blob: f74360dee4becc4298bf7e7b8685a284550e552d [file] [log] [blame]
Jens Axboe71bfa162006-10-25 11:08:19 +02001Table of contents
2-----------------
3
41. Overview
52. How fio works
63. Running fio
74. Job file format
85. Detailed list of parameters
96. Normal output
107. Terse output
Paul Dubs25c8b9d2011-07-21 17:26:02 +0200118. Trace file format
Bruce Cran43f09da2013-02-24 11:09:11 +0000129. CPU idleness profiling
Jens Axboe71bfa162006-10-25 11:08:19 +020013
141.0 Overview and history
15------------------------
16fio was originally written to save me the hassle of writing special test
17case programs when I wanted to test a specific workload, either for
18performance reasons or to find/reproduce a bug. The process of writing
19such a test app can be tiresome, especially if you have to do it often.
20Hence I needed a tool that would be able to simulate a given io workload
21without resorting to writing a tailored test case again and again.
22
23A test work load is difficult to define, though. There can be any number
24of processes or threads involved, and they can each be using their own
25way of generating io. You could have someone dirtying large amounts of
26memory in an memory mapped file, or maybe several threads issuing
27reads using asynchronous io. fio needed to be flexible enough to
28simulate both of these cases, and many more.
29
302.0 How fio works
31-----------------
32The first step in getting fio to simulate a desired io workload, is
33writing a job file describing that specific setup. A job file may contain
34any number of threads and/or files - the typical contents of the job file
35is a global section defining shared parameters, and one or more job
36sections describing the jobs involved. When run, fio parses this file
37and sets everything up as described. If we break down a job from top to
38bottom, it contains the following basic parameters:
39
40 IO type Defines the io pattern issued to the file(s).
41 We may only be reading sequentially from this
42 file(s), or we may be writing randomly. Or even
43 mixing reads and writes, sequentially or randomly.
44
45 Block size In how large chunks are we issuing io? This may be
46 a single value, or it may describe a range of
47 block sizes.
48
49 IO size How much data are we going to be reading/writing.
50
51 IO engine How do we issue io? We could be memory mapping the
52 file, we could be using regular read/write, we
Jens Axboed0ff85d2007-02-14 01:19:41 +010053 could be using splice, async io, syslet, or even
Jens Axboe71bfa162006-10-25 11:08:19 +020054 SG (SCSI generic sg).
55
Jens Axboe6c219762006-11-03 15:51:45 +010056 IO depth If the io engine is async, how large a queuing
Jens Axboe71bfa162006-10-25 11:08:19 +020057 depth do we want to maintain?
58
59 IO type Should we be doing buffered io, or direct/raw io?
60
61 Num files How many files are we spreading the workload over.
62
63 Num threads How many threads or processes should we spread
64 this workload over.
Bruce Cran66c098b2012-11-27 12:16:07 +000065
Jens Axboe71bfa162006-10-25 11:08:19 +020066The above are the basic parameters defined for a workload, in addition
67there's a multitude of parameters that modify other aspects of how this
68job behaves.
69
70
713.0 Running fio
72---------------
73See the README file for command line parameters, there are only a few
74of them.
75
76Running fio is normally the easiest part - you just give it the job file
77(or job files) as parameters:
78
79$ fio job_file
80
81and it will start doing what the job_file tells it to do. You can give
82more than one job file on the command line, fio will serialize the running
83of those files. Internally that is the same as using the 'stonewall'
84parameter described the the parameter section.
85
Jens Axboeb4692822006-10-27 13:43:22 +020086If the job file contains only one job, you may as well just give the
87parameters on the command line. The command line parameters are identical
88to the job parameters, with a few extra that control global parameters
89(see README). For example, for the job file parameter iodepth=2, the
Jens Axboec2b1e752006-10-30 09:03:13 +010090mirror command line option would be --iodepth 2 or --iodepth=2. You can
91also use the command line for giving more than one job entry. For each
92--name option that fio sees, it will start a new job with that name.
93Command line entries following a --name entry will apply to that job,
94until there are no more entries or a new --name entry is seen. This is
95similar to the job file options, where each option applies to the current
96job until a new [] job entry is seen.
Jens Axboeb4692822006-10-27 13:43:22 +020097
Jens Axboe71bfa162006-10-25 11:08:19 +020098fio does not need to run as root, except if the files or devices specified
99in the job section requires that. Some other options may also be restricted,
Jens Axboe6c219762006-11-03 15:51:45 +0100100such as memory locking, io scheduler switching, and decreasing the nice value.
Jens Axboe71bfa162006-10-25 11:08:19 +0200101
102
1034.0 Job file format
104-------------------
105As previously described, fio accepts one or more job files describing
106what it is supposed to do. The job file format is the classic ini file,
107where the names enclosed in [] brackets define the job name. You are free
108to use any ascii name you want, except 'global' which has special meaning.
109A global section sets defaults for the jobs described in that file. A job
110may override a global section parameter, and a job file may even have
111several global sections if so desired. A job is only affected by a global
Jens Axboe65db0852007-02-20 10:22:01 +0100112section residing above it. If the first character in a line is a ';' or a
113'#', the entire line is discarded as a comment.
Jens Axboe71bfa162006-10-25 11:08:19 +0200114
Aaron Carroll3c54bc42008-10-07 11:25:38 +0200115So let's look at a really simple job file that defines two processes, each
Jens Axboeb22989b2009-07-17 22:29:23 +0200116randomly reading from a 128MB file.
Jens Axboe71bfa162006-10-25 11:08:19 +0200117
118; -- start job file --
119[global]
120rw=randread
121size=128m
122
123[job1]
124
125[job2]
126
127; -- end job file --
128
129As you can see, the job file sections themselves are empty as all the
130described parameters are shared. As no filename= option is given, fio
Jens Axboec2b1e752006-10-30 09:03:13 +0100131makes up a filename for each of the jobs as it sees fit. On the command
132line, this job would look as follows:
133
134$ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
135
Jens Axboe71bfa162006-10-25 11:08:19 +0200136
Aaron Carroll3c54bc42008-10-07 11:25:38 +0200137Let's look at an example that has a number of processes writing randomly
Jens Axboe71bfa162006-10-25 11:08:19 +0200138to files.
139
140; -- start job file --
141[random-writers]
142ioengine=libaio
143iodepth=4
144rw=randwrite
145bs=32k
146direct=0
147size=64m
148numjobs=4
149
150; -- end job file --
151
152Here we have no global section, as we only have one job defined anyway.
153We want to use async io here, with a depth of 4 for each file. We also
Jens Axboeb22989b2009-07-17 22:29:23 +0200154increased the buffer size used to 32KB and define numjobs to 4 to
Jens Axboe71bfa162006-10-25 11:08:19 +0200155fork 4 identical jobs. The result is 4 processes each randomly writing
Jens Axboeb22989b2009-07-17 22:29:23 +0200156to their own 64MB file. Instead of using the above job file, you could
Jens Axboeb4692822006-10-27 13:43:22 +0200157have given the parameters on the command line. For this case, you would
158specify:
159
160$ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
Jens Axboe71bfa162006-10-25 11:08:19 +0200161
Jens Axboe74929ac2009-08-05 11:42:37 +02001624.1 Environment variables
163-------------------------
164
Aaron Carroll3c54bc42008-10-07 11:25:38 +0200165fio also supports environment variable expansion in job files. Any
166substring of the form "${VARNAME}" as part of an option value (in other
167words, on the right of the `='), will be expanded to the value of the
168environment variable called VARNAME. If no such environment variable
169is defined, or VARNAME is the empty string, the empty string will be
170substituted.
171
172As an example, let's look at a sample fio invocation and job file:
173
174$ SIZE=64m NUMJOBS=4 fio jobfile.fio
175
176; -- start job file --
177[random-writers]
178rw=randwrite
179size=${SIZE}
180numjobs=${NUMJOBS}
181; -- end job file --
182
183This will expand to the following equivalent job file at runtime:
184
185; -- start job file --
186[random-writers]
187rw=randwrite
188size=64m
189numjobs=4
190; -- end job file --
191
Jens Axboe71bfa162006-10-25 11:08:19 +0200192fio ships with a few example job files, you can also look there for
193inspiration.
194
Jens Axboe74929ac2009-08-05 11:42:37 +02001954.2 Reserved keywords
196---------------------
197
198Additionally, fio has a set of reserved keywords that will be replaced
199internally with the appropriate value. Those keywords are:
200
201$pagesize The architecture page size of the running system
202$mb_memory Megabytes of total memory in the system
203$ncpus Number of online available CPUs
204
205These can be used on the command line or in the job file, and will be
206automatically substituted with the current system values when the job
Jens Axboe892a6ff2009-11-13 12:19:49 +0100207is run. Simple math is also supported on these keywords, so you can
208perform actions like:
209
210size=8*$mb_memory
211
212and get that properly expanded to 8 times the size of memory in the
213machine.
Jens Axboe74929ac2009-08-05 11:42:37 +0200214
Jens Axboe71bfa162006-10-25 11:08:19 +0200215
2165.0 Detailed list of parameters
217-------------------------------
218
219This section describes in details each parameter associated with a job.
220Some parameters take an option of a given type, such as an integer or
221a string. The following types are used:
222
223str String. This is a sequence of alpha characters.
Jens Axboeb09da8f2009-07-17 23:16:17 +0200224time Integer with possible time suffix. In seconds unless otherwise
Jens Axboee417fd62008-09-11 09:27:15 +0200225 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
Jens Axboe0de5b262014-02-21 15:26:01 -0800226 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
227 and 'us' (or 'usec') for microseconds.
Jens Axboeb09da8f2009-07-17 23:16:17 +0200228int SI integer. A whole number value, which may contain a suffix
229 describing the base of the number. Accepted suffixes are k/m/g/t/p,
230 meaning kilo, mega, giga, tera, and peta. The suffix is not case
Jens Axboe57fc29f2010-06-23 22:24:07 +0200231 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
232 as 'k'). So if you want to specify 4096, you could either write
Jens Axboeb09da8f2009-07-17 23:16:17 +0200233 out '4096' or just give 4k. The suffixes signify base 2 values, so
Jens Axboe57fc29f2010-06-23 22:24:07 +0200234 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
235 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
236 case, then 1000 is used as the multiplier. This can be handy for
237 disks, since manufacturers generally use base 10 values when listing
238 the capacity of a drive. If the option accepts an upper and lower
239 range, use a colon ':' or minus '-' to separate such values. May also
240 include a prefix to indicate numbers base. If 0x is used, the number
241 is assumed to be hexadecimal. See irange.
Jens Axboe71bfa162006-10-25 11:08:19 +0200242bool Boolean. Usually parsed as an integer, however only defined for
243 true and false (1 and 0).
Jens Axboeb09da8f2009-07-17 23:16:17 +0200244irange Integer range with suffix. Allows value range to be given, such
Jens Axboebf9a3ed2008-06-05 11:53:08 +0200245 as 1024-4096. A colon may also be used as the separator, eg
Jens Axboe0c9baf92007-01-11 15:59:26 +0100246 1k:4k. If the option allows two sets of ranges, they can be
247 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
Jens Axboef7fa2652009-03-09 14:20:20 +0100248 int.
Yu-ju Hong83349192011-08-13 00:53:44 +0200249float_list A list of floating numbers, separated by a ':' character.
Jens Axboe71bfa162006-10-25 11:08:19 +0200250
251With the above in mind, here follows the complete list of fio job
252parameters.
253
254name=str ASCII name of the job. This may be used to override the
255 name printed by fio for this job. Otherwise the job
Jens Axboec2b1e752006-10-30 09:03:13 +0100256 name is used. On the command line this parameter has the
Jens Axboe6c219762006-11-03 15:51:45 +0100257 special purpose of also signaling the start of a new
Jens Axboec2b1e752006-10-30 09:03:13 +0100258 job.
Jens Axboe71bfa162006-10-25 11:08:19 +0200259
Jens Axboe61697c32007-02-05 15:04:46 +0100260description=str Text description of the job. Doesn't do anything except
261 dump this text description when this job is run. It's
262 not parsed.
263
Randy Dunlap37760412009-05-13 07:51:05 +0200264directory=str Prefix filenames with this directory. Used to place files
Jens Axboe67445b62014-03-12 10:49:36 -0600265 in a different location than "./". See the 'filename' option
266 for escaping certain characters.
Jens Axboe71bfa162006-10-25 11:08:19 +0200267
268filename=str Fio normally makes up a filename based on the job name,
269 thread number, and file number. If you want to share
270 files between threads in a job or several jobs, specify
Jens Axboeed92ac02007-02-06 14:43:52 +0100271 a filename for each of them to override the default. If
Jens Axboe414c2a32009-01-16 13:21:15 +0100272 the ioengine used is 'net', the filename is the host, port,
Jens Axboe0fd666b2011-10-06 20:08:53 +0200273 and protocol to use in the format of =host,port,protocol.
Jens Axboe414c2a32009-01-16 13:21:15 +0100274 See ioengine=net for more. If the ioengine is file based, you
275 can specify a number of files by separating the names with a
276 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
277 as the two working files, you would use
Jens Axboe30a45882013-01-30 12:53:55 +0100278 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
279 accessed as \\.\PhysicalDrive0 for the first device,
280 \\.\PhysicalDrive1 for the second etc. Note: Windows and
281 FreeBSD prevent write access to areas of the disk containing
282 in-use data (e.g. filesystems).
283 If the wanted filename does need to include a colon, then
284 escape that with a '\' character. For instance, if the filename
285 is "/dev/dsk/foo@3,0:c", then you would use
286 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
287 stdin or stdout. Which of the two depends on the read/write
288 direction set.
Jens Axboe71bfa162006-10-25 11:08:19 +0200289
Jens Axboede98bd32013-04-05 11:09:20 +0200290filename_format=str
291 If sharing multiple files between jobs, it is usually necessary
292 to have fio generate the exact names that you want. By default,
293 fio will name a file based on the default file format
294 specification of jobname.jobnumber.filenumber. With this
295 option, that can be customized. Fio will recognize and replace
296 the following keywords in this string:
297
298 $jobname
299 The name of the worker thread or process.
300
301 $jobnum
302 The incremental number of the worker thread or
303 process.
304
305 $filenum
306 The incremental number of the file for that worker
307 thread or process.
308
309 To have dependent jobs share a set of files, this option can
310 be set to have fio generate filenames that are shared between
311 the two. For instance, if testfiles.$filenum is specified,
312 file number 4 for any job will be named testfiles.4. The
313 default of $jobname.$jobnum.$filenum will be used if
314 no other format specifier is given.
315
Jens Axboebbf6b542007-03-13 15:28:55 +0100316opendir=str Tell fio to recursively add any file it can find in this
317 directory and down the file system tree.
318
Randy Dunlap37760412009-05-13 07:51:05 +0200319lockfile=str Fio defaults to not locking any files before it does
Jens Axboe4d4e80f2008-03-04 10:18:56 +0100320 IO to them. If a file or file descriptor is shared, fio
321 can serialize IO to that file to make the end result
322 consistent. This is usual for emulating real workloads that
323 share files. The lock modes are:
Jens Axboe29c13492008-03-01 19:25:20 +0100324
Jens Axboe4d4e80f2008-03-04 10:18:56 +0100325 none No locking. The default.
326 exclusive Only one thread/process may do IO,
327 excluding all others.
328 readwrite Read-write locking on the file. Many
329 readers may access the file at the
330 same time, but writes get exclusive
331 access.
332
Jens Axboed3aad8f2007-03-15 14:12:05 +0100333readwrite=str
Jens Axboe71bfa162006-10-25 11:08:19 +0200334rw=str Type of io pattern. Accepted values are:
335
336 read Sequential reads
337 write Sequential writes
338 randwrite Random writes
339 randread Random reads
Jens Axboe10b023d2012-03-23 13:40:06 +0100340 rw,readwrite Sequential mixed reads and writes
Jens Axboe71bfa162006-10-25 11:08:19 +0200341 randrw Random mixed reads and writes
342
343 For the mixed io types, the default is to split them 50/50.
344 For certain types of io the result may still be skewed a bit,
Jens Axboe211097b2007-03-22 18:56:45 +0100345 since the speed may be different. It is possible to specify
Jens Axboe38dad622010-07-20 14:46:00 -0600346 a number of IO's to do before getting a new offset, this is
347 one by appending a ':<nr>' to the end of the string given.
348 For a random read, it would look like 'rw=randread:8' for
Jens Axboe059b0802011-08-25 09:09:37 +0200349 passing in an offset modifier with a value of 8. If the
Lucian Adrian Grijincuddb754d2012-04-05 18:18:35 -0600350 suffix is used with a sequential IO pattern, then the value
Jens Axboe059b0802011-08-25 09:09:37 +0200351 specified will be added to the generated offset for each IO.
352 For instance, using rw=write:4k will skip 4k for every
353 write. It turns sequential IO into sequential IO with holes.
354 See the 'rw_sequencer' option.
Jens Axboe38dad622010-07-20 14:46:00 -0600355
356rw_sequencer=str If an offset modifier is given by appending a number to
357 the rw=<str> line, then this option controls how that
358 number modifies the IO offset being generated. Accepted
359 values are:
360
361 sequential Generate sequential offset
362 identical Generate the same offset
363
364 'sequential' is only useful for random IO, where fio would
365 normally generate a new random offset for every IO. If you
366 append eg 8 to randread, you would get a new random offset for
Jens Axboe211097b2007-03-22 18:56:45 +0100367 every 8 IO's. The result would be a seek for only every 8
368 IO's, instead of for every IO. Use rw=randread:8 to specify
Jens Axboe38dad622010-07-20 14:46:00 -0600369 that. As sequential IO is already sequential, setting
370 'sequential' for that would not result in any differences.
371 'identical' behaves in a similar fashion, except it sends
372 the same offset 8 number of times before generating a new
373 offset.
Jens Axboe71bfa162006-10-25 11:08:19 +0200374
Jens Axboe90fef2d2009-07-17 22:33:32 +0200375kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
376 Storage manufacturers like to use 10^3 or 1000 as a base
377 ten unit instead, for obvious reasons. Allow values are
378 1024 or 1000, with 1024 being the default.
379
Jens Axboe771e58b2013-01-30 12:56:23 +0100380unified_rw_reporting=bool Fio normally reports statistics on a per
381 data direction basis, meaning that read, write, and trim are
382 accounted and reported separately. If this option is set,
383 the fio will sum the results and report them as "mixed"
384 instead.
385
Jens Axboeee738492007-01-10 11:23:16 +0100386randrepeat=bool For random IO workloads, seed the generator in a predictable
387 way so that results are repeatable across repetitions.
388
Jens Axboe04778ba2014-01-10 20:57:01 -0700389randseed=int Seed the random number generators based on this seed value, to
390 be able to control what sequence of output is being generated.
391 If not set, the random sequence depends on the randrepeat
392 setting.
393
Jens Axboe2615cc42011-03-28 09:35:09 +0200394use_os_rand=bool Fio can either use the random generator supplied by the OS
395 to generator random offsets, or it can use it's own internal
396 generator (based on Tausworthe). Default is to use the
397 internal generator, which is often of better quality and
398 faster.
399
Eric Gourioua596f042011-06-17 09:11:45 +0200400fallocate=str Whether pre-allocation is performed when laying down files.
401 Accepted values are:
402
403 none Do not pre-allocate space
404 posix Pre-allocate via posix_fallocate()
405 keep Pre-allocate via fallocate() with
406 FALLOC_FL_KEEP_SIZE set
407 0 Backward-compatible alias for 'none'
408 1 Backward-compatible alias for 'posix'
409
410 May not be available on all supported platforms. 'keep' is only
411 available on Linux.If using ZFS on Solaris this must be set to
412 'none' because ZFS doesn't support it. Default: 'posix'.
Jens Axboe7bc8c2c2010-01-28 11:31:31 +0100413
Jens Axboed2f3ac32007-03-22 19:24:09 +0100414fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
415 on what IO patterns it is likely to issue. Sometimes you
416 want to test specific IO patterns without telling the
417 kernel about it, in which case you can disable this option.
418 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
419 IO and POSIX_FADV_RANDOM for random IO.
420
Jens Axboef7fa2652009-03-09 14:20:20 +0100421size=int The total size of file io for this job. Fio will run until
Jens Axboe7616caf2007-05-25 09:26:05 +0200422 this many bytes has been transferred, unless runtime is
423 limited by other options (such as 'runtime', for instance).
Randy Dunlap37760412009-05-13 07:51:05 +0200424 Unless specific nrfiles and filesize options are given,
Jens Axboe7616caf2007-05-25 09:26:05 +0200425 fio will divide this size between the available files
Jens Axboed6667262010-06-25 11:32:48 +0200426 specified by the job. If not set, fio will use the full
427 size of the given files or devices. If the the files
Jens Axboe7bb59102011-07-12 19:47:03 +0200428 do not exist, size must be given. It is also possible to
429 give size as a percentage between 1 and 100. If size=20%
430 is given, fio will use 20% of the full size of the given
431 files or devices.
Jens Axboe71bfa162006-10-25 11:08:19 +0200432
Jens Axboef7fa2652009-03-09 14:20:20 +0100433filesize=int Individual file sizes. May be a range, in which case fio
Jens Axboe9c60ce62007-03-15 09:14:47 +0100434 will select sizes for files at random within the given range
435 and limited to 'size' in total (if that is given). If not
436 given, each created file is the same size.
437
Jens Axboebedc9dc2014-03-17 12:51:09 -0600438file_append=bool Perform IO after the end of the file. Normally fio will
439 operate within the size of a file. If this option is set, then
440 fio will append to the file instead. This has identical
Jens Axboe0aae4ce2014-03-17 12:55:08 -0600441 behavior to setting offset to the size of a file. This option
442 is ignored on non-regular files.
Jens Axboebedc9dc2014-03-17 12:51:09 -0600443
Jens Axboe74586c12011-01-20 10:16:03 -0700444fill_device=bool
445fill_fs=bool Sets size to something really large and waits for ENOSPC (no
Shawn Lewisaa31f1f2008-01-11 09:45:11 +0100446 space left on device) as the terminating condition. Only makes
Jens Axboede98bd32013-04-05 11:09:20 +0200447 sense with sequential write. For a read workload, the mount
Jens Axboe4f124322011-01-19 15:35:26 -0700448 point will be filled first then IO started on the result. This
449 option doesn't make sense if operating on a raw device node,
450 since the size of that is already known by the file system.
451 Additionally, writing beyond end-of-device will not return
452 ENOSPC there.
Shawn Lewisaa31f1f2008-01-11 09:45:11 +0100453
Jens Axboef7fa2652009-03-09 14:20:20 +0100454blocksize=int
455bs=int The block size used for the io units. Defaults to 4k. Values
456 can be given for both read and writes. If a single int is
457 given, it will apply to both. If a second int is specified
Jens Axboef90eff52006-11-06 11:08:21 +0100458 after a comma, it will apply to writes only. In other words,
Jens Axboed9472272013-07-25 10:20:45 -0600459 the format is either bs=read_and_write or bs=read,write,trim.
460 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
461 writes, and 8k for trims. You can terminate the list with
462 a trailing comma. bs=4k,8k, would use the default value for
463 trims.. If you only wish to set the write size, you
Jens Axboe787f7e92006-11-06 13:26:29 +0100464 can do so by passing an empty read size - bs=,8k will set
465 8k for writes and leave the read default value.
Jens Axboea00735e2006-11-03 08:58:08 +0100466
Jens Axboe2b7a01d2009-03-11 11:00:13 +0100467blockalign=int
468ba=int At what boundary to align random IO offsets. Defaults to
469 the same as 'blocksize' the minimum blocksize given.
470 Minimum alignment is typically 512b for using direct IO,
471 though it usually depends on the hardware block size. This
472 option is mutually exclusive with using a random map for
473 files, so it will turn off that option.
474
Jens Axboed3aad8f2007-03-15 14:12:05 +0100475blocksize_range=irange
Jens Axboe71bfa162006-10-25 11:08:19 +0200476bsrange=irange Instead of giving a single block size, specify a range
477 and fio will mix the issued io block sizes. The issued
478 io unit will always be a multiple of the minimum value
Jens Axboef90eff52006-11-06 11:08:21 +0100479 given (also see bs_unaligned). Applies to both reads and
480 writes, however a second range can be given after a comma.
481 See bs=.
Jens Axboea00735e2006-11-03 08:58:08 +0100482
Jens Axboe564ca972007-12-14 12:21:19 +0100483bssplit=str Sometimes you want even finer grained control of the
484 block sizes issued, not just an even split between them.
485 This option allows you to weight various block sizes,
486 so that you are able to define a specific amount of
487 block sizes issued. The format for this option is:
488
489 bssplit=blocksize/percentage:blocksize/percentage
490
491 for as many block sizes as needed. So if you want to define
492 a workload that has 50% 64k blocks, 10% 4k blocks, and
493 40% 32k blocks, you would write:
494
495 bssplit=4k/10:64k/50:32k/40
496
497 Ordering does not matter. If the percentage is left blank,
498 fio will fill in the remaining values evenly. So a bssplit
499 option like this one:
500
501 bssplit=4k/50:1k/:32k/
502
503 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
504 always add up to 100, if bssplit is given a range that adds
505 up to more, it will error out.
506
Jens Axboe720e84a2009-04-21 08:29:55 +0200507 bssplit also supports giving separate splits to reads and
508 writes. The format is identical to what bs= accepts. You
509 have to separate the read and write parts with a comma. So
510 if you want a workload that has 50% 2k reads and 50% 4k reads,
511 while having 90% 4k writes and 10% 8k writes, you would
512 specify:
513
514 bssplit=2k/50:4k/50,4k/90,8k/10
515
Jens Axboed3aad8f2007-03-15 14:12:05 +0100516blocksize_unaligned
Jens Axboe690adba2006-10-30 15:25:09 +0100517bs_unaligned If this option is given, any byte size value within bsrange
518 may be used as a block range. This typically wont work with
519 direct IO, as that normally requires sector alignment.
Jens Axboe71bfa162006-10-25 11:08:19 +0200520
Jens Axboe6aca9b32013-07-25 12:45:26 -0600521bs_is_seq_rand If this option is set, fio will use the normal read,write
522 blocksize settings as sequential,random instead. Any random
523 read or write will use the WRITE blocksize settings, and any
524 sequential read or write will use the READ blocksize setting.
525
Jens Axboee9459e52007-04-17 15:46:32 +0200526zero_buffers If this option is given, fio will init the IO buffers to
527 all zeroes. The default is to fill them with random data.
Jens Axboe7750aac2014-03-14 19:41:07 -0600528 The resulting IO buffers will not be completely zeroed,
529 unless scramble_buffers is also turned off.
Jens Axboee9459e52007-04-17 15:46:32 +0200530
Jens Axboe5973caf2008-05-21 19:52:35 +0200531refill_buffers If this option is given, fio will refill the IO buffers
532 on every submit. The default is to only fill it at init
533 time and reuse that data. Only makes sense if zero_buffers
Jens Axboe41ccd842008-05-22 09:17:33 +0200534 isn't specified, naturally. If data verification is enabled,
535 refill_buffers is also automatically enabled.
Jens Axboe5973caf2008-05-21 19:52:35 +0200536
Jens Axboefd684182011-09-19 09:24:44 +0200537scramble_buffers=bool If refill_buffers is too costly and the target is
538 using data deduplication, then setting this option will
539 slightly modify the IO buffer contents to defeat normal
540 de-dupe attempts. This is not enough to defeat more clever
541 block compression attempts, but it will stop naive dedupe of
542 blocks. Default: true.
543
Jens Axboec5751c62012-03-15 15:02:56 +0100544buffer_compress_percentage=int If this is set, then fio will attempt to
545 provide IO buffer content (on WRITEs) that compress to
546 the specified level. Fio does this by providing a mix of
547 random data and zeroes. Note that this is per block size
548 unit, for file/disk wide compression level that matches
549 this setting, you'll also want to set refill_buffers.
550
551buffer_compress_chunk=int See buffer_compress_percentage. This
552 setting allows fio to manage how big the ranges of random
553 data and zeroed data is. Without this set, fio will
554 provide buffer_compress_percentage of blocksize random
555 data, followed by the remaining zeroed. With this set
556 to some chunk size smaller than the block size, fio can
557 alternate random and zeroed data throughout the IO
558 buffer.
559
Jens Axboece35b1e2014-01-14 15:35:58 -0700560buffer_pattern=str If set, fio will fill the io buffers with this pattern.
561 If not set, the contents of io buffers is defined by the other
562 options related to buffer contents. The setting can be any
563 pattern of bytes, and can be prefixed with 0x for hex values.
564
Jens Axboe71bfa162006-10-25 11:08:19 +0200565nrfiles=int Number of files to use for this job. Defaults to 1.
566
Jens Axboe390b1532007-03-09 13:03:00 +0100567openfiles=int Number of files to keep open at the same time. Defaults to
568 the same as nrfiles, can be set smaller to limit the number
569 simultaneous opens.
570
Jens Axboe5af1c6f2007-03-01 10:06:10 +0100571file_service_type=str Defines how fio decides which file from a job to
572 service next. The following types are defined:
573
574 random Just choose a file at random.
575
576 roundrobin Round robin over open files. This
577 is the default.
578
Jens Axboea086c252009-03-04 08:27:37 +0100579 sequential Finish one file before moving on to
580 the next. Multiple files can still be
581 open depending on 'openfiles'.
582
Jens Axboe1907dbc2007-03-12 11:44:28 +0100583 The string can have a number appended, indicating how
584 often to switch to a new file. So if option random:4 is
585 given, fio will switch to a new random file after 4 ios
586 have been issued.
587
Jens Axboe71bfa162006-10-25 11:08:19 +0200588ioengine=str Defines how the job issues io to the file. The following
589 types are defined:
590
591 sync Basic read(2) or write(2) io. lseek(2) is
592 used to position the io location.
593
gurudas paia31041e2007-10-23 15:12:30 +0200594 psync Basic pread(2) or pwrite(2) io.
595
Gurudas Paie05af9e2008-02-06 11:16:15 +0100596 vsync Basic readv(2) or writev(2) IO.
Jens Axboe1d2af022008-02-04 10:59:07 +0100597
Jens Axboea46c5e02013-05-16 20:38:09 +0200598 psyncv Basic preadv(2) or pwritev(2) IO.
599
Jens Axboe15d182a2009-01-16 19:15:07 +0100600 libaio Linux native asynchronous io. Note that Linux
601 may only support queued behaviour with
602 non-buffered IO (set direct=1 or buffered=0).
Steven Langde890a12011-11-09 14:03:34 +0100603 This engine defines engine specific options.
Jens Axboe71bfa162006-10-25 11:08:19 +0200604
605 posixaio glibc posix asynchronous io.
606
Jens Axboe417f0062008-06-02 11:59:30 +0200607 solarisaio Solaris native asynchronous io.
608
Bruce Cran03e20d62011-01-02 20:14:54 +0100609 windowsaio Windows native asynchronous io.
610
Jens Axboe71bfa162006-10-25 11:08:19 +0200611 mmap File is memory mapped and data copied
612 to/from using memcpy(3).
613
614 splice splice(2) is used to transfer the data and
615 vmsplice(2) to transfer data from user
616 space to the kernel.
617
Jens Axboed0ff85d2007-02-14 01:19:41 +0100618 syslet-rw Use the syslet system calls to make
619 regular read/write async.
620
Jens Axboe71bfa162006-10-25 11:08:19 +0200621 sg SCSI generic sg v3 io. May either be
Jens Axboe6c219762006-11-03 15:51:45 +0100622 synchronous using the SG_IO ioctl, or if
Jens Axboe71bfa162006-10-25 11:08:19 +0200623 the target is an sg character device
624 we use read(2) and write(2) for asynchronous
625 io.
626
Jens Axboea94ea282006-11-24 12:37:34 +0100627 null Doesn't transfer any data, just pretends
628 to. This is mainly used to exercise fio
629 itself and for debugging/testing purposes.
630
Jens Axboeed92ac02007-02-06 14:43:52 +0100631 net Transfer over the network to given host:port.
Steven Langde890a12011-11-09 14:03:34 +0100632 Depending on the protocol used, the hostname,
633 port, listen and filename options are used to
634 specify what sort of connection to make, while
635 the protocol option determines which protocol
636 will be used.
637 This engine defines engine specific options.
Jens Axboeed92ac02007-02-06 14:43:52 +0100638
Jens Axboe9cce02e2007-06-22 15:42:21 +0200639 netsplice Like net, but uses splice/vmsplice to
640 map data and send/receive.
Steven Langde890a12011-11-09 14:03:34 +0100641 This engine defines engine specific options.
Jens Axboe9cce02e2007-06-22 15:42:21 +0200642
gurudas pai53aec0a2007-10-05 13:20:18 +0200643 cpuio Doesn't transfer any data, but burns CPU
Jens Axboeba0fbe12007-03-09 14:34:23 +0100644 cycles according to the cpuload= and
645 cpucycle= options. Setting cpuload=85
646 will cause that job to do nothing but burn
Gurudas Pai36ecec82008-02-08 08:50:14 +0100647 85% of the CPU. In case of SMP machines,
648 use numjobs=<no_of_cpu> to get desired CPU
649 usage, as the cpuload only loads a single
650 CPU at the desired rate.
Jens Axboeba0fbe12007-03-09 14:34:23 +0100651
Jens Axboee9a18062007-03-21 08:51:56 +0100652 guasi The GUASI IO engine is the Generic Userspace
653 Asyncronous Syscall Interface approach
654 to async IO. See
655
656 http://www.xmailserver.org/guasi-lib.html
657
658 for more info on GUASI.
659
ren yufei21b8aee2011-08-01 10:01:57 +0200660 rdma The RDMA I/O engine supports both RDMA
Bart Van Asscheeb52fa32011-08-15 09:01:05 +0200661 memory semantics (RDMA_WRITE/RDMA_READ) and
662 channel semantics (Send/Recv) for the
663 InfiniBand, RoCE and iWARP protocols.
ren yufei21b8aee2011-08-01 10:01:57 +0200664
Dmitry Monakhovd54fce82012-09-20 15:37:17 +0400665 falloc IO engine that does regular fallocate to
666 simulate data transfer as fio ioengine.
667 DDIR_READ does fallocate(,mode = keep_size,)
Jens Axboe0981fd72012-09-20 19:23:02 +0200668 DDIR_WRITE does fallocate(,mode = 0)
Dmitry Monakhovd54fce82012-09-20 15:37:17 +0400669 DDIR_TRIM does fallocate(,mode = punch_hole)
670
671 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
672 ioctls to simulate defragment activity in
673 request to DDIR_WRITE event
Jens Axboe0981fd72012-09-20 19:23:02 +0200674
Jens Axboe8a7bd872007-02-28 11:12:25 +0100675 external Prefix to specify loading an external
676 IO engine object file. Append the engine
677 filename, eg ioengine=external:/tmp/foo.o
678 to load ioengine foo.o in /tmp.
679
Jens Axboe71bfa162006-10-25 11:08:19 +0200680iodepth=int This defines how many io units to keep in flight against
681 the file. The default is 1 for each file defined in this
682 job, can be overridden with a larger value for higher
Jens Axboeee72ca02010-12-02 20:05:37 +0100683 concurrency. Note that increasing iodepth beyond 1 will not
684 affect synchronous ioengines (except for small degress when
Bruce Cran9b836562011-01-08 19:49:54 +0100685 verify_async is in use). Even async engines may impose OS
Jens Axboeee72ca02010-12-02 20:05:37 +0100686 restrictions causing the desired depth not to be achieved.
687 This may happen on Linux when using libaio and not setting
688 direct=1, since buffered IO is not async on that OS. Keep an
689 eye on the IO depth distribution in the fio output to verify
690 that the achieved depth is as expected. Default: 1.
Jens Axboe71bfa162006-10-25 11:08:19 +0200691
Jens Axboe49504212008-06-05 09:03:30 +0200692iodepth_batch_submit=int
Jens Axboecb5ab512007-02-26 12:57:09 +0100693iodepth_batch=int This defines how many pieces of IO to submit at once.
Jens Axboe89e820f2008-01-18 10:30:07 +0100694 It defaults to 1 which means that we submit each IO
695 as soon as it is available, but can be raised to submit
696 bigger batches of IO at the time.
Jens Axboecb5ab512007-02-26 12:57:09 +0100697
Jens Axboe49504212008-06-05 09:03:30 +0200698iodepth_batch_complete=int This defines how many pieces of IO to retrieve
699 at once. It defaults to 1 which means that we'll ask
700 for a minimum of 1 IO in the retrieval process from
701 the kernel. The IO retrieval will go on until we
702 hit the limit set by iodepth_low. If this variable is
703 set to 0, then fio will always check for completed
704 events before queuing more IO. This helps reduce
705 IO latency, at the cost of more retrieval system calls.
706
Jens Axboee916b392007-02-20 14:37:26 +0100707iodepth_low=int The low water mark indicating when to start filling
708 the queue again. Defaults to the same as iodepth, meaning
709 that fio will attempt to keep the queue full at all times.
710 If iodepth is set to eg 16 and iodepth_low is set to 4, then
711 after fio has filled the queue of 16 requests, it will let
712 the depth drain down to 4 before starting to fill it again.
713
Jens Axboe71bfa162006-10-25 11:08:19 +0200714direct=bool If value is true, use non-buffered io. This is usually
Bruce Cran9b836562011-01-08 19:49:54 +0100715 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
Bruce Cran93bcfd22012-02-20 20:18:19 +0100716 On Windows the synchronous ioengines don't support direct io.
Jens Axboe76a43db2007-01-11 13:24:44 +0100717
Chris Masond01612f2013-11-15 15:52:58 -0700718atomic=bool If value is true, attempt to use atomic direct IO. Atomic
719 writes are guaranteed to be stable once acknowledged by
720 the operating system. Only Linux supports O_ATOMIC right
721 now.
722
Jens Axboe76a43db2007-01-11 13:24:44 +0100723buffered=bool If value is true, use buffered io. This is the opposite
724 of the 'direct' option. Defaults to true.
Jens Axboe71bfa162006-10-25 11:08:19 +0200725
Jens Axboef7fa2652009-03-09 14:20:20 +0100726offset=int Start io at the given offset in the file. The data before
Jens Axboe71bfa162006-10-25 11:08:19 +0200727 the given offset will not be touched. This effectively
728 caps the file size at real_size - offset.
729
Dan Ehrenberg214ac7e2012-03-15 14:44:26 +0100730offset_increment=int If this is provided, then the real offset becomes
731 the offset + offset_increment * thread_number, where the
732 thread number is a counter that starts at 0 and is incremented
733 for each job. This option is useful if there are several jobs
734 which are intended to operate on a file in parallel in disjoint
735 segments, with even spacing between the starting points.
736
Jens Axboeddf24e42013-08-09 12:53:44 -0600737number_ios=int Fio will normally perform IOs until it has exhausted the size
738 of the region set by size=, or if it exhaust the allocated
739 time (or hits an error condition). With this setting, the
740 range/size can be set independently of the number of IOs to
741 perform. When fio reaches this number, it will exit normally
742 and report status.
743
Jens Axboe71bfa162006-10-25 11:08:19 +0200744fsync=int If writing to a file, issue a sync of the dirty data
745 for every number of blocks given. For example, if you give
746 32 as a parameter, fio will sync the file for every 32
747 writes issued. If fio is using non-buffered io, we may
748 not sync the file. The exception is the sg io engine, which
Jens Axboe6c219762006-11-03 15:51:45 +0100749 synchronizes the disk cache anyway.
Jens Axboe71bfa162006-10-25 11:08:19 +0200750
Jens Axboee76b1da2010-03-09 20:49:54 +0100751fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
Jens Axboe5f9099e2009-06-16 22:40:26 +0200752 metadata blocks.
Bruce Cran93bcfd22012-02-20 20:18:19 +0100753 In FreeBSD and Windows there is no fdatasync(), this falls back to
Joshua Aunee72fa4d2010-02-11 00:59:18 -0700754 using fsync()
Jens Axboe5f9099e2009-06-16 22:40:26 +0200755
Jens Axboee76b1da2010-03-09 20:49:54 +0100756sync_file_range=str:val Use sync_file_range() for every 'val' number of
757 write operations. Fio will track range of writes that
758 have happened since the last sync_file_range() call. 'str'
759 can currently be one or more of:
760
761 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
762 write SYNC_FILE_RANGE_WRITE
763 wait_after SYNC_FILE_RANGE_WAIT_AFTER
764
765 So if you do sync_file_range=wait_before,write:8, fio would
766 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
767 every 8 writes. Also see the sync_file_range(2) man page.
768 This option is Linux specific.
769
Jens Axboe5036fc12008-04-15 09:20:46 +0200770overwrite=bool If true, writes to a file will always overwrite existing
771 data. If the file doesn't already exist, it will be
772 created before the write phase begins. If the file exists
773 and is large enough for the specified write phase, nothing
774 will be done.
Jens Axboe71bfa162006-10-25 11:08:19 +0200775
Jens Axboedbd11ea2013-01-13 17:16:46 +0100776end_fsync=bool If true, fsync file contents when a write stage has completed.
Jens Axboe71bfa162006-10-25 11:08:19 +0200777
Jens Axboeebb14152007-03-13 14:42:15 +0100778fsync_on_close=bool If true, fio will fsync() a dirty file on close.
779 This differs from end_fsync in that it will happen on every
780 file close, not just at the end of the job.
781
Jens Axboe71bfa162006-10-25 11:08:19 +0200782rwmixread=int How large a percentage of the mix should be reads.
783
784rwmixwrite=int How large a percentage of the mix should be writes. If both
785 rwmixread and rwmixwrite is given and the values do not add
786 up to 100%, the latter of the two will be used to override
Jens Axboec35dd7a2009-06-10 08:39:16 +0200787 the first. This may interfere with a given rate setting,
788 if fio is asked to limit reads or writes to a certain rate.
789 If that is the case, then the distribution may be skewed.
Jens Axboe71bfa162006-10-25 11:08:19 +0200790
Jens Axboe92d42d62012-11-15 15:38:32 -0700791random_distribution=str:float By default, fio will use a completely uniform
792 random distribution when asked to perform random IO. Sometimes
793 it is useful to skew the distribution in specific ways,
794 ensuring that some parts of the data is more hot than others.
795 fio includes the following distribution models:
796
797 random Uniform random distribution
798 zipf Zipf distribution
799 pareto Pareto distribution
800
801 When using a zipf or pareto distribution, an input value
802 is also needed to define the access pattern. For zipf, this
803 is the zipf theta. For pareto, it's the pareto power. Fio
804 includes a test program, genzipf, that can be used visualize
805 what the given input values will yield in terms of hit rates.
806 If you wanted to use zipf with a theta of 1.2, you would use
807 random_distribution=zipf:1.2 as the option. If a non-uniform
808 model is used, fio will disable use of the random map.
809
Jens Axboe211c9b82013-04-26 08:56:17 -0600810percentage_random=int For a random workload, set how big a percentage should
811 be random. This defaults to 100%, in which case the workload
812 is fully random. It can be set from anywhere from 0 to 100.
813 Setting it to 0 would make the workload fully sequential. Any
814 setting in between will result in a random mix of sequential
Jens Axboed9472272013-07-25 10:20:45 -0600815 and random IO, at the given percentages. It is possible to
816 set different values for reads, writes, and trim. To do so,
817 simply use a comma separated list. See blocksize.
Jens Axboe211c9b82013-04-26 08:56:17 -0600818
Jens Axboebb8895e2006-10-30 15:14:48 +0100819norandommap Normally fio will cover every block of the file when doing
820 random IO. If this option is given, fio will just get a
821 new random offset without looking at past io history. This
822 means that some blocks may not be read or written, and that
823 some blocks may be read/written more than once. This option
Jens Axboe83472392009-02-19 21:32:12 +0100824 is mutually exclusive with verify= if and only if multiple
825 blocksizes (via bsrange=) are used, since fio only tracks
826 complete rewrites of blocks.
Jens Axboebb8895e2006-10-30 15:14:48 +0100827
Jens Axboe0408c202011-08-08 09:07:28 +0200828softrandommap=bool See norandommap. If fio runs with the random block map
829 enabled and it fails to allocate the map, if this option is
830 set it will continue without a random block map. As coverage
831 will not be as complete as with random maps, this option is
Jens Axboe2b386d22008-03-26 10:32:57 +0100832 disabled by default.
833
Jens Axboee8b19612012-12-05 10:28:08 +0100834random_generator=str Fio supports the following engines for generating
835 IO offsets for random IO:
836
837 tausworthe Strong 2^88 cycle random number generator
838 lfsr Linear feedback shift register generator
839
840 Tausworthe is a strong random number generator, but it
841 requires tracking on the side if we want to ensure that
842 blocks are only read or written once. LFSR guarantees
843 that we never generate the same offset twice, and it's
844 also less computationally expensive. It's not a true
845 random generator, however, though for IO purposes it's
846 typically good enough. LFSR only works with single
847 block sizes, not with workloads that use multiple block
848 sizes. If used with such a workload, fio may read or write
849 some blocks multiple times.
Bruce Cran43f09da2013-02-24 11:09:11 +0000850
Jens Axboe71bfa162006-10-25 11:08:19 +0200851nice=int Run the job with the given nice value. See man nice(2).
852
853prio=int Set the io priority value of this job. Linux limits us to
854 a positive value between 0 and 7, with 0 being the highest.
855 See man ionice(1).
856
857prioclass=int Set the io priority class. See man ionice(1).
858
859thinktime=int Stall the job x microseconds after an io has completed before
860 issuing the next. May be used to simulate processing being
Jens Axboe48097d52007-02-17 06:30:44 +0100861 done by an application. See thinktime_blocks and
862 thinktime_spin.
863
864thinktime_spin=int
865 Only valid if thinktime is set - pretend to spend CPU time
866 doing something with the data received, before falling back
867 to sleeping for the rest of the period specified by
868 thinktime.
Jens Axboe9c1f7432007-01-03 20:43:19 +0100869
Jens Axboe4d01ece2013-05-17 12:47:11 +0200870thinktime_blocks=int
Jens Axboe9c1f7432007-01-03 20:43:19 +0100871 Only valid if thinktime is set - control how many blocks
872 to issue, before waiting 'thinktime' usecs. If not set,
873 defaults to 1 which will make fio wait 'thinktime' usecs
Jens Axboe4d01ece2013-05-17 12:47:11 +0200874 after every block. This effectively makes any queue depth
875 setting redundant, since no more than 1 IO will be queued
876 before we have to complete it and do our thinktime. In
877 other words, this setting effectively caps the queue depth
878 if the latter is larger.
Jens Axboe71bfa162006-10-25 11:08:19 +0200879
Jens Axboe581e7142009-06-09 12:47:16 +0200880rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
Jens Axboeb09da8f2009-07-17 23:16:17 +0200881 the normal suffix rules apply. You can use rate=500k to limit
Jens Axboe581e7142009-06-09 12:47:16 +0200882 reads and writes to 500k each, or you can specify read and
883 writes separately. Using rate=1m,500k would limit reads to
884 1MB/sec and writes to 500KB/sec. Capping only reads or
885 writes can be done with rate=,500k or rate=500k,. The former
886 will only limit writes (to 500KB/sec), the latter will only
887 limit reads.
Jens Axboe71bfa162006-10-25 11:08:19 +0200888
889ratemin=int Tell fio to do whatever it can to maintain at least this
Jens Axboe4e991c22007-03-15 11:41:11 +0100890 bandwidth. Failing to meet this requirement, will cause
Jens Axboe581e7142009-06-09 12:47:16 +0200891 the job to exit. The same format as rate is used for
892 read vs write separation.
Jens Axboe4e991c22007-03-15 11:41:11 +0100893
894rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
895 as rate, just specified independently of bandwidth. If the
896 job is given a block size range instead of a fixed value,
Jens Axboe581e7142009-06-09 12:47:16 +0200897 the smallest block size is used as the metric. The same format
Anatol Pomozovde8f6de2013-09-26 16:31:34 -0700898 as rate is used for read vs write separation.
Jens Axboe4e991c22007-03-15 11:41:11 +0100899
900rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
Jens Axboe581e7142009-06-09 12:47:16 +0200901 the job to exit. The same format as rate is used for read vs
Anatol Pomozovde8f6de2013-09-26 16:31:34 -0700902 write separation.
Jens Axboe71bfa162006-10-25 11:08:19 +0200903
Jens Axboe3e260a42013-12-09 12:38:53 -0700904latency_target=int If set, fio will attempt to find the max performance
905 point that the given workload will run at while maintaining a
906 latency below this target. The values is given in microseconds.
907 See latency_window and latency_percentile
908
909latency_window=int Used with latency_target to specify the sample window
910 that the job is run at varying queue depths to test the
911 performance. The value is given in microseconds.
912
913latency_percentile=float The percentage of IOs that must fall within the
914 criteria specified by latency_target and latency_window. If not
915 set, this defaults to 100.0, meaning that all IOs must be equal
916 or below to the value set by latency_target.
917
Jens Axboe15501532012-10-24 16:37:45 +0200918max_latency=int If set, fio will exit the job if it exceeds this maximum
919 latency. It will exit with an ETIME error.
920
Jens Axboe71bfa162006-10-25 11:08:19 +0200921ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
Jens Axboe6c219762006-11-03 15:51:45 +0100922 of milliseconds.
Jens Axboe71bfa162006-10-25 11:08:19 +0200923
924cpumask=int Set the CPU affinity of this job. The parameter given is a
Jens Axboea08bc172007-06-13 21:00:46 +0200925 bitmask of allowed CPU's the job may run on. So if you want
926 the allowed CPUs to be 1 and 5, you would pass the decimal
927 value of (1 << 1 | 1 << 5), or 34. See man
Jens Axboe7dbb6eb2007-05-22 09:13:31 +0200928 sched_setaffinity(2). This may not work on all supported
Jens Axboeb0ea08c2008-12-05 12:57:11 +0100929 operating systems or kernel versions. This option doesn't
930 work well for a higher CPU count than what you can store in
931 an integer mask, so it can only control cpus 1-32. For
932 boxes with larger CPU counts, use cpus_allowed.
Jens Axboe71bfa162006-10-25 11:08:19 +0200933
Jens Axboed2e268b2007-06-15 10:33:49 +0200934cpus_allowed=str Controls the same options as cpumask, but it allows a text
935 setting of the permitted CPUs instead. So to use CPUs 1 and
Jens Axboe62a72732008-12-08 11:37:01 +0100936 5, you would specify cpus_allowed=1,5. This options also
937 allows a range of CPUs. Say you wanted a binding to CPUs
938 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
Jens Axboed2e268b2007-06-15 10:33:49 +0200939
Jens Axboec2acfba2014-02-27 15:52:02 -0800940cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
941 specified by cpus_allowed or cpumask. Two policies are
942 supported:
943
944 shared All jobs will share the CPU set specified.
945 split Each job will get a unique CPU from the CPU set.
946
947 'shared' is the default behaviour, if the option isn't
Jens Axboeada083c2014-02-28 16:43:57 -0800948 specified. If split is specified, then fio will will assign
949 one cpu per job. If not enough CPUs are given for the jobs
950 listed, then fio will roundrobin the CPUs in the set.
Jens Axboec2acfba2014-02-27 15:52:02 -0800951
Yufei Rend0b937e2012-10-19 23:11:52 -0400952numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
953 arguments allow comma delimited list of cpu numbers,
954 A-B ranges, or 'all'. Note, to enable numa options support,
Jens Axboe67bf9822013-01-10 11:23:19 +0100955 fio must be built on a system with libnuma-dev(el) installed.
Yufei Rend0b937e2012-10-19 23:11:52 -0400956
957numa_mem_policy=str Set this job's memory policy and corresponding NUMA
958 nodes. Format of the argements:
959 <mode>[:<nodelist>]
960 `mode' is one of the following memory policy:
961 default, prefer, bind, interleave, local
962 For `default' and `local' memory policy, no node is
963 needed to be specified.
964 For `prefer', only one node is allowed.
965 For `bind' and `interleave', it allow comma delimited
966 list of numbers, A-B ranges, or 'all'.
967
Jens Axboee417fd62008-09-11 09:27:15 +0200968startdelay=time Start this job the specified number of seconds after fio
Jens Axboe71bfa162006-10-25 11:08:19 +0200969 has started. Only useful if the job file contains several
970 jobs, and you want to delay starting some jobs to a certain
971 time.
972
Jens Axboee417fd62008-09-11 09:27:15 +0200973runtime=time Tell fio to terminate processing after the specified number
Jens Axboe71bfa162006-10-25 11:08:19 +0200974 of seconds. It can be quite hard to determine for how long
975 a specified job will run, so this parameter is handy to
976 cap the total runtime to a given time.
977
Jens Axboecf4464c2007-04-17 20:14:42 +0200978time_based If set, fio will run for the duration of the runtime
Jens Axboebf9a3ed2008-06-05 11:53:08 +0200979 specified even if the file(s) are completely read or
Jens Axboecf4464c2007-04-17 20:14:42 +0200980 written. It will simply loop over the same workload
981 as many times as the runtime allows.
982
Jens Axboee417fd62008-09-11 09:27:15 +0200983ramp_time=time If set, fio will run the specified workload for this amount
Jens Axboe721938a2008-09-10 09:46:16 +0200984 of time before logging any performance numbers. Useful for
985 letting performance settle before logging results, thus
Jens Axboeb29ee5b2008-09-11 10:17:26 +0200986 minimizing the runtime required for stable results. Note
987 that the ramp_time is considered lead in time for a job,
988 thus it will increase the total runtime if a special timeout
989 or runtime is specified.
Jens Axboe721938a2008-09-10 09:46:16 +0200990
Jens Axboe71bfa162006-10-25 11:08:19 +0200991invalidate=bool Invalidate the buffer/page cache parts for this file prior
992 to starting io. Defaults to true.
993
994sync=bool Use sync io for buffered writes. For the majority of the
995 io engines, this means using O_SYNC.
996
Jens Axboed3aad8f2007-03-15 14:12:05 +0100997iomem=str
Jens Axboe71bfa162006-10-25 11:08:19 +0200998mem=str Fio can use various types of memory as the io unit buffer.
999 The allowed values are:
1000
1001 malloc Use memory from malloc(3) as the buffers.
1002
1003 shm Use shared memory as the buffers. Allocated
1004 through shmget(2).
1005
Jens Axboe74b025b2006-12-19 15:18:14 +01001006 shmhuge Same as shm, but use huge pages as backing.
1007
Jens Axboe313cb202006-12-21 09:50:00 +01001008 mmap Use mmap to allocate buffers. May either be
1009 anonymous memory, or can be file backed if
1010 a filename is given after the option. The
1011 format is mem=mmap:/path/to/file.
Jens Axboe71bfa162006-10-25 11:08:19 +02001012
Jens Axboed0bdaf42006-12-20 14:40:44 +01001013 mmaphuge Use a memory mapped huge file as the buffer
1014 backing. Append filename after mmaphuge, ala
1015 mem=mmaphuge:/hugetlbfs/file
1016
Jens Axboe71bfa162006-10-25 11:08:19 +02001017 The area allocated is a function of the maximum allowed
Jens Axboe5394ae52006-12-20 20:15:41 +01001018 bs size for the job, multiplied by the io depth given. Note
1019 that for shmhuge and mmaphuge to work, the system must have
1020 free huge pages allocated. This can normally be checked
1021 and set by reading/writing /proc/sys/vm/nr_hugepages on a
Jens Axboeb22989b2009-07-17 22:29:23 +02001022 Linux system. Fio assumes a huge page is 4MB in size. So
Jens Axboe5394ae52006-12-20 20:15:41 +01001023 to calculate the number of huge pages you need for a given
1024 job file, add up the io depth of all jobs (normally one unless
1025 iodepth= is used) and multiply by the maximum bs set. Then
1026 divide that number by the huge page size. You can see the
1027 size of the huge pages in /proc/meminfo. If no huge pages
1028 are allocated by having a non-zero number in nr_hugepages,
Jens Axboe56bb17f2006-12-20 20:27:36 +01001029 using mmaphuge or shmhuge will fail. Also see hugepage-size.
Jens Axboe5394ae52006-12-20 20:15:41 +01001030
1031 mmaphuge also needs to have hugetlbfs mounted and the file
1032 location should point there. So if it's mounted in /huge,
1033 you would use mem=mmaphuge:/huge/somefile.
Jens Axboe71bfa162006-10-25 11:08:19 +02001034
Jens Axboed529ee12009-07-01 10:33:03 +02001035iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1036 Note that the given alignment is applied to the first IO unit
1037 buffer, if using iodepth the alignment of the following buffers
1038 are given by the bs used. In other words, if using a bs that is
1039 a multiple of the page sized in the system, all buffers will
1040 be aligned to this value. If using a bs that is not page
1041 aligned, the alignment of subsequent IO memory buffers is the
1042 sum of the iomem_align and bs used.
1043
Jens Axboef7fa2652009-03-09 14:20:20 +01001044hugepage-size=int
Jens Axboe56bb17f2006-12-20 20:27:36 +01001045 Defines the size of a huge page. Must at least be equal
Jens Axboeb22989b2009-07-17 22:29:23 +02001046 to the system setting, see /proc/meminfo. Defaults to 4MB.
Jens Axboec51074e2006-12-20 20:28:33 +01001047 Should probably always be a multiple of megabytes, so using
1048 hugepage-size=Xm is the preferred way to set this to avoid
1049 setting a non-pow-2 bad value.
Jens Axboe56bb17f2006-12-20 20:27:36 +01001050
Jens Axboe71bfa162006-10-25 11:08:19 +02001051exitall When one job finishes, terminate the rest. The default is
1052 to wait for each job to finish, sometimes that is not the
1053 desired action.
1054
1055bwavgtime=int Average the calculated bandwidth over the given time. Value
Jens Axboe6c219762006-11-03 15:51:45 +01001056 is specified in milliseconds.
Jens Axboe71bfa162006-10-25 11:08:19 +02001057
Jens Axboec8eeb9d2011-10-05 14:02:22 +02001058iopsavgtime=int Average the calculated IOPS over the given time. Value
1059 is specified in milliseconds.
1060
Jens Axboe71bfa162006-10-25 11:08:19 +02001061create_serialize=bool If true, serialize the file creating for the jobs.
1062 This may be handy to avoid interleaving of data
1063 files, which may greatly depend on the filesystem
1064 used and even the number of processors in the system.
1065
1066create_fsync=bool fsync the data file after creation. This is the
1067 default.
1068
Jens Axboe814452b2009-03-04 12:53:13 +01001069create_on_open=bool Don't pre-setup the files for IO, just create open()
1070 when it's time to do IO to that file.
1071
Jens Axboe25460cf2012-05-02 13:58:02 +02001072create_only=bool If true, fio will only run the setup phase of the job.
1073 If files need to be laid out or updated on disk, only
1074 that will be done. The actual job contents are not
1075 executed.
1076
Zhang, Yanminafad68f2009-05-20 11:30:55 +02001077pre_read=bool If this is given, files will be pre-read into memory before
Jens Axboe34f1c042009-06-02 14:19:25 +02001078 starting the given IO operation. This will also clear
1079 the 'invalidate' flag, since it is pointless to pre-read
Jens Axboe9c0d2242009-07-01 12:26:28 +02001080 and then drop the cache. This will only work for IO engines
1081 that are seekable, since they allow you to read the same data
1082 multiple times. Thus it will not work on eg network or splice
1083 IO.
Zhang, Yanminafad68f2009-05-20 11:30:55 +02001084
Jens Axboee545a6c2007-01-14 00:00:29 +01001085unlink=bool Unlink the job files when done. Not the default, as repeated
Jens Axboebf9a3ed2008-06-05 11:53:08 +02001086 runs of that job would then waste time recreating the file
1087 set again and again.
Jens Axboe71bfa162006-10-25 11:08:19 +02001088
1089loops=int Run the specified number of iterations of this job. Used
1090 to repeat the same workload a given number of times. Defaults
1091 to 1.
1092
Juan Casse62167762013-09-17 14:06:13 -07001093verify_only Do not perform specified workload---only verify data still
1094 matches previous invocation of this workload. This option
1095 allows one to check data multiple times at a later date
1096 without overwriting it. This option makes sense only for
1097 workloads that write data, and does not support workloads
1098 with the time_based option set.
1099
Jens Axboe68e1f292007-08-10 10:32:14 +02001100do_verify=bool Run the verify phase after a write phase. Only makes sense if
Shawn Lewise84c73a2007-08-02 22:19:32 +02001101 verify is set. Defaults to 1.
1102
Jens Axboe71bfa162006-10-25 11:08:19 +02001103verify=str If writing to a file, fio can verify the file contents
1104 after each iteration of the job. The allowed values are:
1105
1106 md5 Use an md5 sum of the data area and store
1107 it in the header of each block.
1108
Jens Axboe17dc34d2007-07-27 15:36:02 +02001109 crc64 Use an experimental crc64 sum of the data
1110 area and store it in the header of each
1111 block.
1112
Jens Axboebac39e02008-06-11 20:46:19 +02001113 crc32c Use a crc32c sum of the data area and store
1114 it in the header of each block.
1115
Jens Axboe38455912008-08-04 15:35:26 +02001116 crc32c-intel Use hardware assisted crc32c calcuation
Jens Axboe0539d752010-06-21 15:22:56 +02001117 provided on SSE4.2 enabled processors. Falls
1118 back to regular software crc32c, if not
1119 supported by the system.
Jens Axboe38455912008-08-04 15:35:26 +02001120
Jens Axboe71bfa162006-10-25 11:08:19 +02001121 crc32 Use a crc32 sum of the data area and store
1122 it in the header of each block.
1123
Jens Axboe969f7ed2007-07-27 09:07:17 +02001124 crc16 Use a crc16 sum of the data area and store
1125 it in the header of each block.
1126
Jens Axboe17dc34d2007-07-27 15:36:02 +02001127 crc7 Use a crc7 sum of the data area and store
1128 it in the header of each block.
1129
Jens Axboe844ea602014-02-20 13:21:45 -08001130 xxhash Use xxhash as the checksum function. Generally
1131 the fastest software checksum that fio
1132 supports.
1133
Jens Axboecd14cc12007-07-30 10:59:33 +02001134 sha512 Use sha512 as the checksum function.
1135
1136 sha256 Use sha256 as the checksum function.
1137
Jens Axboe7c353ce2009-08-09 22:40:33 +02001138 sha1 Use optimized sha1 as the checksum function.
1139
Shawn Lewis7437ee82007-08-02 21:05:58 +02001140 meta Write extra information about each io
1141 (timestamp, block number etc.). The block
Juan Casse62167762013-09-17 14:06:13 -07001142 number is verified. The io sequence number is
1143 verified for workloads that write data.
1144 See also verify_pattern.
Shawn Lewis7437ee82007-08-02 21:05:58 +02001145
Jens Axboe36690c92007-03-26 10:23:34 +02001146 null Only pretend to verify. Useful for testing
1147 internals with ioengine=null, not for much
1148 else.
1149
Jens Axboe6c219762006-11-03 15:51:45 +01001150 This option can be used for repeated burn-in tests of a
Jens Axboe71bfa162006-10-25 11:08:19 +02001151 system to make sure that the written data is also
Jens Axboeb892dc02009-09-05 20:37:35 +02001152 correctly read back. If the data direction given is
1153 a read or random read, fio will assume that it should
1154 verify a previously written file. If the data direction
1155 includes any form of write, the verify will be of the
1156 newly written data.
Jens Axboe71bfa162006-10-25 11:08:19 +02001157
Jens Axboe160b9662007-03-27 10:59:49 +02001158verifysort=bool If set, fio will sort written verify blocks when it deems
1159 it faster to read them back in a sorted manner. This is
1160 often the case when overwriting an existing file, since
1161 the blocks are already laid out in the file system. You
1162 can ignore this option unless doing huge amounts of really
1163 fast IO where the red-black tree sorting CPU time becomes
1164 significant.
Shawn Lewis3f9f4e22007-07-28 21:10:37 +02001165
Jens Axboef7fa2652009-03-09 14:20:20 +01001166verify_offset=int Swap the verification header with data somewhere else
Shawn Lewis546a9142007-07-28 21:11:37 +02001167 in the block before writing. Its swapped back before
1168 verifying.
1169
Jens Axboef7fa2652009-03-09 14:20:20 +01001170verify_interval=int Write the verification header at a finer granularity
Shawn Lewis3f9f4e22007-07-28 21:10:37 +02001171 than the blocksize. It will be written for chunks the
1172 size of header_interval. blocksize should divide this
1173 evenly.
Jens Axboe90059d62007-07-30 09:33:12 +02001174
Radha Ramachandran0e92f872009-10-27 20:14:27 +01001175verify_pattern=str If set, fio will fill the io buffers with this
Shawn Lewise28218f2008-01-16 11:01:33 +01001176 pattern. Fio defaults to filling with totally random
1177 bytes, but sometimes it's interesting to fill with a known
1178 pattern for io verification purposes. Depending on the
1179 width of the pattern, fio will fill 1/2/3/4 bytes of the
Radha Ramachandran0e92f872009-10-27 20:14:27 +01001180 buffer at the time(it can be either a decimal or a hex number).
1181 The verify_pattern if larger than a 32-bit quantity has to
Jens Axboe996093b2010-06-24 08:37:13 +02001182 be a hex number that starts with either "0x" or "0X". Use
1183 with verify=meta.
Shawn Lewise28218f2008-01-16 11:01:33 +01001184
Jens Axboe68e1f292007-08-10 10:32:14 +02001185verify_fatal=bool Normally fio will keep checking the entire contents
Jens Axboea12a3b42007-08-09 10:20:54 +02001186 before quitting on a block verification failure. If this
1187 option is set, fio will exit the job on the first observed
1188 failure.
Jens Axboee8462bd2009-07-06 12:59:04 +02001189
Jens Axboeb463e932011-01-12 09:03:23 +01001190verify_dump=bool If set, dump the contents of both the original data
1191 block and the data block we read off disk to files. This
1192 allows later analysis to inspect just what kind of data
Jens Axboeef71e312011-10-25 22:43:36 +02001193 corruption occurred. Off by default.
Jens Axboeb463e932011-01-12 09:03:23 +01001194
Jens Axboee8462bd2009-07-06 12:59:04 +02001195verify_async=int Fio will normally verify IO inline from the submitting
1196 thread. This option takes an integer describing how many
1197 async offload threads to create for IO verification instead,
1198 causing fio to offload the duty of verifying IO contents
Jens Axboec85c3242009-07-06 14:12:57 +02001199 to one or more separate threads. If using this offload
1200 option, even sync IO engines can benefit from using an
1201 iodepth setting higher than 1, as it allows them to have
1202 IO in flight while verifies are running.
Jens Axboee8462bd2009-07-06 12:59:04 +02001203
1204verify_async_cpus=str Tell fio to set the given CPU affinity on the
1205 async IO verification threads. See cpus_allowed for the
1206 format used.
Jens Axboe6f874182010-06-21 12:53:26 +02001207
1208verify_backlog=int Fio will normally verify the written contents of a
1209 job that utilizes verify once that job has completed. In
1210 other words, everything is written then everything is read
1211 back and verified. You may want to verify continually
1212 instead for a variety of reasons. Fio stores the meta data
1213 associated with an IO block in memory, so for large
1214 verify workloads, quite a bit of memory would be used up
1215 holding this meta data. If this option is enabled, fio
Jens Axboef42195a2010-10-26 08:10:58 -06001216 will write only N blocks before verifying these blocks.
1217
Jens Axboe6f874182010-06-21 12:53:26 +02001218verify_backlog_batch=int Control how many blocks fio will verify
1219 if verify_backlog is set. If not set, will default to
1220 the value of verify_backlog (meaning the entire queue
Jens Axboef42195a2010-10-26 08:10:58 -06001221 is read back and verified). If verify_backlog_batch is
1222 less than verify_backlog then not all blocks will be verified,
1223 if verify_backlog_batch is larger than verify_backlog, some
1224 blocks will be verified more than once.
Bruce Cran66c098b2012-11-27 12:16:07 +00001225
Jens Axboed3923652011-08-03 12:38:39 +02001226stonewall
Anatol Pomozovde8f6de2013-09-26 16:31:34 -07001227wait_for_previous Wait for preceding jobs in the job file to exit, before
Jens Axboe71bfa162006-10-25 11:08:19 +02001228 starting this one. Can be used to insert serialization
Jens Axboeb3d62a72007-03-20 14:23:26 +01001229 points in the job file. A stone wall also implies starting
1230 a new reporting group.
1231
Akash Vermaabcab6a2012-10-04 15:58:28 -07001232new_group Start a new reporting group. See: group_reporting.
Jens Axboe71bfa162006-10-25 11:08:19 +02001233
1234numjobs=int Create the specified number of clones of this job. May be
1235 used to setup a larger number of threads/processes doing
Akash Vermaabcab6a2012-10-04 15:58:28 -07001236 the same thing. Each thread is reported separately; to see
1237 statistics for all clones as a whole, use group_reporting in
1238 conjunction with new_group.
Jens Axboefa28c852007-03-06 15:40:49 +01001239
Akash Vermaabcab6a2012-10-04 15:58:28 -07001240group_reporting It may sometimes be interesting to display statistics for
Jens Axboe04b2f792012-10-10 09:09:59 -06001241 groups of jobs as a whole instead of for each individual job.
1242 This is especially true if 'numjobs' is used; looking at
1243 individual thread/process output quickly becomes unwieldy.
1244 To see the final report per-group instead of per-job, use
1245 'group_reporting'. Jobs in a file will be part of the same
1246 reporting group, unless if separated by a stonewall, or by
1247 using 'new_group'.
Jens Axboe71bfa162006-10-25 11:08:19 +02001248
1249thread fio defaults to forking jobs, however if this option is
1250 given, fio will use pthread_create(3) to create threads
1251 instead.
1252
Jens Axboef7fa2652009-03-09 14:20:20 +01001253zonesize=int Divide a file into zones of the specified size. See zoneskip.
Jens Axboe71bfa162006-10-25 11:08:19 +02001254
Jens Axboef7fa2652009-03-09 14:20:20 +01001255zoneskip=int Skip the specified number of bytes when zonesize data has
Jens Axboe71bfa162006-10-25 11:08:19 +02001256 been read. The two zone options can be used to only do
1257 io on zones of a file.
1258
Jens Axboe076efc72006-10-27 11:24:25 +02001259write_iolog=str Write the issued io patterns to the specified file. See
Stefan Hajnoczi5b42a482011-01-08 20:28:41 +01001260 read_iolog. Specify a separate file for each job, otherwise
1261 the iologs will be interspersed and the file may be corrupt.
Jens Axboe71bfa162006-10-25 11:08:19 +02001262
Jens Axboe076efc72006-10-27 11:24:25 +02001263read_iolog=str Open an iolog with the specified file name and replay the
Jens Axboe71bfa162006-10-25 11:08:19 +02001264 io patterns it contains. This can be used to store a
Jens Axboe6df8ada2007-05-15 13:23:19 +02001265 workload and replay it sometime later. The iolog given
1266 may also be a blktrace binary file, which allows fio
1267 to replay a workload captured by blktrace. See blktrace
1268 for how to capture such logging data. For blktrace replay,
1269 the file needs to be turned into a blkparse binary data
Jens Axboeea3e51c2010-05-17 19:51:45 +02001270 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
Bruce Cran66c098b2012-11-27 12:16:07 +00001271
David Nellans64bbb862010-08-24 22:13:30 +02001272replay_no_stall=int When replaying I/O with read_iolog the default behavior
Jens Axboe62776222010-09-02 15:30:16 +02001273 is to attempt to respect the time stamps within the log and
1274 replay them with the appropriate delay between IOPS. By
1275 setting this variable fio will not respect the timestamps and
1276 attempt to replay them as fast as possible while still
1277 respecting ordering. The result is the same I/O pattern to a
1278 given device, but different timings.
Jens Axboe71bfa162006-10-25 11:08:19 +02001279
David Nellansd1c46c02010-08-31 21:20:47 +02001280replay_redirect=str While replaying I/O patterns using read_iolog the
1281 default behavior is to replay the IOPS onto the major/minor
1282 device that each IOP was recorded from. This is sometimes
Anatol Pomozovde8f6de2013-09-26 16:31:34 -07001283 undesirable because on a different machine those major/minor
David Nellansd1c46c02010-08-31 21:20:47 +02001284 numbers can map to a different device. Changing hardware on
1285 the same system can also result in a different major/minor
1286 mapping. Replay_redirect causes all IOPS to be replayed onto
1287 the single specified device regardless of the device it was
1288 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1289 IO in the blktrace to be replayed onto /dev/sdc. This means
1290 multiple devices will be replayed onto a single, if the trace
1291 contains multiple devices. If you want multiple devices to be
1292 replayed concurrently to multiple redirected devices you must
1293 blkparse your trace into separate traces and replay them with
1294 independent fio invocations. Unfortuantely this also breaks
1295 the strict time ordering between multiple device accesses.
1296
Jens Axboee3cedca2008-11-19 19:57:52 +01001297write_bw_log=str If given, write a bandwidth log of the jobs in this job
Jens Axboe71bfa162006-10-25 11:08:19 +02001298 file. Can be used to store data of the bandwidth of the
Jens Axboee0da9bc2006-10-25 13:08:57 +02001299 jobs in their lifetime. The included fio_generate_plots
1300 script uses gnuplot to turn these text files into nice
Lucian Adrian Grijincuddb754d2012-04-05 18:18:35 -06001301 graphs. See write_lat_log for behaviour of given
1302 filename. For this option, the suffix is _bw.log.
Jens Axboe71bfa162006-10-25 11:08:19 +02001303
Jens Axboee3cedca2008-11-19 19:57:52 +01001304write_lat_log=str Same as write_bw_log, except that this option stores io
Jens Axboe02af0982010-06-24 09:59:34 +02001305 submission, completion, and total latencies instead. If no
1306 filename is given with this option, the default filename of
1307 "jobname_type.log" is used. Even if the filename is given,
1308 fio will still append the type of log. So if one specifies
Jens Axboee3cedca2008-11-19 19:57:52 +01001309
1310 write_lat_log=foo
1311
Jens Axboed5d94592013-05-09 21:10:58 +02001312 The actual log names will be foo_slat.log, foo_clat.log,
Jens Axboe02af0982010-06-24 09:59:34 +02001313 and foo_lat.log. This helps fio_generate_plot fine the logs
1314 automatically.
Jens Axboe71bfa162006-10-25 11:08:19 +02001315
Jens Axboeb8bc8cb2011-12-01 09:04:31 +01001316write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1317 given with this option, the default filename of
1318 "jobname_type.log" is used. Even if the filename is given,
1319 fio will still append the type of log.
1320
1321log_avg_msec=int By default, fio will log an entry in the iops, latency,
1322 or bw log for every IO that completes. When writing to the
1323 disk log, that can quickly grow to a very large size. Setting
1324 this option makes fio average the each log entry over the
1325 specified period of time, reducing the resolution of the log.
1326 Defaults to 0.
1327
Jens Axboef7fa2652009-03-09 14:20:20 +01001328lockmem=int Pin down the specified amount of memory with mlock(2). Can
Jens Axboe71bfa162006-10-25 11:08:19 +02001329 potentially be used instead of removing memory or booting
1330 with less memory to simulate a smaller amount of memory.
Jens Axboe81c6b6c2013-04-10 19:30:50 +02001331 The amount specified is per worker.
Jens Axboe71bfa162006-10-25 11:08:19 +02001332
1333exec_prerun=str Before running this job, issue the command specified
Jens Axboe74c8c482013-07-17 22:15:09 -06001334 through system(3). Output is redirected in a file called
1335 jobname.prerun.txt.
Jens Axboe71bfa162006-10-25 11:08:19 +02001336
1337exec_postrun=str After the job completes, issue the command specified
Jens Axboe74c8c482013-07-17 22:15:09 -06001338 though system(3). Output is redirected in a file called
1339 jobname.postrun.txt.
Jens Axboe71bfa162006-10-25 11:08:19 +02001340
1341ioscheduler=str Attempt to switch the device hosting the file to the specified
1342 io scheduler before running.
1343
Jens Axboe0a839f32007-04-26 09:02:34 +02001344disk_util=bool Generate disk utilization statistics, if the platform
1345 supports it. Defaults to on.
1346
Jens Axboe02af0982010-06-24 09:59:34 +02001347disable_lat=bool Disable measurements of total latency numbers. Useful
Jens Axboe9520ebb2008-10-16 21:03:27 +02001348 only for cutting back the number of calls to gettimeofday,
1349 as that does impact performance at really high IOPS rates.
1350 Note that to really get rid of a large amount of these
1351 calls, this option must be used with disable_slat and
1352 disable_bw as well.
1353
Jens Axboe02af0982010-06-24 09:59:34 +02001354disable_clat=bool Disable measurements of completion latency numbers. See
1355 disable_lat.
1356
Jens Axboe9520ebb2008-10-16 21:03:27 +02001357disable_slat=bool Disable measurements of submission latency numbers. See
Jens Axboe02af0982010-06-24 09:59:34 +02001358 disable_slat.
Jens Axboe9520ebb2008-10-16 21:03:27 +02001359
1360disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
Jens Axboe02af0982010-06-24 09:59:34 +02001361 disable_lat.
Jens Axboe9520ebb2008-10-16 21:03:27 +02001362
Yu-ju Hong83349192011-08-13 00:53:44 +02001363clat_percentiles=bool Enable the reporting of percentiles of
1364 completion latencies.
1365
1366percentile_list=float_list Overwrite the default list of percentiles
1367 for completion latencies. Each number is a floating
1368 number in the range (0,100], and the maximum length of
1369 the list is 20. Use ':' to separate the numbers, and
1370 list the numbers in ascending order. For example,
1371 --percentile_list=99.5:99.9 will cause fio to report
1372 the values of completion latency below which 99.5% and
1373 99.9% of the observed latencies fell, respectively.
1374
Jens Axboe23893642012-12-17 14:44:08 +01001375clocksource=str Use the given clocksource as the base of timing. The
1376 supported options are:
1377
1378 gettimeofday gettimeofday(2)
1379
1380 clock_gettime clock_gettime(2)
1381
1382 cpu Internal CPU clock source
1383
1384 cpu is the preferred clocksource if it is reliable, as it
1385 is very fast (and fio is heavy on time calls). Fio will
1386 automatically use this clocksource if it's supported and
1387 considered reliable on the system it is running on, unless
1388 another clocksource is specifically set. For x86/x86-64 CPUs,
1389 this means supporting TSC Invariant.
1390
Jens Axboe993bf482008-11-14 13:04:53 +01001391gtod_reduce=bool Enable all of the gettimeofday() reducing options
1392 (disable_clat, disable_slat, disable_bw) plus reduce
1393 precision of the timeout somewhat to really shrink
1394 the gettimeofday() call count. With this option enabled,
1395 we only do about 0.4% of the gtod() calls we would have
1396 done if all time keeping was enabled.
1397
Jens Axboebe4ecfd2008-12-08 14:10:52 +01001398gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1399 execution to just getting the current time. Fio (and
1400 databases, for instance) are very intensive on gettimeofday()
1401 calls. With this option, you can set one CPU aside for
1402 doing nothing but logging current time to a shared memory
1403 location. Then the other threads/processes that run IO
1404 workloads need only copy that segment, instead of entering
1405 the kernel with a gettimeofday() call. The CPU set aside
1406 for doing these time calls will be excluded from other
1407 uses. Fio will manually clear it from the CPU mask of other
1408 jobs.
Jens Axboea696fa22009-12-04 10:05:02 +01001409
Steven Lang06842022011-11-17 09:45:17 +01001410continue_on_error=str Normally fio will exit the job on the first observed
Radha Ramachandranf2bba182009-06-15 08:40:16 +02001411 failure. If this option is set, fio will continue the job when
1412 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1413 is exceeded or the I/O size specified is completed. If this
1414 option is used, there are two more stats that are appended,
1415 the total error count and the first error. The error field
1416 given in the stats is the first error that was hit during the
1417 run.
Jens Axboebe4ecfd2008-12-08 14:10:52 +01001418
Steven Lang06842022011-11-17 09:45:17 +01001419 The allowed values are:
1420
1421 none Exit on any IO or verify errors.
1422
1423 read Continue on read errors, exit on all others.
1424
1425 write Continue on write errors, exit on all others.
1426
1427 io Continue on any IO error, exit on all others.
1428
1429 verify Continue on verify errors, exit on all others.
1430
1431 all Continue on all errors.
1432
1433 0 Backward-compatible alias for 'none'.
1434
1435 1 Backward-compatible alias for 'all'.
1436
Dmitry Monakhov8b28bd42012-09-23 15:46:09 +04001437ignore_error=str Sometimes you want to ignore some errors during test
1438 in that case you can specify error list for each error type.
1439 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1440 errors for given error type is separated with ':'. Error
1441 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1442 Example:
1443 ignore_error=EAGAIN,ENOSPC:122
Bruce Cran66c098b2012-11-27 12:16:07 +00001444 This option will ignore EAGAIN from READ, and ENOSPC and
1445 122(EDQUOT) from WRITE.
Dmitry Monakhov8b28bd42012-09-23 15:46:09 +04001446
1447error_dump=bool If set dump every error even if it is non fatal, true
1448 by default. If disabled only fatal error will be dumped
Bruce Cran66c098b2012-11-27 12:16:07 +00001449
Jens Axboe6adb38a2009-12-07 08:01:26 +01001450cgroup=str Add job to this control group. If it doesn't exist, it will
1451 be created. The system must have a mounted cgroup blkio
1452 mount point for this to work. If your system doesn't have it
1453 mounted, you can do so with:
Jens Axboea696fa22009-12-04 10:05:02 +01001454
1455 # mount -t cgroup -o blkio none /cgroup
1456
Jens Axboea696fa22009-12-04 10:05:02 +01001457cgroup_weight=int Set the weight of the cgroup to this value. See
1458 the documentation that comes with the kernel, allowed values
1459 are in the range of 100..1000.
Jens Axboe71bfa162006-10-25 11:08:19 +02001460
Vivek Goyal7de87092010-03-31 22:55:15 +02001461cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1462 the job completion. To override this behavior and to leave
1463 cgroups around after the job completion, set cgroup_nodelete=1.
1464 This can be useful if one wants to inspect various cgroup
1465 files after job completion. Default: false
1466
Jens Axboee0b0d892009-12-08 10:10:14 +01001467uid=int Instead of running as the invoking user, set the user ID to
1468 this value before the thread/process does any work.
1469
1470gid=int Set group ID, see uid.
1471
Dan Ehrenberg9e684a42012-02-20 11:05:14 +01001472flow_id=int The ID of the flow. If not specified, it defaults to being a
1473 global flow. See flow.
1474
1475flow=int Weight in token-based flow control. If this value is used, then
1476 there is a 'flow counter' which is used to regulate the
1477 proportion of activity between two or more jobs. fio attempts
1478 to keep this flow counter near zero. The 'flow' parameter
1479 stands for how much should be added or subtracted to the flow
1480 counter on each iteration of the main I/O loop. That is, if
1481 one job has flow=8 and another job has flow=-1, then there
1482 will be a roughly 1:8 ratio in how much one runs vs the other.
1483
1484flow_watermark=int The maximum value that the absolute value of the flow
1485 counter is allowed to reach before the job must wait for a
1486 lower value of the counter.
1487
1488flow_sleep=int The period of time, in microseconds, to wait after the flow
1489 watermark has been exceeded before retrying operations
1490
Steven Langde890a12011-11-09 14:03:34 +01001491In addition, there are some parameters which are only valid when a specific
1492ioengine is in use. These are used identically to normal parameters, with the
1493caveat that when used on the command line, they must come after the ioengine
1494that defines them is selected.
1495
1496[libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1497 the io_getevents system call to reap newly returned events.
1498 With this flag turned on, the AIO ring will be read directly
1499 from user-space to reap events. The reaping mode is only
1500 enabled when polling for a minimum of 0 events (eg when
1501 iodepth_batch_complete=0).
1502
Jens Axboe03530502012-03-19 21:45:12 +01001503[cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1504
1505[cpu] cpuchunks=int Split the load into cycles of the given time. In
1506 microseconds.
1507
Jens Axboe046395d2014-04-09 13:57:38 -06001508[cpu] exit_on_io_done=bool Detect when IO threads are done, then exit.
1509
Steven Langde890a12011-11-09 14:03:34 +01001510[netsplice] hostname=str
1511[net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1512 If the job is a TCP listener or UDP reader, the hostname is not
Shawn Bohrerb511c9a2013-07-19 13:24:06 -05001513 used and must be omitted unless it is a valid UDP multicast
1514 address.
Steven Langde890a12011-11-09 14:03:34 +01001515
1516[netsplice] port=int
1517[net] port=int The TCP or UDP port to bind to or connect to.
1518
Shawn Bohrerb93b6a22013-07-19 13:24:07 -05001519[netsplice] interface=str
1520[net] interface=str The IP address of the network interface used to send or
1521 receive UDP multicast
1522
Shawn Bohrerd3a623d2013-07-19 13:24:08 -05001523[netsplice] ttl=int
1524[net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1525 Default: 1
1526
Jens Axboe1d360ff2013-01-31 13:33:45 +01001527[netsplice] nodelay=bool
1528[net] nodelay=bool Set TCP_NODELAY on TCP connections.
1529
Steven Langde890a12011-11-09 14:03:34 +01001530[netsplice] protocol=str
1531[netsplice] proto=str
1532[net] protocol=str
1533[net] proto=str The network protocol to use. Accepted values are:
1534
1535 tcp Transmission control protocol
Jens Axboe49ccb8c2014-01-23 16:49:37 -08001536 tcpv6 Transmission control protocol V6
Bruce Cranf5cc3d02012-10-10 08:17:44 -06001537 udp User datagram protocol
Jens Axboe49ccb8c2014-01-23 16:49:37 -08001538 udpv6 User datagram protocol V6
Steven Langde890a12011-11-09 14:03:34 +01001539 unix UNIX domain socket
1540
1541 When the protocol is TCP or UDP, the port must also be given,
1542 as well as the hostname if the job is a TCP listener or UDP
1543 reader. For unix sockets, the normal filename option should be
1544 used and the port is invalid.
1545
1546[net] listen For TCP network connections, tell fio to listen for incoming
1547 connections rather than initiating an outgoing connection. The
1548 hostname must be omitted if this option is used.
Shawn Bohrerb511c9a2013-07-19 13:24:06 -05001549[net] pingpong Normaly a network writer will just continue writing data, and
Jens Axboe7aeb1e92012-12-06 20:53:57 +01001550 a network reader will just consume packages. If pingpong=1
1551 is set, a writer will send its normal payload to the reader,
1552 then wait for the reader to send the same payload back. This
1553 allows fio to measure network latencies. The submission
1554 and completion latencies then measure local time spent
1555 sending or receiving, and the completion latency measures
1556 how long it took for the other end to receive and send back.
Shawn Bohrerb511c9a2013-07-19 13:24:06 -05001557 For UDP multicast traffic pingpong=1 should only be set for a
1558 single reader when multiple readers are listening to the same
1559 address.
Jens Axboe7aeb1e92012-12-06 20:53:57 +01001560
Dmitry Monakhovd54fce82012-09-20 15:37:17 +04001561[e4defrag] donorname=str
1562 File will be used as a block donor(swap extents between files)
1563[e4defrag] inplace=int
Bruce Cran66c098b2012-11-27 12:16:07 +00001564 Configure donor file blocks allocation strategy
Dmitry Monakhovd54fce82012-09-20 15:37:17 +04001565 0(default): Preallocate donor's file on init
1566 1 : allocate space immidietly inside defragment event,
1567 and free right after event
1568
Steven Langde890a12011-11-09 14:03:34 +01001569
1570
Jens Axboe71bfa162006-10-25 11:08:19 +020015716.0 Interpreting the output
1572---------------------------
1573
1574fio spits out a lot of output. While running, fio will display the
1575status of the jobs created. An example of that would be:
1576
Jens Axboe73c8b082007-01-11 19:25:52 +01001577Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
Jens Axboe71bfa162006-10-25 11:08:19 +02001578
1579The characters inside the square brackets denote the current status of
1580each thread. The possible values (in typical life cycle order) are:
1581
1582Idle Run
1583---- ---
1584P Thread setup, but not started.
1585C Thread created.
Jens Axboe9c6f6312012-11-07 09:15:45 +01001586I Thread initialized, waiting or generating necessary data.
Jens Axboeb0f65862009-05-20 11:52:15 +02001587 p Thread running pre-reading file(s).
Jens Axboe71bfa162006-10-25 11:08:19 +02001588 R Running, doing sequential reads.
1589 r Running, doing random reads.
1590 W Running, doing sequential writes.
1591 w Running, doing random writes.
1592 M Running, doing mixed sequential reads/writes.
1593 m Running, doing mixed random reads/writes.
1594 F Running, currently waiting for fsync()
Jens Axboe3d434052014-04-02 15:46:22 -06001595 f Running, finishing up (writing IO logs, etc)
Jens Axboefc6bd432009-04-29 09:52:10 +02001596 V Running, doing verification of written data.
Jens Axboe71bfa162006-10-25 11:08:19 +02001597E Thread exited, not reaped by main thread yet.
Jens Axboe4f7e57a2012-03-30 21:21:20 +02001598_ Thread reaped, or
1599X Thread reaped, exited with an error.
Jens Axboea5e371a2012-04-02 09:47:09 -07001600K Thread reaped, exited due to signal.
Jens Axboe71bfa162006-10-25 11:08:19 +02001601
1602The other values are fairly self explanatory - number of threads
Jens Axboec9f60302007-07-20 12:43:05 +02001603currently running and doing io, rate of io since last check (read speed
1604listed first, then write speed), and the estimated completion percentage
1605and time for the running group. It's impossible to estimate runtime of
Jens Axboe4f7e57a2012-03-30 21:21:20 +02001606the following groups (if any). Note that the string is displayed in order,
1607so it's possible to tell which of the jobs are currently doing what. The
1608first character is the first job defined in the job file, and so forth.
Jens Axboe71bfa162006-10-25 11:08:19 +02001609
1610When fio is done (or interrupted by ctrl-c), it will show the data for
1611each thread, group of threads, and disks in that order. For each data
1612direction, the output looks like:
1613
1614Client1 (g=0): err= 0:
Paul Dubs35649e52011-07-21 16:04:52 +02001615 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
Jens Axboe6104ddb2007-01-11 14:24:29 +01001616 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1617 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
Jens Axboeb22989b2009-07-17 22:29:23 +02001618 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
Jens Axboee7823a92007-09-07 20:33:33 +02001619 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
Jens Axboe71619dc2007-01-13 23:56:33 +01001620 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
Jens Axboe838bc702008-05-22 13:08:23 +02001621 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1622 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
Jens Axboe30061b92007-04-17 13:31:34 +02001623 issued r/w: total=0/32768, short=0/0
Jens Axboe8abdce62007-02-21 10:22:55 +01001624 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1625 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
Jens Axboe71bfa162006-10-25 11:08:19 +02001626
1627The client number is printed, along with the group id and error of that
1628thread. Below is the io statistics, here for writes. In the order listed,
1629they denote:
1630
1631io= Number of megabytes io performed
1632bw= Average bandwidth rate
Paul Dubs35649e52011-07-21 16:04:52 +02001633iops= Average IOs performed per second
Jens Axboe71bfa162006-10-25 11:08:19 +02001634runt= The runtime of that thread
Jens Axboe72fbda22007-03-20 10:02:06 +01001635 slat= Submission latency (avg being the average, stdev being the
Jens Axboe71bfa162006-10-25 11:08:19 +02001636 standard deviation). This is the time it took to submit
1637 the io. For sync io, the slat is really the completion
Jens Axboe8a35c712007-06-19 09:53:31 +02001638 latency, since queue/complete is one operation there. This
Jens Axboebf9a3ed2008-06-05 11:53:08 +02001639 value can be in milliseconds or microseconds, fio will choose
Jens Axboe8a35c712007-06-19 09:53:31 +02001640 the most appropriate base and print that. In the example
Lucian Adrian Grijincu0d237712012-04-03 14:42:48 -06001641 above, milliseconds is the best scale. Note: in --minimal mode
1642 latencies are always expressed in microseconds.
Jens Axboe71bfa162006-10-25 11:08:19 +02001643 clat= Completion latency. Same names as slat, this denotes the
1644 time from submission to completion of the io pieces. For
1645 sync io, clat will usually be equal (or very close) to 0,
1646 as the time from submit to complete is basically just
1647 CPU time (io has already been done, see slat explanation).
1648 bw= Bandwidth. Same names as the xlat stats, but also includes
1649 an approximate percentage of total aggregate bandwidth
1650 this thread received in this group. This last value is
1651 only really useful if the threads in this group are on the
1652 same disk, since they are then competing for disk access.
1653cpu= CPU usage. User and system time, along with the number
Jens Axboee7823a92007-09-07 20:33:33 +02001654 of context switches this thread went through, usage of
1655 system and user time, and finally the number of major
1656 and minor page faults.
Jens Axboe71619dc2007-01-13 23:56:33 +01001657IO depths= The distribution of io depths over the job life time. The
1658 numbers are divided into powers of 2, so for example the
1659 16= entries includes depths up to that value but higher
1660 than the previous entry. In other words, it covers the
1661 range from 16 to 31.
Jens Axboe838bc702008-05-22 13:08:23 +02001662IO submit= How many pieces of IO were submitting in a single submit
1663 call. Each entry denotes that amount and below, until
1664 the previous entry - eg, 8=100% mean that we submitted
1665 anywhere in between 5-8 ios per submit call.
1666IO complete= Like the above submit number, but for completions instead.
Jens Axboe30061b92007-04-17 13:31:34 +02001667IO issued= The number of read/write requests issued, and how many
1668 of them were short.
Jens Axboeec118302007-02-17 04:38:20 +01001669IO latencies= The distribution of IO completion latencies. This is the
1670 time from when IO leaves fio and when it gets completed.
1671 The numbers follow the same pattern as the IO depths,
1672 meaning that 2=1.6% means that 1.6% of the IO completed
Jens Axboe8abdce62007-02-21 10:22:55 +01001673 within 2 msecs, 20=12.8% means that 12.8% of the IO
1674 took more than 10 msecs, but less than (or equal to) 20 msecs.
Jens Axboe71bfa162006-10-25 11:08:19 +02001675
1676After each client has been listed, the group statistics are printed. They
1677will look like this:
1678
1679Run status group 0 (all jobs):
Jens Axboeb22989b2009-07-17 22:29:23 +02001680 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1681 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
Jens Axboe71bfa162006-10-25 11:08:19 +02001682
1683For each data direction, it prints:
1684
1685io= Number of megabytes io performed.
1686aggrb= Aggregate bandwidth of threads in this group.
1687minb= The minimum average bandwidth a thread saw.
1688maxb= The maximum average bandwidth a thread saw.
1689mint= The smallest runtime of the threads in that group.
1690maxt= The longest runtime of the threads in that group.
1691
1692And finally, the disk statistics are printed. They will look like this:
1693
1694Disk stats (read/write):
1695 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1696
1697Each value is printed for both reads and writes, with reads first. The
1698numbers denote:
1699
1700ios= Number of ios performed by all groups.
1701merge= Number of merges io the io scheduler.
1702ticks= Number of ticks we kept the disk busy.
1703io_queue= Total time spent in the disk queue.
1704util= The disk utilization. A value of 100% means we kept the disk
1705 busy constantly, 50% would be a disk idling half of the time.
1706
Jens Axboe8423bd12012-04-12 09:18:38 +02001707It is also possible to get fio to dump the current output while it is
1708running, without terminating the job. To do that, send fio the USR1 signal.
Jens Axboe06464902013-04-24 20:38:54 -06001709You can also get regularly timed dumps by using the --status-interval
1710parameter, or by creating a file in /tmp named fio-dump-status. If fio
1711sees this file, it will unlink it and dump the current output status.
Jens Axboe8423bd12012-04-12 09:18:38 +02001712
Jens Axboe71bfa162006-10-25 11:08:19 +02001713
17147.0 Terse output
1715----------------
1716
1717For scripted usage where you typically want to generate tables or graphs
Jens Axboe6af019c2007-03-06 19:50:58 +01001718of the results, fio can output the results in a semicolon separated format.
Jens Axboe71bfa162006-10-25 11:08:19 +02001719The format is one long line of values, such as:
1720
David Nellans562c2d22010-09-23 08:38:17 +020017212;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%
1722A description of this job goes here.
1723
1724The job description (if provided) follows on a second line.
Jens Axboe71bfa162006-10-25 11:08:19 +02001725
Jens Axboe525c2bf2010-06-30 15:22:21 +02001726To enable terse output, use the --minimal command line option. The first
1727value is the version of the terse output format. If the output has to
1728be changed for some reason, this number will be incremented by 1 to
1729signify that change.
Jens Axboe6820cb32008-09-27 12:33:53 +02001730
Jens Axboe71bfa162006-10-25 11:08:19 +02001731Split up, the format is as follows:
1732
Jens Axboe5e726d02011-10-14 08:08:10 +02001733 terse version, fio version, jobname, groupid, error
Jens Axboe71bfa162006-10-25 11:08:19 +02001734 READ status:
Jens Axboe312b4af2011-10-13 13:11:42 +02001735 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
Jens Axboede196b82012-04-02 07:03:26 -07001736 Submission latency: min, max, mean, deviation (usec)
1737 Completion latency: min, max, mean, deviation (usec)
Jens Axboe1db92cb2011-10-13 13:43:36 +02001738 Completion latency percentiles: 20 fields (see below)
Jens Axboede196b82012-04-02 07:03:26 -07001739 Total latency: min, max, mean, deviation (usec)
Lucian Adrian Grijincu0d237712012-04-03 14:42:48 -06001740 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
Jens Axboe71bfa162006-10-25 11:08:19 +02001741 WRITE status:
Jens Axboe312b4af2011-10-13 13:11:42 +02001742 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
Jens Axboede196b82012-04-02 07:03:26 -07001743 Submission latency: min, max, mean, deviation (usec)
1744 Completion latency: min, max, mean, deviation (usec)
Jens Axboe1db92cb2011-10-13 13:43:36 +02001745 Completion latency percentiles: 20 fields (see below)
Jens Axboede196b82012-04-02 07:03:26 -07001746 Total latency: min, max, mean, deviation (usec)
Lucian Adrian Grijincu0d237712012-04-03 14:42:48 -06001747 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
Shawn Lewis046ee302007-11-21 09:38:34 +01001748 CPU usage: user, system, context switches, major faults, minor faults
Jens Axboe22708902007-03-06 17:05:32 +01001749 IO depths: <=1, 2, 4, 8, 16, 32, >=64
David Nellans562c2d22010-09-23 08:38:17 +02001750 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1751 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
Jens Axboef2f788d2011-10-13 14:03:52 +02001752 Disk utilization: Disk name, Read ios, write ios,
1753 Read merges, write merges,
1754 Read ticks, write ticks,
Jens Axboe3d7cd9b2011-10-18 08:31:01 +02001755 Time spent in queue, disk utilization percentage
Anatol Pomozovde8f6de2013-09-26 16:31:34 -07001756 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
Bruce Cran66c098b2012-11-27 12:16:07 +00001757
Anatol Pomozovde8f6de2013-09-26 16:31:34 -07001758 Additional Info (dependent on description being set): Text description
Paul Dubs25c8b9d2011-07-21 17:26:02 +02001759
Jens Axboe1db92cb2011-10-13 13:43:36 +02001760Completion latency percentiles can be a grouping of up to 20 sets, so
1761for the terse output fio writes all of them. Each field will look like this:
1762
1763 1.00%=6112
1764
1765which is the Xth percentile, and the usec latency associated with it.
1766
Jens Axboef2f788d2011-10-13 14:03:52 +02001767For disk utilization, all disks used by fio are shown. So for each disk
1768there will be a disk utilization section.
1769
Paul Dubs25c8b9d2011-07-21 17:26:02 +02001770
17718.0 Trace file format
1772---------------------
Bruce Cran66c098b2012-11-27 12:16:07 +00001773There are two trace file format that you can encounter. The older (v1) format
Paul Dubs25c8b9d2011-07-21 17:26:02 +02001774is unsupported since version 1.20-rc3 (March 2008). It will still be described
1775below in case that you get an old trace and want to understand it.
1776
1777In any case the trace is a simple text file with a single action per line.
1778
1779
17808.1 Trace file format v1
1781------------------------
1782Each line represents a single io action in the following format:
1783
1784rw, offset, length
1785
1786where rw=0/1 for read/write, and the offset and length entries being in bytes.
1787
1788This format is not supported in Fio versions => 1.20-rc3.
1789
1790
17918.2 Trace file format v2
1792------------------------
1793The second version of the trace file format was added in Fio version 1.17.
1794It allows to access more then one file per trace and has a bigger set of
1795possible file actions.
1796
1797The first line of the trace file has to be:
1798
1799fio version 2 iolog
1800
1801Following this can be lines in two different formats, which are described below.
1802
1803The file management format:
1804
1805filename action
1806
1807The filename is given as an absolute path. The action can be one of these:
1808
1809add Add the given filename to the trace
Bruce Cran66c098b2012-11-27 12:16:07 +00001810open Open the file with the given filename. The filename has to have
Paul Dubs25c8b9d2011-07-21 17:26:02 +02001811 been added with the add action before.
1812close Close the file with the given filename. The file has to have been
1813 opened before.
1814
1815
1816The file io action format:
1817
1818filename action offset length
1819
1820The filename is given as an absolute path, and has to have been added and opened
Bruce Cran66c098b2012-11-27 12:16:07 +00001821before it can be used with this format. The offset and length are given in
Paul Dubs25c8b9d2011-07-21 17:26:02 +02001822bytes. The action can be one of these:
1823
1824wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1825read Read 'length' bytes beginning from 'offset'
1826write Write 'length' bytes beginning from 'offset'
1827sync fsync() the file
1828datasync fdatasync() the file
1829trim trim the given file from the given 'offset' for 'length' bytes
Huadong Liuf2a2ce02013-01-30 13:22:24 +01001830
1831
18329.0 CPU idleness profiling
Jens Axboe06464902013-04-24 20:38:54 -06001833--------------------------
Huadong Liuf2a2ce02013-01-30 13:22:24 +01001834In some cases, we want to understand CPU overhead in a test. For example,
1835we test patches for the specific goodness of whether they reduce CPU usage.
1836fio implements a balloon approach to create a thread per CPU that runs at
1837idle priority, meaning that it only runs when nobody else needs the cpu.
1838By measuring the amount of work completed by the thread, idleness of each
1839CPU can be derived accordingly.
1840
1841An unit work is defined as touching a full page of unsigned characters. Mean
1842and standard deviation of time to complete an unit work is reported in "unit
1843work" section. Options can be chosen to report detailed percpu idleness or
1844overall system idleness by aggregating percpu stats.