or written. The same workload will be repeated as many times
1070as \fBruntime\fR allows.
1071.TP
901bb994
JA
1072.BI ramp_time \fR=\fPint
1073If set, fio will run the specified workload for this amount of time before
1074logging any performance numbers. Useful for letting performance settle before
1075logging results, thus minimizing the runtime required for stable results. Note
c35dd7a6
JA
1076that the \fBramp_time\fR is considered lead in time for a job, thus it will
1077increase the total runtime if a special timeout or runtime is specified.
901bb994 1078.TP
d60e92d1
AC
1079.BI invalidate \fR=\fPbool
1080Invalidate buffer-cache for the file prior to starting I/O. Default: true.
1081.TP
1082.BI sync \fR=\fPbool
1083Use synchronous I/O for buffered writes. For the majority of I/O engines,
d1429b5c 1084this means using O_SYNC. Default: false.
d60e92d1
AC
1085.TP
1086.BI iomem \fR=\fPstr "\fR,\fP mem" \fR=\fPstr
1087Allocation method for I/O unit buffer. Allowed values are:
1088.RS
1089.RS
1090.TP
1091.B malloc
ccc2b328 1092Allocate memory with \fBmalloc\fR\|(3).
d60e92d1
AC
1093.TP
1094.B shm
ccc2b328 1095Use shared memory buffers allocated through \fBshmget\fR\|(2).
d60e92d1
AC
1096.TP
1097.B shmhuge
1098Same as \fBshm\fR, but use huge pages as backing.
1099.TP
1100.B mmap
ccc2b328 1101Use \fBmmap\fR\|(2) for allocation. Uses anonymous memory unless a filename
d60e92d1
AC
1102is given after the option in the format `:\fIfile\fR'.
1103.TP
1104.B mmaphuge
1105Same as \fBmmap\fR, but use huge files as backing.
09c782bb
JA
1106.TP
1107.B mmapshared
1108Same as \fBmmap\fR, but use a MMAP_SHARED mapping.
d60e92d1
AC
1109.RE
1110.P
1111The amount of memory allocated is the maximum allowed \fBblocksize\fR for the
1112job multiplied by \fBiodepth\fR. For \fBshmhuge\fR or \fBmmaphuge\fR to work,
1113the system must have free huge pages allocated. \fBmmaphuge\fR also needs to
2e266ba6
JA
1114have hugetlbfs mounted, and \fIfile\fR must point there. At least on Linux,
1115huge pages must be manually allocated. See \fB/proc/sys/vm/nr_hugehages\fR
1116and the documentation for that. Normally you just need to echo an appropriate
1117number, eg echoing 8 will ensure that the OS has 8 huge pages ready for
1118use.
d60e92d1
AC
1119.RE
1120.TP
d392365e 1121.BI iomem_align \fR=\fPint "\fR,\fP mem_align" \fR=\fPint
cecbfd47 1122This indicates the memory alignment of the IO memory buffers. Note that the
d529ee19
JA
1123given alignment is applied to the first IO unit buffer, if using \fBiodepth\fR
1124the alignment of the following buffers are given by the \fBbs\fR used. In
1125other words, if using a \fBbs\fR that is a multiple of the page sized in the
1126system, all buffers will be aligned to this value. If using a \fBbs\fR that
1127is not page aligned, the alignment of subsequent IO memory buffers is the
1128sum of the \fBiomem_align\fR and \fBbs\fR used.
1129.TP
f7fa2653 1130.BI hugepage\-size \fR=\fPint
d60e92d1 1131Defines the size of a huge page. Must be at least equal to the system setting.
b22989b9 1132Should be a multiple of 1MB. Default: 4MB.
d60e92d1
AC
1133.TP
1134.B exitall
1135Terminate all jobs when one finishes. Default: wait for each job to finish.
1136.TP
f9cafb12
JA
1137.B exitall_on_error \fR=\fPbool
1138Terminate all jobs if one job finishes in error. Default: wait for each job
1139to finish.
1140.TP
d60e92d1
AC
1141.BI bwavgtime \fR=\fPint
1142Average bandwidth calculations over the given time in milliseconds. Default:
1143500ms.
1144.TP
c8eeb9df
JA
1145.BI iopsavgtime \fR=\fPint
1146Average IOPS calculations over the given time in milliseconds. Default:
1147500ms.
1148.TP
d60e92d1 1149.BI create_serialize \fR=\fPbool
d1429b5c 1150If true, serialize file creation for the jobs. Default: true.
d60e92d1
AC
1151.TP
1152.BI create_fsync \fR=\fPbool
ccc2b328 1153\fBfsync\fR\|(2) data file after creation. Default: true.
d60e92d1 1154.TP
6b7f6851
JA
1155.BI create_on_open \fR=\fPbool
1156If true, the files are not created until they are opened for IO by the job.
1157.TP
25460cf6
JA
1158.BI create_only \fR=\fPbool
1159If true, fio will only run the setup phase of the job. If files need to be
1160laid out or updated on disk, only that will be done. The actual job contents
1161are not executed.
1162.TP
2378826d
JA
1163.BI allow_file_create \fR=\fPbool
1164If true, fio is permitted to create files as part of its workload. This is
1165the default behavior. If this option is false, then fio will error out if the
1166files it needs to use don't already exist. Default: true.
1167.TP
e81ecca3
JA
1168.BI allow_mounted_write \fR=\fPbool
1169If this isn't set, fio will abort jobs that are destructive (eg that write)
1170to what appears to be a mounted device or partition. This should help catch
1171creating inadvertently destructive tests, not realizing that the test will
1172destroy data on the mounted file system. Default: false.
1173.TP
e9f48479
JA
1174.BI pre_read \fR=\fPbool
1175If this is given, files will be pre-read into memory before starting the given
1176IO operation. This will also clear the \fR \fBinvalidate\fR flag, since it is
9c0d2241
JA
1177pointless to pre-read and then drop the cache. This will only work for IO
1178engines that are seekable, since they allow you to read the same data
1179multiple times. Thus it will not work on eg network or splice IO.
e9f48479 1180.TP
d60e92d1
AC
1181.BI unlink \fR=\fPbool
1182Unlink job files when done. Default: false.
1183.TP
1184.BI loops \fR=\fPint
1185Specifies the number of iterations (runs of the same workload) of this job.
1186Default: 1.
1187.TP
5e4c7118
JA
1188.BI verify_only \fR=\fPbool
1189Do not perform the specified workload, only verify data still matches previous
1190invocation of this workload. This option allows one to check data multiple
1191times at a later date without overwriting it. This option makes sense only for
1192workloads that write data, and does not support workloads with the
1193\fBtime_based\fR option set.
1194.TP
d60e92d1
AC
1195.BI do_verify \fR=\fPbool
1196Run the verify phase after a write phase. Only valid if \fBverify\fR is set.
1197Default: true.
1198.TP
1199.BI verify \fR=\fPstr
b638d82f
RP
1200Method of verifying file contents after each iteration of the job. Each
1201verification method also implies verification of special header, which is
1202written to the beginning of each block. This header also includes meta
1203information, like offset of the block, block number, timestamp when block
1204was written, etc. \fBverify\fR=str can be combined with \fBverify_pattern\fR=str
1205option. The allowed values are:
d60e92d1
AC
1206.RS
1207.RS
1208.TP
844ea602 1209.B md5 crc16 crc32 crc32c crc32c-intel crc64 crc7 sha256 sha512 sha1 xxhash
0539d758
JA
1210Store appropriate checksum in the header of each block. crc32c-intel is
1211hardware accelerated SSE4.2 driven, falls back to regular crc32c if
1212not supported by the system.
d60e92d1
AC
1213.TP
1214.B meta
b638d82f
RP
1215This option is deprecated, since now meta information is included in generic
1216verification header and meta verification happens by default. For detailed
1217information see the description of the \fBverify\fR=str setting. This option
1218is kept because of compatibility's sake with old configurations. Do not use it.
d60e92d1 1219.TP
59245381
JA
1220.B pattern
1221Verify a strict pattern. Normally fio includes a header with some basic
1222information and checksumming, but if this option is set, only the
1223specific pattern set with \fBverify_pattern\fR is verified.
1224.TP
d60e92d1
AC
1225.B null
1226Pretend to verify. Used for testing internals.
1227.RE
b892dc08
JA
1228
1229This option can be used for repeated burn-in tests of a system to make sure
1230that the written data is also correctly read back. If the data direction given
1231is a read or random read, fio will assume that it should verify a previously
1232written file. If the data direction includes any form of write, the verify will
1233be of the newly written data.
d60e92d1
AC
1234.RE
1235.TP
5c9323fb 1236.BI verifysort \fR=\fPbool
d60e92d1
AC
1237If true, written verify blocks are sorted if \fBfio\fR deems it to be faster to
1238read them back in a sorted manner. Default: true.
1239.TP
fa769d44
SW
1240.BI verifysort_nr \fR=\fPint
1241Pre-load and sort verify blocks for a read workload.
1242.TP
f7fa2653 1243.BI verify_offset \fR=\fPint
d60e92d1 1244Swap the verification header with data somewhere else in the block before
d1429b5c 1245writing. It is swapped back before verifying.
d60e92d1 1246.TP
f7fa2653 1247.BI verify_interval \fR=\fPint
d60e92d1
AC
1248Write the verification header for this number of bytes, which should divide
1249\fBblocksize\fR. Default: \fBblocksize\fR.
1250.TP
996093bb
JA
1251.BI verify_pattern \fR=\fPstr
1252If set, fio will fill the io buffers with this pattern. Fio defaults to filling
1253with totally random bytes, but sometimes it's interesting to fill with a known
1254pattern for io verification purposes. Depending on the width of the pattern,
1255fio will fill 1/2/3/4 bytes of the buffer at the time(it can be either a
1256decimal or a hex number). The verify_pattern if larger than a 32-bit quantity
1257has to be a hex number that starts with either "0x" or "0X". Use with
b638d82f
RP
1258\fBverify\fP=str. Also, verify_pattern supports %o format, which means that for
1259each block offset will be written and then verifyied back, e.g.:
2fa5a241
RP
1260.RS
1261.RS
1262\fBverify_pattern\fR=%o
1263.RE
1264Or use combination of everything:
1265.LP
1266.RS
1267\fBverify_pattern\fR=0xff%o"abcd"-21
1268.RE
1269.RE
996093bb 1270.TP
d60e92d1
AC
1271.BI verify_fatal \fR=\fPbool
1272If true, exit the job on the first observed verification failure. Default:
1273false.
1274.TP
b463e936
JA
1275.BI verify_dump \fR=\fPbool
1276If set, dump the contents of both the original data block and the data block we
1277read off disk to files. This allows later analysis to inspect just what kind of
ef71e317 1278data corruption occurred. Off by default.
b463e936 1279.TP
e8462bd8
JA
1280.BI verify_async \fR=\fPint
1281Fio will normally verify IO inline from the submitting thread. This option
1282takes an integer describing how many async offload threads to create for IO
1283verification instead, causing fio to offload the duty of verifying IO contents
c85c324c
JA
1284to one or more separate threads. If using this offload option, even sync IO
1285engines can benefit from using an \fBiodepth\fR setting higher than 1, as it
1286allows them to have IO in flight while verifies are running.
e8462bd8
JA
1287.TP
1288.BI verify_async_cpus \fR=\fPstr
1289Tell fio to set the given CPU affinity on the async IO verification threads.
1290See \fBcpus_allowed\fP for the format used.
1291.TP
6f87418f
JA
1292.BI verify_backlog \fR=\fPint
1293Fio will normally verify the written contents of a job that utilizes verify
1294once that job has completed. In other words, everything is written then
1295everything is read back and verified. You may want to verify continually
1296instead for a variety of reasons. Fio stores the meta data associated with an
1297IO block in memory, so for large verify workloads, quite a bit of memory would
092f707f
DN
1298be used up holding this meta data. If this option is enabled, fio will write
1299only N blocks before verifying these blocks.
6f87418f
JA
1300.TP
1301.BI verify_backlog_batch \fR=\fPint
1302Control how many blocks fio will verify if verify_backlog is set. If not set,
1303will default to the value of \fBverify_backlog\fR (meaning the entire queue is
ff6bb260
SL
1304read back and verified). If \fBverify_backlog_batch\fR is less than
1305\fBverify_backlog\fR then not all blocks will be verified, if
092f707f
DN
1306\fBverify_backlog_batch\fR is larger than \fBverify_backlog\fR, some blocks
1307will be verified more than once.
6f87418f 1308.TP
fa769d44
SW
1309.BI trim_percentage \fR=\fPint
1310Number of verify blocks to discard/trim.
1311.TP
1312.BI trim_verify_zero \fR=\fPbool
1313Verify that trim/discarded blocks are returned as zeroes.
1314.TP
1315.BI trim_backlog \fR=\fPint
1316Trim after this number of blocks are written.
1317.TP
1318.BI trim_backlog_batch \fR=\fPint
1319Trim this number of IO blocks.
1320.TP
1321.BI experimental_verify \fR=\fPbool
1322Enable experimental verification.
1323.TP
ca09be4b
JA
1324.BI verify_state_save \fR=\fPbool
1325When a job exits during the write phase of a verify workload, save its
1326current state. This allows fio to replay up until that point, if the
1327verify state is loaded for the verify read phase.
1328.TP
1329.BI verify_state_load \fR=\fPbool
1330If a verify termination trigger was used, fio stores the current write
1331state of each thread. This can be used at verification time so that fio
1332knows how far it should verify. Without this information, fio will run
1333a full verification pass, according to the settings in the job file used.
1334.TP
d392365e 1335.B stonewall "\fR,\fP wait_for_previous"
5982a925 1336Wait for preceding jobs in the job file to exit before starting this one.
d60e92d1
AC
1337\fBstonewall\fR implies \fBnew_group\fR.
1338.TP
1339.B new_group
1340Start a new reporting group. If not given, all jobs in a file will be part
1341of the same reporting group, unless separated by a stonewall.
1342.TP
1343.BI numjobs \fR=\fPint
ff6bb260 1344Number of clones (processes/threads performing the same workload) of this job.
d60e92d1
AC
1345Default: 1.
1346.TP
1347.B group_reporting
1348If set, display per-group reports instead of per-job when \fBnumjobs\fR is
1349specified.
1350.TP
1351.B thread
1352Use threads created with \fBpthread_create\fR\|(3) instead of processes created
1353with \fBfork\fR\|(2).
1354.TP
f7fa2653 1355.BI zonesize \fR=\fPint
d60e92d1
AC
1356Divide file into zones of the specified size in bytes. See \fBzoneskip\fR.
1357.TP
fa769d44
SW
1358.BI zonerange \fR=\fPint
1359Give size of an IO zone. See \fBzoneskip\fR.
1360.TP
f7fa2653 1361.BI zoneskip \fR=\fPint
d1429b5c 1362Skip the specified number of bytes when \fBzonesize\fR bytes of data have been
d60e92d1
AC
1363read.
1364.TP
1365.BI write_iolog \fR=\fPstr
5b42a488
SH
1366Write the issued I/O patterns to the specified file. Specify a separate file
1367for each job, otherwise the iologs will be interspersed and the file may be
1368corrupt.
d60e92d1
AC
1369.TP
1370.BI read_iolog \fR=\fPstr
1371Replay the I/O patterns contained in the specified file generated by
1372\fBwrite_iolog\fR, or may be a \fBblktrace\fR binary file.
1373.TP
64bbb865
DN
1374.BI replay_no_stall \fR=\fPint
1375While replaying I/O patterns using \fBread_iolog\fR the default behavior
1376attempts to respect timing information between I/Os. Enabling
1377\fBreplay_no_stall\fR causes I/Os to be replayed as fast as possible while
1378still respecting ordering.
1379.TP
d1c46c04
DN
1380.BI replay_redirect \fR=\fPstr
1381While replaying I/O patterns using \fBread_iolog\fR the default behavior
1382is to replay the IOPS onto the major/minor device that each IOP was recorded
1383from. Setting \fBreplay_redirect\fR causes all IOPS to be replayed onto the
1384single specified device regardless of the device it was recorded from.
1385.TP
0c63576e
JA
1386.BI replay_align \fR=\fPint
1387Force alignment of IO offsets and lengths in a trace to this power of 2 value.
1388.TP
1389.BI replay_scale \fR=\fPint
1390Scale sector offsets down by this factor when replaying traces.
1391.TP
3a5db920
JA
1392.BI per_job_logs \fR=\fPbool
1393If set, this generates bw/clat/iops log with per file private filenames. If
1394not set, jobs with identical names will share the log filename. Default: true.
1395.TP
836bad52 1396.BI write_bw_log \fR=\fPstr
901bb994
JA
1397If given, write a bandwidth log of the jobs in this job file. Can be used to
1398store data of the bandwidth of the jobs in their lifetime. The included
1399fio_generate_plots script uses gnuplot to turn these text files into nice
26b26fca 1400graphs. See \fBwrite_lat_log\fR for behaviour of given filename. For this
8ad3b3dd 1401option, the postfix is _bw.x.log, where x is the index of the job (1..N,
3a5db920
JA
1402where N is the number of jobs). If \fBper_job_logs\fR is false, then the
1403filename will not include the job index.
d60e92d1 1404.TP
836bad52 1405.BI write_lat_log \fR=\fPstr
901bb994 1406Same as \fBwrite_bw_log\fR, but writes I/O completion latencies. If no
8ad3b3dd
JA
1407filename is given with this option, the default filename of
1408"jobname_type.x.log" is used, where x is the index of the job (1..N, where
1409N is the number of jobs). Even if the filename is given, fio will still
3a5db920
JA
1410append the type of log. If \fBper_job_logs\fR is false, then the filename will
1411not include the job index.
901bb994 1412.TP
c8eeb9df
JA
1413.BI write_iops_log \fR=\fPstr
1414Same as \fBwrite_bw_log\fR, but writes IOPS. If no filename is given with this
8ad3b3dd
JA
1415option, the default filename of "jobname_type.x.log" is used, where x is the
1416index of the job (1..N, where N is the number of jobs). Even if the filename
3a5db920
JA
1417is given, fio will still append the type of log. If \fBper_job_logs\fR is false,
1418then the filename will not include the job index.
c8eeb9df 1419.TP
b8bc8cba
JA
1420.BI log_avg_msec \fR=\fPint
1421By default, fio will log an entry in the iops, latency, or bw log for every
1422IO that completes. When writing to the disk log, that can quickly grow to a
1423very large size. Setting this option makes fio average the each log entry
1424over the specified period of time, reducing the resolution of the log.
1425Defaults to 0.
1426.TP
ae588852
JA
1427.BI log_offset \fR=\fPbool
1428If this is set, the iolog options will include the byte offset for the IO
1429entry as well as the other data values.
1430.TP
aee2ab67
JA
1431.BI log_compression \fR=\fPint
1432If this is set, fio will compress the IO logs as it goes, to keep the memory
1433footprint lower. When a log reaches the specified size, that chunk is removed
1434and compressed in the background. Given that IO logs are fairly highly
1435compressible, this yields a nice memory savings for longer runs. The downside
1436is that the compression will consume some background CPU cycles, so it may
1437impact the run. This, however, is also true if the logging ends up consuming
1438most of the system memory. So pick your poison. The IO logs are saved
1439normally at the end of a run, by decompressing the chunks and storing them
1440in the specified log file. This feature depends on the availability of zlib.
1441.TP
c08f9fe2
JA
1442.BI log_compression_cpus \fR=\fPstr
1443Define the set of CPUs that are allowed to handle online log compression
1444for the IO jobs. This can provide better isolation between performance
1445sensitive jobs, and background compression work.
1446.TP
b26317c9 1447.BI log_store_compressed \fR=\fPbool
c08f9fe2
JA
1448If set, fio will store the log files in a compressed format. They can be
1449decompressed with fio, using the \fB\-\-inflate-log\fR command line parameter.
1450The files will be stored with a \fB\.fz\fR suffix.
b26317c9 1451.TP
66347cfa
DE
1452.BI block_error_percentiles \fR=\fPbool
1453If set, record errors in trim block-sized units from writes and trims and output
1454a histogram of how many trims it took to get to errors, and what kind of error
1455was encountered.
1456.TP
836bad52 1457.BI disable_lat \fR=\fPbool
02af0988 1458Disable measurements of total latency numbers. Useful only for cutting
ccc2b328 1459back the number of calls to \fBgettimeofday\fR\|(2), as that does impact performance at
901bb994
JA
1460really high IOPS rates. Note that to really get rid of a large amount of these
1461calls, this option must be used with disable_slat and disable_bw as well.
1462.TP
836bad52 1463.BI disable_clat \fR=\fPbool
c95f9daf 1464Disable measurements of completion latency numbers. See \fBdisable_lat\fR.
02af0988 1465.TP
836bad52 1466.BI disable_slat \fR=\fPbool
02af0988 1467Disable measurements of submission latency numbers. See \fBdisable_lat\fR.
901bb994 1468.TP
836bad52 1469.BI disable_bw_measurement \fR=\fPbool
02af0988 1470Disable measurements of throughput/bandwidth numbers. See \fBdisable_lat\fR.
d60e92d1 1471.TP
f7fa2653 1472.BI lockmem \fR=\fPint
d60e92d1 1473Pin the specified amount of memory with \fBmlock\fR\|(2). Can be used to
81c6b6cd 1474simulate a smaller amount of memory. The amount specified is per worker.
d60e92d1
AC
1475.TP
1476.BI exec_prerun \fR=\fPstr
1477Before running the job, execute the specified command with \fBsystem\fR\|(3).
ce486495
EV
1478.RS
1479Output is redirected in a file called \fBjobname.prerun.txt\fR
1480.RE
d60e92d1
AC
1481.TP
1482.BI exec_postrun \fR=\fPstr
1483Same as \fBexec_prerun\fR, but the command is executed after the job completes.
ce486495
EV
1484.RS
1485Output is redirected in a file called \fBjobname.postrun.txt\fR
1486.RE
d60e92d1
AC
1487.TP
1488.BI ioscheduler \fR=\fPstr
1489Attempt to switch the device hosting the file to the specified I/O scheduler.
1490.TP
d60e92d1 1491.BI disk_util \fR=\fPbool
d1429b5c 1492Generate disk utilization statistics if the platform supports it. Default: true.
901bb994 1493.TP
23893646
JA
1494.BI clocksource \fR=\fPstr
1495Use the given clocksource as the base of timing. The supported options are:
1496.RS
1497.TP
1498.B gettimeofday
ccc2b328 1499\fBgettimeofday\fR\|(2)
23893646
JA
1500.TP
1501.B clock_gettime
ccc2b328 1502\fBclock_gettime\fR\|(2)
23893646
JA
1503.TP
1504.B cpu
1505Internal CPU clock source
1506.TP
1507.RE
1508.P
1509\fBcpu\fR is the preferred clocksource if it is reliable, as it is very fast
1510(and fio is heavy on time calls). Fio will automatically use this clocksource
1511if it's supported and considered reliable on the system it is running on,
1512unless another clocksource is specifically set. For x86/x86-64 CPUs, this
1513means supporting TSC Invariant.
1514.TP
901bb994 1515.BI gtod_reduce \fR=\fPbool
ccc2b328 1516Enable all of the \fBgettimeofday\fR\|(2) reducing options (disable_clat, disable_slat,
901bb994 1517disable_bw) plus reduce precision of the timeout somewhat to really shrink the
ccc2b328 1518\fBgettimeofday\fR\|(2) call count. With this option enabled, we only do about 0.4% of
901bb994
JA
1519the gtod() calls we would have done if all time keeping was enabled.
1520.TP
1521.BI gtod_cpu \fR=\fPint
1522Sometimes it's cheaper to dedicate a single thread of execution to just getting
1523the current time. Fio (and databases, for instance) are very intensive on
ccc2b328 1524\fBgettimeofday\fR\|(2) calls. With this option, you can set one CPU aside for doing
901bb994
JA
1525nothing but logging current time to a shared memory location. Then the other
1526threads/processes that run IO workloads need only copy that segment, instead of
ccc2b328 1527entering the kernel with a \fBgettimeofday\fR\|(2) call. The CPU set aside for doing
901bb994
JA
1528these time calls will be excluded from other uses. Fio will manually clear it
1529from the CPU mask of other jobs.
f2bba182 1530.TP
8b28bd41
DM
1531.BI ignore_error \fR=\fPstr
1532Sometimes you want to ignore some errors during test in that case you can specify
1533error list for each error type.
1534.br
1535ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1536.br
1537errors for given error type is separated with ':'.
1538Error may be symbol ('ENOSPC', 'ENOMEM') or an integer.
1539.br
1540Example: ignore_error=EAGAIN,ENOSPC:122 .
ff6bb260
SL
1541.br
1542This option will ignore EAGAIN from READ, and ENOSPC and 122(EDQUOT) from WRITE.
8b28bd41
DM
1543.TP
1544.BI error_dump \fR=\fPbool
1545If set dump every error even if it is non fatal, true by default. If disabled
1546only fatal error will be dumped
1547.TP
fa769d44
SW
1548.BI profile \fR=\fPstr
1549Select a specific builtin performance test.
1550.TP
a696fa2a
JA
1551.BI cgroup \fR=\fPstr
1552Add job to this control group. If it doesn't exist, it will be created.
6adb38a1
JA
1553The system must have a mounted cgroup blkio mount point for this to work. If
1554your system doesn't have it mounted, you can do so with:
1555
5982a925 1556# mount \-t cgroup \-o blkio none /cgroup
a696fa2a
JA
1557.TP
1558.BI cgroup_weight \fR=\fPint
1559Set the weight of the cgroup to this value. See the documentation that comes
1560with the kernel, allowed values are in the range of 100..1000.
e0b0d892 1561.TP
7de87099
VG
1562.BI cgroup_nodelete \fR=\fPbool
1563Normally fio will delete the cgroups it has created after the job completion.
1564To override this behavior and to leave cgroups around after the job completion,
1565set cgroup_nodelete=1. This can be useful if one wants to inspect various
1566cgroup files after job completion. Default: false
1567.TP
e0b0d892
JA
1568.BI uid \fR=\fPint
1569Instead of running as the invoking user, set the user ID to this value before
1570the thread/process does any work.
1571.TP
1572.BI gid \fR=\fPint
1573Set group ID, see \fBuid\fR.
83349190 1574.TP
fa769d44
SW
1575.BI unit_base \fR=\fPint
1576Base unit for reporting. Allowed values are:
1577.RS
1578.TP
1579.B 0
1580Use auto-detection (default).
1581.TP
1582.B 8
1583Byte based.
1584.TP
1585.B 1
1586Bit based.
1587.RE
1588.P
1589.TP
9e684a49
DE
1590.BI flow_id \fR=\fPint
1591The ID of the flow. If not specified, it defaults to being a global flow. See
1592\fBflow\fR.
1593.TP
1594.BI flow \fR=\fPint
1595Weight in token-based flow control. If this value is used, then there is a
1596\fBflow counter\fR which is used to regulate the proportion of activity between
1597two or more jobs. fio attempts to keep this flow counter near zero. The
1598\fBflow\fR parameter stands for how much should be added or subtracted to the
1599flow counter on each iteration of the main I/O loop. That is, if one job has
1600\fBflow=8\fR and another job has \fBflow=-1\fR, then there will be a roughly
16011:8 ratio in how much one runs vs the other.
1602.TP
1603.BI flow_watermark \fR=\fPint
1604The maximum value that the absolute value of the flow counter is allowed to
1605reach before the job must wait for a lower value of the counter.
1606.TP
1607.BI flow_sleep \fR=\fPint
1608The period of time, in microseconds, to wait after the flow watermark has been
1609exceeded before retrying operations
1610.TP
83349190
YH
1611.BI clat_percentiles \fR=\fPbool
1612Enable the reporting of percentiles of completion latencies.
1613.TP
1614.BI percentile_list \fR=\fPfloat_list
66347cfa
DE
1615Overwrite the default list of percentiles for completion latencies and the
1616block error histogram. Each number is a floating number in the range (0,100],
1617and the maximum length of the list is 20. Use ':' to separate the
3eb07285 1618numbers. For example, \-\-percentile_list=99.5:99.9 will cause fio to
83349190
YH
1619report the values of completion latency below which 99.5% and 99.9% of
1620the observed latencies fell, respectively.
de890a1e
SL
1621.SS "Ioengine Parameters List"
1622Some parameters are only valid when a specific ioengine is in use. These are
1623used identically to normal parameters, with the caveat that when used on the
cf145d90 1624command line, they must come after the ioengine.
de890a1e 1625.TP
e4585935
JA
1626.BI (cpu)cpuload \fR=\fPint
1627Attempt to use the specified percentage of CPU cycles.
1628.TP
1629.BI (cpu)cpuchunks \fR=\fPint
1630Split the load into cycles of the given time. In microseconds.
1631.TP
046395d7
JA
1632.BI (cpu)exit_on_io_done \fR=\fPbool
1633Detect when IO threads are done, then exit.
1634.TP
de890a1e
SL
1635.BI (libaio)userspace_reap
1636Normally, with the libaio engine in use, fio will use
1637the io_getevents system call to reap newly returned events.
1638With this flag turned on, the AIO ring will be read directly
1639from user-space to reap events. The reaping mode is only
1640enabled when polling for a minimum of 0 events (eg when
1641iodepth_batch_complete=0).
1642.TP
1643.BI (net,netsplice)hostname \fR=\fPstr
1644The host name or IP address to use for TCP or UDP based IO.
1645If the job is a TCP listener or UDP reader, the hostname is not
b511c9aa 1646used and must be omitted unless it is a valid UDP multicast address.
de890a1e
SL
1647.TP
1648.BI (net,netsplice)port \fR=\fPint
6315af9d
JA
1649The TCP or UDP port to bind to or connect to. If this is used with
1650\fBnumjobs\fR to spawn multiple instances of the same job type, then
1651this will be the starting port number since fio will use a range of ports.
de890a1e 1652.TP
b93b6a2e
SB
1653.BI (net,netsplice)interface \fR=\fPstr
1654The IP address of the network interface used to send or receive UDP multicast
1655packets.
1656.TP
d3a623de
SB
1657.BI (net,netsplice)ttl \fR=\fPint
1658Time-to-live value for outgoing UDP multicast packets. Default: 1
1659.TP
1d360ffb
JA
1660.BI (net,netsplice)nodelay \fR=\fPbool
1661Set TCP_NODELAY on TCP connections.
1662.TP
de890a1e
SL
1663.BI (net,netsplice)protocol \fR=\fPstr "\fR,\fP proto" \fR=\fPstr
1664The network protocol to use. Accepted values are:
1665.RS
1666.RS
1667.TP
1668.B tcp
1669Transmission control protocol
1670.TP
49ccb8c1
JA
1671.B tcpv6
1672Transmission control protocol V6
1673.TP
de890a1e 1674.B udp
f5cc3d0e 1675User datagram protocol
de890a1e 1676.TP
49ccb8c1
JA
1677.B udpv6
1678User datagram protocol V6
1679.TP
de890a1e
SL
1680.B unix
1681UNIX domain socket
1682.RE
1683.P
1684When the protocol is TCP or UDP, the port must also be given,
1685as well as the hostname if the job is a TCP listener or UDP
1686reader. For unix sockets, the normal filename option should be
1687used and the port is invalid.
1688.RE
1689.TP
1690.BI (net,netsplice)listen
1691For TCP network connections, tell fio to listen for incoming
1692connections rather than initiating an outgoing connection. The
1693hostname must be omitted if this option is used.
d54fce84 1694.TP
7aeb1e94 1695.BI (net, pingpong) \fR=\fPbool
cecbfd47 1696Normally a network writer will just continue writing data, and a network reader
cf145d90 1697will just consume packets. If pingpong=1 is set, a writer will send its normal
7aeb1e94
JA
1698payload to the reader, then wait for the reader to send the same payload back.
1699This allows fio to measure network latencies. The submission and completion
1700latencies then measure local time spent sending or receiving, and the
1701completion latency measures how long it took for the other end to receive and
b511c9aa
SB
1702send back. For UDP multicast traffic pingpong=1 should only be set for a single
1703reader when multiple readers are listening to the same address.
7aeb1e94 1704.TP
1008602c
JA
1705.BI (net, window_size) \fR=\fPint
1706Set the desired socket buffer size for the connection.
1707.TP
e5f34d95
JA
1708.BI (net, mss) \fR=\fPint
1709Set the TCP maximum segment size (TCP_MAXSEG).
1710.TP
d54fce84
DM
1711.BI (e4defrag,donorname) \fR=\fPstr
1712File will be used as a block donor (swap extents between files)
1713.TP
1714.BI (e4defrag,inplace) \fR=\fPint
ff6bb260 1715Configure donor file block allocation strategy
d54fce84
DM
1716.RS
1717.BI 0(default) :
1718Preallocate donor's file on init
1719.TP
1720.BI 1:
cecbfd47 1721allocate space immediately inside defragment event, and free right after event
d54fce84 1722.RE
0d978694
DAG
1723.TP
1724.BI (rbd)rbdname \fR=\fPstr
1725Specifies the name of the RBD.
1726.TP
1727.BI (rbd)pool \fR=\fPstr
1728Specifies the name of the Ceph pool containing the RBD.
1729.TP
1730.BI (rbd)clientname \fR=\fPstr
1731Specifies the username (without the 'client.' prefix) used to access the Ceph cluster.
65fa28ca
DE
1732.TP
1733.BI (mtd)skipbad \fR=\fPbool
1734Skip operations against known bad blocks.
d60e92d1 1735.SH OUTPUT
d1429b5c
AC
1736While running, \fBfio\fR will display the status of the created jobs. For
1737example:
d60e92d1 1738.RS
d1429b5c 1739.P
d60e92d1
AC
1740Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1741.RE
1742.P
d1429b5c
AC
1743The characters in the first set of brackets denote the current status of each
1744threads. The possible values are:
1745.P
1746.PD 0
d60e92d1
AC
1747.RS
1748.TP
1749.B P
1750Setup but not started.
1751.TP
1752.B C
1753Thread created.
1754.TP
1755.B I
1756Initialized, waiting.
1757.TP
1758.B R
1759Running, doing sequential reads.
1760.TP
1761.B r
1762Running, doing random reads.
1763.TP
1764.B W
1765Running, doing sequential writes.
1766.TP
1767.B w
1768Running, doing random writes.
1769.TP
1770.B M
1771Running, doing mixed sequential reads/writes.
1772.TP
1773.B m
1774Running, doing mixed random reads/writes.
1775.TP
1776.B F
1777Running, currently waiting for \fBfsync\fR\|(2).
1778.TP
1779.B V
1780Running, verifying written data.
1781.TP
1782.B E
1783Exited, not reaped by main thread.
1784.TP
1785.B \-
1786Exited, thread reaped.
1787.RE
d1429b5c 1788.PD
d60e92d1
AC
1789.P
1790The second set of brackets shows the estimated completion percentage of
1791the current group. The third set shows the read and write I/O rate,
1792respectively. Finally, the estimated run time of the job is displayed.
1793.P
1794When \fBfio\fR completes (or is interrupted by Ctrl-C), it will show data
1795for each thread, each group of threads, and each disk, in that order.
1796.P
1797Per-thread statistics first show the threads client number, group-id, and
1798error code. The remaining figures are as follows:
1799.RS
d60e92d1
AC
1800.TP
1801.B io
1802Number of megabytes of I/O performed.
1803.TP
1804.B bw
1805Average data rate (bandwidth).
1806.TP
1807.B runt
1808Threads run time.
1809.TP
1810.B slat
1811Submission latency minimum, maximum, average and standard deviation. This is
1812the time it took to submit the I/O.
1813.TP
1814.B clat
1815Completion latency minimum, maximum, average and standard deviation. This
1816is the time between submission and completion.
1817.TP
1818.B bw
1819Bandwidth minimum, maximum, percentage of aggregate bandwidth received, average
1820and standard deviation.
1821.TP
1822.B cpu
1823CPU usage statistics. Includes user and system time, number of context switches
1824this thread went through and number of major and minor page faults.
1825.TP
1826.B IO depths
1827Distribution of I/O depths. Each depth includes everything less than (or equal)
1828to it, but greater than the previous depth.
1829.TP
1830.B IO issued
1831Number of read/write requests issued, and number of short read/write requests.
1832.TP
1833.B IO latencies
1834Distribution of I/O completion latencies. The numbers follow the same pattern
1835as \fBIO depths\fR.
1836.RE
d60e92d1
AC
1837.P
1838The group statistics show:
d1429b5c 1839.PD 0
d60e92d1
AC
1840.RS
1841.TP
1842.B io
1843Number of megabytes I/O performed.
1844.TP
1845.B aggrb
1846Aggregate bandwidth of threads in the group.
1847.TP
1848.B minb
1849Minimum average bandwidth a thread saw.
1850.TP
1851.B maxb
1852Maximum average bandwidth a thread saw.
1853.TP
1854.B mint
d1429b5c 1855Shortest runtime of threads in the group.
d60e92d1
AC
1856.TP
1857.B maxt
1858Longest runtime of threads in the group.
1859.RE
d1429b5c 1860.PD
d60e92d1
AC
1861.P
1862Finally, disk statistics are printed with reads first:
d1429b5c 1863.PD 0
d60e92d1
AC
1864.RS
1865.TP
1866.B ios
1867Number of I/Os performed by all groups.
1868.TP
1869.B merge
1870Number of merges in the I/O scheduler.
1871.TP
1872.B ticks
1873Number of ticks we kept the disk busy.
1874.TP
1875.B io_queue
1876Total time spent in the disk queue.
1877.TP
1878.B util
1879Disk utilization.
1880.RE
d1429b5c 1881.PD
8423bd11
JA
1882.P
1883It is also possible to get fio to dump the current output while it is
1884running, without terminating the job. To do that, send fio the \fBUSR1\fR
1885signal.
d60e92d1 1886.SH TERSE OUTPUT
2b8c71b0
CE
1887If the \fB\-\-minimal\fR / \fB\-\-append-terse\fR options are given, the
1888results will be printed/appended in a semicolon-delimited format suitable for
1889scripted use.
1890A job description (if provided) follows on a new line. Note that the first
525c2bfa
JA
1891number in the line is the version number. If the output has to be changed
1892for some reason, this number will be incremented by 1 to signify that
1893change. The fields are:
d60e92d1
AC
1894.P
1895.RS
5e726d0a 1896.B terse version, fio version, jobname, groupid, error
d60e92d1
AC
1897.P
1898Read status:
1899.RS
312b4af2 1900.B Total I/O \fR(KB)\fP, bandwidth \fR(KB/s)\fP, IOPS, runtime \fR(ms)\fP
d60e92d1
AC
1901.P
1902Submission latency:
1903.RS
1904.B min, max, mean, standard deviation
1905.RE
1906Completion latency:
1907.RS
1908.B min, max, mean, standard deviation
1909.RE
1db92cb6
JA
1910Completion latency percentiles (20 fields):
1911.RS
1912.B Xth percentile=usec
1913.RE
525c2bfa
JA
1914Total latency:
1915.RS
1916.B min, max, mean, standard deviation
1917.RE
d60e92d1
AC
1918Bandwidth:
1919.RS
1920.B min, max, aggregate percentage of total, mean, standard deviation
1921.RE
1922.RE
1923.P
1924Write status:
1925.RS
312b4af2 1926.B Total I/O \fR(KB)\fP, bandwidth \fR(KB/s)\fP, IOPS, runtime \fR(ms)\fP
d60e92d1
AC
1927.P
1928Submission latency:
1929.RS
1930.B min, max, mean, standard deviation
1931.RE
1932Completion latency:
1933.RS
1934.B min, max, mean, standard deviation
1935.RE
1db92cb6
JA
1936Completion latency percentiles (20 fields):
1937.RS
1938.B Xth percentile=usec
1939.RE
525c2bfa
JA
1940Total latency:
1941.RS
1942.B min, max, mean, standard deviation
1943.RE
d60e92d1
AC
1944Bandwidth:
1945.RS
1946.B min, max, aggregate percentage of total, mean, standard deviation
1947.RE
1948.RE
1949.P
d1429b5c 1950CPU usage:
d60e92d1 1951.RS
bd2626f0 1952.B user, system, context switches, major page faults, minor page faults
d60e92d1
AC
1953.RE
1954.P
1955IO depth distribution:
1956.RS
1957.B <=1, 2, 4, 8, 16, 32, >=64
1958.RE
1959.P
562c2d2f 1960IO latency distribution:
d60e92d1 1961.RS
562c2d2f
DN
1962Microseconds:
1963.RS
1964.B <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1965.RE
1966Milliseconds:
1967.RS
1968.B <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1969.RE
1970.RE
1971.P
f2f788dd
JA
1972Disk utilization (1 for each disk used):
1973.RS
1974.B name, read ios, write ios, read merges, write merges, read ticks, write ticks, read in-queue time, write in-queue time, disk utilization percentage
1975.RE
1976.P
5982a925 1977Error Info (dependent on continue_on_error, default off):
562c2d2f 1978.RS
ff6bb260 1979.B total # errors, first error code
d60e92d1
AC
1980.RE
1981.P
562c2d2f 1982.B text description (if provided in config - appears on newline)
d60e92d1 1983.RE
49da1240
JA
1984.SH CLIENT / SERVER
1985Normally you would run fio as a stand-alone application on the machine
1986where the IO workload should be generated. However, it is also possible to
1987run the frontend and backend of fio separately. This makes it possible to
1988have a fio server running on the machine(s) where the IO workload should
1989be running, while controlling it from another machine.
1990
1991To start the server, you would do:
1992
1993\fBfio \-\-server=args\fR
1994
1995on that machine, where args defines what fio listens to. The arguments
811826be 1996are of the form 'type:hostname or IP:port'. 'type' is either 'ip' (or ip4)
20c67f10
MS
1997for TCP/IP v4, 'ip6' for TCP/IP v6, or 'sock' for a local unix domain
1998socket. 'hostname' is either a hostname or IP address, and 'port' is the port to
811826be 1999listen to (only valid for TCP/IP, not a local socket). Some examples:
49da1240 2000
e01e9745 20011) fio \-\-server
49da1240
JA
2002
2003 Start a fio server, listening on all interfaces on the default port (8765).
2004
e01e9745 20052) fio \-\-server=ip:hostname,4444
49da1240
JA
2006
2007 Start a fio server, listening on IP belonging to hostname and on port 4444.
2008
e01e9745 20093) fio \-\-server=ip6:::1,4444
811826be
JA
2010
2011 Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
2012
e01e9745 20134) fio \-\-server=,4444
49da1240
JA
2014
2015 Start a fio server, listening on all interfaces on port 4444.
2016
e01e9745 20175) fio \-\-server=1.2.3.4
49da1240
JA
2018
2019 Start a fio server, listening on IP 1.2.3.4 on the default port.
2020
e01e9745 20216) fio \-\-server=sock:/tmp/fio.sock
49da1240
JA
2022
2023 Start a fio server, listening on the local socket /tmp/fio.sock.
2024
2025When a server is running, you can connect to it from a client. The client
2026is run with:
2027
e01e9745 2028fio \-\-local-args \-\-client=server \-\-remote-args <job file(s)>
49da1240 2029
e01e9745
MS
2030where \-\-local-args are arguments that are local to the client where it is
2031running, 'server' is the connect string, and \-\-remote-args and <job file(s)>
49da1240
JA
2032are sent to the server. The 'server' string follows the same format as it
2033does on the server side, to allow IP/hostname/socket and port strings.
2034You can connect to multiple clients as well, to do that you could run:
2035
e01e9745 2036fio \-\-client=server2 \-\-client=server2 <job file(s)>
323255cc
JA
2037
2038If the job file is located on the fio server, then you can tell the server
2039to load a local file as well. This is done by using \-\-remote-config:
2040
2041fio \-\-client=server \-\-remote-config /path/to/file.fio
2042
39b5f61e 2043Then fio will open this local (to the server) job file instead
323255cc 2044of being passed one from the client.
39b5f61e 2045
ff6bb260 2046If you have many servers (example: 100 VMs/containers), you can input a pathname
39b5f61e
BE
2047of a file containing host IPs/names as the parameter value for the \-\-client option.
2048For example, here is an example "host.list" file containing 2 hostnames:
2049
2050host1.your.dns.domain
2051.br
2052host2.your.dns.domain
2053
2054The fio command would then be:
2055
2056fio \-\-client=host.list <job file>
2057
2058In this mode, you cannot input server-specific parameters or job files, and all
2059servers receive the same job file.
2060
2061In order to enable fio \-\-client runs utilizing a shared filesystem from multiple hosts,
ff6bb260
SL
2062fio \-\-client now prepends the IP address of the server to the filename. For example,
2063if fio is using directory /mnt/nfs/fio and is writing filename fileio.tmp,
39b5f61e
BE
2064with a \-\-client hostfile
2065containing two hostnames h1 and h2 with IP addresses 192.168.10.120 and 192.168.10.121, then
2066fio will create two files:
2067
2068/mnt/nfs/fio/192.168.10.120.fileio.tmp
2069.br
2070/mnt/nfs/fio/192.168.10.121.fileio.tmp
2071
d60e92d1 2072.SH AUTHORS
49da1240 2073
d60e92d1 2074.B fio
aa58d252 2075was written by Jens Axboe <jens.axboe@oracle.com>,
f8b8f7da 2076now Jens Axboe <axboe@fb.com>.
d1429b5c
AC
2077.br
2078This man page was written by Aaron Carroll <aaronc@cse.unsw.edu.au> based
d60e92d1
AC
2079on documentation by Jens Axboe.
2080.SH "REPORTING BUGS"
482900c9 2081Report bugs to the \fBfio\fR mailing list <fio@vger.kernel.org>.
d1429b5c 2082See \fBREADME\fR.
d60e92d1 2083.SH "SEE ALSO"
d1429b5c
AC
2084For further documentation see \fBHOWTO\fR and \fBREADME\fR.
2085.br
2086Sample jobfiles are available in the \fBexamples\fR directory.