HOWTO: make it clear that deviation is standard deviation
[fio.git] / HOWTO
CommitLineData
71bfa161
JA
1Table 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
25c8b9d7 118. Trace file format
43f09da1 129. CPU idleness profiling
71bfa161
JA
13
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
d0ff85df 53 could be using splice, async io, syslet, or even
71bfa161
JA
54 SG (SCSI generic sg).
55
6c219763 56 IO depth If the io engine is async, how large a queuing
71bfa161
JA
57 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.
66c098b8 65
71bfa161
JA
66The 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'
550b1db6 84parameter described in the parameter section.
71bfa161 85
b4692828
JA
86If 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
c2b1e753
JA
90mirror 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.
b4692828 97
71bfa161
JA
98fio 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,
6c219763 100such as memory locking, io scheduler switching, and decreasing the nice value.
71bfa161
JA
101
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
65db0851
JA
112section residing above it. If the first character in a line is a ';' or a
113'#', the entire line is discarded as a comment.
71bfa161 114
3c54bc46 115So let's look at a really simple job file that defines two processes, each
b22989b9 116randomly reading from a 128MB file.
71bfa161
JA
117
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
c2b1e753
JA
131makes 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
71bfa161 136
3c54bc46 137Let's look at an example that has a number of processes writing randomly
71bfa161
JA
138to 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
b22989b9 154increased the buffer size used to 32KB and define numjobs to 4 to
71bfa161 155fork 4 identical jobs. The result is 4 processes each randomly writing
b22989b9 156to their own 64MB file. Instead of using the above job file, you could
b4692828
JA
157have 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
71bfa161 161
df5ad464
AK
162When fio is utilized as a basis of any reasonably large test suite, it might be
163desirable to share a set of standardized settings across multiple job files.
164Instead of copy/pasting such settings, any section may pull in an external
165.fio file with 'include filename' directive, as in the following example:
166
167; -- start job file including.fio --
168[global]
169filename=/tmp/test
170filesize=1m
171include glob-include.fio
172
173[test]
174rw=randread
175bs=4k
176time_based=1
177runtime=10
178include test-include.fio
179; -- end job file including.fio --
180
181; -- start job file glob-include.fio --
182thread=1
183group_reporting=1
184; -- end job file glob-include.fio --
185
186; -- start job file test-include.fio --
187ioengine=libaio
188iodepth=4
189; -- end job file test-include.fio --
190
191Settings pulled into a section apply to that section only (except global
192section). Include directives may be nested in that any included file may
73568e1b
JA
193contain further include directive(s). Include files may not contain []
194sections.
df5ad464
AK
195
196
74929ac2
JA
1974.1 Environment variables
198-------------------------
199
3c54bc46
AC
200fio also supports environment variable expansion in job files. Any
201substring of the form "${VARNAME}" as part of an option value (in other
202words, on the right of the `='), will be expanded to the value of the
203environment variable called VARNAME. If no such environment variable
204is defined, or VARNAME is the empty string, the empty string will be
205substituted.
206
207As an example, let's look at a sample fio invocation and job file:
208
209$ SIZE=64m NUMJOBS=4 fio jobfile.fio
210
211; -- start job file --
212[random-writers]
213rw=randwrite
214size=${SIZE}
215numjobs=${NUMJOBS}
216; -- end job file --
217
218This will expand to the following equivalent job file at runtime:
219
220; -- start job file --
221[random-writers]
222rw=randwrite
223size=64m
224numjobs=4
225; -- end job file --
226
71bfa161
JA
227fio ships with a few example job files, you can also look there for
228inspiration.
229
74929ac2
JA
2304.2 Reserved keywords
231---------------------
232
233Additionally, fio has a set of reserved keywords that will be replaced
234internally with the appropriate value. Those keywords are:
235
236$pagesize The architecture page size of the running system
237$mb_memory Megabytes of total memory in the system
238$ncpus Number of online available CPUs
239
240These can be used on the command line or in the job file, and will be
241automatically substituted with the current system values when the job
892a6ffc
JA
242is run. Simple math is also supported on these keywords, so you can
243perform actions like:
244
245size=8*$mb_memory
246
247and get that properly expanded to 8 times the size of memory in the
248machine.
74929ac2 249
71bfa161
JA
250
2515.0 Detailed list of parameters
252-------------------------------
253
254This section describes in details each parameter associated with a job.
255Some parameters take an option of a given type, such as an integer or
d59aa780
JA
256a string. Anywhere a numeric value is required, an arithmetic expression
257may be used, provided it is surrounded by parentheses. Supported operators
258are:
259
260 addition (+)
261 subtraction (-)
262 multiplication (*)
263 division (/)
264 modulus (%)
265 exponentiation (^)
266
267For time values in expressions, units are microseconds by default. This is
268different than for time values not in expressions (not enclosed in
269parentheses). The following types are used:
71bfa161
JA
270
271str String. This is a sequence of alpha characters.
b09da8fa 272time Integer with possible time suffix. In seconds unless otherwise
e417fd66 273 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
0de5b26f
JA
274 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
275 and 'us' (or 'usec') for microseconds.
b09da8fa
JA
276int SI integer. A whole number value, which may contain a suffix
277 describing the base of the number. Accepted suffixes are k/m/g/t/p,
278 meaning kilo, mega, giga, tera, and peta. The suffix is not case
57fc29fa
JA
279 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
280 as 'k'). So if you want to specify 4096, you could either write
b09da8fa 281 out '4096' or just give 4k. The suffixes signify base 2 values, so
57fc29fa
JA
282 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
283 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
284 case, then 1000 is used as the multiplier. This can be handy for
285 disks, since manufacturers generally use base 10 values when listing
286 the capacity of a drive. If the option accepts an upper and lower
287 range, use a colon ':' or minus '-' to separate such values. May also
288 include a prefix to indicate numbers base. If 0x is used, the number
289 is assumed to be hexadecimal. See irange.
71bfa161
JA
290bool Boolean. Usually parsed as an integer, however only defined for
291 true and false (1 and 0).
b09da8fa 292irange Integer range with suffix. Allows value range to be given, such
bf9a3edb 293 as 1024-4096. A colon may also be used as the separator, eg
0c9baf91
JA
294 1k:4k. If the option allows two sets of ranges, they can be
295 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
f7fa2653 296 int.
83349190 297float_list A list of floating numbers, separated by a ':' character.
71bfa161
JA
298
299With the above in mind, here follows the complete list of fio job
300parameters.
301
302name=str ASCII name of the job. This may be used to override the
303 name printed by fio for this job. Otherwise the job
c2b1e753 304 name is used. On the command line this parameter has the
6c219763 305 special purpose of also signaling the start of a new
c2b1e753 306 job.
71bfa161 307
61697c37
JA
308description=str Text description of the job. Doesn't do anything except
309 dump this text description when this job is run. It's
310 not parsed.
311
3776041e 312directory=str Prefix filenames with this directory. Used to place files
67445b63
JA
313 in a different location than "./". See the 'filename' option
314 for escaping certain characters.
71bfa161
JA
315
316filename=str Fio normally makes up a filename based on the job name,
317 thread number, and file number. If you want to share
318 files between threads in a job or several jobs, specify
ed92ac0c 319 a filename for each of them to override the default. If
414c2a3e 320 the ioengine used is 'net', the filename is the host, port,
0fd666bf 321 and protocol to use in the format of =host,port,protocol.
414c2a3e
JA
322 See ioengine=net for more. If the ioengine is file based, you
323 can specify a number of files by separating the names with a
324 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
325 as the two working files, you would use
30a4588a
JA
326 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
327 accessed as \\.\PhysicalDrive0 for the first device,
328 \\.\PhysicalDrive1 for the second etc. Note: Windows and
329 FreeBSD prevent write access to areas of the disk containing
330 in-use data (e.g. filesystems).
331 If the wanted filename does need to include a colon, then
332 escape that with a '\' character. For instance, if the filename
333 is "/dev/dsk/foo@3,0:c", then you would use
334 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
335 stdin or stdout. Which of the two depends on the read/write
336 direction set.
71bfa161 337
de98bd30
JA
338filename_format=str
339 If sharing multiple files between jobs, it is usually necessary
340 to have fio generate the exact names that you want. By default,
341 fio will name a file based on the default file format
342 specification of jobname.jobnumber.filenumber. With this
343 option, that can be customized. Fio will recognize and replace
344 the following keywords in this string:
345
346 $jobname
347 The name of the worker thread or process.
348
349 $jobnum
350 The incremental number of the worker thread or
351 process.
352
353 $filenum
354 The incremental number of the file for that worker
355 thread or process.
356
357 To have dependent jobs share a set of files, this option can
358 be set to have fio generate filenames that are shared between
359 the two. For instance, if testfiles.$filenum is specified,
360 file number 4 for any job will be named testfiles.4. The
361 default of $jobname.$jobnum.$filenum will be used if
362 no other format specifier is given.
363
bbf6b540
JA
364opendir=str Tell fio to recursively add any file it can find in this
365 directory and down the file system tree.
366
3776041e 367lockfile=str Fio defaults to not locking any files before it does
4d4e80f2
JA
368 IO to them. If a file or file descriptor is shared, fio
369 can serialize IO to that file to make the end result
370 consistent. This is usual for emulating real workloads that
371 share files. The lock modes are:
372
373 none No locking. The default.
374 exclusive Only one thread/process may do IO,
375 excluding all others.
376 readwrite Read-write locking on the file. Many
377 readers may access the file at the
378 same time, but writes get exclusive
379 access.
380
d3aad8f2 381readwrite=str
71bfa161
JA
382rw=str Type of io pattern. Accepted values are:
383
384 read Sequential reads
385 write Sequential writes
386 randwrite Random writes
387 randread Random reads
10b023db 388 rw,readwrite Sequential mixed reads and writes
71bfa161 389 randrw Random mixed reads and writes
82a90686
JA
390 trimwrite Mixed trims and writes. Blocks will be
391 trimmed first, then written to.
71bfa161
JA
392
393 For the mixed io types, the default is to split them 50/50.
394 For certain types of io the result may still be skewed a bit,
211097b2 395 since the speed may be different. It is possible to specify
38dad62d 396 a number of IO's to do before getting a new offset, this is
892ea9bd 397 done by appending a ':<nr>' to the end of the string given.
38dad62d 398 For a random read, it would look like 'rw=randread:8' for
059b0802 399 passing in an offset modifier with a value of 8. If the
ddb754db 400 suffix is used with a sequential IO pattern, then the value
059b0802
JA
401 specified will be added to the generated offset for each IO.
402 For instance, using rw=write:4k will skip 4k for every
403 write. It turns sequential IO into sequential IO with holes.
404 See the 'rw_sequencer' option.
38dad62d
JA
405
406rw_sequencer=str If an offset modifier is given by appending a number to
407 the rw=<str> line, then this option controls how that
408 number modifies the IO offset being generated. Accepted
409 values are:
410
411 sequential Generate sequential offset
412 identical Generate the same offset
413
414 'sequential' is only useful for random IO, where fio would
415 normally generate a new random offset for every IO. If you
416 append eg 8 to randread, you would get a new random offset for
211097b2
JA
417 every 8 IO's. The result would be a seek for only every 8
418 IO's, instead of for every IO. Use rw=randread:8 to specify
38dad62d
JA
419 that. As sequential IO is already sequential, setting
420 'sequential' for that would not result in any differences.
421 'identical' behaves in a similar fashion, except it sends
422 the same offset 8 number of times before generating a new
423 offset.
71bfa161 424
90fef2d1
JA
425kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
426 Storage manufacturers like to use 10^3 or 1000 as a base
427 ten unit instead, for obvious reasons. Allow values are
428 1024 or 1000, with 1024 being the default.
429
771e58be
JA
430unified_rw_reporting=bool Fio normally reports statistics on a per
431 data direction basis, meaning that read, write, and trim are
432 accounted and reported separately. If this option is set,
433 the fio will sum the results and report them as "mixed"
434 instead.
435
ee738499
JA
436randrepeat=bool For random IO workloads, seed the generator in a predictable
437 way so that results are repeatable across repetitions.
40fe5e7b 438 Defaults to true.
ee738499 439
04778baf
JA
440randseed=int Seed the random number generators based on this seed value, to
441 be able to control what sequence of output is being generated.
442 If not set, the random sequence depends on the randrepeat
443 setting.
444
a596f047
EG
445fallocate=str Whether pre-allocation is performed when laying down files.
446 Accepted values are:
447
448 none Do not pre-allocate space
449 posix Pre-allocate via posix_fallocate()
450 keep Pre-allocate via fallocate() with
451 FALLOC_FL_KEEP_SIZE set
452 0 Backward-compatible alias for 'none'
453 1 Backward-compatible alias for 'posix'
454
455 May not be available on all supported platforms. 'keep' is only
456 available on Linux.If using ZFS on Solaris this must be set to
457 'none' because ZFS doesn't support it. Default: 'posix'.
7bc8c2cf 458
d2f3ac35
JA
459fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
460 on what IO patterns it is likely to issue. Sometimes you
461 want to test specific IO patterns without telling the
462 kernel about it, in which case you can disable this option.
463 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
464 IO and POSIX_FADV_RANDOM for random IO.
465
37659335
JA
466fadvise_stream=int Notify the kernel what write stream ID to place these
467 writes under. Only supported on Linux. Note, this option
468 may change going forward.
469
f7fa2653 470size=int The total size of file io for this job. Fio will run until
7616cafe 471 this many bytes has been transferred, unless runtime is
a4d3b4db
JA
472 limited by other options (such as 'runtime', for instance,
473 or increased/decreased by 'io_size'). Unless specific nrfiles
474 and filesize options are given, fio will divide this size
475 between the available files specified by the job. If not set,
476 fio will use the full size of the given files or devices.
477 If the files do not exist, size must be given. It is also
478 possible to give size as a percentage between 1 and 100. If
479 size=20% is given, fio will use 20% of the full size of the
480 given files or devices.
481
482io_size=int
77731b29
JA
483io_limit=int Normally fio operates within the region set by 'size', which
484 means that the 'size' option sets both the region and size of
485 IO to be performed. Sometimes that is not what you want. With
486 this option, it is possible to define just the amount of IO
487 that fio should do. For instance, if 'size' is set to 20G and
a4d3b4db
JA
488 'io_size' is set to 5G, fio will perform IO within the first
489 20G but exit when 5G have been done. The opposite is also
490 possible - if 'size' is set to 20G, and 'io_size' is set to
491 40G, then fio will do 40G of IO within the 0..20G region.
77731b29 492
f7fa2653 493filesize=int Individual file sizes. May be a range, in which case fio
9c60ce64
JA
494 will select sizes for files at random within the given range
495 and limited to 'size' in total (if that is given). If not
496 given, each created file is the same size.
497
bedc9dc2
JA
498file_append=bool Perform IO after the end of the file. Normally fio will
499 operate within the size of a file. If this option is set, then
500 fio will append to the file instead. This has identical
0aae4ce7
JA
501 behavior to setting offset to the size of a file. This option
502 is ignored on non-regular files.
bedc9dc2 503
74586c1e
JA
504fill_device=bool
505fill_fs=bool Sets size to something really large and waits for ENOSPC (no
aa31f1f1 506 space left on device) as the terminating condition. Only makes
de98bd30 507 sense with sequential write. For a read workload, the mount
4f12432e
JA
508 point will be filled first then IO started on the result. This
509 option doesn't make sense if operating on a raw device node,
510 since the size of that is already known by the file system.
511 Additionally, writing beyond end-of-device will not return
512 ENOSPC there.
aa31f1f1 513
f7fa2653
JA
514blocksize=int
515bs=int The block size used for the io units. Defaults to 4k. Values
516 can be given for both read and writes. If a single int is
517 given, it will apply to both. If a second int is specified
f90eff5a 518 after a comma, it will apply to writes only. In other words,
d9472271
JA
519 the format is either bs=read_and_write or bs=read,write,trim.
520 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
521 writes, and 8k for trims. You can terminate the list with
522 a trailing comma. bs=4k,8k, would use the default value for
523 trims.. If you only wish to set the write size, you
787f7e95
JA
524 can do so by passing an empty read size - bs=,8k will set
525 8k for writes and leave the read default value.
a00735e6 526
2b7a01d0
JA
527blockalign=int
528ba=int At what boundary to align random IO offsets. Defaults to
529 the same as 'blocksize' the minimum blocksize given.
530 Minimum alignment is typically 512b for using direct IO,
531 though it usually depends on the hardware block size. This
532 option is mutually exclusive with using a random map for
533 files, so it will turn off that option.
534
d3aad8f2 535blocksize_range=irange
71bfa161
JA
536bsrange=irange Instead of giving a single block size, specify a range
537 and fio will mix the issued io block sizes. The issued
538 io unit will always be a multiple of the minimum value
f90eff5a
JA
539 given (also see bs_unaligned). Applies to both reads and
540 writes, however a second range can be given after a comma.
541 See bs=.
a00735e6 542
564ca972
JA
543bssplit=str Sometimes you want even finer grained control of the
544 block sizes issued, not just an even split between them.
545 This option allows you to weight various block sizes,
546 so that you are able to define a specific amount of
547 block sizes issued. The format for this option is:
548
549 bssplit=blocksize/percentage:blocksize/percentage
550
551 for as many block sizes as needed. So if you want to define
552 a workload that has 50% 64k blocks, 10% 4k blocks, and
553 40% 32k blocks, you would write:
554
555 bssplit=4k/10:64k/50:32k/40
556
557 Ordering does not matter. If the percentage is left blank,
558 fio will fill in the remaining values evenly. So a bssplit
559 option like this one:
560
561 bssplit=4k/50:1k/:32k/
562
563 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
564 always add up to 100, if bssplit is given a range that adds
565 up to more, it will error out.
566
720e84ad
JA
567 bssplit also supports giving separate splits to reads and
568 writes. The format is identical to what bs= accepts. You
569 have to separate the read and write parts with a comma. So
570 if you want a workload that has 50% 2k reads and 50% 4k reads,
571 while having 90% 4k writes and 10% 8k writes, you would
572 specify:
573
892ea9bd 574 bssplit=2k/50:4k/50,4k/90:8k/10
720e84ad 575
d3aad8f2 576blocksize_unaligned
690adba3
JA
577bs_unaligned If this option is given, any byte size value within bsrange
578 may be used as a block range. This typically wont work with
579 direct IO, as that normally requires sector alignment.
71bfa161 580
6aca9b3d
JA
581bs_is_seq_rand If this option is set, fio will use the normal read,write
582 blocksize settings as sequential,random instead. Any random
583 read or write will use the WRITE blocksize settings, and any
584 sequential read or write will use the READ blocksize setting.
585
e9459e5a
JA
586zero_buffers If this option is given, fio will init the IO buffers to
587 all zeroes. The default is to fill them with random data.
588
5973cafb
JA
589refill_buffers If this option is given, fio will refill the IO buffers
590 on every submit. The default is to only fill it at init
591 time and reuse that data. Only makes sense if zero_buffers
41ccd845
JA
592 isn't specified, naturally. If data verification is enabled,
593 refill_buffers is also automatically enabled.
5973cafb 594
fd68418e
JA
595scramble_buffers=bool If refill_buffers is too costly and the target is
596 using data deduplication, then setting this option will
597 slightly modify the IO buffer contents to defeat normal
598 de-dupe attempts. This is not enough to defeat more clever
599 block compression attempts, but it will stop naive dedupe of
600 blocks. Default: true.
601
c5751c62
JA
602buffer_compress_percentage=int If this is set, then fio will attempt to
603 provide IO buffer content (on WRITEs) that compress to
604 the specified level. Fio does this by providing a mix of
d1af2894
JA
605 random data and a fixed pattern. The fixed pattern is either
606 zeroes, or the pattern specified by buffer_pattern. If the
607 pattern option is used, it might skew the compression ratio
608 slightly. Note that this is per block size unit, for file/disk
609 wide compression level that matches this setting, you'll also
610 want to set refill_buffers.
c5751c62
JA
611
612buffer_compress_chunk=int See buffer_compress_percentage. This
613 setting allows fio to manage how big the ranges of random
614 data and zeroed data is. Without this set, fio will
615 provide buffer_compress_percentage of blocksize random
616 data, followed by the remaining zeroed. With this set
617 to some chunk size smaller than the block size, fio can
618 alternate random and zeroed data throughout the IO
619 buffer.
620
5c94b008
JA
621buffer_pattern=str If set, fio will fill the io buffers with this
622 pattern. If not set, the contents of io buffers is defined by
623 the other options related to buffer contents. The setting can
624 be any pattern of bytes, and can be prefixed with 0x for hex
625 values. It may also be a string, where the string must then
626 be wrapped with "".
627
628dedupe_percentage=int If set, fio will generate this percentage of
629 identical buffers when writing. These buffers will be
630 naturally dedupable. The contents of the buffers depend on
631 what other buffer compression settings have been set. It's
632 possible to have the individual buffers either fully
633 compressible, or not at all. This option only controls the
634 distribution of unique buffers.
ce35b1ec 635
71bfa161
JA
636nrfiles=int Number of files to use for this job. Defaults to 1.
637
390b1537
JA
638openfiles=int Number of files to keep open at the same time. Defaults to
639 the same as nrfiles, can be set smaller to limit the number
640 simultaneous opens.
641
5af1c6f3
JA
642file_service_type=str Defines how fio decides which file from a job to
643 service next. The following types are defined:
644
645 random Just choose a file at random.
646
647 roundrobin Round robin over open files. This
648 is the default.
649
a086c257
JA
650 sequential Finish one file before moving on to
651 the next. Multiple files can still be
652 open depending on 'openfiles'.
653
1907dbc6
JA
654 The string can have a number appended, indicating how
655 often to switch to a new file. So if option random:4 is
656 given, fio will switch to a new random file after 4 ios
657 have been issued.
658
71bfa161
JA
659ioengine=str Defines how the job issues io to the file. The following
660 types are defined:
661
662 sync Basic read(2) or write(2) io. lseek(2) is
663 used to position the io location.
664
a31041ea 665 psync Basic pread(2) or pwrite(2) io.
666
e05af9e5 667 vsync Basic readv(2) or writev(2) IO.
1d2af02a 668
a46c5e01
JA
669 psyncv Basic preadv(2) or pwritev(2) IO.
670
15d182aa
JA
671 libaio Linux native asynchronous io. Note that Linux
672 may only support queued behaviour with
673 non-buffered IO (set direct=1 or buffered=0).
de890a1e 674 This engine defines engine specific options.
71bfa161
JA
675
676 posixaio glibc posix asynchronous io.
677
417f0068
JA
678 solarisaio Solaris native asynchronous io.
679
03e20d68
BC
680 windowsaio Windows native asynchronous io.
681
71bfa161
JA
682 mmap File is memory mapped and data copied
683 to/from using memcpy(3).
684
685 splice splice(2) is used to transfer the data and
686 vmsplice(2) to transfer data from user
687 space to the kernel.
688
d0ff85df
JA
689 syslet-rw Use the syslet system calls to make
690 regular read/write async.
691
71bfa161 692 sg SCSI generic sg v3 io. May either be
6c219763 693 synchronous using the SG_IO ioctl, or if
71bfa161
JA
694 the target is an sg character device
695 we use read(2) and write(2) for asynchronous
696 io.
697
a94ea28b
JA
698 null Doesn't transfer any data, just pretends
699 to. This is mainly used to exercise fio
700 itself and for debugging/testing purposes.
701
ed92ac0c 702 net Transfer over the network to given host:port.
de890a1e
SL
703 Depending on the protocol used, the hostname,
704 port, listen and filename options are used to
705 specify what sort of connection to make, while
706 the protocol option determines which protocol
707 will be used.
708 This engine defines engine specific options.
ed92ac0c 709
9cce02e8
JA
710 netsplice Like net, but uses splice/vmsplice to
711 map data and send/receive.
de890a1e 712 This engine defines engine specific options.
9cce02e8 713
53aec0a4 714 cpuio Doesn't transfer any data, but burns CPU
ba0fbe10
JA
715 cycles according to the cpuload= and
716 cpucycle= options. Setting cpuload=85
717 will cause that job to do nothing but burn
36ecec83
GP
718 85% of the CPU. In case of SMP machines,
719 use numjobs=<no_of_cpu> to get desired CPU
720 usage, as the cpuload only loads a single
721 CPU at the desired rate.
ba0fbe10 722
e9a1806f
JA
723 guasi The GUASI IO engine is the Generic Userspace
724 Asyncronous Syscall Interface approach
725 to async IO. See
726
727 http://www.xmailserver.org/guasi-lib.html
728
729 for more info on GUASI.
730
21b8aee8 731 rdma The RDMA I/O engine supports both RDMA
eb52fa3f
BVA
732 memory semantics (RDMA_WRITE/RDMA_READ) and
733 channel semantics (Send/Recv) for the
734 InfiniBand, RoCE and iWARP protocols.
21b8aee8 735
b861be9f
JA
736 falloc IO engine that does regular fallocate to
737 simulate data transfer as fio ioengine.
738 DDIR_READ does fallocate(,mode = keep_size,)
739 DDIR_WRITE does fallocate(,mode = 0)
740 DDIR_TRIM does fallocate(,mode = punch_hole)
d54fce84
DM
741
742 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
b861be9f
JA
743 ioctls to simulate defragment activity in
744 request to DDIR_WRITE event
745
746 rbd IO engine supporting direct access to Ceph
747 Rados Block Devices (RBD) via librbd without
748 the need to use the kernel rbd driver. This
749 ioengine defines engine specific options.
750
751 gfapi Using Glusterfs libgfapi sync interface to
752 direct access to Glusterfs volumes without
753 options.
754
755 gfapi_async Using Glusterfs libgfapi async interface
756 to direct access to Glusterfs volumes without
757 having to go through FUSE. This ioengine
758 defines engine specific options.
0981fd71 759
b74e419e
MM
760 libhdfs Read and write through Hadoop (HDFS).
761 The 'filename' option is used to specify host,
762 port of the hdfs name-node to connect. This
763 engine interprets offsets a little
764 differently. In HDFS, files once created
765 cannot be modified. So random writes are not
766 possible. To imitate this, libhdfs engine
767 expects bunch of small files to be created
768 over HDFS, and engine will randomly pick a
769 file out of those files based on the offset
770 generated by fio backend. (see the example
771 job file to create such files, use rw=write
772 option). Please note, you might want to set
773 necessary environment variables to work with
774 hdfs/libhdfs properly.
1b10477b 775
65fa28ca
DE
776 mtd Read, write and erase an MTD character device
777 (e.g., /dev/mtd0). Discards are treated as
778 erases. Depending on the underlying device
779 type, the I/O may have to go in a certain
780 pattern, e.g., on NAND, writing sequentially
781 to erase blocks and discarding before
782 overwriting. The writetrim mode works well
783 for this constraint.
784
8a7bd877
JA
785 external Prefix to specify loading an external
786 IO engine object file. Append the engine
787 filename, eg ioengine=external:/tmp/foo.o
788 to load ioengine foo.o in /tmp.
789
71bfa161
JA
790iodepth=int This defines how many io units to keep in flight against
791 the file. The default is 1 for each file defined in this
792 job, can be overridden with a larger value for higher
ee72ca09
JA
793 concurrency. Note that increasing iodepth beyond 1 will not
794 affect synchronous ioengines (except for small degress when
9b836561 795 verify_async is in use). Even async engines may impose OS
ee72ca09
JA
796 restrictions causing the desired depth not to be achieved.
797 This may happen on Linux when using libaio and not setting
798 direct=1, since buffered IO is not async on that OS. Keep an
799 eye on the IO depth distribution in the fio output to verify
800 that the achieved depth is as expected. Default: 1.
71bfa161 801
4950421a 802iodepth_batch_submit=int
cb5ab512 803iodepth_batch=int This defines how many pieces of IO to submit at once.
89e820f6
JA
804 It defaults to 1 which means that we submit each IO
805 as soon as it is available, but can be raised to submit
806 bigger batches of IO at the time.
cb5ab512 807
4950421a
JA
808iodepth_batch_complete=int This defines how many pieces of IO to retrieve
809 at once. It defaults to 1 which means that we'll ask
810 for a minimum of 1 IO in the retrieval process from
811 the kernel. The IO retrieval will go on until we
812 hit the limit set by iodepth_low. If this variable is
813 set to 0, then fio will always check for completed
814 events before queuing more IO. This helps reduce
815 IO latency, at the cost of more retrieval system calls.
816
e916b390
JA
817iodepth_low=int The low water mark indicating when to start filling
818 the queue again. Defaults to the same as iodepth, meaning
819 that fio will attempt to keep the queue full at all times.
820 If iodepth is set to eg 16 and iodepth_low is set to 4, then
821 after fio has filled the queue of 16 requests, it will let
822 the depth drain down to 4 before starting to fill it again.
823
1ad01bd1
JA
824io_submit_mode=str This option controls how fio submits the IO to
825 the IO engine. The default is 'inline', which means that the
826 fio job threads submit and reap IO directly. If set to
827 'offload', the job threads will offload IO submission to a
828 dedicated pool of IO threads. This requires some coordination
829 and thus has a bit of extra overhead, especially for lower
830 queue depth IO where it can increase latencies. The benefit
831 is that fio can manage submission rates independently of
832 the device completion rates. This avoids skewed latency
833 reporting if IO gets back up on the device side (the
834 coordinated omission problem).
835
71bfa161 836direct=bool If value is true, use non-buffered io. This is usually
9b836561 837 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
93bcfd20 838 On Windows the synchronous ioengines don't support direct io.
76a43db4 839
d01612f3
CM
840atomic=bool If value is true, attempt to use atomic direct IO. Atomic
841 writes are guaranteed to be stable once acknowledged by
842 the operating system. Only Linux supports O_ATOMIC right
843 now.
844
76a43db4
JA
845buffered=bool If value is true, use buffered io. This is the opposite
846 of the 'direct' option. Defaults to true.
71bfa161 847
f7fa2653 848offset=int Start io at the given offset in the file. The data before
71bfa161
JA
849 the given offset will not be touched. This effectively
850 caps the file size at real_size - offset.
851
214ac7e0 852offset_increment=int If this is provided, then the real offset becomes
69bdd6ba
JH
853 offset + offset_increment * thread_number, where the thread
854 number is a counter that starts at 0 and is incremented for
855 each sub-job (i.e. when numjobs option is specified). This
856 option is useful if there are several jobs which are intended
857 to operate on a file in parallel disjoint segments, with
858 even spacing between the starting points.
214ac7e0 859
ddf24e42
JA
860number_ios=int Fio will normally perform IOs until it has exhausted the size
861 of the region set by size=, or if it exhaust the allocated
862 time (or hits an error condition). With this setting, the
863 range/size can be set independently of the number of IOs to
864 perform. When fio reaches this number, it will exit normally
be3fec7d
JA
865 and report status. Note that this does not extend the amount
866 of IO that will be done, it will only stop fio if this
867 condition is met before other end-of-job criteria.
ddf24e42 868
71bfa161
JA
869fsync=int If writing to a file, issue a sync of the dirty data
870 for every number of blocks given. For example, if you give
871 32 as a parameter, fio will sync the file for every 32
872 writes issued. If fio is using non-buffered io, we may
873 not sync the file. The exception is the sg io engine, which
6c219763 874 synchronizes the disk cache anyway.
71bfa161 875
e76b1da4 876fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
5f9099ea 877 metadata blocks.
93bcfd20 878 In FreeBSD and Windows there is no fdatasync(), this falls back to
e72fa4d4 879 using fsync()
5f9099ea 880
e76b1da4
JA
881sync_file_range=str:val Use sync_file_range() for every 'val' number of
882 write operations. Fio will track range of writes that
883 have happened since the last sync_file_range() call. 'str'
884 can currently be one or more of:
885
886 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
887 write SYNC_FILE_RANGE_WRITE
888 wait_after SYNC_FILE_RANGE_WAIT_AFTER
889
890 So if you do sync_file_range=wait_before,write:8, fio would
891 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
892 every 8 writes. Also see the sync_file_range(2) man page.
893 This option is Linux specific.
894
5036fc1e
JA
895overwrite=bool If true, writes to a file will always overwrite existing
896 data. If the file doesn't already exist, it will be
897 created before the write phase begins. If the file exists
898 and is large enough for the specified write phase, nothing
899 will be done.
71bfa161 900
dbd11ead 901end_fsync=bool If true, fsync file contents when a write stage has completed.
71bfa161 902
ebb1415f
JA
903fsync_on_close=bool If true, fio will fsync() a dirty file on close.
904 This differs from end_fsync in that it will happen on every
905 file close, not just at the end of the job.
906
71bfa161
JA
907rwmixread=int How large a percentage of the mix should be reads.
908
909rwmixwrite=int How large a percentage of the mix should be writes. If both
910 rwmixread and rwmixwrite is given and the values do not add
911 up to 100%, the latter of the two will be used to override
c35dd7a6
JA
912 the first. This may interfere with a given rate setting,
913 if fio is asked to limit reads or writes to a certain rate.
914 If that is the case, then the distribution may be skewed.
71bfa161 915
92d42d69
JA
916random_distribution=str:float By default, fio will use a completely uniform
917 random distribution when asked to perform random IO. Sometimes
918 it is useful to skew the distribution in specific ways,
919 ensuring that some parts of the data is more hot than others.
920 fio includes the following distribution models:
921
922 random Uniform random distribution
923 zipf Zipf distribution
924 pareto Pareto distribution
925
926 When using a zipf or pareto distribution, an input value
927 is also needed to define the access pattern. For zipf, this
928 is the zipf theta. For pareto, it's the pareto power. Fio
929 includes a test program, genzipf, that can be used visualize
930 what the given input values will yield in terms of hit rates.
931 If you wanted to use zipf with a theta of 1.2, you would use
932 random_distribution=zipf:1.2 as the option. If a non-uniform
933 model is used, fio will disable use of the random map.
934
211c9b89
JA
935percentage_random=int For a random workload, set how big a percentage should
936 be random. This defaults to 100%, in which case the workload
937 is fully random. It can be set from anywhere from 0 to 100.
938 Setting it to 0 would make the workload fully sequential. Any
939 setting in between will result in a random mix of sequential
d9472271
JA
940 and random IO, at the given percentages. It is possible to
941 set different values for reads, writes, and trim. To do so,
942 simply use a comma separated list. See blocksize.
211c9b89 943
bb8895e0
JA
944norandommap Normally fio will cover every block of the file when doing
945 random IO. If this option is given, fio will just get a
946 new random offset without looking at past io history. This
947 means that some blocks may not be read or written, and that
83da8fbf
JE
948 some blocks may be read/written more than once. If this option
949 is used with verify= and multiple blocksizes (via bsrange=),
950 only intact blocks are verified, i.e., partially-overwritten
951 blocks are ignored.
bb8895e0 952
0408c206
JA
953softrandommap=bool See norandommap. If fio runs with the random block map
954 enabled and it fails to allocate the map, if this option is
955 set it will continue without a random block map. As coverage
956 will not be as complete as with random maps, this option is
2b386d25
JA
957 disabled by default.
958
e8b1961d
JA
959random_generator=str Fio supports the following engines for generating
960 IO offsets for random IO:
961
962 tausworthe Strong 2^88 cycle random number generator
963 lfsr Linear feedback shift register generator
c3546b53
JA
964 tausworthe64 Strong 64-bit 2^258 cycle random number
965 generator
e8b1961d
JA
966
967 Tausworthe is a strong random number generator, but it
968 requires tracking on the side if we want to ensure that
969 blocks are only read or written once. LFSR guarantees
970 that we never generate the same offset twice, and it's
971 also less computationally expensive. It's not a true
972 random generator, however, though for IO purposes it's
973 typically good enough. LFSR only works with single
974 block sizes, not with workloads that use multiple block
975 sizes. If used with such a workload, fio may read or write
976 some blocks multiple times.
43f09da1 977
71bfa161
JA
978nice=int Run the job with the given nice value. See man nice(2).
979
980prio=int Set the io priority value of this job. Linux limits us to
981 a positive value between 0 and 7, with 0 being the highest.
982 See man ionice(1).
983
984prioclass=int Set the io priority class. See man ionice(1).
985
986thinktime=int Stall the job x microseconds after an io has completed before
987 issuing the next. May be used to simulate processing being
48097d5c
JA
988 done by an application. See thinktime_blocks and
989 thinktime_spin.
990
991thinktime_spin=int
992 Only valid if thinktime is set - pretend to spend CPU time
993 doing something with the data received, before falling back
994 to sleeping for the rest of the period specified by
995 thinktime.
9c1f7434 996
4d01ece6 997thinktime_blocks=int
9c1f7434
JA
998 Only valid if thinktime is set - control how many blocks
999 to issue, before waiting 'thinktime' usecs. If not set,
1000 defaults to 1 which will make fio wait 'thinktime' usecs
4d01ece6
JA
1001 after every block. This effectively makes any queue depth
1002 setting redundant, since no more than 1 IO will be queued
1003 before we have to complete it and do our thinktime. In
1004 other words, this setting effectively caps the queue depth
1005 if the latter is larger.
71bfa161 1006
581e7141 1007rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
b09da8fa 1008 the normal suffix rules apply. You can use rate=500k to limit
581e7141
JA
1009 reads and writes to 500k each, or you can specify read and
1010 writes separately. Using rate=1m,500k would limit reads to
1011 1MB/sec and writes to 500KB/sec. Capping only reads or
1012 writes can be done with rate=,500k or rate=500k,. The former
1013 will only limit writes (to 500KB/sec), the latter will only
1014 limit reads.
71bfa161
JA
1015
1016ratemin=int Tell fio to do whatever it can to maintain at least this
4e991c23 1017 bandwidth. Failing to meet this requirement, will cause
581e7141
JA
1018 the job to exit. The same format as rate is used for
1019 read vs write separation.
4e991c23
JA
1020
1021rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
1022 as rate, just specified independently of bandwidth. If the
1023 job is given a block size range instead of a fixed value,
581e7141 1024 the smallest block size is used as the metric. The same format
de8f6de9 1025 as rate is used for read vs write separation.
4e991c23
JA
1026
1027rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
581e7141 1028 the job to exit. The same format as rate is used for read vs
de8f6de9 1029 write separation.
71bfa161 1030
3e260a46
JA
1031latency_target=int If set, fio will attempt to find the max performance
1032 point that the given workload will run at while maintaining a
1033 latency below this target. The values is given in microseconds.
1034 See latency_window and latency_percentile
1035
1036latency_window=int Used with latency_target to specify the sample window
1037 that the job is run at varying queue depths to test the
1038 performance. The value is given in microseconds.
1039
1040latency_percentile=float The percentage of IOs that must fall within the
1041 criteria specified by latency_target and latency_window. If not
1042 set, this defaults to 100.0, meaning that all IOs must be equal
1043 or below to the value set by latency_target.
1044
15501535
JA
1045max_latency=int If set, fio will exit the job if it exceeds this maximum
1046 latency. It will exit with an ETIME error.
1047
71bfa161 1048ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
6c219763 1049 of milliseconds.
71bfa161
JA
1050
1051cpumask=int Set the CPU affinity of this job. The parameter given is a
a08bc17f
JA
1052 bitmask of allowed CPU's the job may run on. So if you want
1053 the allowed CPUs to be 1 and 5, you would pass the decimal
1054 value of (1 << 1 | 1 << 5), or 34. See man
7dbb6eba 1055 sched_setaffinity(2). This may not work on all supported
b0ea08ce
JA
1056 operating systems or kernel versions. This option doesn't
1057 work well for a higher CPU count than what you can store in
1058 an integer mask, so it can only control cpus 1-32. For
1059 boxes with larger CPU counts, use cpus_allowed.
71bfa161 1060
d2e268b0
JA
1061cpus_allowed=str Controls the same options as cpumask, but it allows a text
1062 setting of the permitted CPUs instead. So to use CPUs 1 and
62a7273d
JA
1063 5, you would specify cpus_allowed=1,5. This options also
1064 allows a range of CPUs. Say you wanted a binding to CPUs
1065 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
d2e268b0 1066
c2acfbac
JA
1067cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
1068 specified by cpus_allowed or cpumask. Two policies are
1069 supported:
1070
1071 shared All jobs will share the CPU set specified.
1072 split Each job will get a unique CPU from the CPU set.
1073
1074 'shared' is the default behaviour, if the option isn't
ada083cd
JA
1075 specified. If split is specified, then fio will will assign
1076 one cpu per job. If not enough CPUs are given for the jobs
1077 listed, then fio will roundrobin the CPUs in the set.
c2acfbac 1078
d0b937ed
YR
1079numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
1080 arguments allow comma delimited list of cpu numbers,
1081 A-B ranges, or 'all'. Note, to enable numa options support,
67bf9823 1082 fio must be built on a system with libnuma-dev(el) installed.
d0b937ed
YR
1083
1084numa_mem_policy=str Set this job's memory policy and corresponding NUMA
1085 nodes. Format of the argements:
1086 <mode>[:<nodelist>]
1087 `mode' is one of the following memory policy:
1088 default, prefer, bind, interleave, local
1089 For `default' and `local' memory policy, no node is
1090 needed to be specified.
1091 For `prefer', only one node is allowed.
1092 For `bind' and `interleave', it allow comma delimited
1093 list of numbers, A-B ranges, or 'all'.
1094
e417fd66 1095startdelay=time Start this job the specified number of seconds after fio
71bfa161
JA
1096 has started. Only useful if the job file contains several
1097 jobs, and you want to delay starting some jobs to a certain
1098 time.
1099
e417fd66 1100runtime=time Tell fio to terminate processing after the specified number
71bfa161
JA
1101 of seconds. It can be quite hard to determine for how long
1102 a specified job will run, so this parameter is handy to
1103 cap the total runtime to a given time.
1104
cf4464ca 1105time_based If set, fio will run for the duration of the runtime
bf9a3edb 1106 specified even if the file(s) are completely read or
cf4464ca
JA
1107 written. It will simply loop over the same workload
1108 as many times as the runtime allows.
1109
e417fd66 1110ramp_time=time If set, fio will run the specified workload for this amount
721938ae
JA
1111 of time before logging any performance numbers. Useful for
1112 letting performance settle before logging results, thus
b29ee5b3
JA
1113 minimizing the runtime required for stable results. Note
1114 that the ramp_time is considered lead in time for a job,
1115 thus it will increase the total runtime if a special timeout
1116 or runtime is specified.
721938ae 1117
71bfa161
JA
1118invalidate=bool Invalidate the buffer/page cache parts for this file prior
1119 to starting io. Defaults to true.
1120
1121sync=bool Use sync io for buffered writes. For the majority of the
1122 io engines, this means using O_SYNC.
1123
d3aad8f2 1124iomem=str
71bfa161
JA
1125mem=str Fio can use various types of memory as the io unit buffer.
1126 The allowed values are:
1127
1128 malloc Use memory from malloc(3) as the buffers.
1129
1130 shm Use shared memory as the buffers. Allocated
1131 through shmget(2).
1132
74b025b0
JA
1133 shmhuge Same as shm, but use huge pages as backing.
1134
313cb206
JA
1135 mmap Use mmap to allocate buffers. May either be
1136 anonymous memory, or can be file backed if
1137 a filename is given after the option. The
1138 format is mem=mmap:/path/to/file.
71bfa161 1139
d0bdaf49
JA
1140 mmaphuge Use a memory mapped huge file as the buffer
1141 backing. Append filename after mmaphuge, ala
1142 mem=mmaphuge:/hugetlbfs/file
1143
71bfa161 1144 The area allocated is a function of the maximum allowed
5394ae5f
JA
1145 bs size for the job, multiplied by the io depth given. Note
1146 that for shmhuge and mmaphuge to work, the system must have
1147 free huge pages allocated. This can normally be checked
1148 and set by reading/writing /proc/sys/vm/nr_hugepages on a
b22989b9 1149 Linux system. Fio assumes a huge page is 4MB in size. So
5394ae5f
JA
1150 to calculate the number of huge pages you need for a given
1151 job file, add up the io depth of all jobs (normally one unless
1152 iodepth= is used) and multiply by the maximum bs set. Then
1153 divide that number by the huge page size. You can see the
1154 size of the huge pages in /proc/meminfo. If no huge pages
1155 are allocated by having a non-zero number in nr_hugepages,
56bb17f2 1156 using mmaphuge or shmhuge will fail. Also see hugepage-size.
5394ae5f
JA
1157
1158 mmaphuge also needs to have hugetlbfs mounted and the file
1159 location should point there. So if it's mounted in /huge,
1160 you would use mem=mmaphuge:/huge/somefile.
71bfa161 1161
d529ee19
JA
1162iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1163 Note that the given alignment is applied to the first IO unit
1164 buffer, if using iodepth the alignment of the following buffers
1165 are given by the bs used. In other words, if using a bs that is
1166 a multiple of the page sized in the system, all buffers will
1167 be aligned to this value. If using a bs that is not page
1168 aligned, the alignment of subsequent IO memory buffers is the
1169 sum of the iomem_align and bs used.
1170
f7fa2653 1171hugepage-size=int
56bb17f2 1172 Defines the size of a huge page. Must at least be equal
b22989b9 1173 to the system setting, see /proc/meminfo. Defaults to 4MB.
c51074e7
JA
1174 Should probably always be a multiple of megabytes, so using
1175 hugepage-size=Xm is the preferred way to set this to avoid
1176 setting a non-pow-2 bad value.
56bb17f2 1177
71bfa161
JA
1178exitall When one job finishes, terminate the rest. The default is
1179 to wait for each job to finish, sometimes that is not the
1180 desired action.
1181
1182bwavgtime=int Average the calculated bandwidth over the given time. Value
6c219763 1183 is specified in milliseconds.
71bfa161 1184
c8eeb9df
JA
1185iopsavgtime=int Average the calculated IOPS over the given time. Value
1186 is specified in milliseconds.
1187
71bfa161
JA
1188create_serialize=bool If true, serialize the file creating for the jobs.
1189 This may be handy to avoid interleaving of data
1190 files, which may greatly depend on the filesystem
1191 used and even the number of processors in the system.
1192
1193create_fsync=bool fsync the data file after creation. This is the
1194 default.
1195
814452bd
JA
1196create_on_open=bool Don't pre-setup the files for IO, just create open()
1197 when it's time to do IO to that file.
1198
25460cf6
JA
1199create_only=bool If true, fio will only run the setup phase of the job.
1200 If files need to be laid out or updated on disk, only
1201 that will be done. The actual job contents are not
1202 executed.
1203
2378826d
JA
1204allow_file_create=bool If true, fio is permitted to create files as part
1205 of its workload. This is the default behavior. If this
1206 option is false, then fio will error out if the files it
1207 needs to use don't already exist. Default: true.
1208
e81ecca3
JA
1209allow_mounted_write=bool If this isn't set, fio will abort jobs that
1210 are destructive (eg that write) to what appears to be a
1211 mounted device or partition. This should help catch creating
1212 inadvertently destructive tests, not realizing that the test
1213 will destroy data on the mounted file system. Default: false.
1214
afad68f7 1215pre_read=bool If this is given, files will be pre-read into memory before
34f1c044
JA
1216 starting the given IO operation. This will also clear
1217 the 'invalidate' flag, since it is pointless to pre-read
9c0d2241
JA
1218 and then drop the cache. This will only work for IO engines
1219 that are seekable, since they allow you to read the same data
1220 multiple times. Thus it will not work on eg network or splice
1221 IO.
afad68f7 1222
e545a6ce 1223unlink=bool Unlink the job files when done. Not the default, as repeated
bf9a3edb
JA
1224 runs of that job would then waste time recreating the file
1225 set again and again.
71bfa161
JA
1226
1227loops=int Run the specified number of iterations of this job. Used
1228 to repeat the same workload a given number of times. Defaults
1229 to 1.
1230
62167762
JC
1231verify_only Do not perform specified workload---only verify data still
1232 matches previous invocation of this workload. This option
1233 allows one to check data multiple times at a later date
1234 without overwriting it. This option makes sense only for
1235 workloads that write data, and does not support workloads
1236 with the time_based option set.
1237
68e1f29a 1238do_verify=bool Run the verify phase after a write phase. Only makes sense if
e84c73a8
SL
1239 verify is set. Defaults to 1.
1240
71bfa161
JA
1241verify=str If writing to a file, fio can verify the file contents
1242 after each iteration of the job. The allowed values are:
1243
1244 md5 Use an md5 sum of the data area and store
1245 it in the header of each block.
1246
17dc34df
JA
1247 crc64 Use an experimental crc64 sum of the data
1248 area and store it in the header of each
1249 block.
1250
bac39e0e
JA
1251 crc32c Use a crc32c sum of the data area and store
1252 it in the header of each block.
1253
3845591f 1254 crc32c-intel Use hardware assisted crc32c calcuation
0539d758
JA
1255 provided on SSE4.2 enabled processors. Falls
1256 back to regular software crc32c, if not
1257 supported by the system.
3845591f 1258
71bfa161
JA
1259 crc32 Use a crc32 sum of the data area and store
1260 it in the header of each block.
1261
969f7ed3
JA
1262 crc16 Use a crc16 sum of the data area and store
1263 it in the header of each block.
1264
17dc34df
JA
1265 crc7 Use a crc7 sum of the data area and store
1266 it in the header of each block.
1267
844ea602
JA
1268 xxhash Use xxhash as the checksum function. Generally
1269 the fastest software checksum that fio
1270 supports.
1271
cd14cc10
JA
1272 sha512 Use sha512 as the checksum function.
1273
1274 sha256 Use sha256 as the checksum function.
1275
7c353ceb
JA
1276 sha1 Use optimized sha1 as the checksum function.
1277
7437ee87
SL
1278 meta Write extra information about each io
1279 (timestamp, block number etc.). The block
62167762
JC
1280 number is verified. The io sequence number is
1281 verified for workloads that write data.
1282 See also verify_pattern.
7437ee87 1283
59245381
JA
1284 pattern Verify a strict pattern. Normally fio includes
1285 a header with some basic information and
1286 checksumming, but if this option is set, only
1287 the specific pattern set with 'verify_pattern'
1288 is verified.
1289
36690c9b
JA
1290 null Only pretend to verify. Useful for testing
1291 internals with ioengine=null, not for much
1292 else.
1293
6c219763 1294 This option can be used for repeated burn-in tests of a
71bfa161 1295 system to make sure that the written data is also
b892dc08
JA
1296 correctly read back. If the data direction given is
1297 a read or random read, fio will assume that it should
1298 verify a previously written file. If the data direction
1299 includes any form of write, the verify will be of the
1300 newly written data.
71bfa161 1301
160b966d
JA
1302verifysort=bool If set, fio will sort written verify blocks when it deems
1303 it faster to read them back in a sorted manner. This is
1304 often the case when overwriting an existing file, since
1305 the blocks are already laid out in the file system. You
1306 can ignore this option unless doing huge amounts of really
1307 fast IO where the red-black tree sorting CPU time becomes
1308 significant.
3f9f4e26 1309
f7fa2653 1310verify_offset=int Swap the verification header with data somewhere else
546a9142
SL
1311 in the block before writing. Its swapped back before
1312 verifying.
1313
f7fa2653 1314verify_interval=int Write the verification header at a finer granularity
3f9f4e26
SL
1315 than the blocksize. It will be written for chunks the
1316 size of header_interval. blocksize should divide this
1317 evenly.
90059d65 1318
0e92f873 1319verify_pattern=str If set, fio will fill the io buffers with this
e28218f3
SL
1320 pattern. Fio defaults to filling with totally random
1321 bytes, but sometimes it's interesting to fill with a known
1322 pattern for io verification purposes. Depending on the
1323 width of the pattern, fio will fill 1/2/3/4 bytes of the
0e92f873
RR
1324 buffer at the time(it can be either a decimal or a hex number).
1325 The verify_pattern if larger than a 32-bit quantity has to
996093bb
JA
1326 be a hex number that starts with either "0x" or "0X". Use
1327 with verify=meta.
e28218f3 1328
68e1f29a 1329verify_fatal=bool Normally fio will keep checking the entire contents
a12a3b4d
JA
1330 before quitting on a block verification failure. If this
1331 option is set, fio will exit the job on the first observed
1332 failure.
e8462bd8 1333
b463e936
JA
1334verify_dump=bool If set, dump the contents of both the original data
1335 block and the data block we read off disk to files. This
1336 allows later analysis to inspect just what kind of data
ef71e317 1337 corruption occurred. Off by default.
b463e936 1338
e8462bd8
JA
1339verify_async=int Fio will normally verify IO inline from the submitting
1340 thread. This option takes an integer describing how many
1341 async offload threads to create for IO verification instead,
1342 causing fio to offload the duty of verifying IO contents
c85c324c
JA
1343 to one or more separate threads. If using this offload
1344 option, even sync IO engines can benefit from using an
1345 iodepth setting higher than 1, as it allows them to have
1346 IO in flight while verifies are running.
e8462bd8
JA
1347
1348verify_async_cpus=str Tell fio to set the given CPU affinity on the
1349 async IO verification threads. See cpus_allowed for the
1350 format used.
6f87418f
JA
1351
1352verify_backlog=int Fio will normally verify the written contents of a
1353 job that utilizes verify once that job has completed. In
1354 other words, everything is written then everything is read
1355 back and verified. You may want to verify continually
1356 instead for a variety of reasons. Fio stores the meta data
1357 associated with an IO block in memory, so for large
1358 verify workloads, quite a bit of memory would be used up
1359 holding this meta data. If this option is enabled, fio
f42195a3
JA
1360 will write only N blocks before verifying these blocks.
1361
6f87418f
JA
1362verify_backlog_batch=int Control how many blocks fio will verify
1363 if verify_backlog is set. If not set, will default to
1364 the value of verify_backlog (meaning the entire queue
f42195a3
JA
1365 is read back and verified). If verify_backlog_batch is
1366 less than verify_backlog then not all blocks will be verified,
1367 if verify_backlog_batch is larger than verify_backlog, some
1368 blocks will be verified more than once.
66c098b8 1369
ca09be4b
JA
1370verify_state_save=bool When a job exits during the write phase of a verify
1371 workload, save its current state. This allows fio to replay
1372 up until that point, if the verify state is loaded for the
1373 verify read phase. The format of the filename is, roughly,
1374 <type>-<jobname>-<jobindex>-verify.state. <type> is "local"
1375 for a local run, "sock" for a client/server socket connection,
1376 and "ip" (192.168.0.1, for instance) for a networked
1377 client/server connection.
1378
1379verify_state_load=bool If a verify termination trigger was used, fio stores
1380 the current write state of each thread. This can be used at
1381 verification time so that fio knows how far it should verify.
1382 Without this information, fio will run a full verification
1383 pass, according to the settings in the job file used.
1384
d392365e 1385stonewall
de8f6de9 1386wait_for_previous Wait for preceding jobs in the job file to exit, before
71bfa161 1387 starting this one. Can be used to insert serialization
b3d62a75
JA
1388 points in the job file. A stone wall also implies starting
1389 a new reporting group.
1390
abcab6af 1391new_group Start a new reporting group. See: group_reporting.
71bfa161
JA
1392
1393numjobs=int Create the specified number of clones of this job. May be
1394 used to setup a larger number of threads/processes doing
abcab6af
AV
1395 the same thing. Each thread is reported separately; to see
1396 statistics for all clones as a whole, use group_reporting in
1397 conjunction with new_group.
1398
1399group_reporting It may sometimes be interesting to display statistics for
04b2f799
JA
1400 groups of jobs as a whole instead of for each individual job.
1401 This is especially true if 'numjobs' is used; looking at
1402 individual thread/process output quickly becomes unwieldy.
1403 To see the final report per-group instead of per-job, use
1404 'group_reporting'. Jobs in a file will be part of the same
1405 reporting group, unless if separated by a stonewall, or by
1406 using 'new_group'.
71bfa161
JA
1407
1408thread fio defaults to forking jobs, however if this option is
1409 given, fio will use pthread_create(3) to create threads
1410 instead.
1411
f7fa2653 1412zonesize=int Divide a file into zones of the specified size. See zoneskip.
71bfa161 1413
f7fa2653 1414zoneskip=int Skip the specified number of bytes when zonesize data has
71bfa161
JA
1415 been read. The two zone options can be used to only do
1416 io on zones of a file.
1417
076efc7c 1418write_iolog=str Write the issued io patterns to the specified file. See
5b42a488
SH
1419 read_iolog. Specify a separate file for each job, otherwise
1420 the iologs will be interspersed and the file may be corrupt.
71bfa161 1421
076efc7c 1422read_iolog=str Open an iolog with the specified file name and replay the
71bfa161 1423 io patterns it contains. This can be used to store a
6df8adaa
JA
1424 workload and replay it sometime later. The iolog given
1425 may also be a blktrace binary file, which allows fio
1426 to replay a workload captured by blktrace. See blktrace
1427 for how to capture such logging data. For blktrace replay,
1428 the file needs to be turned into a blkparse binary data
ea3e51c3 1429 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
66c098b8 1430
64bbb865 1431replay_no_stall=int When replaying I/O with read_iolog the default behavior
62776229
JA
1432 is to attempt to respect the time stamps within the log and
1433 replay them with the appropriate delay between IOPS. By
1434 setting this variable fio will not respect the timestamps and
1435 attempt to replay them as fast as possible while still
1436 respecting ordering. The result is the same I/O pattern to a
1437 given device, but different timings.
71bfa161 1438
d1c46c04
DN
1439replay_redirect=str While replaying I/O patterns using read_iolog the
1440 default behavior is to replay the IOPS onto the major/minor
1441 device that each IOP was recorded from. This is sometimes
de8f6de9 1442 undesirable because on a different machine those major/minor
d1c46c04
DN
1443 numbers can map to a different device. Changing hardware on
1444 the same system can also result in a different major/minor
1445 mapping. Replay_redirect causes all IOPS to be replayed onto
1446 the single specified device regardless of the device it was
1447 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1448 IO in the blktrace to be replayed onto /dev/sdc. This means
1449 multiple devices will be replayed onto a single, if the trace
1450 contains multiple devices. If you want multiple devices to be
1451 replayed concurrently to multiple redirected devices you must
1452 blkparse your trace into separate traces and replay them with
1453 independent fio invocations. Unfortuantely this also breaks
1454 the strict time ordering between multiple device accesses.
1455
0c63576e
JA
1456replay_align=int Force alignment of IO offsets and lengths in a trace
1457 to this power of 2 value.
1458
1459replay_scale=int Scale sector offsets down by this factor when
1460 replaying traces.
1461
3a5db920
JA
1462per_job_logs=bool If set, this generates bw/clat/iops log with per
1463 file private filenames. If not set, jobs with identical names
1464 will share the log filename. Default: true.
1465
e3cedca7 1466write_bw_log=str If given, write a bandwidth log of the jobs in this job
71bfa161 1467 file. Can be used to store data of the bandwidth of the
e0da9bc2
JA
1468 jobs in their lifetime. The included fio_generate_plots
1469 script uses gnuplot to turn these text files into nice
ddb754db 1470 graphs. See write_lat_log for behaviour of given
8ad3b3dd
JA
1471 filename. For this option, the suffix is _bw.x.log, where
1472 x is the index of the job (1..N, where N is the number of
3a5db920
JA
1473 jobs). If 'per_job_logs' is false, then the filename will not
1474 include the job index.
71bfa161 1475
e3cedca7 1476write_lat_log=str Same as write_bw_log, except that this option stores io
02af0988
JA
1477 submission, completion, and total latencies instead. If no
1478 filename is given with this option, the default filename of
1479 "jobname_type.log" is used. Even if the filename is given,
1480 fio will still append the type of log. So if one specifies
e3cedca7
JA
1481
1482 write_lat_log=foo
1483
8ad3b3dd
JA
1484 The actual log names will be foo_slat.x.log, foo_clat.x.log,
1485 and foo_lat.x.log, where x is the index of the job (1..N,
1486 where N is the number of jobs). This helps fio_generate_plot
3a5db920
JA
1487 fine the logs automatically. If 'per_job_logs' is false, then
1488 the filename will not include the job index.
1489
71bfa161 1490
b8bc8cba
JA
1491write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1492 given with this option, the default filename of
8ad3b3dd
JA
1493 "jobname_type.x.log" is used,where x is the index of the job
1494 (1..N, where N is the number of jobs). Even if the filename
3a5db920
JA
1495 is given, fio will still append the type of log. If
1496 'per_job_logs' is false, then the filename will not include
1497 the job index.
b8bc8cba
JA
1498
1499log_avg_msec=int By default, fio will log an entry in the iops, latency,
1500 or bw log for every IO that completes. When writing to the
1501 disk log, that can quickly grow to a very large size. Setting
1502 this option makes fio average the each log entry over the
1503 specified period of time, reducing the resolution of the log.
1504 Defaults to 0.
1505
ae588852
JA
1506log_offset=int If this is set, the iolog options will include the byte
1507 offset for the IO entry as well as the other data values.
1508
aee2ab67
JA
1509log_compression=int If this is set, fio will compress the IO logs as
1510 it goes, to keep the memory footprint lower. When a log
1511 reaches the specified size, that chunk is removed and
1512 compressed in the background. Given that IO logs are
1513 fairly highly compressible, this yields a nice memory
1514 savings for longer runs. The downside is that the
1515 compression will consume some background CPU cycles, so
1516 it may impact the run. This, however, is also true if
1517 the logging ends up consuming most of the system memory.
1518 So pick your poison. The IO logs are saved normally at the
1519 end of a run, by decompressing the chunks and storing them
1520 in the specified log file. This feature depends on the
1521 availability of zlib.
1522
b26317c9
JA
1523log_store_compressed=bool If set, and log_compression is also set,
1524 fio will store the log files in a compressed format. They
1525 can be decompressed with fio, using the --inflate-log
1526 command line parameter. The files will be stored with a
1527 .fz suffix.
1528
66347cfa
DE
1529block_error_percentiles=bool If set, record errors in trim block-sized
1530 units from writes and trims and output a histogram of
1531 how many trims it took to get to errors, and what kind
1532 of error was encountered.
1533
f7fa2653 1534lockmem=int Pin down the specified amount of memory with mlock(2). Can
71bfa161
JA
1535 potentially be used instead of removing memory or booting
1536 with less memory to simulate a smaller amount of memory.
81c6b6cd 1537 The amount specified is per worker.
71bfa161
JA
1538
1539exec_prerun=str Before running this job, issue the command specified
74c8c488
JA
1540 through system(3). Output is redirected in a file called
1541 jobname.prerun.txt.
71bfa161
JA
1542
1543exec_postrun=str After the job completes, issue the command specified
74c8c488
JA
1544 though system(3). Output is redirected in a file called
1545 jobname.postrun.txt.
71bfa161
JA
1546
1547ioscheduler=str Attempt to switch the device hosting the file to the specified
1548 io scheduler before running.
1549
0a839f30
JA
1550disk_util=bool Generate disk utilization statistics, if the platform
1551 supports it. Defaults to on.
1552
02af0988 1553disable_lat=bool Disable measurements of total latency numbers. Useful
9520ebb9
JA
1554 only for cutting back the number of calls to gettimeofday,
1555 as that does impact performance at really high IOPS rates.
1556 Note that to really get rid of a large amount of these
1557 calls, this option must be used with disable_slat and
1558 disable_bw as well.
1559
02af0988
JA
1560disable_clat=bool Disable measurements of completion latency numbers. See
1561 disable_lat.
1562
9520ebb9 1563disable_slat=bool Disable measurements of submission latency numbers. See
02af0988 1564 disable_slat.
9520ebb9
JA
1565
1566disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
02af0988 1567 disable_lat.
9520ebb9 1568
83349190
YH
1569clat_percentiles=bool Enable the reporting of percentiles of
1570 completion latencies.
1571
1572percentile_list=float_list Overwrite the default list of percentiles
66347cfa
DE
1573 for completion latencies and the block error histogram.
1574 Each number is a floating number in the range (0,100],
1575 and the maximum length of the list is 20. Use ':'
1576 to separate the numbers, and list the numbers in ascending
1577 order. For example, --percentile_list=99.5:99.9 will cause
1578 fio to report the values of completion latency below which
1579 99.5% and 99.9% of the observed latencies fell, respectively.
83349190 1580
23893646
JA
1581clocksource=str Use the given clocksource as the base of timing. The
1582 supported options are:
1583
1584 gettimeofday gettimeofday(2)
1585
1586 clock_gettime clock_gettime(2)
1587
1588 cpu Internal CPU clock source
1589
1590 cpu is the preferred clocksource if it is reliable, as it
1591 is very fast (and fio is heavy on time calls). Fio will
1592 automatically use this clocksource if it's supported and
1593 considered reliable on the system it is running on, unless
1594 another clocksource is specifically set. For x86/x86-64 CPUs,
1595 this means supporting TSC Invariant.
1596
993bf48b
JA
1597gtod_reduce=bool Enable all of the gettimeofday() reducing options
1598 (disable_clat, disable_slat, disable_bw) plus reduce
1599 precision of the timeout somewhat to really shrink
1600 the gettimeofday() call count. With this option enabled,
1601 we only do about 0.4% of the gtod() calls we would have
1602 done if all time keeping was enabled.
1603
be4ecfdf
JA
1604gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1605 execution to just getting the current time. Fio (and
1606 databases, for instance) are very intensive on gettimeofday()
1607 calls. With this option, you can set one CPU aside for
1608 doing nothing but logging current time to a shared memory
1609 location. Then the other threads/processes that run IO
1610 workloads need only copy that segment, instead of entering
1611 the kernel with a gettimeofday() call. The CPU set aside
1612 for doing these time calls will be excluded from other
1613 uses. Fio will manually clear it from the CPU mask of other
1614 jobs.
a696fa2a 1615
06842027 1616continue_on_error=str Normally fio will exit the job on the first observed
f2bba182
RR
1617 failure. If this option is set, fio will continue the job when
1618 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1619 is exceeded or the I/O size specified is completed. If this
1620 option is used, there are two more stats that are appended,
1621 the total error count and the first error. The error field
1622 given in the stats is the first error that was hit during the
1623 run.
be4ecfdf 1624
06842027
SL
1625 The allowed values are:
1626
1627 none Exit on any IO or verify errors.
1628
1629 read Continue on read errors, exit on all others.
1630
1631 write Continue on write errors, exit on all others.
1632
1633 io Continue on any IO error, exit on all others.
1634
1635 verify Continue on verify errors, exit on all others.
1636
1637 all Continue on all errors.
1638
1639 0 Backward-compatible alias for 'none'.
1640
1641 1 Backward-compatible alias for 'all'.
1642
8b28bd41
DM
1643ignore_error=str Sometimes you want to ignore some errors during test
1644 in that case you can specify error list for each error type.
1645 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1646 errors for given error type is separated with ':'. Error
1647 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1648 Example:
1649 ignore_error=EAGAIN,ENOSPC:122
66c098b8
BC
1650 This option will ignore EAGAIN from READ, and ENOSPC and
1651 122(EDQUOT) from WRITE.
8b28bd41
DM
1652
1653error_dump=bool If set dump every error even if it is non fatal, true
1654 by default. If disabled only fatal error will be dumped
66c098b8 1655
6adb38a1
JA
1656cgroup=str Add job to this control group. If it doesn't exist, it will
1657 be created. The system must have a mounted cgroup blkio
1658 mount point for this to work. If your system doesn't have it
1659 mounted, you can do so with:
a696fa2a
JA
1660
1661 # mount -t cgroup -o blkio none /cgroup
1662
a696fa2a
JA
1663cgroup_weight=int Set the weight of the cgroup to this value. See
1664 the documentation that comes with the kernel, allowed values
1665 are in the range of 100..1000.
71bfa161 1666
7de87099
VG
1667cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1668 the job completion. To override this behavior and to leave
1669 cgroups around after the job completion, set cgroup_nodelete=1.
1670 This can be useful if one wants to inspect various cgroup
1671 files after job completion. Default: false
1672
e0b0d892
JA
1673uid=int Instead of running as the invoking user, set the user ID to
1674 this value before the thread/process does any work.
1675
1676gid=int Set group ID, see uid.
1677
9e684a49
DE
1678flow_id=int The ID of the flow. If not specified, it defaults to being a
1679 global flow. See flow.
1680
1681flow=int Weight in token-based flow control. If this value is used, then
1682 there is a 'flow counter' which is used to regulate the
1683 proportion of activity between two or more jobs. fio attempts
1684 to keep this flow counter near zero. The 'flow' parameter
1685 stands for how much should be added or subtracted to the flow
1686 counter on each iteration of the main I/O loop. That is, if
1687 one job has flow=8 and another job has flow=-1, then there
1688 will be a roughly 1:8 ratio in how much one runs vs the other.
1689
1690flow_watermark=int The maximum value that the absolute value of the flow
1691 counter is allowed to reach before the job must wait for a
1692 lower value of the counter.
1693
1694flow_sleep=int The period of time, in microseconds, to wait after the flow
1695 watermark has been exceeded before retrying operations
1696
de890a1e
SL
1697In addition, there are some parameters which are only valid when a specific
1698ioengine is in use. These are used identically to normal parameters, with the
1699caveat that when used on the command line, they must come after the ioengine
1700that defines them is selected.
1701
1702[libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1703 the io_getevents system call to reap newly returned events.
1704 With this flag turned on, the AIO ring will be read directly
1705 from user-space to reap events. The reaping mode is only
1706 enabled when polling for a minimum of 0 events (eg when
1707 iodepth_batch_complete=0).
1708
0353050f
JA
1709[cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1710
1711[cpu] cpuchunks=int Split the load into cycles of the given time. In
1712 microseconds.
1713
046395d7
JA
1714[cpu] exit_on_io_done=bool Detect when IO threads are done, then exit.
1715
de890a1e
SL
1716[netsplice] hostname=str
1717[net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1718 If the job is a TCP listener or UDP reader, the hostname is not
b511c9aa
SB
1719 used and must be omitted unless it is a valid UDP multicast
1720 address.
de890a1e
SL
1721
1722[netsplice] port=int
6315af9d
JA
1723[net] port=int The TCP or UDP port to bind to or connect to. If this is used
1724with numjobs to spawn multiple instances of the same job type, then this will
1725be the starting port number since fio will use a range of ports.
de890a1e 1726
b93b6a2e
SB
1727[netsplice] interface=str
1728[net] interface=str The IP address of the network interface used to send or
1729 receive UDP multicast
1730
d3a623de
SB
1731[netsplice] ttl=int
1732[net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1733 Default: 1
1734
1d360ffb
JA
1735[netsplice] nodelay=bool
1736[net] nodelay=bool Set TCP_NODELAY on TCP connections.
1737
de890a1e
SL
1738[netsplice] protocol=str
1739[netsplice] proto=str
1740[net] protocol=str
1741[net] proto=str The network protocol to use. Accepted values are:
1742
1743 tcp Transmission control protocol
49ccb8c1 1744 tcpv6 Transmission control protocol V6
f5cc3d0e 1745 udp User datagram protocol
49ccb8c1 1746 udpv6 User datagram protocol V6
de890a1e
SL
1747 unix UNIX domain socket
1748
1749 When the protocol is TCP or UDP, the port must also be given,
1750 as well as the hostname if the job is a TCP listener or UDP
1751 reader. For unix sockets, the normal filename option should be
1752 used and the port is invalid.
1753
1754[net] listen For TCP network connections, tell fio to listen for incoming
1755 connections rather than initiating an outgoing connection. The
1756 hostname must be omitted if this option is used.
1008602c 1757
b511c9aa 1758[net] pingpong Normaly a network writer will just continue writing data, and
7aeb1e94
JA
1759 a network reader will just consume packages. If pingpong=1
1760 is set, a writer will send its normal payload to the reader,
1761 then wait for the reader to send the same payload back. This
1762 allows fio to measure network latencies. The submission
1763 and completion latencies then measure local time spent
1764 sending or receiving, and the completion latency measures
1765 how long it took for the other end to receive and send back.
b511c9aa
SB
1766 For UDP multicast traffic pingpong=1 should only be set for a
1767 single reader when multiple readers are listening to the same
1768 address.
7aeb1e94 1769
1008602c
JA
1770[net] window_size Set the desired socket buffer size for the connection.
1771
e5f34d95
JA
1772[net] mss Set the TCP maximum segment size (TCP_MAXSEG).
1773
d54fce84
DM
1774[e4defrag] donorname=str
1775 File will be used as a block donor(swap extents between files)
1776[e4defrag] inplace=int
66c098b8 1777 Configure donor file blocks allocation strategy
d54fce84
DM
1778 0(default): Preallocate donor's file on init
1779 1 : allocate space immidietly inside defragment event,
1780 and free right after event
1781
65fa28ca 1782[mtd] skip_bad=bool Skip operations against known bad blocks.
de890a1e
SL
1783
1784
71bfa161
JA
17856.0 Interpreting the output
1786---------------------------
1787
1788fio spits out a lot of output. While running, fio will display the
1789status of the jobs created. An example of that would be:
1790
73c8b082 1791Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
71bfa161
JA
1792
1793The characters inside the square brackets denote the current status of
1794each thread. The possible values (in typical life cycle order) are:
1795
1796Idle Run
1797---- ---
1798P Thread setup, but not started.
1799C Thread created.
9c6f6316 1800I Thread initialized, waiting or generating necessary data.
b0f65863 1801 p Thread running pre-reading file(s).
71bfa161
JA
1802 R Running, doing sequential reads.
1803 r Running, doing random reads.
1804 W Running, doing sequential writes.
1805 w Running, doing random writes.
1806 M Running, doing mixed sequential reads/writes.
1807 m Running, doing mixed random reads/writes.
1808 F Running, currently waiting for fsync()
3d434057 1809 f Running, finishing up (writing IO logs, etc)
fc6bd43c 1810 V Running, doing verification of written data.
71bfa161 1811E Thread exited, not reaped by main thread yet.
4f7e57a4
JA
1812_ Thread reaped, or
1813X Thread reaped, exited with an error.
a5e371a6 1814K Thread reaped, exited due to signal.
71bfa161 1815
3e2e48a7
JA
1816Fio will condense the thread string as not to take up more space on the
1817command line as is needed. For instance, if you have 10 readers and 10
1818writers running, the output would look like this:
1819
1820Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s]
1821
1822Fio will still maintain the ordering, though. So the above means that jobs
18231..10 are readers, and 11..20 are writers.
1824
71bfa161 1825The other values are fairly self explanatory - number of threads
c9f60304
JA
1826currently running and doing io, rate of io since last check (read speed
1827listed first, then write speed), and the estimated completion percentage
1828and time for the running group. It's impossible to estimate runtime of
4f7e57a4
JA
1829the following groups (if any). Note that the string is displayed in order,
1830so it's possible to tell which of the jobs are currently doing what. The
1831first character is the first job defined in the job file, and so forth.
71bfa161
JA
1832
1833When fio is done (or interrupted by ctrl-c), it will show the data for
1834each thread, group of threads, and disks in that order. For each data
1835direction, the output looks like:
1836
1837Client1 (g=0): err= 0:
35649e58 1838 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
6104ddb6
JA
1839 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1840 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
b22989b9 1841 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
e7823a94 1842 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
71619dc2 1843 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
838bc709
JA
1844 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1845 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
30061b97 1846 issued r/w: total=0/32768, short=0/0
8abdce66
JA
1847 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1848 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
71bfa161
JA
1849
1850The client number is printed, along with the group id and error of that
1851thread. Below is the io statistics, here for writes. In the order listed,
1852they denote:
1853
1854io= Number of megabytes io performed
1855bw= Average bandwidth rate
35649e58 1856iops= Average IOs performed per second
71bfa161 1857runt= The runtime of that thread
72fbda2a 1858 slat= Submission latency (avg being the average, stdev being the
71bfa161
JA
1859 standard deviation). This is the time it took to submit
1860 the io. For sync io, the slat is really the completion
8a35c71e 1861 latency, since queue/complete is one operation there. This
bf9a3edb 1862 value can be in milliseconds or microseconds, fio will choose
8a35c71e 1863 the most appropriate base and print that. In the example
0d237712
LAG
1864 above, milliseconds is the best scale. Note: in --minimal mode
1865 latencies are always expressed in microseconds.
71bfa161
JA
1866 clat= Completion latency. Same names as slat, this denotes the
1867 time from submission to completion of the io pieces. For
1868 sync io, clat will usually be equal (or very close) to 0,
1869 as the time from submit to complete is basically just
1870 CPU time (io has already been done, see slat explanation).
1871 bw= Bandwidth. Same names as the xlat stats, but also includes
1872 an approximate percentage of total aggregate bandwidth
1873 this thread received in this group. This last value is
1874 only really useful if the threads in this group are on the
1875 same disk, since they are then competing for disk access.
1876cpu= CPU usage. User and system time, along with the number
e7823a94
JA
1877 of context switches this thread went through, usage of
1878 system and user time, and finally the number of major
1879 and minor page faults.
71619dc2
JA
1880IO depths= The distribution of io depths over the job life time. The
1881 numbers are divided into powers of 2, so for example the
1882 16= entries includes depths up to that value but higher
1883 than the previous entry. In other words, it covers the
1884 range from 16 to 31.
838bc709
JA
1885IO submit= How many pieces of IO were submitting in a single submit
1886 call. Each entry denotes that amount and below, until
1887 the previous entry - eg, 8=100% mean that we submitted
1888 anywhere in between 5-8 ios per submit call.
1889IO complete= Like the above submit number, but for completions instead.
30061b97
JA
1890IO issued= The number of read/write requests issued, and how many
1891 of them were short.
ec118304
JA
1892IO latencies= The distribution of IO completion latencies. This is the
1893 time from when IO leaves fio and when it gets completed.
1894 The numbers follow the same pattern as the IO depths,
1895 meaning that 2=1.6% means that 1.6% of the IO completed
8abdce66
JA
1896 within 2 msecs, 20=12.8% means that 12.8% of the IO
1897 took more than 10 msecs, but less than (or equal to) 20 msecs.
71bfa161
JA
1898
1899After each client has been listed, the group statistics are printed. They
1900will look like this:
1901
1902Run status group 0 (all jobs):
b22989b9
JA
1903 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1904 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
71bfa161
JA
1905
1906For each data direction, it prints:
1907
1908io= Number of megabytes io performed.
1909aggrb= Aggregate bandwidth of threads in this group.
1910minb= The minimum average bandwidth a thread saw.
1911maxb= The maximum average bandwidth a thread saw.
1912mint= The smallest runtime of the threads in that group.
1913maxt= The longest runtime of the threads in that group.
1914
1915And finally, the disk statistics are printed. They will look like this:
1916
1917Disk stats (read/write):
1918 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1919
1920Each value is printed for both reads and writes, with reads first. The
1921numbers denote:
1922
1923ios= Number of ios performed by all groups.
1924merge= Number of merges io the io scheduler.
1925ticks= Number of ticks we kept the disk busy.
1926io_queue= Total time spent in the disk queue.
1927util= The disk utilization. A value of 100% means we kept the disk
1928 busy constantly, 50% would be a disk idling half of the time.
1929
8423bd11
JA
1930It is also possible to get fio to dump the current output while it is
1931running, without terminating the job. To do that, send fio the USR1 signal.
06464907
JA
1932You can also get regularly timed dumps by using the --status-interval
1933parameter, or by creating a file in /tmp named fio-dump-status. If fio
1934sees this file, it will unlink it and dump the current output status.
8423bd11 1935
71bfa161
JA
1936
19377.0 Terse output
1938----------------
1939
1940For scripted usage where you typically want to generate tables or graphs
6af019c9 1941of the results, fio can output the results in a semicolon separated format.
71bfa161
JA
1942The format is one long line of values, such as:
1943
562c2d2f
DN
19442;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%
1945A description of this job goes here.
1946
1947The job description (if provided) follows on a second line.
71bfa161 1948
525c2bfa
JA
1949To enable terse output, use the --minimal command line option. The first
1950value is the version of the terse output format. If the output has to
1951be changed for some reason, this number will be incremented by 1 to
1952signify that change.
6820cb3b 1953
71bfa161
JA
1954Split up, the format is as follows:
1955
5e726d0a 1956 terse version, fio version, jobname, groupid, error
71bfa161 1957 READ status:
312b4af2 1958 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
d86ae56c
CW
1959 Submission latency: min, max, mean, stdev (usec)
1960 Completion latency: min, max, mean, stdev (usec)
1db92cb6 1961 Completion latency percentiles: 20 fields (see below)
d86ae56c
CW
1962 Total latency: min, max, mean, stdev (usec)
1963 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
71bfa161 1964 WRITE status:
312b4af2 1965 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
d86ae56c
CW
1966 Submission latency: min, max, mean, stdev (usec)
1967 Completion latency: min, max, mean, stdev(usec)
1db92cb6 1968 Completion latency percentiles: 20 fields (see below)
d86ae56c
CW
1969 Total latency: min, max, mean, stdev (usec)
1970 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
046ee302 1971 CPU usage: user, system, context switches, major faults, minor faults
2270890c 1972 IO depths: <=1, 2, 4, 8, 16, 32, >=64
562c2d2f
DN
1973 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1974 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
f2f788dd
JA
1975 Disk utilization: Disk name, Read ios, write ios,
1976 Read merges, write merges,
1977 Read ticks, write ticks,
3d7cd9b4 1978 Time spent in queue, disk utilization percentage
de8f6de9 1979 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
66c098b8 1980
de8f6de9 1981 Additional Info (dependent on description being set): Text description
25c8b9d7 1982
1db92cb6
JA
1983Completion latency percentiles can be a grouping of up to 20 sets, so
1984for the terse output fio writes all of them. Each field will look like this:
1985
1986 1.00%=6112
1987
1988which is the Xth percentile, and the usec latency associated with it.
1989
f2f788dd
JA
1990For disk utilization, all disks used by fio are shown. So for each disk
1991there will be a disk utilization section.
1992
25c8b9d7
PD
1993
19948.0 Trace file format
1995---------------------
66c098b8 1996There are two trace file format that you can encounter. The older (v1) format
25c8b9d7
PD
1997is unsupported since version 1.20-rc3 (March 2008). It will still be described
1998below in case that you get an old trace and want to understand it.
1999
2000In any case the trace is a simple text file with a single action per line.
2001
2002
20038.1 Trace file format v1
2004------------------------
2005Each line represents a single io action in the following format:
2006
2007rw, offset, length
2008
2009where rw=0/1 for read/write, and the offset and length entries being in bytes.
2010
2011This format is not supported in Fio versions => 1.20-rc3.
2012
2013
20148.2 Trace file format v2
2015------------------------
2016The second version of the trace file format was added in Fio version 1.17.
2017It allows to access more then one file per trace and has a bigger set of
2018possible file actions.
2019
2020The first line of the trace file has to be:
2021
2022fio version 2 iolog
2023
2024Following this can be lines in two different formats, which are described below.
2025
2026The file management format:
2027
2028filename action
2029
2030The filename is given as an absolute path. The action can be one of these:
2031
2032add Add the given filename to the trace
66c098b8 2033open Open the file with the given filename. The filename has to have
25c8b9d7
PD
2034 been added with the add action before.
2035close Close the file with the given filename. The file has to have been
2036 opened before.
2037
2038
2039The file io action format:
2040
2041filename action offset length
2042
2043The filename is given as an absolute path, and has to have been added and opened
66c098b8 2044before it can be used with this format. The offset and length are given in
25c8b9d7
PD
2045bytes. The action can be one of these:
2046
2047wait Wait for 'offset' microseconds. Everything below 100 is discarded.
5c7808fe 2048 The time is relative to the previous wait statement.
25c8b9d7
PD
2049read Read 'length' bytes beginning from 'offset'
2050write Write 'length' bytes beginning from 'offset'
2051sync fsync() the file
2052datasync fdatasync() the file
2053trim trim the given file from the given 'offset' for 'length' bytes
f2a2ce0e
HL
2054
2055
20569.0 CPU idleness profiling
06464907 2057--------------------------
f2a2ce0e
HL
2058In some cases, we want to understand CPU overhead in a test. For example,
2059we test patches for the specific goodness of whether they reduce CPU usage.
2060fio implements a balloon approach to create a thread per CPU that runs at
2061idle priority, meaning that it only runs when nobody else needs the cpu.
2062By measuring the amount of work completed by the thread, idleness of each
2063CPU can be derived accordingly.
2064
2065An unit work is defined as touching a full page of unsigned characters. Mean
2066and standard deviation of time to complete an unit work is reported in "unit
2067work" section. Options can be chosen to report detailed percpu idleness or
2068overall system idleness by aggregating percpu stats.
99b9a85a
JA
2069
2070
207110.0 Verification and triggers
2072------------------------------
2073Fio is usually run in one of two ways, when data verification is done. The
2074first is a normal write job of some sort with verify enabled. When the
2075write phase has completed, fio switches to reads and verifies everything
2076it wrote. The second model is running just the write phase, and then later
2077on running the same job (but with reads instead of writes) to repeat the
2078same IO patterns and verify the contents. Both of these methods depend
2079on the write phase being completed, as fio otherwise has no idea how much
2080data was written.
2081
2082With verification triggers, fio supports dumping the current write state
2083to local files. Then a subsequent read verify workload can load this state
2084and know exactly where to stop. This is useful for testing cases where
2085power is cut to a server in a managed fashion, for instance.
2086
2087A verification trigger consists of two things:
2088
20891) Storing the write state of each job
20902) Executing a trigger command
2091
2092The write state is relatively small, on the order of hundreds of bytes
2093to single kilobytes. It contains information on the number of completions
2094done, the last X completions, etc.
2095
2096A trigger is invoked either through creation ('touch') of a specified
2097file in the system, or through a timeout setting. If fio is run with
2098--trigger-file=/tmp/trigger-file, then it will continually check for
2099the existence of /tmp/trigger-file. When it sees this file, it will
2100fire off the trigger (thus saving state, and executing the trigger
2101command).
2102
2103For client/server runs, there's both a local and remote trigger. If
2104fio is running as a server backend, it will send the job states back
2105to the client for safe storage, then execute the remote trigger, if
2106specified. If a local trigger is specified, the server will still send
2107back the write state, but the client will then execute the trigger.
2108
210910.1 Verification trigger example
2110---------------------------------
2111Lets say we want to run a powercut test on the remote machine 'server'.
2112Our write workload is in write-test.fio. We want to cut power to 'server'
2113at some point during the run, and we'll run this test from the safety
2114or our local machine, 'localbox'. On the server, we'll start the fio
2115backend normally:
2116
2117server# fio --server
2118
2119and on the client, we'll fire off the workload:
2120
2121localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
2122
2123We set /tmp/my-trigger as the trigger file, and we tell fio to execute
2124
2125echo b > /proc/sysrq-trigger
2126
2127on the server once it has received the trigger and sent us the write
2128state. This will work, but it's not _really_ cutting power to the server,
2129it's merely abruptly rebooting it. If we have a remote way of cutting
2130power to the server through IPMI or similar, we could do that through
2131a local trigger command instead. Lets assume we have a script that does
2132IPMI reboot of a given hostname, ipmi-reboot. On localbox, we could
2133then have run fio with a local trigger instead:
2134
2135localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
2136
2137For this case, fio would wait for the server to send us the write state,
2138then execute 'ipmi-reboot server' when that happened.
2139
214010.1 Loading verify state
2141-------------------------
2142To load store write state, read verification job file must contain
2143the verify_state_load option. If that is set, fio will load the previously
2144stored state. For a local fio run this is done by loading the files directly,
2145and on a client/server run, the server backend will ask the client to send
2146the files over and load them from there.