backend: dump state of stuck thread
[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
29dbd1e5 1310. Verification and triggers
a3ae5b05
JA
1411. Log File Formats
15
71bfa161
JA
16
171.0 Overview and history
18------------------------
19fio was originally written to save me the hassle of writing special test
20case programs when I wanted to test a specific workload, either for
21performance reasons or to find/reproduce a bug. The process of writing
22such a test app can be tiresome, especially if you have to do it often.
23Hence I needed a tool that would be able to simulate a given io workload
24without resorting to writing a tailored test case again and again.
25
26A test work load is difficult to define, though. There can be any number
27of processes or threads involved, and they can each be using their own
28way of generating io. You could have someone dirtying large amounts of
29memory in an memory mapped file, or maybe several threads issuing
30reads using asynchronous io. fio needed to be flexible enough to
31simulate both of these cases, and many more.
32
332.0 How fio works
34-----------------
35The first step in getting fio to simulate a desired io workload, is
36writing a job file describing that specific setup. A job file may contain
37any number of threads and/or files - the typical contents of the job file
38is a global section defining shared parameters, and one or more job
39sections describing the jobs involved. When run, fio parses this file
40and sets everything up as described. If we break down a job from top to
41bottom, it contains the following basic parameters:
42
43 IO type Defines the io pattern issued to the file(s).
44 We may only be reading sequentially from this
45 file(s), or we may be writing randomly. Or even
46 mixing reads and writes, sequentially or randomly.
47
48 Block size In how large chunks are we issuing io? This may be
49 a single value, or it may describe a range of
50 block sizes.
51
52 IO size How much data are we going to be reading/writing.
53
54 IO engine How do we issue io? We could be memory mapping the
55 file, we could be using regular read/write, we
d0ff85df 56 could be using splice, async io, syslet, or even
71bfa161
JA
57 SG (SCSI generic sg).
58
6c219763 59 IO depth If the io engine is async, how large a queuing
71bfa161
JA
60 depth do we want to maintain?
61
62 IO type Should we be doing buffered io, or direct/raw io?
63
64 Num files How many files are we spreading the workload over.
65
66 Num threads How many threads or processes should we spread
67 this workload over.
66c098b8 68
71bfa161
JA
69The above are the basic parameters defined for a workload, in addition
70there's a multitude of parameters that modify other aspects of how this
71job behaves.
72
73
743.0 Running fio
75---------------
76See the README file for command line parameters, there are only a few
77of them.
78
79Running fio is normally the easiest part - you just give it the job file
80(or job files) as parameters:
81
82$ fio job_file
83
84and it will start doing what the job_file tells it to do. You can give
85more than one job file on the command line, fio will serialize the running
86of those files. Internally that is the same as using the 'stonewall'
550b1db6 87parameter described in the parameter section.
71bfa161 88
b4692828
JA
89If the job file contains only one job, you may as well just give the
90parameters on the command line. The command line parameters are identical
91to the job parameters, with a few extra that control global parameters
92(see README). For example, for the job file parameter iodepth=2, the
c2b1e753
JA
93mirror command line option would be --iodepth 2 or --iodepth=2. You can
94also use the command line for giving more than one job entry. For each
95--name option that fio sees, it will start a new job with that name.
96Command line entries following a --name entry will apply to that job,
97until there are no more entries or a new --name entry is seen. This is
98similar to the job file options, where each option applies to the current
99job until a new [] job entry is seen.
b4692828 100
71bfa161
JA
101fio does not need to run as root, except if the files or devices specified
102in the job section requires that. Some other options may also be restricted,
6c219763 103such as memory locking, io scheduler switching, and decreasing the nice value.
71bfa161
JA
104
105
1064.0 Job file format
107-------------------
108As previously described, fio accepts one or more job files describing
109what it is supposed to do. The job file format is the classic ini file,
110where the names enclosed in [] brackets define the job name. You are free
111to use any ascii name you want, except 'global' which has special meaning.
112A global section sets defaults for the jobs described in that file. A job
113may override a global section parameter, and a job file may even have
114several global sections if so desired. A job is only affected by a global
65db0851
JA
115section residing above it. If the first character in a line is a ';' or a
116'#', the entire line is discarded as a comment.
71bfa161 117
3c54bc46 118So let's look at a really simple job file that defines two processes, each
b22989b9 119randomly reading from a 128MB file.
71bfa161
JA
120
121; -- start job file --
122[global]
123rw=randread
124size=128m
125
126[job1]
127
128[job2]
129
130; -- end job file --
131
132As you can see, the job file sections themselves are empty as all the
133described parameters are shared. As no filename= option is given, fio
c2b1e753
JA
134makes up a filename for each of the jobs as it sees fit. On the command
135line, this job would look as follows:
136
137$ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
138
71bfa161 139
3c54bc46 140Let's look at an example that has a number of processes writing randomly
71bfa161
JA
141to files.
142
143; -- start job file --
144[random-writers]
145ioengine=libaio
146iodepth=4
147rw=randwrite
148bs=32k
149direct=0
150size=64m
151numjobs=4
152
153; -- end job file --
154
155Here we have no global section, as we only have one job defined anyway.
156We want to use async io here, with a depth of 4 for each file. We also
b22989b9 157increased the buffer size used to 32KB and define numjobs to 4 to
71bfa161 158fork 4 identical jobs. The result is 4 processes each randomly writing
b22989b9 159to their own 64MB file. Instead of using the above job file, you could
b4692828
JA
160have given the parameters on the command line. For this case, you would
161specify:
162
163$ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
71bfa161 164
df5ad464
AK
165When fio is utilized as a basis of any reasonably large test suite, it might be
166desirable to share a set of standardized settings across multiple job files.
167Instead of copy/pasting such settings, any section may pull in an external
168.fio file with 'include filename' directive, as in the following example:
169
170; -- start job file including.fio --
171[global]
172filename=/tmp/test
173filesize=1m
174include glob-include.fio
175
176[test]
177rw=randread
178bs=4k
179time_based=1
180runtime=10
181include test-include.fio
182; -- end job file including.fio --
183
184; -- start job file glob-include.fio --
185thread=1
186group_reporting=1
187; -- end job file glob-include.fio --
188
189; -- start job file test-include.fio --
190ioengine=libaio
191iodepth=4
192; -- end job file test-include.fio --
193
194Settings pulled into a section apply to that section only (except global
195section). Include directives may be nested in that any included file may
73568e1b
JA
196contain further include directive(s). Include files may not contain []
197sections.
df5ad464
AK
198
199
74929ac2
JA
2004.1 Environment variables
201-------------------------
202
3c54bc46 203fio also supports environment variable expansion in job files. Any
4fbe1860 204sub-string of the form "${VARNAME}" as part of an option value (in other
3c54bc46
AC
205words, on the right of the `='), will be expanded to the value of the
206environment variable called VARNAME. If no such environment variable
207is defined, or VARNAME is the empty string, the empty string will be
208substituted.
209
210As an example, let's look at a sample fio invocation and job file:
211
212$ SIZE=64m NUMJOBS=4 fio jobfile.fio
213
214; -- start job file --
215[random-writers]
216rw=randwrite
217size=${SIZE}
218numjobs=${NUMJOBS}
219; -- end job file --
220
221This will expand to the following equivalent job file at runtime:
222
223; -- start job file --
224[random-writers]
225rw=randwrite
226size=64m
227numjobs=4
228; -- end job file --
229
71bfa161
JA
230fio ships with a few example job files, you can also look there for
231inspiration.
232
74929ac2
JA
2334.2 Reserved keywords
234---------------------
235
236Additionally, fio has a set of reserved keywords that will be replaced
237internally with the appropriate value. Those keywords are:
238
239$pagesize The architecture page size of the running system
240$mb_memory Megabytes of total memory in the system
241$ncpus Number of online available CPUs
242
243These can be used on the command line or in the job file, and will be
244automatically substituted with the current system values when the job
892a6ffc
JA
245is run. Simple math is also supported on these keywords, so you can
246perform actions like:
247
248size=8*$mb_memory
249
250and get that properly expanded to 8 times the size of memory in the
251machine.
74929ac2 252
71bfa161
JA
253
2545.0 Detailed list of parameters
255-------------------------------
256
257This section describes in details each parameter associated with a job.
258Some parameters take an option of a given type, such as an integer or
d59aa780
JA
259a string. Anywhere a numeric value is required, an arithmetic expression
260may be used, provided it is surrounded by parentheses. Supported operators
261are:
262
263 addition (+)
264 subtraction (-)
265 multiplication (*)
266 division (/)
267 modulus (%)
268 exponentiation (^)
269
270For time values in expressions, units are microseconds by default. This is
271different than for time values not in expressions (not enclosed in
272parentheses). The following types are used:
71bfa161
JA
273
274str String. This is a sequence of alpha characters.
b09da8fa 275time Integer with possible time suffix. In seconds unless otherwise
e417fd66 276 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
0de5b26f
JA
277 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
278 and 'us' (or 'usec') for microseconds.
b09da8fa
JA
279int SI integer. A whole number value, which may contain a suffix
280 describing the base of the number. Accepted suffixes are k/m/g/t/p,
281 meaning kilo, mega, giga, tera, and peta. The suffix is not case
57fc29fa
JA
282 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
283 as 'k'). So if you want to specify 4096, you could either write
b09da8fa 284 out '4096' or just give 4k. The suffixes signify base 2 values, so
57fc29fa
JA
285 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
286 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
287 case, then 1000 is used as the multiplier. This can be handy for
288 disks, since manufacturers generally use base 10 values when listing
289 the capacity of a drive. If the option accepts an upper and lower
290 range, use a colon ':' or minus '-' to separate such values. May also
291 include a prefix to indicate numbers base. If 0x is used, the number
292 is assumed to be hexadecimal. See irange.
71bfa161
JA
293bool Boolean. Usually parsed as an integer, however only defined for
294 true and false (1 and 0).
b09da8fa 295irange Integer range with suffix. Allows value range to be given, such
bf9a3edb 296 as 1024-4096. A colon may also be used as the separator, eg
0c9baf91
JA
297 1k:4k. If the option allows two sets of ranges, they can be
298 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
f7fa2653 299 int.
83349190 300float_list A list of floating numbers, separated by a ':' character.
71bfa161
JA
301
302With the above in mind, here follows the complete list of fio job
303parameters.
304
305name=str ASCII name of the job. This may be used to override the
306 name printed by fio for this job. Otherwise the job
c2b1e753 307 name is used. On the command line this parameter has the
6c219763 308 special purpose of also signaling the start of a new
c2b1e753 309 job.
71bfa161 310
9cc8cb91
AK
311wait_for=str Specifies the name of the already defined job to wait
312 for. Single waitee name only may be specified. If set, the job
313 won't be started until all workers of the waitee job are done.
314
315 Wait_for operates on the job name basis, so there are a few
316 limitations. First, the waitee must be defined prior to the
317 waiter job (meaning no forward references). Second, if a job
318 is being referenced as a waitee, it must have a unique name
319 (no duplicate waitees).
320
61697c37
JA
321description=str Text description of the job. Doesn't do anything except
322 dump this text description when this job is run. It's
323 not parsed.
324
3776041e 325directory=str Prefix filenames with this directory. Used to place files
67445b63
JA
326 in a different location than "./". See the 'filename' option
327 for escaping certain characters.
71bfa161
JA
328
329filename=str Fio normally makes up a filename based on the job name,
330 thread number, and file number. If you want to share
331 files between threads in a job or several jobs, specify
ed92ac0c 332 a filename for each of them to override the default. If
414c2a3e 333 the ioengine used is 'net', the filename is the host, port,
0fd666bf 334 and protocol to use in the format of =host,port,protocol.
414c2a3e
JA
335 See ioengine=net for more. If the ioengine is file based, you
336 can specify a number of files by separating the names with a
337 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
338 as the two working files, you would use
30a4588a
JA
339 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
340 accessed as \\.\PhysicalDrive0 for the first device,
341 \\.\PhysicalDrive1 for the second etc. Note: Windows and
342 FreeBSD prevent write access to areas of the disk containing
343 in-use data (e.g. filesystems).
344 If the wanted filename does need to include a colon, then
345 escape that with a '\' character. For instance, if the filename
346 is "/dev/dsk/foo@3,0:c", then you would use
347 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
348 stdin or stdout. Which of the two depends on the read/write
349 direction set.
71bfa161 350
de98bd30
JA
351filename_format=str
352 If sharing multiple files between jobs, it is usually necessary
353 to have fio generate the exact names that you want. By default,
354 fio will name a file based on the default file format
355 specification of jobname.jobnumber.filenumber. With this
356 option, that can be customized. Fio will recognize and replace
357 the following keywords in this string:
358
359 $jobname
360 The name of the worker thread or process.
361
362 $jobnum
363 The incremental number of the worker thread or
364 process.
365
366 $filenum
367 The incremental number of the file for that worker
368 thread or process.
369
370 To have dependent jobs share a set of files, this option can
371 be set to have fio generate filenames that are shared between
372 the two. For instance, if testfiles.$filenum is specified,
373 file number 4 for any job will be named testfiles.4. The
374 default of $jobname.$jobnum.$filenum will be used if
375 no other format specifier is given.
376
bbf6b540
JA
377opendir=str Tell fio to recursively add any file it can find in this
378 directory and down the file system tree.
379
3776041e 380lockfile=str Fio defaults to not locking any files before it does
4d4e80f2
JA
381 IO to them. If a file or file descriptor is shared, fio
382 can serialize IO to that file to make the end result
383 consistent. This is usual for emulating real workloads that
384 share files. The lock modes are:
385
386 none No locking. The default.
387 exclusive Only one thread/process may do IO,
388 excluding all others.
389 readwrite Read-write locking on the file. Many
390 readers may access the file at the
391 same time, but writes get exclusive
392 access.
393
d3aad8f2 394readwrite=str
71bfa161
JA
395rw=str Type of io pattern. Accepted values are:
396
397 read Sequential reads
398 write Sequential writes
399 randwrite Random writes
400 randread Random reads
10b023db 401 rw,readwrite Sequential mixed reads and writes
71bfa161 402 randrw Random mixed reads and writes
82a90686
JA
403 trimwrite Mixed trims and writes. Blocks will be
404 trimmed first, then written to.
71bfa161
JA
405
406 For the mixed io types, the default is to split them 50/50.
407 For certain types of io the result may still be skewed a bit,
211097b2 408 since the speed may be different. It is possible to specify
38dad62d 409 a number of IO's to do before getting a new offset, this is
892ea9bd 410 done by appending a ':<nr>' to the end of the string given.
38dad62d 411 For a random read, it would look like 'rw=randread:8' for
059b0802 412 passing in an offset modifier with a value of 8. If the
ddb754db 413 suffix is used with a sequential IO pattern, then the value
059b0802
JA
414 specified will be added to the generated offset for each IO.
415 For instance, using rw=write:4k will skip 4k for every
416 write. It turns sequential IO into sequential IO with holes.
417 See the 'rw_sequencer' option.
38dad62d
JA
418
419rw_sequencer=str If an offset modifier is given by appending a number to
420 the rw=<str> line, then this option controls how that
421 number modifies the IO offset being generated. Accepted
422 values are:
423
424 sequential Generate sequential offset
425 identical Generate the same offset
426
427 'sequential' is only useful for random IO, where fio would
428 normally generate a new random offset for every IO. If you
429 append eg 8 to randread, you would get a new random offset for
211097b2
JA
430 every 8 IO's. The result would be a seek for only every 8
431 IO's, instead of for every IO. Use rw=randread:8 to specify
38dad62d
JA
432 that. As sequential IO is already sequential, setting
433 'sequential' for that would not result in any differences.
434 'identical' behaves in a similar fashion, except it sends
435 the same offset 8 number of times before generating a new
436 offset.
71bfa161 437
90fef2d1
JA
438kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
439 Storage manufacturers like to use 10^3 or 1000 as a base
440 ten unit instead, for obvious reasons. Allow values are
441 1024 or 1000, with 1024 being the default.
442
771e58be
JA
443unified_rw_reporting=bool Fio normally reports statistics on a per
444 data direction basis, meaning that read, write, and trim are
445 accounted and reported separately. If this option is set,
446 the fio will sum the results and report them as "mixed"
447 instead.
448
ee738499
JA
449randrepeat=bool For random IO workloads, seed the generator in a predictable
450 way so that results are repeatable across repetitions.
40fe5e7b 451 Defaults to true.
ee738499 452
04778baf
JA
453randseed=int Seed the random number generators based on this seed value, to
454 be able to control what sequence of output is being generated.
455 If not set, the random sequence depends on the randrepeat
456 setting.
457
a596f047
EG
458fallocate=str Whether pre-allocation is performed when laying down files.
459 Accepted values are:
460
461 none Do not pre-allocate space
462 posix Pre-allocate via posix_fallocate()
463 keep Pre-allocate via fallocate() with
464 FALLOC_FL_KEEP_SIZE set
465 0 Backward-compatible alias for 'none'
466 1 Backward-compatible alias for 'posix'
467
468 May not be available on all supported platforms. 'keep' is only
469 available on Linux.If using ZFS on Solaris this must be set to
470 'none' because ZFS doesn't support it. Default: 'posix'.
7bc8c2cf 471
d2f3ac35
JA
472fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
473 on what IO patterns it is likely to issue. Sometimes you
474 want to test specific IO patterns without telling the
475 kernel about it, in which case you can disable this option.
476 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
477 IO and POSIX_FADV_RANDOM for random IO.
478
37659335
JA
479fadvise_stream=int Notify the kernel what write stream ID to place these
480 writes under. Only supported on Linux. Note, this option
481 may change going forward.
482
f7fa2653 483size=int The total size of file io for this job. Fio will run until
7616cafe 484 this many bytes has been transferred, unless runtime is
a4d3b4db
JA
485 limited by other options (such as 'runtime', for instance,
486 or increased/decreased by 'io_size'). Unless specific nrfiles
487 and filesize options are given, fio will divide this size
488 between the available files specified by the job. If not set,
489 fio will use the full size of the given files or devices.
490 If the files do not exist, size must be given. It is also
491 possible to give size as a percentage between 1 and 100. If
492 size=20% is given, fio will use 20% of the full size of the
493 given files or devices.
494
495io_size=int
77731b29
JA
496io_limit=int Normally fio operates within the region set by 'size', which
497 means that the 'size' option sets both the region and size of
498 IO to be performed. Sometimes that is not what you want. With
499 this option, it is possible to define just the amount of IO
500 that fio should do. For instance, if 'size' is set to 20G and
a4d3b4db
JA
501 'io_size' is set to 5G, fio will perform IO within the first
502 20G but exit when 5G have been done. The opposite is also
503 possible - if 'size' is set to 20G, and 'io_size' is set to
504 40G, then fio will do 40G of IO within the 0..20G region.
77731b29 505
f7fa2653 506filesize=int Individual file sizes. May be a range, in which case fio
9c60ce64
JA
507 will select sizes for files at random within the given range
508 and limited to 'size' in total (if that is given). If not
509 given, each created file is the same size.
510
bedc9dc2
JA
511file_append=bool Perform IO after the end of the file. Normally fio will
512 operate within the size of a file. If this option is set, then
513 fio will append to the file instead. This has identical
0aae4ce7
JA
514 behavior to setting offset to the size of a file. This option
515 is ignored on non-regular files.
bedc9dc2 516
74586c1e
JA
517fill_device=bool
518fill_fs=bool Sets size to something really large and waits for ENOSPC (no
aa31f1f1 519 space left on device) as the terminating condition. Only makes
de98bd30 520 sense with sequential write. For a read workload, the mount
4f12432e
JA
521 point will be filled first then IO started on the result. This
522 option doesn't make sense if operating on a raw device node,
523 since the size of that is already known by the file system.
524 Additionally, writing beyond end-of-device will not return
525 ENOSPC there.
aa31f1f1 526
f7fa2653
JA
527blocksize=int
528bs=int The block size used for the io units. Defaults to 4k. Values
529 can be given for both read and writes. If a single int is
530 given, it will apply to both. If a second int is specified
f90eff5a 531 after a comma, it will apply to writes only. In other words,
d9472271
JA
532 the format is either bs=read_and_write or bs=read,write,trim.
533 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
534 writes, and 8k for trims. You can terminate the list with
535 a trailing comma. bs=4k,8k, would use the default value for
536 trims.. If you only wish to set the write size, you
787f7e95
JA
537 can do so by passing an empty read size - bs=,8k will set
538 8k for writes and leave the read default value.
a00735e6 539
2b7a01d0
JA
540blockalign=int
541ba=int At what boundary to align random IO offsets. Defaults to
542 the same as 'blocksize' the minimum blocksize given.
543 Minimum alignment is typically 512b for using direct IO,
544 though it usually depends on the hardware block size. This
545 option is mutually exclusive with using a random map for
546 files, so it will turn off that option.
547
d3aad8f2 548blocksize_range=irange
71bfa161
JA
549bsrange=irange Instead of giving a single block size, specify a range
550 and fio will mix the issued io block sizes. The issued
551 io unit will always be a multiple of the minimum value
f90eff5a
JA
552 given (also see bs_unaligned). Applies to both reads and
553 writes, however a second range can be given after a comma.
554 See bs=.
a00735e6 555
564ca972
JA
556bssplit=str Sometimes you want even finer grained control of the
557 block sizes issued, not just an even split between them.
558 This option allows you to weight various block sizes,
559 so that you are able to define a specific amount of
560 block sizes issued. The format for this option is:
561
562 bssplit=blocksize/percentage:blocksize/percentage
563
564 for as many block sizes as needed. So if you want to define
565 a workload that has 50% 64k blocks, 10% 4k blocks, and
566 40% 32k blocks, you would write:
567
568 bssplit=4k/10:64k/50:32k/40
569
570 Ordering does not matter. If the percentage is left blank,
571 fio will fill in the remaining values evenly. So a bssplit
572 option like this one:
573
574 bssplit=4k/50:1k/:32k/
575
576 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
577 always add up to 100, if bssplit is given a range that adds
578 up to more, it will error out.
579
720e84ad
JA
580 bssplit also supports giving separate splits to reads and
581 writes. The format is identical to what bs= accepts. You
582 have to separate the read and write parts with a comma. So
583 if you want a workload that has 50% 2k reads and 50% 4k reads,
584 while having 90% 4k writes and 10% 8k writes, you would
585 specify:
586
892ea9bd 587 bssplit=2k/50:4k/50,4k/90:8k/10
720e84ad 588
d3aad8f2 589blocksize_unaligned
690adba3
JA
590bs_unaligned If this option is given, any byte size value within bsrange
591 may be used as a block range. This typically wont work with
592 direct IO, as that normally requires sector alignment.
71bfa161 593
6aca9b3d
JA
594bs_is_seq_rand If this option is set, fio will use the normal read,write
595 blocksize settings as sequential,random instead. Any random
596 read or write will use the WRITE blocksize settings, and any
597 sequential read or write will use the READ blocksize setting.
598
e9459e5a
JA
599zero_buffers If this option is given, fio will init the IO buffers to
600 all zeroes. The default is to fill them with random data.
601
5973cafb
JA
602refill_buffers If this option is given, fio will refill the IO buffers
603 on every submit. The default is to only fill it at init
604 time and reuse that data. Only makes sense if zero_buffers
41ccd845
JA
605 isn't specified, naturally. If data verification is enabled,
606 refill_buffers is also automatically enabled.
5973cafb 607
fd68418e
JA
608scramble_buffers=bool If refill_buffers is too costly and the target is
609 using data deduplication, then setting this option will
610 slightly modify the IO buffer contents to defeat normal
611 de-dupe attempts. This is not enough to defeat more clever
612 block compression attempts, but it will stop naive dedupe of
613 blocks. Default: true.
614
c5751c62
JA
615buffer_compress_percentage=int If this is set, then fio will attempt to
616 provide IO buffer content (on WRITEs) that compress to
617 the specified level. Fio does this by providing a mix of
d1af2894
JA
618 random data and a fixed pattern. The fixed pattern is either
619 zeroes, or the pattern specified by buffer_pattern. If the
620 pattern option is used, it might skew the compression ratio
621 slightly. Note that this is per block size unit, for file/disk
622 wide compression level that matches this setting, you'll also
623 want to set refill_buffers.
c5751c62
JA
624
625buffer_compress_chunk=int See buffer_compress_percentage. This
626 setting allows fio to manage how big the ranges of random
627 data and zeroed data is. Without this set, fio will
628 provide buffer_compress_percentage of blocksize random
629 data, followed by the remaining zeroed. With this set
630 to some chunk size smaller than the block size, fio can
631 alternate random and zeroed data throughout the IO
632 buffer.
633
5c94b008
JA
634buffer_pattern=str If set, fio will fill the io buffers with this
635 pattern. If not set, the contents of io buffers is defined by
636 the other options related to buffer contents. The setting can
637 be any pattern of bytes, and can be prefixed with 0x for hex
638 values. It may also be a string, where the string must then
61b9861d
RP
639 be wrapped with "", e.g.:
640
641 buffer_pattern="abcd"
642 or
643 buffer_pattern=-12
644 or
645 buffer_pattern=0xdeadface
646
647 Also you can combine everything together in any order:
648 buffer_pattern=0xdeadface"abcd"-12
5c94b008
JA
649
650dedupe_percentage=int If set, fio will generate this percentage of
651 identical buffers when writing. These buffers will be
652 naturally dedupable. The contents of the buffers depend on
653 what other buffer compression settings have been set. It's
654 possible to have the individual buffers either fully
655 compressible, or not at all. This option only controls the
656 distribution of unique buffers.
ce35b1ec 657
71bfa161
JA
658nrfiles=int Number of files to use for this job. Defaults to 1.
659
390b1537
JA
660openfiles=int Number of files to keep open at the same time. Defaults to
661 the same as nrfiles, can be set smaller to limit the number
662 simultaneous opens.
663
5af1c6f3
JA
664file_service_type=str Defines how fio decides which file from a job to
665 service next. The following types are defined:
666
667 random Just choose a file at random.
668
669 roundrobin Round robin over open files. This
670 is the default.
671
a086c257
JA
672 sequential Finish one file before moving on to
673 the next. Multiple files can still be
674 open depending on 'openfiles'.
675
8c07860d
JA
676 zipf Use a zipfian distribution to decide what file
677 to access.
678
679 pareto Use a pareto distribution to decide what file
680 to access.
681
682 gauss Use a gaussian (normal) distribution to decide
683 what file to access.
684
685 For random, roundrobin, and sequential, a postfix can be
686 appended to tell fio how many I/Os to issue before switching
687 to a new file. For example, specifying
688 'file_service_type=random:8' would cause fio to issue 8 I/Os
689 before selecting a new file at random. For the non-uniform
690 distributions, a floating point postfix can be given to
691 influence how the distribution is skewed. See
692 'random_distribution' for a description of how that would work.
1907dbc6 693
71bfa161
JA
694ioengine=str Defines how the job issues io to the file. The following
695 types are defined:
696
697 sync Basic read(2) or write(2) io. lseek(2) is
698 used to position the io location.
699
a31041ea 700 psync Basic pread(2) or pwrite(2) io.
701
e05af9e5 702 vsync Basic readv(2) or writev(2) IO.
1d2af02a 703
a46c5e01
JA
704 psyncv Basic preadv(2) or pwritev(2) IO.
705
15d182aa
JA
706 libaio Linux native asynchronous io. Note that Linux
707 may only support queued behaviour with
708 non-buffered IO (set direct=1 or buffered=0).
de890a1e 709 This engine defines engine specific options.
71bfa161
JA
710
711 posixaio glibc posix asynchronous io.
712
417f0068
JA
713 solarisaio Solaris native asynchronous io.
714
03e20d68
BC
715 windowsaio Windows native asynchronous io.
716
71bfa161
JA
717 mmap File is memory mapped and data copied
718 to/from using memcpy(3).
719
720 splice splice(2) is used to transfer the data and
721 vmsplice(2) to transfer data from user
722 space to the kernel.
723
d0ff85df
JA
724 syslet-rw Use the syslet system calls to make
725 regular read/write async.
726
71bfa161 727 sg SCSI generic sg v3 io. May either be
6c219763 728 synchronous using the SG_IO ioctl, or if
71bfa161
JA
729 the target is an sg character device
730 we use read(2) and write(2) for asynchronous
731 io.
732
a94ea28b
JA
733 null Doesn't transfer any data, just pretends
734 to. This is mainly used to exercise fio
735 itself and for debugging/testing purposes.
736
ed92ac0c 737 net Transfer over the network to given host:port.
de890a1e
SL
738 Depending on the protocol used, the hostname,
739 port, listen and filename options are used to
740 specify what sort of connection to make, while
741 the protocol option determines which protocol
742 will be used.
743 This engine defines engine specific options.
ed92ac0c 744
9cce02e8
JA
745 netsplice Like net, but uses splice/vmsplice to
746 map data and send/receive.
de890a1e 747 This engine defines engine specific options.
9cce02e8 748
53aec0a4 749 cpuio Doesn't transfer any data, but burns CPU
ba0fbe10
JA
750 cycles according to the cpuload= and
751 cpucycle= options. Setting cpuload=85
752 will cause that job to do nothing but burn
36ecec83
GP
753 85% of the CPU. In case of SMP machines,
754 use numjobs=<no_of_cpu> to get desired CPU
755 usage, as the cpuload only loads a single
756 CPU at the desired rate.
ba0fbe10 757
e9a1806f
JA
758 guasi The GUASI IO engine is the Generic Userspace
759 Asyncronous Syscall Interface approach
760 to async IO. See
761
762 http://www.xmailserver.org/guasi-lib.html
763
764 for more info on GUASI.
765
21b8aee8 766 rdma The RDMA I/O engine supports both RDMA
eb52fa3f
BVA
767 memory semantics (RDMA_WRITE/RDMA_READ) and
768 channel semantics (Send/Recv) for the
769 InfiniBand, RoCE and iWARP protocols.
21b8aee8 770
b861be9f
JA
771 falloc IO engine that does regular fallocate to
772 simulate data transfer as fio ioengine.
773 DDIR_READ does fallocate(,mode = keep_size,)
774 DDIR_WRITE does fallocate(,mode = 0)
775 DDIR_TRIM does fallocate(,mode = punch_hole)
d54fce84
DM
776
777 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
b861be9f
JA
778 ioctls to simulate defragment activity in
779 request to DDIR_WRITE event
780
781 rbd IO engine supporting direct access to Ceph
782 Rados Block Devices (RBD) via librbd without
783 the need to use the kernel rbd driver. This
784 ioengine defines engine specific options.
785
786 gfapi Using Glusterfs libgfapi sync interface to
787 direct access to Glusterfs volumes without
788 options.
789
790 gfapi_async Using Glusterfs libgfapi async interface
791 to direct access to Glusterfs volumes without
792 having to go through FUSE. This ioengine
793 defines engine specific options.
0981fd71 794
b74e419e 795 libhdfs Read and write through Hadoop (HDFS).
a3f001f5 796 This engine interprets offsets a little
b74e419e
MM
797 differently. In HDFS, files once created
798 cannot be modified. So random writes are not
799 possible. To imitate this, libhdfs engine
a3f001f5
FB
800 creates bunch of small files, and engine will
801 pick a file out of those files based on the
802 offset enerated by fio backend. Each jobs uses
803 it's own connection to HDFS.
1b10477b 804
65fa28ca
DE
805 mtd Read, write and erase an MTD character device
806 (e.g., /dev/mtd0). Discards are treated as
807 erases. Depending on the underlying device
808 type, the I/O may have to go in a certain
809 pattern, e.g., on NAND, writing sequentially
810 to erase blocks and discarding before
811 overwriting. The writetrim mode works well
812 for this constraint.
813
5c4ef02e
JA
814 pmemblk Read and write through the NVML libpmemblk
815 interface.
816
8a7bd877
JA
817 external Prefix to specify loading an external
818 IO engine object file. Append the engine
819 filename, eg ioengine=external:/tmp/foo.o
820 to load ioengine foo.o in /tmp.
821
71bfa161
JA
822iodepth=int This defines how many io units to keep in flight against
823 the file. The default is 1 for each file defined in this
824 job, can be overridden with a larger value for higher
ee72ca09
JA
825 concurrency. Note that increasing iodepth beyond 1 will not
826 affect synchronous ioengines (except for small degress when
9b836561 827 verify_async is in use). Even async engines may impose OS
ee72ca09
JA
828 restrictions causing the desired depth not to be achieved.
829 This may happen on Linux when using libaio and not setting
830 direct=1, since buffered IO is not async on that OS. Keep an
831 eye on the IO depth distribution in the fio output to verify
832 that the achieved depth is as expected. Default: 1.
71bfa161 833
4950421a 834iodepth_batch_submit=int
cb5ab512 835iodepth_batch=int This defines how many pieces of IO to submit at once.
89e820f6
JA
836 It defaults to 1 which means that we submit each IO
837 as soon as it is available, but can be raised to submit
e63a0b2f
RP
838 bigger batches of IO at the time. If it is set to 0 the iodepth
839 value will be used.
cb5ab512 840
82407585 841iodepth_batch_complete_min=int
4950421a
JA
842iodepth_batch_complete=int This defines how many pieces of IO to retrieve
843 at once. It defaults to 1 which means that we'll ask
844 for a minimum of 1 IO in the retrieval process from
845 the kernel. The IO retrieval will go on until we
846 hit the limit set by iodepth_low. If this variable is
847 set to 0, then fio will always check for completed
848 events before queuing more IO. This helps reduce
849 IO latency, at the cost of more retrieval system calls.
850
82407585
RP
851iodepth_batch_complete_max=int This defines maximum pieces of IO to
852 retrieve at once. This variable should be used along with
853 iodepth_batch_complete_min=int variable, specifying the range
854 of min and max amount of IO which should be retrieved. By default
855 it is equal to iodepth_batch_complete_min value.
856
857 Example #1:
858
859 iodepth_batch_complete_min=1
860 iodepth_batch_complete_max=<iodepth>
861
862 which means that we will retrieve at leat 1 IO and up to the
863 whole submitted queue depth. If none of IO has been completed
864 yet, we will wait.
865
866 Example #2:
867
868 iodepth_batch_complete_min=0
869 iodepth_batch_complete_max=<iodepth>
870
871 which means that we can retrieve up to the whole submitted
872 queue depth, but if none of IO has been completed yet, we will
873 NOT wait and immediately exit the system call. In this example
874 we simply do polling.
875
e916b390
JA
876iodepth_low=int The low water mark indicating when to start filling
877 the queue again. Defaults to the same as iodepth, meaning
878 that fio will attempt to keep the queue full at all times.
879 If iodepth is set to eg 16 and iodepth_low is set to 4, then
880 after fio has filled the queue of 16 requests, it will let
881 the depth drain down to 4 before starting to fill it again.
882
1ad01bd1
JA
883io_submit_mode=str This option controls how fio submits the IO to
884 the IO engine. The default is 'inline', which means that the
885 fio job threads submit and reap IO directly. If set to
886 'offload', the job threads will offload IO submission to a
887 dedicated pool of IO threads. This requires some coordination
888 and thus has a bit of extra overhead, especially for lower
889 queue depth IO where it can increase latencies. The benefit
890 is that fio can manage submission rates independently of
891 the device completion rates. This avoids skewed latency
892 reporting if IO gets back up on the device side (the
893 coordinated omission problem).
894
71bfa161 895direct=bool If value is true, use non-buffered io. This is usually
9b836561 896 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
93bcfd20 897 On Windows the synchronous ioengines don't support direct io.
76a43db4 898
d01612f3
CM
899atomic=bool If value is true, attempt to use atomic direct IO. Atomic
900 writes are guaranteed to be stable once acknowledged by
901 the operating system. Only Linux supports O_ATOMIC right
902 now.
903
76a43db4
JA
904buffered=bool If value is true, use buffered io. This is the opposite
905 of the 'direct' option. Defaults to true.
71bfa161 906
f7fa2653 907offset=int Start io at the given offset in the file. The data before
71bfa161
JA
908 the given offset will not be touched. This effectively
909 caps the file size at real_size - offset.
910
214ac7e0 911offset_increment=int If this is provided, then the real offset becomes
69bdd6ba
JH
912 offset + offset_increment * thread_number, where the thread
913 number is a counter that starts at 0 and is incremented for
914 each sub-job (i.e. when numjobs option is specified). This
915 option is useful if there are several jobs which are intended
916 to operate on a file in parallel disjoint segments, with
917 even spacing between the starting points.
214ac7e0 918
ddf24e42
JA
919number_ios=int Fio will normally perform IOs until it has exhausted the size
920 of the region set by size=, or if it exhaust the allocated
921 time (or hits an error condition). With this setting, the
922 range/size can be set independently of the number of IOs to
923 perform. When fio reaches this number, it will exit normally
be3fec7d
JA
924 and report status. Note that this does not extend the amount
925 of IO that will be done, it will only stop fio if this
926 condition is met before other end-of-job criteria.
ddf24e42 927
71bfa161
JA
928fsync=int If writing to a file, issue a sync of the dirty data
929 for every number of blocks given. For example, if you give
930 32 as a parameter, fio will sync the file for every 32
931 writes issued. If fio is using non-buffered io, we may
932 not sync the file. The exception is the sg io engine, which
6c219763 933 synchronizes the disk cache anyway.
71bfa161 934
e76b1da4 935fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
5f9099ea 936 metadata blocks.
37db59d6
JA
937 In FreeBSD and Windows there is no fdatasync(), this falls back
938 to using fsync()
5f9099ea 939
e76b1da4
JA
940sync_file_range=str:val Use sync_file_range() for every 'val' number of
941 write operations. Fio will track range of writes that
942 have happened since the last sync_file_range() call. 'str'
943 can currently be one or more of:
944
945 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
946 write SYNC_FILE_RANGE_WRITE
947 wait_after SYNC_FILE_RANGE_WAIT_AFTER
948
949 So if you do sync_file_range=wait_before,write:8, fio would
950 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
951 every 8 writes. Also see the sync_file_range(2) man page.
952 This option is Linux specific.
953
5036fc1e
JA
954overwrite=bool If true, writes to a file will always overwrite existing
955 data. If the file doesn't already exist, it will be
956 created before the write phase begins. If the file exists
957 and is large enough for the specified write phase, nothing
958 will be done.
71bfa161 959
dbd11ead 960end_fsync=bool If true, fsync file contents when a write stage has completed.
71bfa161 961
ebb1415f
JA
962fsync_on_close=bool If true, fio will fsync() a dirty file on close.
963 This differs from end_fsync in that it will happen on every
964 file close, not just at the end of the job.
965
71bfa161
JA
966rwmixread=int How large a percentage of the mix should be reads.
967
968rwmixwrite=int How large a percentage of the mix should be writes. If both
969 rwmixread and rwmixwrite is given and the values do not add
970 up to 100%, the latter of the two will be used to override
c35dd7a6
JA
971 the first. This may interfere with a given rate setting,
972 if fio is asked to limit reads or writes to a certain rate.
973 If that is the case, then the distribution may be skewed.
71bfa161 974
92d42d69
JA
975random_distribution=str:float By default, fio will use a completely uniform
976 random distribution when asked to perform random IO. Sometimes
977 it is useful to skew the distribution in specific ways,
978 ensuring that some parts of the data is more hot than others.
979 fio includes the following distribution models:
980
981 random Uniform random distribution
982 zipf Zipf distribution
983 pareto Pareto distribution
8116fd24 984 gauss Normal (guassian) distribution
e0a04ac1 985 zoned Zoned random distribution
92d42d69
JA
986
987 When using a zipf or pareto distribution, an input value
988 is also needed to define the access pattern. For zipf, this
989 is the zipf theta. For pareto, it's the pareto power. Fio
990 includes a test program, genzipf, that can be used visualize
991 what the given input values will yield in terms of hit rates.
992 If you wanted to use zipf with a theta of 1.2, you would use
993 random_distribution=zipf:1.2 as the option. If a non-uniform
8116fd24
JA
994 model is used, fio will disable use of the random map. For
995 the gauss distribution, a normal deviation is supplied as
996 a value between 0 and 100.
92d42d69 997
e0a04ac1
JA
998 For a zoned distribution, fio supports specifying percentages
999 of IO access that should fall within what range of the file or
1000 device. For example, given a criteria of:
1001
1002 60% of accesses should be to the first 10%
1003 30% of accesses should be to the next 20%
1004 8% of accesses should be to to the next 30%
1005 2% of accesses should be to the next 40%
1006
1007 we can define that through zoning of the random accesses. For
1008 the above example, the user would do:
1009
1010 random_distribution=zoned:60/10:30/20:8/30:2/40
1011
1012 similarly to how bssplit works for setting ranges and
1013 percentages of block sizes. Like bssplit, it's possible to
1014 specify separate zones for reads, writes, and trims. If just
1015 one set is given, it'll apply to all of them.
1016
211c9b89
JA
1017percentage_random=int For a random workload, set how big a percentage should
1018 be random. This defaults to 100%, in which case the workload
1019 is fully random. It can be set from anywhere from 0 to 100.
1020 Setting it to 0 would make the workload fully sequential. Any
1021 setting in between will result in a random mix of sequential
d9472271
JA
1022 and random IO, at the given percentages. It is possible to
1023 set different values for reads, writes, and trim. To do so,
1024 simply use a comma separated list. See blocksize.
211c9b89 1025
bb8895e0
JA
1026norandommap Normally fio will cover every block of the file when doing
1027 random IO. If this option is given, fio will just get a
1028 new random offset without looking at past io history. This
1029 means that some blocks may not be read or written, and that
83da8fbf
JE
1030 some blocks may be read/written more than once. If this option
1031 is used with verify= and multiple blocksizes (via bsrange=),
1032 only intact blocks are verified, i.e., partially-overwritten
1033 blocks are ignored.
bb8895e0 1034
0408c206
JA
1035softrandommap=bool See norandommap. If fio runs with the random block map
1036 enabled and it fails to allocate the map, if this option is
1037 set it will continue without a random block map. As coverage
1038 will not be as complete as with random maps, this option is
2b386d25
JA
1039 disabled by default.
1040
e8b1961d
JA
1041random_generator=str Fio supports the following engines for generating
1042 IO offsets for random IO:
1043
1044 tausworthe Strong 2^88 cycle random number generator
1045 lfsr Linear feedback shift register generator
c3546b53
JA
1046 tausworthe64 Strong 64-bit 2^258 cycle random number
1047 generator
e8b1961d
JA
1048
1049 Tausworthe is a strong random number generator, but it
1050 requires tracking on the side if we want to ensure that
1051 blocks are only read or written once. LFSR guarantees
1052 that we never generate the same offset twice, and it's
1053 also less computationally expensive. It's not a true
1054 random generator, however, though for IO purposes it's
1055 typically good enough. LFSR only works with single
1056 block sizes, not with workloads that use multiple block
1057 sizes. If used with such a workload, fio may read or write
3bb85e84
JA
1058 some blocks multiple times. The default value is tausworthe,
1059 unless the required space exceeds 2^32 blocks. If it does,
1060 then tausworthe64 is selected automatically.
43f09da1 1061
71bfa161
JA
1062nice=int Run the job with the given nice value. See man nice(2).
1063
1064prio=int Set the io priority value of this job. Linux limits us to
1065 a positive value between 0 and 7, with 0 being the highest.
1066 See man ionice(1).
1067
1068prioclass=int Set the io priority class. See man ionice(1).
1069
1070thinktime=int Stall the job x microseconds after an io has completed before
1071 issuing the next. May be used to simulate processing being
48097d5c
JA
1072 done by an application. See thinktime_blocks and
1073 thinktime_spin.
1074
1075thinktime_spin=int
1076 Only valid if thinktime is set - pretend to spend CPU time
1077 doing something with the data received, before falling back
1078 to sleeping for the rest of the period specified by
1079 thinktime.
9c1f7434 1080
4d01ece6 1081thinktime_blocks=int
9c1f7434
JA
1082 Only valid if thinktime is set - control how many blocks
1083 to issue, before waiting 'thinktime' usecs. If not set,
1084 defaults to 1 which will make fio wait 'thinktime' usecs
4d01ece6
JA
1085 after every block. This effectively makes any queue depth
1086 setting redundant, since no more than 1 IO will be queued
1087 before we have to complete it and do our thinktime. In
1088 other words, this setting effectively caps the queue depth
1089 if the latter is larger.
71bfa161 1090
581e7141 1091rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
b09da8fa 1092 the normal suffix rules apply. You can use rate=500k to limit
581e7141
JA
1093 reads and writes to 500k each, or you can specify read and
1094 writes separately. Using rate=1m,500k would limit reads to
1095 1MB/sec and writes to 500KB/sec. Capping only reads or
1096 writes can be done with rate=,500k or rate=500k,. The former
1097 will only limit writes (to 500KB/sec), the latter will only
1098 limit reads.
71bfa161 1099
6d428bcd 1100rate_min=int Tell fio to do whatever it can to maintain at least this
4e991c23 1101 bandwidth. Failing to meet this requirement, will cause
581e7141
JA
1102 the job to exit. The same format as rate is used for
1103 read vs write separation.
4e991c23
JA
1104
1105rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
1106 as rate, just specified independently of bandwidth. If the
1107 job is given a block size range instead of a fixed value,
581e7141 1108 the smallest block size is used as the metric. The same format
de8f6de9 1109 as rate is used for read vs write separation.
4e991c23
JA
1110
1111rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
581e7141 1112 the job to exit. The same format as rate is used for read vs
de8f6de9 1113 write separation.
71bfa161 1114
6de65959
JA
1115rate_process=str This option controls how fio manages rated IO
1116 submissions. The default is 'linear', which submits IO in a
1117 linear fashion with fixed delays between IOs that gets
1118 adjusted based on IO completion rates. If this is set to
1119 'poisson', fio will submit IO based on a more real world
1120 random request flow, known as the Poisson process
5d02b083
JA
1121 (https://en.wikipedia.org/wiki/Poisson_process). The lambda
1122 will be 10^6 / IOPS for the given workload.
e7b24047 1123
3e260a46
JA
1124latency_target=int If set, fio will attempt to find the max performance
1125 point that the given workload will run at while maintaining a
1126 latency below this target. The values is given in microseconds.
1127 See latency_window and latency_percentile
1128
1129latency_window=int Used with latency_target to specify the sample window
1130 that the job is run at varying queue depths to test the
1131 performance. The value is given in microseconds.
1132
1133latency_percentile=float The percentage of IOs that must fall within the
1134 criteria specified by latency_target and latency_window. If not
1135 set, this defaults to 100.0, meaning that all IOs must be equal
1136 or below to the value set by latency_target.
1137
15501535
JA
1138max_latency=int If set, fio will exit the job if it exceeds this maximum
1139 latency. It will exit with an ETIME error.
1140
6d428bcd 1141rate_cycle=int Average bandwidth for 'rate' and 'rate_min' over this number
6c219763 1142 of milliseconds.
71bfa161
JA
1143
1144cpumask=int Set the CPU affinity of this job. The parameter given is a
a08bc17f
JA
1145 bitmask of allowed CPU's the job may run on. So if you want
1146 the allowed CPUs to be 1 and 5, you would pass the decimal
1147 value of (1 << 1 | 1 << 5), or 34. See man
7dbb6eba 1148 sched_setaffinity(2). This may not work on all supported
b0ea08ce
JA
1149 operating systems or kernel versions. This option doesn't
1150 work well for a higher CPU count than what you can store in
1151 an integer mask, so it can only control cpus 1-32. For
1152 boxes with larger CPU counts, use cpus_allowed.
71bfa161 1153
d2e268b0
JA
1154cpus_allowed=str Controls the same options as cpumask, but it allows a text
1155 setting of the permitted CPUs instead. So to use CPUs 1 and
62a7273d
JA
1156 5, you would specify cpus_allowed=1,5. This options also
1157 allows a range of CPUs. Say you wanted a binding to CPUs
1158 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
d2e268b0 1159
c2acfbac
JA
1160cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
1161 specified by cpus_allowed or cpumask. Two policies are
1162 supported:
1163
1164 shared All jobs will share the CPU set specified.
1165 split Each job will get a unique CPU from the CPU set.
1166
1167 'shared' is the default behaviour, if the option isn't
ada083cd
JA
1168 specified. If split is specified, then fio will will assign
1169 one cpu per job. If not enough CPUs are given for the jobs
1170 listed, then fio will roundrobin the CPUs in the set.
c2acfbac 1171
d0b937ed
YR
1172numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
1173 arguments allow comma delimited list of cpu numbers,
1174 A-B ranges, or 'all'. Note, to enable numa options support,
67bf9823 1175 fio must be built on a system with libnuma-dev(el) installed.
d0b937ed
YR
1176
1177numa_mem_policy=str Set this job's memory policy and corresponding NUMA
1178 nodes. Format of the argements:
1179 <mode>[:<nodelist>]
1180 `mode' is one of the following memory policy:
1181 default, prefer, bind, interleave, local
1182 For `default' and `local' memory policy, no node is
1183 needed to be specified.
1184 For `prefer', only one node is allowed.
1185 For `bind' and `interleave', it allow comma delimited
1186 list of numbers, A-B ranges, or 'all'.
1187
e417fd66 1188startdelay=time Start this job the specified number of seconds after fio
71bfa161
JA
1189 has started. Only useful if the job file contains several
1190 jobs, and you want to delay starting some jobs to a certain
1191 time.
1192
e417fd66 1193runtime=time Tell fio to terminate processing after the specified number
71bfa161
JA
1194 of seconds. It can be quite hard to determine for how long
1195 a specified job will run, so this parameter is handy to
1196 cap the total runtime to a given time.
1197
cf4464ca 1198time_based If set, fio will run for the duration of the runtime
bf9a3edb 1199 specified even if the file(s) are completely read or
cf4464ca
JA
1200 written. It will simply loop over the same workload
1201 as many times as the runtime allows.
1202
e417fd66 1203ramp_time=time If set, fio will run the specified workload for this amount
721938ae
JA
1204 of time before logging any performance numbers. Useful for
1205 letting performance settle before logging results, thus
b29ee5b3
JA
1206 minimizing the runtime required for stable results. Note
1207 that the ramp_time is considered lead in time for a job,
1208 thus it will increase the total runtime if a special timeout
1209 or runtime is specified.
721938ae 1210
71bfa161
JA
1211invalidate=bool Invalidate the buffer/page cache parts for this file prior
1212 to starting io. Defaults to true.
1213
1214sync=bool Use sync io for buffered writes. For the majority of the
1215 io engines, this means using O_SYNC.
1216
d3aad8f2 1217iomem=str
71bfa161
JA
1218mem=str Fio can use various types of memory as the io unit buffer.
1219 The allowed values are:
1220
1221 malloc Use memory from malloc(3) as the buffers.
1222
1223 shm Use shared memory as the buffers. Allocated
1224 through shmget(2).
1225
74b025b0
JA
1226 shmhuge Same as shm, but use huge pages as backing.
1227
313cb206
JA
1228 mmap Use mmap to allocate buffers. May either be
1229 anonymous memory, or can be file backed if
1230 a filename is given after the option. The
1231 format is mem=mmap:/path/to/file.
71bfa161 1232
d0bdaf49
JA
1233 mmaphuge Use a memory mapped huge file as the buffer
1234 backing. Append filename after mmaphuge, ala
1235 mem=mmaphuge:/hugetlbfs/file
1236
09c782bb
JA
1237 mmapshared Same as mmap, but use a MMAP_SHARED
1238 mapping.
1239
71bfa161 1240 The area allocated is a function of the maximum allowed
5394ae5f
JA
1241 bs size for the job, multiplied by the io depth given. Note
1242 that for shmhuge and mmaphuge to work, the system must have
1243 free huge pages allocated. This can normally be checked
1244 and set by reading/writing /proc/sys/vm/nr_hugepages on a
b22989b9 1245 Linux system. Fio assumes a huge page is 4MB in size. So
5394ae5f
JA
1246 to calculate the number of huge pages you need for a given
1247 job file, add up the io depth of all jobs (normally one unless
1248 iodepth= is used) and multiply by the maximum bs set. Then
1249 divide that number by the huge page size. You can see the
1250 size of the huge pages in /proc/meminfo. If no huge pages
1251 are allocated by having a non-zero number in nr_hugepages,
56bb17f2 1252 using mmaphuge or shmhuge will fail. Also see hugepage-size.
5394ae5f
JA
1253
1254 mmaphuge also needs to have hugetlbfs mounted and the file
1255 location should point there. So if it's mounted in /huge,
1256 you would use mem=mmaphuge:/huge/somefile.
71bfa161 1257
d529ee19
JA
1258iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1259 Note that the given alignment is applied to the first IO unit
1260 buffer, if using iodepth the alignment of the following buffers
1261 are given by the bs used. In other words, if using a bs that is
1262 a multiple of the page sized in the system, all buffers will
1263 be aligned to this value. If using a bs that is not page
1264 aligned, the alignment of subsequent IO memory buffers is the
1265 sum of the iomem_align and bs used.
1266
f7fa2653 1267hugepage-size=int
56bb17f2 1268 Defines the size of a huge page. Must at least be equal
b22989b9 1269 to the system setting, see /proc/meminfo. Defaults to 4MB.
c51074e7
JA
1270 Should probably always be a multiple of megabytes, so using
1271 hugepage-size=Xm is the preferred way to set this to avoid
1272 setting a non-pow-2 bad value.
56bb17f2 1273
71bfa161
JA
1274exitall When one job finishes, terminate the rest. The default is
1275 to wait for each job to finish, sometimes that is not the
1276 desired action.
1277
f9cafb12
JA
1278exitall_on_error When one job finishes in error, terminate the rest. The
1279 default is to wait for each job to finish.
1280
71bfa161 1281bwavgtime=int Average the calculated bandwidth over the given time. Value
a47591e4
JA
1282 is specified in milliseconds. If the job also does bandwidth
1283 logging through 'write_bw_log', then the minimum of this option
1284 and 'log_avg_msec' will be used. Default: 500ms.
71bfa161 1285
c8eeb9df 1286iopsavgtime=int Average the calculated IOPS over the given time. Value
a47591e4
JA
1287 is specified in milliseconds. If the job also does IOPS logging
1288 through 'write_iops_log', then the minimum of this option and
1289 'log_avg_msec' will be used. Default: 500ms.
c8eeb9df 1290
71bfa161
JA
1291create_serialize=bool If true, serialize the file creating for the jobs.
1292 This may be handy to avoid interleaving of data
1293 files, which may greatly depend on the filesystem
1294 used and even the number of processors in the system.
1295
1296create_fsync=bool fsync the data file after creation. This is the
1297 default.
1298
814452bd
JA
1299create_on_open=bool Don't pre-setup the files for IO, just create open()
1300 when it's time to do IO to that file.
1301
25460cf6
JA
1302create_only=bool If true, fio will only run the setup phase of the job.
1303 If files need to be laid out or updated on disk, only
1304 that will be done. The actual job contents are not
1305 executed.
1306
2378826d
JA
1307allow_file_create=bool If true, fio is permitted to create files as part
1308 of its workload. This is the default behavior. If this
1309 option is false, then fio will error out if the files it
1310 needs to use don't already exist. Default: true.
1311
e81ecca3
JA
1312allow_mounted_write=bool If this isn't set, fio will abort jobs that
1313 are destructive (eg that write) to what appears to be a
1314 mounted device or partition. This should help catch creating
1315 inadvertently destructive tests, not realizing that the test
1316 will destroy data on the mounted file system. Default: false.
1317
afad68f7 1318pre_read=bool If this is given, files will be pre-read into memory before
34f1c044
JA
1319 starting the given IO operation. This will also clear
1320 the 'invalidate' flag, since it is pointless to pre-read
9c0d2241
JA
1321 and then drop the cache. This will only work for IO engines
1322 that are seekable, since they allow you to read the same data
1323 multiple times. Thus it will not work on eg network or splice
1324 IO.
afad68f7 1325
e545a6ce 1326unlink=bool Unlink the job files when done. Not the default, as repeated
bf9a3edb
JA
1327 runs of that job would then waste time recreating the file
1328 set again and again.
71bfa161
JA
1329
1330loops=int Run the specified number of iterations of this job. Used
1331 to repeat the same workload a given number of times. Defaults
1332 to 1.
1333
62167762
JC
1334verify_only Do not perform specified workload---only verify data still
1335 matches previous invocation of this workload. This option
1336 allows one to check data multiple times at a later date
1337 without overwriting it. This option makes sense only for
1338 workloads that write data, and does not support workloads
1339 with the time_based option set.
1340
68e1f29a 1341do_verify=bool Run the verify phase after a write phase. Only makes sense if
e84c73a8
SL
1342 verify is set. Defaults to 1.
1343
71bfa161 1344verify=str If writing to a file, fio can verify the file contents
b638d82f
RP
1345 after each iteration of the job. Each verification method also implies
1346 verification of special header, which is written to the beginning of
1347 each block. This header also includes meta information, like offset
1348 of the block, block number, timestamp when block was written, etc.
1349 verify=str can be combined with verify_pattern=str option.
1350 The allowed values are:
71bfa161
JA
1351
1352 md5 Use an md5 sum of the data area and store
1353 it in the header of each block.
1354
17dc34df
JA
1355 crc64 Use an experimental crc64 sum of the data
1356 area and store it in the header of each
1357 block.
1358
bac39e0e
JA
1359 crc32c Use a crc32c sum of the data area and store
1360 it in the header of each block.
1361
3845591f 1362 crc32c-intel Use hardware assisted crc32c calcuation
0539d758
JA
1363 provided on SSE4.2 enabled processors. Falls
1364 back to regular software crc32c, if not
1365 supported by the system.
3845591f 1366
71bfa161
JA
1367 crc32 Use a crc32 sum of the data area and store
1368 it in the header of each block.
1369
969f7ed3
JA
1370 crc16 Use a crc16 sum of the data area and store
1371 it in the header of each block.
1372
17dc34df
JA
1373 crc7 Use a crc7 sum of the data area and store
1374 it in the header of each block.
1375
844ea602
JA
1376 xxhash Use xxhash as the checksum function. Generally
1377 the fastest software checksum that fio
1378 supports.
1379
cd14cc10
JA
1380 sha512 Use sha512 as the checksum function.
1381
1382 sha256 Use sha256 as the checksum function.
1383
7c353ceb
JA
1384 sha1 Use optimized sha1 as the checksum function.
1385
b638d82f
RP
1386 meta This option is deprecated, since now meta information is
1387 included in generic verification header and meta verification
1388 happens by default. For detailed information see the description
1389 of the verify=str setting. This option is kept because of
1390 compatibility's sake with old configurations. Do not use it.
7437ee87 1391
59245381
JA
1392 pattern Verify a strict pattern. Normally fio includes
1393 a header with some basic information and
1394 checksumming, but if this option is set, only
1395 the specific pattern set with 'verify_pattern'
1396 is verified.
1397
36690c9b
JA
1398 null Only pretend to verify. Useful for testing
1399 internals with ioengine=null, not for much
1400 else.
1401
6c219763 1402 This option can be used for repeated burn-in tests of a
71bfa161 1403 system to make sure that the written data is also
b892dc08
JA
1404 correctly read back. If the data direction given is
1405 a read or random read, fio will assume that it should
1406 verify a previously written file. If the data direction
1407 includes any form of write, the verify will be of the
1408 newly written data.
71bfa161 1409
160b966d
JA
1410verifysort=bool If set, fio will sort written verify blocks when it deems
1411 it faster to read them back in a sorted manner. This is
1412 often the case when overwriting an existing file, since
1413 the blocks are already laid out in the file system. You
1414 can ignore this option unless doing huge amounts of really
1415 fast IO where the red-black tree sorting CPU time becomes
1416 significant.
3f9f4e26 1417
f7fa2653 1418verify_offset=int Swap the verification header with data somewhere else
546a9142
SL
1419 in the block before writing. Its swapped back before
1420 verifying.
1421
f7fa2653 1422verify_interval=int Write the verification header at a finer granularity
3f9f4e26
SL
1423 than the blocksize. It will be written for chunks the
1424 size of header_interval. blocksize should divide this
1425 evenly.
90059d65 1426
0e92f873 1427verify_pattern=str If set, fio will fill the io buffers with this
e28218f3
SL
1428 pattern. Fio defaults to filling with totally random
1429 bytes, but sometimes it's interesting to fill with a known
1430 pattern for io verification purposes. Depending on the
1431 width of the pattern, fio will fill 1/2/3/4 bytes of the
0e92f873
RR
1432 buffer at the time(it can be either a decimal or a hex number).
1433 The verify_pattern if larger than a 32-bit quantity has to
996093bb 1434 be a hex number that starts with either "0x" or "0X". Use
b638d82f 1435 with verify=str. Also, verify_pattern supports %o format,
61b9861d
RP
1436 which means that for each block offset will be written and
1437 then verifyied back, e.g.:
1438
1439 verify_pattern=%o
1440
1441 Or use combination of everything:
1442 verify_pattern=0xff%o"abcd"-12
e28218f3 1443
68e1f29a 1444verify_fatal=bool Normally fio will keep checking the entire contents
a12a3b4d
JA
1445 before quitting on a block verification failure. If this
1446 option is set, fio will exit the job on the first observed
1447 failure.
e8462bd8 1448
b463e936
JA
1449verify_dump=bool If set, dump the contents of both the original data
1450 block and the data block we read off disk to files. This
1451 allows later analysis to inspect just what kind of data
ef71e317 1452 corruption occurred. Off by default.
b463e936 1453
e8462bd8
JA
1454verify_async=int Fio will normally verify IO inline from the submitting
1455 thread. This option takes an integer describing how many
1456 async offload threads to create for IO verification instead,
1457 causing fio to offload the duty of verifying IO contents
c85c324c
JA
1458 to one or more separate threads. If using this offload
1459 option, even sync IO engines can benefit from using an
1460 iodepth setting higher than 1, as it allows them to have
1461 IO in flight while verifies are running.
e8462bd8
JA
1462
1463verify_async_cpus=str Tell fio to set the given CPU affinity on the
1464 async IO verification threads. See cpus_allowed for the
1465 format used.
6f87418f
JA
1466
1467verify_backlog=int Fio will normally verify the written contents of a
1468 job that utilizes verify once that job has completed. In
1469 other words, everything is written then everything is read
1470 back and verified. You may want to verify continually
1471 instead for a variety of reasons. Fio stores the meta data
1472 associated with an IO block in memory, so for large
1473 verify workloads, quite a bit of memory would be used up
1474 holding this meta data. If this option is enabled, fio
f42195a3
JA
1475 will write only N blocks before verifying these blocks.
1476
6f87418f
JA
1477verify_backlog_batch=int Control how many blocks fio will verify
1478 if verify_backlog is set. If not set, will default to
1479 the value of verify_backlog (meaning the entire queue
f42195a3
JA
1480 is read back and verified). If verify_backlog_batch is
1481 less than verify_backlog then not all blocks will be verified,
1482 if verify_backlog_batch is larger than verify_backlog, some
1483 blocks will be verified more than once.
66c098b8 1484
ca09be4b
JA
1485verify_state_save=bool When a job exits during the write phase of a verify
1486 workload, save its current state. This allows fio to replay
1487 up until that point, if the verify state is loaded for the
1488 verify read phase. The format of the filename is, roughly,
1489 <type>-<jobname>-<jobindex>-verify.state. <type> is "local"
1490 for a local run, "sock" for a client/server socket connection,
1491 and "ip" (192.168.0.1, for instance) for a networked
1492 client/server connection.
1493
1494verify_state_load=bool If a verify termination trigger was used, fio stores
1495 the current write state of each thread. This can be used at
1496 verification time so that fio knows how far it should verify.
1497 Without this information, fio will run a full verification
1498 pass, according to the settings in the job file used.
1499
d392365e 1500stonewall
de8f6de9 1501wait_for_previous Wait for preceding jobs in the job file to exit, before
71bfa161 1502 starting this one. Can be used to insert serialization
b3d62a75
JA
1503 points in the job file. A stone wall also implies starting
1504 a new reporting group.
1505
abcab6af 1506new_group Start a new reporting group. See: group_reporting.
71bfa161
JA
1507
1508numjobs=int Create the specified number of clones of this job. May be
1509 used to setup a larger number of threads/processes doing
abcab6af
AV
1510 the same thing. Each thread is reported separately; to see
1511 statistics for all clones as a whole, use group_reporting in
1512 conjunction with new_group.
1513
1514group_reporting It may sometimes be interesting to display statistics for
04b2f799
JA
1515 groups of jobs as a whole instead of for each individual job.
1516 This is especially true if 'numjobs' is used; looking at
1517 individual thread/process output quickly becomes unwieldy.
1518 To see the final report per-group instead of per-job, use
1519 'group_reporting'. Jobs in a file will be part of the same
1520 reporting group, unless if separated by a stonewall, or by
1521 using 'new_group'.
71bfa161
JA
1522
1523thread fio defaults to forking jobs, however if this option is
1524 given, fio will use pthread_create(3) to create threads
1525 instead.
1526
f7fa2653 1527zonesize=int Divide a file into zones of the specified size. See zoneskip.
71bfa161 1528
f7fa2653 1529zoneskip=int Skip the specified number of bytes when zonesize data has
71bfa161
JA
1530 been read. The two zone options can be used to only do
1531 io on zones of a file.
1532
076efc7c 1533write_iolog=str Write the issued io patterns to the specified file. See
5b42a488
SH
1534 read_iolog. Specify a separate file for each job, otherwise
1535 the iologs will be interspersed and the file may be corrupt.
71bfa161 1536
076efc7c 1537read_iolog=str Open an iolog with the specified file name and replay the
71bfa161 1538 io patterns it contains. This can be used to store a
6df8adaa
JA
1539 workload and replay it sometime later. The iolog given
1540 may also be a blktrace binary file, which allows fio
1541 to replay a workload captured by blktrace. See blktrace
1542 for how to capture such logging data. For blktrace replay,
1543 the file needs to be turned into a blkparse binary data
ea3e51c3 1544 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
66c098b8 1545
64bbb865 1546replay_no_stall=int When replaying I/O with read_iolog the default behavior
62776229
JA
1547 is to attempt to respect the time stamps within the log and
1548 replay them with the appropriate delay between IOPS. By
1549 setting this variable fio will not respect the timestamps and
1550 attempt to replay them as fast as possible while still
1551 respecting ordering. The result is the same I/O pattern to a
1552 given device, but different timings.
71bfa161 1553
d1c46c04
DN
1554replay_redirect=str While replaying I/O patterns using read_iolog the
1555 default behavior is to replay the IOPS onto the major/minor
1556 device that each IOP was recorded from. This is sometimes
de8f6de9 1557 undesirable because on a different machine those major/minor
d1c46c04
DN
1558 numbers can map to a different device. Changing hardware on
1559 the same system can also result in a different major/minor
1560 mapping. Replay_redirect causes all IOPS to be replayed onto
1561 the single specified device regardless of the device it was
1562 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1563 IO in the blktrace to be replayed onto /dev/sdc. This means
1564 multiple devices will be replayed onto a single, if the trace
1565 contains multiple devices. If you want multiple devices to be
1566 replayed concurrently to multiple redirected devices you must
1567 blkparse your trace into separate traces and replay them with
1568 independent fio invocations. Unfortuantely this also breaks
1569 the strict time ordering between multiple device accesses.
1570
0c63576e
JA
1571replay_align=int Force alignment of IO offsets and lengths in a trace
1572 to this power of 2 value.
1573
1574replay_scale=int Scale sector offsets down by this factor when
1575 replaying traces.
1576
3a5db920
JA
1577per_job_logs=bool If set, this generates bw/clat/iops log with per
1578 file private filenames. If not set, jobs with identical names
1579 will share the log filename. Default: true.
1580
e3cedca7 1581write_bw_log=str If given, write a bandwidth log of the jobs in this job
71bfa161 1582 file. Can be used to store data of the bandwidth of the
e0da9bc2
JA
1583 jobs in their lifetime. The included fio_generate_plots
1584 script uses gnuplot to turn these text files into nice
ddb754db 1585 graphs. See write_lat_log for behaviour of given
8ad3b3dd
JA
1586 filename. For this option, the suffix is _bw.x.log, where
1587 x is the index of the job (1..N, where N is the number of
3a5db920 1588 jobs). If 'per_job_logs' is false, then the filename will not
a3ae5b05 1589 include the job index. See 'Log File Formats'.
71bfa161 1590
e3cedca7 1591write_lat_log=str Same as write_bw_log, except that this option stores io
02af0988
JA
1592 submission, completion, and total latencies instead. If no
1593 filename is given with this option, the default filename of
1594 "jobname_type.log" is used. Even if the filename is given,
1595 fio will still append the type of log. So if one specifies
e3cedca7
JA
1596
1597 write_lat_log=foo
1598
8ad3b3dd
JA
1599 The actual log names will be foo_slat.x.log, foo_clat.x.log,
1600 and foo_lat.x.log, where x is the index of the job (1..N,
1601 where N is the number of jobs). This helps fio_generate_plot
3a5db920 1602 fine the logs automatically. If 'per_job_logs' is false, then
a3ae5b05
JA
1603 the filename will not include the job index. See 'Log File
1604 Formats'.
71bfa161 1605
b8bc8cba
JA
1606write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1607 given with this option, the default filename of
8ad3b3dd
JA
1608 "jobname_type.x.log" is used,where x is the index of the job
1609 (1..N, where N is the number of jobs). Even if the filename
3a5db920
JA
1610 is given, fio will still append the type of log. If
1611 'per_job_logs' is false, then the filename will not include
a3ae5b05 1612 the job index. See 'Log File Formats'.
b8bc8cba
JA
1613
1614log_avg_msec=int By default, fio will log an entry in the iops, latency,
1615 or bw log for every IO that completes. When writing to the
1616 disk log, that can quickly grow to a very large size. Setting
1617 this option makes fio average the each log entry over the
1618 specified period of time, reducing the resolution of the log.
4b1ddb7a
JA
1619 See log_max_value as well. Defaults to 0, logging all entries.
1620
1621log_max_value=bool If log_avg_msec is set, fio logs the average over that
1622 window. If you instead want to log the maximum value, set this
1623 option to 1. Defaults to 0, meaning that averaged values are
1624 logged.
b8bc8cba 1625
ae588852
JA
1626log_offset=int If this is set, the iolog options will include the byte
1627 offset for the IO entry as well as the other data values.
1628
aee2ab67
JA
1629log_compression=int If this is set, fio will compress the IO logs as
1630 it goes, to keep the memory footprint lower. When a log
1631 reaches the specified size, that chunk is removed and
1632 compressed in the background. Given that IO logs are
1633 fairly highly compressible, this yields a nice memory
1634 savings for longer runs. The downside is that the
1635 compression will consume some background CPU cycles, so
1636 it may impact the run. This, however, is also true if
1637 the logging ends up consuming most of the system memory.
1638 So pick your poison. The IO logs are saved normally at the
1639 end of a run, by decompressing the chunks and storing them
1640 in the specified log file. This feature depends on the
1641 availability of zlib.
1642
c08f9fe2
JA
1643log_compression_cpus=str Define the set of CPUs that are allowed to
1644 handle online log compression for the IO jobs. This can
1645 provide better isolation between performance sensitive jobs,
1646 and background compression work.
1647
1648log_store_compressed=bool If set, fio will store the log files in a
1649 compressed format. They can be decompressed with fio, using
1650 the --inflate-log command line parameter. The files will be
1651 stored with a .fz suffix.
b26317c9 1652
66347cfa
DE
1653block_error_percentiles=bool If set, record errors in trim block-sized
1654 units from writes and trims and output a histogram of
1655 how many trims it took to get to errors, and what kind
1656 of error was encountered.
1657
f7fa2653 1658lockmem=int Pin down the specified amount of memory with mlock(2). Can
71bfa161
JA
1659 potentially be used instead of removing memory or booting
1660 with less memory to simulate a smaller amount of memory.
81c6b6cd 1661 The amount specified is per worker.
71bfa161
JA
1662
1663exec_prerun=str Before running this job, issue the command specified
74c8c488
JA
1664 through system(3). Output is redirected in a file called
1665 jobname.prerun.txt.
71bfa161
JA
1666
1667exec_postrun=str After the job completes, issue the command specified
74c8c488
JA
1668 though system(3). Output is redirected in a file called
1669 jobname.postrun.txt.
71bfa161
JA
1670
1671ioscheduler=str Attempt to switch the device hosting the file to the specified
1672 io scheduler before running.
1673
0a839f30
JA
1674disk_util=bool Generate disk utilization statistics, if the platform
1675 supports it. Defaults to on.
1676
02af0988 1677disable_lat=bool Disable measurements of total latency numbers. Useful
9520ebb9
JA
1678 only for cutting back the number of calls to gettimeofday,
1679 as that does impact performance at really high IOPS rates.
1680 Note that to really get rid of a large amount of these
1681 calls, this option must be used with disable_slat and
1682 disable_bw as well.
1683
02af0988
JA
1684disable_clat=bool Disable measurements of completion latency numbers. See
1685 disable_lat.
1686
9520ebb9 1687disable_slat=bool Disable measurements of submission latency numbers. See
02af0988 1688 disable_slat.
9520ebb9
JA
1689
1690disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
02af0988 1691 disable_lat.
9520ebb9 1692
83349190
YH
1693clat_percentiles=bool Enable the reporting of percentiles of
1694 completion latencies.
1695
1696percentile_list=float_list Overwrite the default list of percentiles
66347cfa
DE
1697 for completion latencies and the block error histogram.
1698 Each number is a floating number in the range (0,100],
1699 and the maximum length of the list is 20. Use ':'
1700 to separate the numbers, and list the numbers in ascending
1701 order. For example, --percentile_list=99.5:99.9 will cause
1702 fio to report the values of completion latency below which
1703 99.5% and 99.9% of the observed latencies fell, respectively.
83349190 1704
23893646
JA
1705clocksource=str Use the given clocksource as the base of timing. The
1706 supported options are:
1707
1708 gettimeofday gettimeofday(2)
1709
1710 clock_gettime clock_gettime(2)
1711
1712 cpu Internal CPU clock source
1713
1714 cpu is the preferred clocksource if it is reliable, as it
1715 is very fast (and fio is heavy on time calls). Fio will
1716 automatically use this clocksource if it's supported and
1717 considered reliable on the system it is running on, unless
1718 another clocksource is specifically set. For x86/x86-64 CPUs,
1719 this means supporting TSC Invariant.
1720
993bf48b
JA
1721gtod_reduce=bool Enable all of the gettimeofday() reducing options
1722 (disable_clat, disable_slat, disable_bw) plus reduce
1723 precision of the timeout somewhat to really shrink
1724 the gettimeofday() call count. With this option enabled,
1725 we only do about 0.4% of the gtod() calls we would have
1726 done if all time keeping was enabled.
1727
be4ecfdf
JA
1728gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1729 execution to just getting the current time. Fio (and
1730 databases, for instance) are very intensive on gettimeofday()
1731 calls. With this option, you can set one CPU aside for
1732 doing nothing but logging current time to a shared memory
1733 location. Then the other threads/processes that run IO
1734 workloads need only copy that segment, instead of entering
1735 the kernel with a gettimeofday() call. The CPU set aside
1736 for doing these time calls will be excluded from other
1737 uses. Fio will manually clear it from the CPU mask of other
1738 jobs.
a696fa2a 1739
06842027 1740continue_on_error=str Normally fio will exit the job on the first observed
f2bba182
RR
1741 failure. If this option is set, fio will continue the job when
1742 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1743 is exceeded or the I/O size specified is completed. If this
1744 option is used, there are two more stats that are appended,
1745 the total error count and the first error. The error field
1746 given in the stats is the first error that was hit during the
1747 run.
be4ecfdf 1748
06842027
SL
1749 The allowed values are:
1750
1751 none Exit on any IO or verify errors.
1752
1753 read Continue on read errors, exit on all others.
1754
1755 write Continue on write errors, exit on all others.
1756
1757 io Continue on any IO error, exit on all others.
1758
1759 verify Continue on verify errors, exit on all others.
1760
1761 all Continue on all errors.
1762
1763 0 Backward-compatible alias for 'none'.
1764
1765 1 Backward-compatible alias for 'all'.
1766
8b28bd41
DM
1767ignore_error=str Sometimes you want to ignore some errors during test
1768 in that case you can specify error list for each error type.
1769 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1770 errors for given error type is separated with ':'. Error
1771 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1772 Example:
1773 ignore_error=EAGAIN,ENOSPC:122
66c098b8
BC
1774 This option will ignore EAGAIN from READ, and ENOSPC and
1775 122(EDQUOT) from WRITE.
8b28bd41
DM
1776
1777error_dump=bool If set dump every error even if it is non fatal, true
1778 by default. If disabled only fatal error will be dumped
66c098b8 1779
6adb38a1
JA
1780cgroup=str Add job to this control group. If it doesn't exist, it will
1781 be created. The system must have a mounted cgroup blkio
1782 mount point for this to work. If your system doesn't have it
1783 mounted, you can do so with:
a696fa2a
JA
1784
1785 # mount -t cgroup -o blkio none /cgroup
1786
a696fa2a
JA
1787cgroup_weight=int Set the weight of the cgroup to this value. See
1788 the documentation that comes with the kernel, allowed values
1789 are in the range of 100..1000.
71bfa161 1790
7de87099
VG
1791cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1792 the job completion. To override this behavior and to leave
1793 cgroups around after the job completion, set cgroup_nodelete=1.
1794 This can be useful if one wants to inspect various cgroup
1795 files after job completion. Default: false
1796
e0b0d892
JA
1797uid=int Instead of running as the invoking user, set the user ID to
1798 this value before the thread/process does any work.
1799
1800gid=int Set group ID, see uid.
1801
9e684a49
DE
1802flow_id=int The ID of the flow. If not specified, it defaults to being a
1803 global flow. See flow.
1804
1805flow=int Weight in token-based flow control. If this value is used, then
1806 there is a 'flow counter' which is used to regulate the
1807 proportion of activity between two or more jobs. fio attempts
1808 to keep this flow counter near zero. The 'flow' parameter
1809 stands for how much should be added or subtracted to the flow
1810 counter on each iteration of the main I/O loop. That is, if
1811 one job has flow=8 and another job has flow=-1, then there
1812 will be a roughly 1:8 ratio in how much one runs vs the other.
1813
1814flow_watermark=int The maximum value that the absolute value of the flow
1815 counter is allowed to reach before the job must wait for a
1816 lower value of the counter.
1817
1818flow_sleep=int The period of time, in microseconds, to wait after the flow
1819 watermark has been exceeded before retrying operations
1820
de890a1e
SL
1821In addition, there are some parameters which are only valid when a specific
1822ioengine is in use. These are used identically to normal parameters, with the
1823caveat that when used on the command line, they must come after the ioengine
1824that defines them is selected.
1825
1826[libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1827 the io_getevents system call to reap newly returned events.
1828 With this flag turned on, the AIO ring will be read directly
1829 from user-space to reap events. The reaping mode is only
1830 enabled when polling for a minimum of 0 events (eg when
1831 iodepth_batch_complete=0).
1832
2cafffbe
JA
1833[psyncv2] hipri Set RWF_HIPRI on IO, indicating to the kernel that
1834 it's of higher priority than normal.
1835
0353050f
JA
1836[cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1837
1838[cpu] cpuchunks=int Split the load into cycles of the given time. In
1839 microseconds.
1840
046395d7
JA
1841[cpu] exit_on_io_done=bool Detect when IO threads are done, then exit.
1842
de890a1e
SL
1843[netsplice] hostname=str
1844[net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1845 If the job is a TCP listener or UDP reader, the hostname is not
b511c9aa
SB
1846 used and must be omitted unless it is a valid UDP multicast
1847 address.
a3f001f5 1848[libhdfs] namenode=str The host name or IP address of a HDFS cluster namenode to contact.
de890a1e
SL
1849
1850[netsplice] port=int
6315af9d
JA
1851[net] port=int The TCP or UDP port to bind to or connect to. If this is used
1852with numjobs to spawn multiple instances of the same job type, then this will
1853be the starting port number since fio will use a range of ports.
a3f001f5 1854[libhdfs] port=int the listening port of the HFDS cluster namenode.
de890a1e 1855
b93b6a2e
SB
1856[netsplice] interface=str
1857[net] interface=str The IP address of the network interface used to send or
1858 receive UDP multicast
1859
d3a623de
SB
1860[netsplice] ttl=int
1861[net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1862 Default: 1
1863
1d360ffb
JA
1864[netsplice] nodelay=bool
1865[net] nodelay=bool Set TCP_NODELAY on TCP connections.
1866
de890a1e
SL
1867[netsplice] protocol=str
1868[netsplice] proto=str
1869[net] protocol=str
1870[net] proto=str The network protocol to use. Accepted values are:
1871
1872 tcp Transmission control protocol
49ccb8c1 1873 tcpv6 Transmission control protocol V6
f5cc3d0e 1874 udp User datagram protocol
49ccb8c1 1875 udpv6 User datagram protocol V6
de890a1e
SL
1876 unix UNIX domain socket
1877
1878 When the protocol is TCP or UDP, the port must also be given,
1879 as well as the hostname if the job is a TCP listener or UDP
1880 reader. For unix sockets, the normal filename option should be
1881 used and the port is invalid.
1882
1883[net] listen For TCP network connections, tell fio to listen for incoming
1884 connections rather than initiating an outgoing connection. The
1885 hostname must be omitted if this option is used.
1008602c 1886
b511c9aa 1887[net] pingpong Normaly a network writer will just continue writing data, and
7aeb1e94
JA
1888 a network reader will just consume packages. If pingpong=1
1889 is set, a writer will send its normal payload to the reader,
1890 then wait for the reader to send the same payload back. This
1891 allows fio to measure network latencies. The submission
1892 and completion latencies then measure local time spent
1893 sending or receiving, and the completion latency measures
1894 how long it took for the other end to receive and send back.
b511c9aa
SB
1895 For UDP multicast traffic pingpong=1 should only be set for a
1896 single reader when multiple readers are listening to the same
1897 address.
7aeb1e94 1898
1008602c
JA
1899[net] window_size Set the desired socket buffer size for the connection.
1900
e5f34d95
JA
1901[net] mss Set the TCP maximum segment size (TCP_MAXSEG).
1902
d54fce84
DM
1903[e4defrag] donorname=str
1904 File will be used as a block donor(swap extents between files)
1905[e4defrag] inplace=int
66c098b8 1906 Configure donor file blocks allocation strategy
d54fce84
DM
1907 0(default): Preallocate donor's file on init
1908 1 : allocate space immidietly inside defragment event,
1909 and free right after event
1910
08a2cbf6
JA
1911[rbd] clustername=str Specifies the name of the Ceph cluster.
1912[rbd] rbdname=str Specifies the name of the RBD.
1913[rbd] pool=str Specifies the naem of the Ceph pool containing RBD.
1914[rbd] clientname=str Specifies the username (without the 'client.' prefix)
1915 used to access the Ceph cluster. If the clustername is
1916 specified, the clientmae shall be the full type.id
1917 string. If no type. prefix is given, fio will add
1918 'client.' by default.
1919
65fa28ca 1920[mtd] skip_bad=bool Skip operations against known bad blocks.
de890a1e 1921
a3f001f5
FB
1922[libhdfs] hdfsdirectory libhdfs will create chunk in this HDFS directory
1923[libhdfs] chunck_size the size of the chunck to use for each file.
1924
de890a1e 1925
71bfa161
JA
19266.0 Interpreting the output
1927---------------------------
1928
1929fio spits out a lot of output. While running, fio will display the
1930status of the jobs created. An example of that would be:
1931
73c8b082 1932Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
71bfa161
JA
1933
1934The characters inside the square brackets denote the current status of
1935each thread. The possible values (in typical life cycle order) are:
1936
1937Idle Run
1938---- ---
1939P Thread setup, but not started.
1940C Thread created.
9c6f6316 1941I Thread initialized, waiting or generating necessary data.
b0f65863 1942 p Thread running pre-reading file(s).
71bfa161
JA
1943 R Running, doing sequential reads.
1944 r Running, doing random reads.
1945 W Running, doing sequential writes.
1946 w Running, doing random writes.
1947 M Running, doing mixed sequential reads/writes.
1948 m Running, doing mixed random reads/writes.
1949 F Running, currently waiting for fsync()
3d434057 1950 f Running, finishing up (writing IO logs, etc)
fc6bd43c 1951 V Running, doing verification of written data.
71bfa161 1952E Thread exited, not reaped by main thread yet.
4f7e57a4
JA
1953_ Thread reaped, or
1954X Thread reaped, exited with an error.
a5e371a6 1955K Thread reaped, exited due to signal.
71bfa161 1956
3e2e48a7
JA
1957Fio will condense the thread string as not to take up more space on the
1958command line as is needed. For instance, if you have 10 readers and 10
1959writers running, the output would look like this:
1960
1961Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s]
1962
1963Fio will still maintain the ordering, though. So the above means that jobs
19641..10 are readers, and 11..20 are writers.
1965
71bfa161 1966The other values are fairly self explanatory - number of threads
c9f60304
JA
1967currently running and doing io, rate of io since last check (read speed
1968listed first, then write speed), and the estimated completion percentage
1969and time for the running group. It's impossible to estimate runtime of
4f7e57a4
JA
1970the following groups (if any). Note that the string is displayed in order,
1971so it's possible to tell which of the jobs are currently doing what. The
1972first character is the first job defined in the job file, and so forth.
71bfa161
JA
1973
1974When fio is done (or interrupted by ctrl-c), it will show the data for
1975each thread, group of threads, and disks in that order. For each data
1976direction, the output looks like:
1977
1978Client1 (g=0): err= 0:
35649e58 1979 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
6104ddb6
JA
1980 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1981 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
b22989b9 1982 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
e7823a94 1983 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
71619dc2 1984 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
1985 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1986 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
30061b97 1987 issued r/w: total=0/32768, short=0/0
8abdce66
JA
1988 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1989 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
71bfa161
JA
1990
1991The client number is printed, along with the group id and error of that
1992thread. Below is the io statistics, here for writes. In the order listed,
1993they denote:
1994
1995io= Number of megabytes io performed
1996bw= Average bandwidth rate
35649e58 1997iops= Average IOs performed per second
71bfa161 1998runt= The runtime of that thread
72fbda2a 1999 slat= Submission latency (avg being the average, stdev being the
71bfa161
JA
2000 standard deviation). This is the time it took to submit
2001 the io. For sync io, the slat is really the completion
8a35c71e 2002 latency, since queue/complete is one operation there. This
bf9a3edb 2003 value can be in milliseconds or microseconds, fio will choose
8a35c71e 2004 the most appropriate base and print that. In the example
0d237712
LAG
2005 above, milliseconds is the best scale. Note: in --minimal mode
2006 latencies are always expressed in microseconds.
71bfa161
JA
2007 clat= Completion latency. Same names as slat, this denotes the
2008 time from submission to completion of the io pieces. For
2009 sync io, clat will usually be equal (or very close) to 0,
2010 as the time from submit to complete is basically just
2011 CPU time (io has already been done, see slat explanation).
2012 bw= Bandwidth. Same names as the xlat stats, but also includes
2013 an approximate percentage of total aggregate bandwidth
2014 this thread received in this group. This last value is
2015 only really useful if the threads in this group are on the
2016 same disk, since they are then competing for disk access.
2017cpu= CPU usage. User and system time, along with the number
e7823a94
JA
2018 of context switches this thread went through, usage of
2019 system and user time, and finally the number of major
23a8e176
JA
2020 and minor page faults. The CPU utilization numbers are
2021 averages for the jobs in that reporting group, while the
2022 context and fault counters are summed.
71619dc2
JA
2023IO depths= The distribution of io depths over the job life time. The
2024 numbers are divided into powers of 2, so for example the
2025 16= entries includes depths up to that value but higher
2026 than the previous entry. In other words, it covers the
2027 range from 16 to 31.
838bc709
JA
2028IO submit= How many pieces of IO were submitting in a single submit
2029 call. Each entry denotes that amount and below, until
2030 the previous entry - eg, 8=100% mean that we submitted
2031 anywhere in between 5-8 ios per submit call.
2032IO complete= Like the above submit number, but for completions instead.
30061b97
JA
2033IO issued= The number of read/write requests issued, and how many
2034 of them were short.
ec118304
JA
2035IO latencies= The distribution of IO completion latencies. This is the
2036 time from when IO leaves fio and when it gets completed.
2037 The numbers follow the same pattern as the IO depths,
2038 meaning that 2=1.6% means that 1.6% of the IO completed
8abdce66
JA
2039 within 2 msecs, 20=12.8% means that 12.8% of the IO
2040 took more than 10 msecs, but less than (or equal to) 20 msecs.
71bfa161
JA
2041
2042After each client has been listed, the group statistics are printed. They
2043will look like this:
2044
2045Run status group 0 (all jobs):
b22989b9
JA
2046 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
2047 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
71bfa161
JA
2048
2049For each data direction, it prints:
2050
2051io= Number of megabytes io performed.
2052aggrb= Aggregate bandwidth of threads in this group.
2053minb= The minimum average bandwidth a thread saw.
2054maxb= The maximum average bandwidth a thread saw.
2055mint= The smallest runtime of the threads in that group.
2056maxt= The longest runtime of the threads in that group.
2057
2058And finally, the disk statistics are printed. They will look like this:
2059
2060Disk stats (read/write):
2061 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
2062
2063Each value is printed for both reads and writes, with reads first. The
2064numbers denote:
2065
2066ios= Number of ios performed by all groups.
2067merge= Number of merges io the io scheduler.
2068ticks= Number of ticks we kept the disk busy.
2069io_queue= Total time spent in the disk queue.
2070util= The disk utilization. A value of 100% means we kept the disk
2071 busy constantly, 50% would be a disk idling half of the time.
2072
8423bd11
JA
2073It is also possible to get fio to dump the current output while it is
2074running, without terminating the job. To do that, send fio the USR1 signal.
06464907
JA
2075You can also get regularly timed dumps by using the --status-interval
2076parameter, or by creating a file in /tmp named fio-dump-status. If fio
2077sees this file, it will unlink it and dump the current output status.
8423bd11 2078
71bfa161
JA
2079
20807.0 Terse output
2081----------------
2082
2083For scripted usage where you typically want to generate tables or graphs
6af019c9 2084of the results, fio can output the results in a semicolon separated format.
71bfa161
JA
2085The format is one long line of values, such as:
2086
562c2d2f
DN
20872;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%
2088A description of this job goes here.
2089
2090The job description (if provided) follows on a second line.
71bfa161 2091
525c2bfa
JA
2092To enable terse output, use the --minimal command line option. The first
2093value is the version of the terse output format. If the output has to
2094be changed for some reason, this number will be incremented by 1 to
2095signify that change.
6820cb3b 2096
71bfa161
JA
2097Split up, the format is as follows:
2098
5e726d0a 2099 terse version, fio version, jobname, groupid, error
71bfa161 2100 READ status:
312b4af2 2101 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
d86ae56c
CW
2102 Submission latency: min, max, mean, stdev (usec)
2103 Completion latency: min, max, mean, stdev (usec)
1db92cb6 2104 Completion latency percentiles: 20 fields (see below)
d86ae56c
CW
2105 Total latency: min, max, mean, stdev (usec)
2106 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
71bfa161 2107 WRITE status:
312b4af2 2108 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
d86ae56c
CW
2109 Submission latency: min, max, mean, stdev (usec)
2110 Completion latency: min, max, mean, stdev(usec)
1db92cb6 2111 Completion latency percentiles: 20 fields (see below)
d86ae56c
CW
2112 Total latency: min, max, mean, stdev (usec)
2113 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
046ee302 2114 CPU usage: user, system, context switches, major faults, minor faults
2270890c 2115 IO depths: <=1, 2, 4, 8, 16, 32, >=64
562c2d2f
DN
2116 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
2117 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
f2f788dd
JA
2118 Disk utilization: Disk name, Read ios, write ios,
2119 Read merges, write merges,
2120 Read ticks, write ticks,
3d7cd9b4 2121 Time spent in queue, disk utilization percentage
de8f6de9 2122 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
66c098b8 2123
de8f6de9 2124 Additional Info (dependent on description being set): Text description
25c8b9d7 2125
1db92cb6
JA
2126Completion latency percentiles can be a grouping of up to 20 sets, so
2127for the terse output fio writes all of them. Each field will look like this:
2128
2129 1.00%=6112
2130
2131which is the Xth percentile, and the usec latency associated with it.
2132
f2f788dd
JA
2133For disk utilization, all disks used by fio are shown. So for each disk
2134there will be a disk utilization section.
2135
25c8b9d7
PD
2136
21378.0 Trace file format
2138---------------------
66c098b8 2139There are two trace file format that you can encounter. The older (v1) format
25c8b9d7
PD
2140is unsupported since version 1.20-rc3 (March 2008). It will still be described
2141below in case that you get an old trace and want to understand it.
2142
2143In any case the trace is a simple text file with a single action per line.
2144
2145
21468.1 Trace file format v1
2147------------------------
2148Each line represents a single io action in the following format:
2149
2150rw, offset, length
2151
2152where rw=0/1 for read/write, and the offset and length entries being in bytes.
2153
2154This format is not supported in Fio versions => 1.20-rc3.
2155
2156
21578.2 Trace file format v2
2158------------------------
2159The second version of the trace file format was added in Fio version 1.17.
2160It allows to access more then one file per trace and has a bigger set of
2161possible file actions.
2162
2163The first line of the trace file has to be:
2164
2165fio version 2 iolog
2166
2167Following this can be lines in two different formats, which are described below.
2168
2169The file management format:
2170
2171filename action
2172
2173The filename is given as an absolute path. The action can be one of these:
2174
2175add Add the given filename to the trace
66c098b8 2176open Open the file with the given filename. The filename has to have
25c8b9d7
PD
2177 been added with the add action before.
2178close Close the file with the given filename. The file has to have been
2179 opened before.
2180
2181
2182The file io action format:
2183
2184filename action offset length
2185
2186The filename is given as an absolute path, and has to have been added and opened
66c098b8 2187before it can be used with this format. The offset and length are given in
25c8b9d7
PD
2188bytes. The action can be one of these:
2189
2190wait Wait for 'offset' microseconds. Everything below 100 is discarded.
5c7808fe 2191 The time is relative to the previous wait statement.
25c8b9d7
PD
2192read Read 'length' bytes beginning from 'offset'
2193write Write 'length' bytes beginning from 'offset'
2194sync fsync() the file
2195datasync fdatasync() the file
2196trim trim the given file from the given 'offset' for 'length' bytes
f2a2ce0e
HL
2197
2198
21999.0 CPU idleness profiling
06464907 2200--------------------------
f2a2ce0e
HL
2201In some cases, we want to understand CPU overhead in a test. For example,
2202we test patches for the specific goodness of whether they reduce CPU usage.
2203fio implements a balloon approach to create a thread per CPU that runs at
2204idle priority, meaning that it only runs when nobody else needs the cpu.
2205By measuring the amount of work completed by the thread, idleness of each
2206CPU can be derived accordingly.
2207
2208An unit work is defined as touching a full page of unsigned characters. Mean
2209and standard deviation of time to complete an unit work is reported in "unit
2210work" section. Options can be chosen to report detailed percpu idleness or
2211overall system idleness by aggregating percpu stats.
99b9a85a
JA
2212
2213
221410.0 Verification and triggers
2215------------------------------
2216Fio is usually run in one of two ways, when data verification is done. The
2217first is a normal write job of some sort with verify enabled. When the
2218write phase has completed, fio switches to reads and verifies everything
2219it wrote. The second model is running just the write phase, and then later
2220on running the same job (but with reads instead of writes) to repeat the
2221same IO patterns and verify the contents. Both of these methods depend
2222on the write phase being completed, as fio otherwise has no idea how much
2223data was written.
2224
2225With verification triggers, fio supports dumping the current write state
2226to local files. Then a subsequent read verify workload can load this state
2227and know exactly where to stop. This is useful for testing cases where
2228power is cut to a server in a managed fashion, for instance.
2229
2230A verification trigger consists of two things:
2231
22321) Storing the write state of each job
22332) Executing a trigger command
2234
2235The write state is relatively small, on the order of hundreds of bytes
2236to single kilobytes. It contains information on the number of completions
2237done, the last X completions, etc.
2238
2239A trigger is invoked either through creation ('touch') of a specified
2240file in the system, or through a timeout setting. If fio is run with
2241--trigger-file=/tmp/trigger-file, then it will continually check for
2242the existence of /tmp/trigger-file. When it sees this file, it will
2243fire off the trigger (thus saving state, and executing the trigger
2244command).
2245
2246For client/server runs, there's both a local and remote trigger. If
2247fio is running as a server backend, it will send the job states back
2248to the client for safe storage, then execute the remote trigger, if
2249specified. If a local trigger is specified, the server will still send
2250back the write state, but the client will then execute the trigger.
2251
225210.1 Verification trigger example
2253---------------------------------
2254Lets say we want to run a powercut test on the remote machine 'server'.
2255Our write workload is in write-test.fio. We want to cut power to 'server'
2256at some point during the run, and we'll run this test from the safety
2257or our local machine, 'localbox'. On the server, we'll start the fio
2258backend normally:
2259
2260server# fio --server
2261
2262and on the client, we'll fire off the workload:
2263
2264localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
2265
2266We set /tmp/my-trigger as the trigger file, and we tell fio to execute
2267
2268echo b > /proc/sysrq-trigger
2269
2270on the server once it has received the trigger and sent us the write
2271state. This will work, but it's not _really_ cutting power to the server,
2272it's merely abruptly rebooting it. If we have a remote way of cutting
2273power to the server through IPMI or similar, we could do that through
2274a local trigger command instead. Lets assume we have a script that does
2275IPMI reboot of a given hostname, ipmi-reboot. On localbox, we could
2276then have run fio with a local trigger instead:
2277
2278localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
2279
2280For this case, fio would wait for the server to send us the write state,
2281then execute 'ipmi-reboot server' when that happened.
2282
29dbd1e5 228310.2 Loading verify state
99b9a85a
JA
2284-------------------------
2285To load store write state, read verification job file must contain
2286the verify_state_load option. If that is set, fio will load the previously
2287stored state. For a local fio run this is done by loading the files directly,
2288and on a client/server run, the server backend will ask the client to send
2289the files over and load them from there.
a3ae5b05
JA
2290
2291
229211.0 Log File Formats
2293---------------------
2294
2295Fio supports a variety of log file formats, for logging latencies, bandwidth,
2296and IOPS. The logs share a common format, which looks like this:
2297
2298time (msec), value, data direction, offset
2299
2300Time for the log entry is always in milliseconds. The value logged depends
2301on the type of log, it will be one of the following:
2302
2303 Latency log Value is latency in usecs
2304 Bandwidth log Value is in KB/sec
2305 IOPS log Value is IOPS
2306
2307Data direction is one of the following:
2308
2309 0 IO is a READ
2310 1 IO is a WRITE
2311 2 IO is a TRIM
2312
2313The offset is the offset, in bytes, from the start of the file, for that
2314particular IO. The logging of the offset can be toggled with 'log_offset'.
2315
2316If windowed logging is enabled though 'log_avg_msec', then fio doesn't log
2317individual IOs. Instead of logs the average values over the specified
2318period of time. Since 'data direction' and 'offset' are per-IO values,
2319they aren't applicable if windowed logging is enabled. If windowed logging
2320is enabled and 'log_max_value' is set, then fio logs maximum values in
2321that window instead of averages.
2322