Clear f on error get_next_file_rr()
[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
11
12
131.0 Overview and history
14------------------------
15fio was originally written to save me the hassle of writing special test
16case programs when I wanted to test a specific workload, either for
17performance reasons or to find/reproduce a bug. The process of writing
18such a test app can be tiresome, especially if you have to do it often.
19Hence I needed a tool that would be able to simulate a given io workload
20without resorting to writing a tailored test case again and again.
21
22A test work load is difficult to define, though. There can be any number
23of processes or threads involved, and they can each be using their own
24way of generating io. You could have someone dirtying large amounts of
25memory in an memory mapped file, or maybe several threads issuing
26reads using asynchronous io. fio needed to be flexible enough to
27simulate both of these cases, and many more.
28
292.0 How fio works
30-----------------
31The first step in getting fio to simulate a desired io workload, is
32writing a job file describing that specific setup. A job file may contain
33any number of threads and/or files - the typical contents of the job file
34is a global section defining shared parameters, and one or more job
35sections describing the jobs involved. When run, fio parses this file
36and sets everything up as described. If we break down a job from top to
37bottom, it contains the following basic parameters:
38
39 IO type Defines the io pattern issued to the file(s).
40 We may only be reading sequentially from this
41 file(s), or we may be writing randomly. Or even
42 mixing reads and writes, sequentially or randomly.
43
44 Block size In how large chunks are we issuing io? This may be
45 a single value, or it may describe a range of
46 block sizes.
47
48 IO size How much data are we going to be reading/writing.
49
50 IO engine How do we issue io? We could be memory mapping the
51 file, we could be using regular read/write, we
d0ff85df 52 could be using splice, async io, syslet, or even
71bfa161
JA
53 SG (SCSI generic sg).
54
6c219763 55 IO depth If the io engine is async, how large a queuing
71bfa161
JA
56 depth do we want to maintain?
57
58 IO type Should we be doing buffered io, or direct/raw io?
59
60 Num files How many files are we spreading the workload over.
61
62 Num threads How many threads or processes should we spread
63 this workload over.
64
65The above are the basic parameters defined for a workload, in addition
66there's a multitude of parameters that modify other aspects of how this
67job behaves.
68
69
703.0 Running fio
71---------------
72See the README file for command line parameters, there are only a few
73of them.
74
75Running fio is normally the easiest part - you just give it the job file
76(or job files) as parameters:
77
78$ fio job_file
79
80and it will start doing what the job_file tells it to do. You can give
81more than one job file on the command line, fio will serialize the running
82of those files. Internally that is the same as using the 'stonewall'
83parameter described the the parameter section.
84
b4692828
JA
85If the job file contains only one job, you may as well just give the
86parameters on the command line. The command line parameters are identical
87to the job parameters, with a few extra that control global parameters
88(see README). For example, for the job file parameter iodepth=2, the
c2b1e753
JA
89mirror command line option would be --iodepth 2 or --iodepth=2. You can
90also use the command line for giving more than one job entry. For each
91--name option that fio sees, it will start a new job with that name.
92Command line entries following a --name entry will apply to that job,
93until there are no more entries or a new --name entry is seen. This is
94similar to the job file options, where each option applies to the current
95job until a new [] job entry is seen.
b4692828 96
71bfa161
JA
97fio does not need to run as root, except if the files or devices specified
98in the job section requires that. Some other options may also be restricted,
6c219763 99such as memory locking, io scheduler switching, and decreasing the nice value.
71bfa161
JA
100
101
1024.0 Job file format
103-------------------
104As previously described, fio accepts one or more job files describing
105what it is supposed to do. The job file format is the classic ini file,
106where the names enclosed in [] brackets define the job name. You are free
107to use any ascii name you want, except 'global' which has special meaning.
108A global section sets defaults for the jobs described in that file. A job
109may override a global section parameter, and a job file may even have
110several global sections if so desired. A job is only affected by a global
65db0851
JA
111section residing above it. If the first character in a line is a ';' or a
112'#', the entire line is discarded as a comment.
71bfa161 113
3c54bc46 114So let's look at a really simple job file that defines two processes, each
71bfa161
JA
115randomly reading from a 128MiB file.
116
117; -- start job file --
118[global]
119rw=randread
120size=128m
121
122[job1]
123
124[job2]
125
126; -- end job file --
127
128As you can see, the job file sections themselves are empty as all the
129described parameters are shared. As no filename= option is given, fio
c2b1e753
JA
130makes up a filename for each of the jobs as it sees fit. On the command
131line, this job would look as follows:
132
133$ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
134
71bfa161 135
3c54bc46 136Let's look at an example that has a number of processes writing randomly
71bfa161
JA
137to files.
138
139; -- start job file --
140[random-writers]
141ioengine=libaio
142iodepth=4
143rw=randwrite
144bs=32k
145direct=0
146size=64m
147numjobs=4
148
149; -- end job file --
150
151Here we have no global section, as we only have one job defined anyway.
152We want to use async io here, with a depth of 4 for each file. We also
153increased the buffer size used to 32KiB and define numjobs to 4 to
154fork 4 identical jobs. The result is 4 processes each randomly writing
b4692828
JA
155to their own 64MiB file. Instead of using the above job file, you could
156have given the parameters on the command line. For this case, you would
157specify:
158
159$ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
71bfa161 160
3c54bc46
AC
161fio also supports environment variable expansion in job files. Any
162substring of the form "${VARNAME}" as part of an option value (in other
163words, on the right of the `='), will be expanded to the value of the
164environment variable called VARNAME. If no such environment variable
165is defined, or VARNAME is the empty string, the empty string will be
166substituted.
167
168As an example, let's look at a sample fio invocation and job file:
169
170$ SIZE=64m NUMJOBS=4 fio jobfile.fio
171
172; -- start job file --
173[random-writers]
174rw=randwrite
175size=${SIZE}
176numjobs=${NUMJOBS}
177; -- end job file --
178
179This will expand to the following equivalent job file at runtime:
180
181; -- start job file --
182[random-writers]
183rw=randwrite
184size=64m
185numjobs=4
186; -- end job file --
187
71bfa161
JA
188fio ships with a few example job files, you can also look there for
189inspiration.
190
191
1925.0 Detailed list of parameters
193-------------------------------
194
195This section describes in details each parameter associated with a job.
196Some parameters take an option of a given type, such as an integer or
197a string. The following types are used:
198
199str String. This is a sequence of alpha characters.
e417fd66
JA
200time Integer with possible time postfix. In seconds unless otherwise
201 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
202 minutes, and hours.
f7fa2653 203int SI integer. A whole number value, which may contain a postfix
71bfa161 204 describing the base of the number. Accepted postfixes are k/m/g,
6c219763 205 meaning kilo, mega, and giga. So if you want to specify 4096,
71bfa161
JA
206 you could either write out '4096' or just give 4k. The postfixes
207 signify base 2 values, so 1024 is 1k and 1024k is 1m and so on.
43159d18 208 If the option accepts an upper and lower range, use a colon ':'
ef67a8a3
JA
209 or minus '-' to separate such values. May also include a prefix
210 to indicate numbers base. If 0x is used, the number is assumed to
211 be hexadecimal. See irange.
71bfa161
JA
212bool Boolean. Usually parsed as an integer, however only defined for
213 true and false (1 and 0).
214irange Integer range with postfix. Allows value range to be given, such
bf9a3edb 215 as 1024-4096. A colon may also be used as the separator, eg
0c9baf91
JA
216 1k:4k. If the option allows two sets of ranges, they can be
217 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
f7fa2653 218 int.
71bfa161
JA
219
220With the above in mind, here follows the complete list of fio job
221parameters.
222
223name=str ASCII name of the job. This may be used to override the
224 name printed by fio for this job. Otherwise the job
c2b1e753 225 name is used. On the command line this parameter has the
6c219763 226 special purpose of also signaling the start of a new
c2b1e753 227 job.
71bfa161 228
61697c37
JA
229description=str Text description of the job. Doesn't do anything except
230 dump this text description when this job is run. It's
231 not parsed.
232
3776041e 233directory=str Prefix filenames with this directory. Used to place files
71bfa161
JA
234 in a different location than "./".
235
236filename=str Fio normally makes up a filename based on the job name,
237 thread number, and file number. If you want to share
238 files between threads in a job or several jobs, specify
ed92ac0c 239 a filename for each of them to override the default. If
414c2a3e
JA
240 the ioengine used is 'net', the filename is the host, port,
241 and protocol to use in the format of =host/port/protocol.
242 See ioengine=net for more. If the ioengine is file based, you
243 can specify a number of files by separating the names with a
244 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
245 as the two working files, you would use
246 filename=/dev/sda:/dev/sdb. '-' is a reserved name, meaning
247 stdin or stdout. Which of the two depends on the read/write
248 direction set.
71bfa161 249
bbf6b540
JA
250opendir=str Tell fio to recursively add any file it can find in this
251 directory and down the file system tree.
252
3776041e 253lockfile=str Fio defaults to not locking any files before it does
4d4e80f2
JA
254 IO to them. If a file or file descriptor is shared, fio
255 can serialize IO to that file to make the end result
256 consistent. This is usual for emulating real workloads that
257 share files. The lock modes are:
258
259 none No locking. The default.
260 exclusive Only one thread/process may do IO,
261 excluding all others.
262 readwrite Read-write locking on the file. Many
263 readers may access the file at the
264 same time, but writes get exclusive
265 access.
266
267 The option may be post-fixed with a lock batch number. If
268 set, then each thread/process may do that amount of IOs to
bf9a3edb 269 the file before giving up the lock. Since lock acquisition is
4d4e80f2 270 expensive, batching the lock/unlocks will speed up IO.
29c1349f 271
d3aad8f2 272readwrite=str
71bfa161
JA
273rw=str Type of io pattern. Accepted values are:
274
275 read Sequential reads
276 write Sequential writes
277 randwrite Random writes
278 randread Random reads
279 rw Sequential mixed reads and writes
280 randrw Random mixed reads and writes
281
282 For the mixed io types, the default is to split them 50/50.
283 For certain types of io the result may still be skewed a bit,
211097b2
JA
284 since the speed may be different. It is possible to specify
285 a number of IO's to do before getting a new offset - this
286 is only useful for random IO, where fio would normally
287 generate a new random offset for every IO. If you append
288 eg 8 to randread, you would get a new random offset for
289 every 8 IO's. The result would be a seek for only every 8
290 IO's, instead of for every IO. Use rw=randread:8 to specify
291 that.
71bfa161 292
ee738499
JA
293randrepeat=bool For random IO workloads, seed the generator in a predictable
294 way so that results are repeatable across repetitions.
295
d2f3ac35
JA
296fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
297 on what IO patterns it is likely to issue. Sometimes you
298 want to test specific IO patterns without telling the
299 kernel about it, in which case you can disable this option.
300 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
301 IO and POSIX_FADV_RANDOM for random IO.
302
f7fa2653 303size=int The total size of file io for this job. Fio will run until
7616cafe
JA
304 this many bytes has been transferred, unless runtime is
305 limited by other options (such as 'runtime', for instance).
3776041e 306 Unless specific nrfiles and filesize options are given,
7616cafe
JA
307 fio will divide this size between the available files
308 specified by the job.
71bfa161 309
f7fa2653 310filesize=int Individual file sizes. May be a range, in which case fio
9c60ce64
JA
311 will select sizes for files at random within the given range
312 and limited to 'size' in total (if that is given). If not
313 given, each created file is the same size.
314
aa31f1f1
SL
315fill_device=bool Sets size to something really large and waits for ENOSPC (no
316 space left on device) as the terminating condition. Only makes
317 sense with sequential write.
318
f7fa2653
JA
319blocksize=int
320bs=int The block size used for the io units. Defaults to 4k. Values
321 can be given for both read and writes. If a single int is
322 given, it will apply to both. If a second int is specified
f90eff5a
JA
323 after a comma, it will apply to writes only. In other words,
324 the format is either bs=read_and_write or bs=read,write.
325 bs=4k,8k will thus use 4k blocks for reads, and 8k blocks
787f7e95
JA
326 for writes. If you only wish to set the write size, you
327 can do so by passing an empty read size - bs=,8k will set
328 8k for writes and leave the read default value.
a00735e6 329
2b7a01d0
JA
330blockalign=int
331ba=int At what boundary to align random IO offsets. Defaults to
332 the same as 'blocksize' the minimum blocksize given.
333 Minimum alignment is typically 512b for using direct IO,
334 though it usually depends on the hardware block size. This
335 option is mutually exclusive with using a random map for
336 files, so it will turn off that option.
337
d3aad8f2 338blocksize_range=irange
71bfa161
JA
339bsrange=irange Instead of giving a single block size, specify a range
340 and fio will mix the issued io block sizes. The issued
341 io unit will always be a multiple of the minimum value
f90eff5a
JA
342 given (also see bs_unaligned). Applies to both reads and
343 writes, however a second range can be given after a comma.
344 See bs=.
a00735e6 345
564ca972
JA
346bssplit=str Sometimes you want even finer grained control of the
347 block sizes issued, not just an even split between them.
348 This option allows you to weight various block sizes,
349 so that you are able to define a specific amount of
350 block sizes issued. The format for this option is:
351
352 bssplit=blocksize/percentage:blocksize/percentage
353
354 for as many block sizes as needed. So if you want to define
355 a workload that has 50% 64k blocks, 10% 4k blocks, and
356 40% 32k blocks, you would write:
357
358 bssplit=4k/10:64k/50:32k/40
359
360 Ordering does not matter. If the percentage is left blank,
361 fio will fill in the remaining values evenly. So a bssplit
362 option like this one:
363
364 bssplit=4k/50:1k/:32k/
365
366 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
367 always add up to 100, if bssplit is given a range that adds
368 up to more, it will error out.
369
720e84ad
JA
370 bssplit also supports giving separate splits to reads and
371 writes. The format is identical to what bs= accepts. You
372 have to separate the read and write parts with a comma. So
373 if you want a workload that has 50% 2k reads and 50% 4k reads,
374 while having 90% 4k writes and 10% 8k writes, you would
375 specify:
376
377 bssplit=2k/50:4k/50,4k/90,8k/10
378
d3aad8f2 379blocksize_unaligned
690adba3
JA
380bs_unaligned If this option is given, any byte size value within bsrange
381 may be used as a block range. This typically wont work with
382 direct IO, as that normally requires sector alignment.
71bfa161 383
e9459e5a
JA
384zero_buffers If this option is given, fio will init the IO buffers to
385 all zeroes. The default is to fill them with random data.
386
5973cafb
JA
387refill_buffers If this option is given, fio will refill the IO buffers
388 on every submit. The default is to only fill it at init
389 time and reuse that data. Only makes sense if zero_buffers
41ccd845
JA
390 isn't specified, naturally. If data verification is enabled,
391 refill_buffers is also automatically enabled.
5973cafb 392
71bfa161
JA
393nrfiles=int Number of files to use for this job. Defaults to 1.
394
390b1537
JA
395openfiles=int Number of files to keep open at the same time. Defaults to
396 the same as nrfiles, can be set smaller to limit the number
397 simultaneous opens.
398
5af1c6f3
JA
399file_service_type=str Defines how fio decides which file from a job to
400 service next. The following types are defined:
401
402 random Just choose a file at random.
403
404 roundrobin Round robin over open files. This
405 is the default.
406
a086c257
JA
407 sequential Finish one file before moving on to
408 the next. Multiple files can still be
409 open depending on 'openfiles'.
410
1907dbc6
JA
411 The string can have a number appended, indicating how
412 often to switch to a new file. So if option random:4 is
413 given, fio will switch to a new random file after 4 ios
414 have been issued.
415
71bfa161
JA
416ioengine=str Defines how the job issues io to the file. The following
417 types are defined:
418
419 sync Basic read(2) or write(2) io. lseek(2) is
420 used to position the io location.
421
a31041ea 422 psync Basic pread(2) or pwrite(2) io.
423
e05af9e5 424 vsync Basic readv(2) or writev(2) IO.
1d2af02a 425
15d182aa
JA
426 libaio Linux native asynchronous io. Note that Linux
427 may only support queued behaviour with
428 non-buffered IO (set direct=1 or buffered=0).
71bfa161
JA
429
430 posixaio glibc posix asynchronous io.
431
417f0068
JA
432 solarisaio Solaris native asynchronous io.
433
71bfa161
JA
434 mmap File is memory mapped and data copied
435 to/from using memcpy(3).
436
437 splice splice(2) is used to transfer the data and
438 vmsplice(2) to transfer data from user
439 space to the kernel.
440
d0ff85df
JA
441 syslet-rw Use the syslet system calls to make
442 regular read/write async.
443
71bfa161 444 sg SCSI generic sg v3 io. May either be
6c219763 445 synchronous using the SG_IO ioctl, or if
71bfa161
JA
446 the target is an sg character device
447 we use read(2) and write(2) for asynchronous
448 io.
449
a94ea28b
JA
450 null Doesn't transfer any data, just pretends
451 to. This is mainly used to exercise fio
452 itself and for debugging/testing purposes.
453
ed92ac0c
JA
454 net Transfer over the network to given host:port.
455 'filename' must be set appropriately to
414c2a3e 456 filename=host/port/protocol regardless of send
ed92ac0c 457 or receive, if the latter only the port
414c2a3e
JA
458 argument is used. 'host' may be an IP address
459 or hostname, port is the port number to be used,
460 and protocol may be 'udp' or 'tcp'. If no
461 protocol is given, TCP is used.
ed92ac0c 462
9cce02e8
JA
463 netsplice Like net, but uses splice/vmsplice to
464 map data and send/receive.
465
53aec0a4 466 cpuio Doesn't transfer any data, but burns CPU
ba0fbe10
JA
467 cycles according to the cpuload= and
468 cpucycle= options. Setting cpuload=85
469 will cause that job to do nothing but burn
36ecec83
GP
470 85% of the CPU. In case of SMP machines,
471 use numjobs=<no_of_cpu> to get desired CPU
472 usage, as the cpuload only loads a single
473 CPU at the desired rate.
ba0fbe10 474
e9a1806f
JA
475 guasi The GUASI IO engine is the Generic Userspace
476 Asyncronous Syscall Interface approach
477 to async IO. See
478
479 http://www.xmailserver.org/guasi-lib.html
480
481 for more info on GUASI.
482
8a7bd877
JA
483 external Prefix to specify loading an external
484 IO engine object file. Append the engine
485 filename, eg ioengine=external:/tmp/foo.o
486 to load ioengine foo.o in /tmp.
487
71bfa161
JA
488iodepth=int This defines how many io units to keep in flight against
489 the file. The default is 1 for each file defined in this
490 job, can be overridden with a larger value for higher
491 concurrency.
492
4950421a 493iodepth_batch_submit=int
cb5ab512 494iodepth_batch=int This defines how many pieces of IO to submit at once.
89e820f6
JA
495 It defaults to 1 which means that we submit each IO
496 as soon as it is available, but can be raised to submit
497 bigger batches of IO at the time.
cb5ab512 498
4950421a
JA
499iodepth_batch_complete=int This defines how many pieces of IO to retrieve
500 at once. It defaults to 1 which means that we'll ask
501 for a minimum of 1 IO in the retrieval process from
502 the kernel. The IO retrieval will go on until we
503 hit the limit set by iodepth_low. If this variable is
504 set to 0, then fio will always check for completed
505 events before queuing more IO. This helps reduce
506 IO latency, at the cost of more retrieval system calls.
507
e916b390
JA
508iodepth_low=int The low water mark indicating when to start filling
509 the queue again. Defaults to the same as iodepth, meaning
510 that fio will attempt to keep the queue full at all times.
511 If iodepth is set to eg 16 and iodepth_low is set to 4, then
512 after fio has filled the queue of 16 requests, it will let
513 the depth drain down to 4 before starting to fill it again.
514
71bfa161 515direct=bool If value is true, use non-buffered io. This is usually
76a43db4
JA
516 O_DIRECT.
517
518buffered=bool If value is true, use buffered io. This is the opposite
519 of the 'direct' option. Defaults to true.
71bfa161 520
f7fa2653 521offset=int Start io at the given offset in the file. The data before
71bfa161
JA
522 the given offset will not be touched. This effectively
523 caps the file size at real_size - offset.
524
525fsync=int If writing to a file, issue a sync of the dirty data
526 for every number of blocks given. For example, if you give
527 32 as a parameter, fio will sync the file for every 32
528 writes issued. If fio is using non-buffered io, we may
529 not sync the file. The exception is the sg io engine, which
6c219763 530 synchronizes the disk cache anyway.
71bfa161 531
5036fc1e
JA
532overwrite=bool If true, writes to a file will always overwrite existing
533 data. If the file doesn't already exist, it will be
534 created before the write phase begins. If the file exists
535 and is large enough for the specified write phase, nothing
536 will be done.
71bfa161
JA
537
538end_fsync=bool If true, fsync file contents when the job exits.
539
ebb1415f
JA
540fsync_on_close=bool If true, fio will fsync() a dirty file on close.
541 This differs from end_fsync in that it will happen on every
542 file close, not just at the end of the job.
543
71bfa161
JA
544rwmixread=int How large a percentage of the mix should be reads.
545
546rwmixwrite=int How large a percentage of the mix should be writes. If both
547 rwmixread and rwmixwrite is given and the values do not add
548 up to 100%, the latter of the two will be used to override
549 the first.
550
bb8895e0
JA
551norandommap Normally fio will cover every block of the file when doing
552 random IO. If this option is given, fio will just get a
553 new random offset without looking at past io history. This
554 means that some blocks may not be read or written, and that
555 some blocks may be read/written more than once. This option
8347239a
JA
556 is mutually exclusive with verify= if and only if multiple
557 blocksizes (via bsrange=) are used, since fio only tracks
558 complete rewrites of blocks.
bb8895e0 559
2b386d25
JA
560softrandommap See norandommap. If fio runs with the random block map enabled
561 and it fails to allocate the map, if this option is set it
562 will continue without a random block map. As coverage will
563 not be as complete as with random maps, this option is
564 disabled by default.
565
71bfa161
JA
566nice=int Run the job with the given nice value. See man nice(2).
567
568prio=int Set the io priority value of this job. Linux limits us to
569 a positive value between 0 and 7, with 0 being the highest.
570 See man ionice(1).
571
572prioclass=int Set the io priority class. See man ionice(1).
573
574thinktime=int Stall the job x microseconds after an io has completed before
575 issuing the next. May be used to simulate processing being
48097d5c
JA
576 done by an application. See thinktime_blocks and
577 thinktime_spin.
578
579thinktime_spin=int
580 Only valid if thinktime is set - pretend to spend CPU time
581 doing something with the data received, before falling back
582 to sleeping for the rest of the period specified by
583 thinktime.
9c1f7434
JA
584
585thinktime_blocks
586 Only valid if thinktime is set - control how many blocks
587 to issue, before waiting 'thinktime' usecs. If not set,
588 defaults to 1 which will make fio wait 'thinktime' usecs
589 after every block.
71bfa161
JA
590
591rate=int Cap the bandwidth used by this job to this number of KiB/sec.
592
593ratemin=int Tell fio to do whatever it can to maintain at least this
4e991c23
JA
594 bandwidth. Failing to meet this requirement, will cause
595 the job to exit.
596
597rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
598 as rate, just specified independently of bandwidth. If the
599 job is given a block size range instead of a fixed value,
600 the smallest block size is used as the metric.
601
602rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
603 the job to exit.
71bfa161
JA
604
605ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
6c219763 606 of milliseconds.
71bfa161
JA
607
608cpumask=int Set the CPU affinity of this job. The parameter given is a
a08bc17f
JA
609 bitmask of allowed CPU's the job may run on. So if you want
610 the allowed CPUs to be 1 and 5, you would pass the decimal
611 value of (1 << 1 | 1 << 5), or 34. See man
7dbb6eba 612 sched_setaffinity(2). This may not work on all supported
b0ea08ce
JA
613 operating systems or kernel versions. This option doesn't
614 work well for a higher CPU count than what you can store in
615 an integer mask, so it can only control cpus 1-32. For
616 boxes with larger CPU counts, use cpus_allowed.
71bfa161 617
d2e268b0
JA
618cpus_allowed=str Controls the same options as cpumask, but it allows a text
619 setting of the permitted CPUs instead. So to use CPUs 1 and
62a7273d
JA
620 5, you would specify cpus_allowed=1,5. This options also
621 allows a range of CPUs. Say you wanted a binding to CPUs
622 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
d2e268b0 623
e417fd66 624startdelay=time Start this job the specified number of seconds after fio
71bfa161
JA
625 has started. Only useful if the job file contains several
626 jobs, and you want to delay starting some jobs to a certain
627 time.
628
e417fd66 629runtime=time Tell fio to terminate processing after the specified number
71bfa161
JA
630 of seconds. It can be quite hard to determine for how long
631 a specified job will run, so this parameter is handy to
632 cap the total runtime to a given time.
633
cf4464ca 634time_based If set, fio will run for the duration of the runtime
bf9a3edb 635 specified even if the file(s) are completely read or
cf4464ca
JA
636 written. It will simply loop over the same workload
637 as many times as the runtime allows.
638
e417fd66 639ramp_time=time If set, fio will run the specified workload for this amount
721938ae
JA
640 of time before logging any performance numbers. Useful for
641 letting performance settle before logging results, thus
b29ee5b3
JA
642 minimizing the runtime required for stable results. Note
643 that the ramp_time is considered lead in time for a job,
644 thus it will increase the total runtime if a special timeout
645 or runtime is specified.
721938ae 646
71bfa161
JA
647invalidate=bool Invalidate the buffer/page cache parts for this file prior
648 to starting io. Defaults to true.
649
650sync=bool Use sync io for buffered writes. For the majority of the
651 io engines, this means using O_SYNC.
652
d3aad8f2 653iomem=str
71bfa161
JA
654mem=str Fio can use various types of memory as the io unit buffer.
655 The allowed values are:
656
657 malloc Use memory from malloc(3) as the buffers.
658
659 shm Use shared memory as the buffers. Allocated
660 through shmget(2).
661
74b025b0
JA
662 shmhuge Same as shm, but use huge pages as backing.
663
313cb206
JA
664 mmap Use mmap to allocate buffers. May either be
665 anonymous memory, or can be file backed if
666 a filename is given after the option. The
667 format is mem=mmap:/path/to/file.
71bfa161 668
d0bdaf49
JA
669 mmaphuge Use a memory mapped huge file as the buffer
670 backing. Append filename after mmaphuge, ala
671 mem=mmaphuge:/hugetlbfs/file
672
71bfa161 673 The area allocated is a function of the maximum allowed
5394ae5f
JA
674 bs size for the job, multiplied by the io depth given. Note
675 that for shmhuge and mmaphuge to work, the system must have
676 free huge pages allocated. This can normally be checked
677 and set by reading/writing /proc/sys/vm/nr_hugepages on a
678 Linux system. Fio assumes a huge page is 4MiB in size. So
679 to calculate the number of huge pages you need for a given
680 job file, add up the io depth of all jobs (normally one unless
681 iodepth= is used) and multiply by the maximum bs set. Then
682 divide that number by the huge page size. You can see the
683 size of the huge pages in /proc/meminfo. If no huge pages
684 are allocated by having a non-zero number in nr_hugepages,
56bb17f2 685 using mmaphuge or shmhuge will fail. Also see hugepage-size.
5394ae5f
JA
686
687 mmaphuge also needs to have hugetlbfs mounted and the file
688 location should point there. So if it's mounted in /huge,
689 you would use mem=mmaphuge:/huge/somefile.
71bfa161 690
f7fa2653 691hugepage-size=int
56bb17f2
JA
692 Defines the size of a huge page. Must at least be equal
693 to the system setting, see /proc/meminfo. Defaults to 4MiB.
c51074e7
JA
694 Should probably always be a multiple of megabytes, so using
695 hugepage-size=Xm is the preferred way to set this to avoid
696 setting a non-pow-2 bad value.
56bb17f2 697
71bfa161
JA
698exitall When one job finishes, terminate the rest. The default is
699 to wait for each job to finish, sometimes that is not the
700 desired action.
701
702bwavgtime=int Average the calculated bandwidth over the given time. Value
6c219763 703 is specified in milliseconds.
71bfa161
JA
704
705create_serialize=bool If true, serialize the file creating for the jobs.
706 This may be handy to avoid interleaving of data
707 files, which may greatly depend on the filesystem
708 used and even the number of processors in the system.
709
710create_fsync=bool fsync the data file after creation. This is the
711 default.
712
814452bd
JA
713create_on_open=bool Don't pre-setup the files for IO, just create open()
714 when it's time to do IO to that file.
715
e545a6ce 716unlink=bool Unlink the job files when done. Not the default, as repeated
bf9a3edb
JA
717 runs of that job would then waste time recreating the file
718 set again and again.
71bfa161
JA
719
720loops=int Run the specified number of iterations of this job. Used
721 to repeat the same workload a given number of times. Defaults
722 to 1.
723
68e1f29a 724do_verify=bool Run the verify phase after a write phase. Only makes sense if
e84c73a8
SL
725 verify is set. Defaults to 1.
726
71bfa161
JA
727verify=str If writing to a file, fio can verify the file contents
728 after each iteration of the job. The allowed values are:
729
730 md5 Use an md5 sum of the data area and store
731 it in the header of each block.
732
17dc34df
JA
733 crc64 Use an experimental crc64 sum of the data
734 area and store it in the header of each
735 block.
736
bac39e0e
JA
737 crc32c Use a crc32c sum of the data area and store
738 it in the header of each block.
739
3845591f
JA
740 crc32c-intel Use hardware assisted crc32c calcuation
741 provided on SSE4.2 enabled processors.
742
71bfa161
JA
743 crc32 Use a crc32 sum of the data area and store
744 it in the header of each block.
745
969f7ed3
JA
746 crc16 Use a crc16 sum of the data area and store
747 it in the header of each block.
748
17dc34df
JA
749 crc7 Use a crc7 sum of the data area and store
750 it in the header of each block.
751
cd14cc10
JA
752 sha512 Use sha512 as the checksum function.
753
754 sha256 Use sha256 as the checksum function.
755
7437ee87
SL
756 meta Write extra information about each io
757 (timestamp, block number etc.). The block
758 number is verified.
759
36690c9b
JA
760 null Only pretend to verify. Useful for testing
761 internals with ioengine=null, not for much
762 else.
763
6c219763 764 This option can be used for repeated burn-in tests of a
71bfa161
JA
765 system to make sure that the written data is also
766 correctly read back.
767
160b966d
JA
768verifysort=bool If set, fio will sort written verify blocks when it deems
769 it faster to read them back in a sorted manner. This is
770 often the case when overwriting an existing file, since
771 the blocks are already laid out in the file system. You
772 can ignore this option unless doing huge amounts of really
773 fast IO where the red-black tree sorting CPU time becomes
774 significant.
3f9f4e26 775
f7fa2653 776verify_offset=int Swap the verification header with data somewhere else
546a9142
SL
777 in the block before writing. Its swapped back before
778 verifying.
779
f7fa2653 780verify_interval=int Write the verification header at a finer granularity
3f9f4e26
SL
781 than the blocksize. It will be written for chunks the
782 size of header_interval. blocksize should divide this
783 evenly.
90059d65 784
e28218f3
SL
785verify_pattern=int If set, fio will fill the io buffers with this
786 pattern. Fio defaults to filling with totally random
787 bytes, but sometimes it's interesting to fill with a known
788 pattern for io verification purposes. Depending on the
789 width of the pattern, fio will fill 1/2/3/4 bytes of the
790 buffer at the time. The verify_pattern cannot be larger than
791 a 32-bit quantity.
792
68e1f29a 793verify_fatal=bool Normally fio will keep checking the entire contents
a12a3b4d
JA
794 before quitting on a block verification failure. If this
795 option is set, fio will exit the job on the first observed
796 failure.
160b966d 797
71bfa161
JA
798stonewall Wait for preceeding jobs in the job file to exit, before
799 starting this one. Can be used to insert serialization
b3d62a75
JA
800 points in the job file. A stone wall also implies starting
801 a new reporting group.
802
803new_group Start a new reporting group. If this option isn't given,
804 jobs in a file will be part of the same reporting group
bf9a3edb 805 unless separated by a stone wall (or if it's a group
b3d62a75 806 by itself, with the numjobs option).
71bfa161
JA
807
808numjobs=int Create the specified number of clones of this job. May be
809 used to setup a larger number of threads/processes doing
fa28c85a
JA
810 the same thing. We regard that grouping of jobs as a
811 specific group.
812
813group_reporting If 'numjobs' is set, it may be interesting to display
814 statistics for the group as a whole instead of for each
815 individual job. This is especially true of 'numjobs' is
816 large, looking at individual thread/process output quickly
817 becomes unwieldy. If 'group_reporting' is specified, fio
818 will show the final report per-group instead of per-job.
71bfa161
JA
819
820thread fio defaults to forking jobs, however if this option is
821 given, fio will use pthread_create(3) to create threads
822 instead.
823
f7fa2653 824zonesize=int Divide a file into zones of the specified size. See zoneskip.
71bfa161 825
f7fa2653 826zoneskip=int Skip the specified number of bytes when zonesize data has
71bfa161
JA
827 been read. The two zone options can be used to only do
828 io on zones of a file.
829
076efc7c
JA
830write_iolog=str Write the issued io patterns to the specified file. See
831 read_iolog.
71bfa161 832
076efc7c 833read_iolog=str Open an iolog with the specified file name and replay the
71bfa161 834 io patterns it contains. This can be used to store a
6df8adaa
JA
835 workload and replay it sometime later. The iolog given
836 may also be a blktrace binary file, which allows fio
837 to replay a workload captured by blktrace. See blktrace
838 for how to capture such logging data. For blktrace replay,
839 the file needs to be turned into a blkparse binary data
840 file first (blktrace <device> -d file_for_fio.bin).
71bfa161 841
e3cedca7 842write_bw_log=str If given, write a bandwidth log of the jobs in this job
71bfa161 843 file. Can be used to store data of the bandwidth of the
e0da9bc2
JA
844 jobs in their lifetime. The included fio_generate_plots
845 script uses gnuplot to turn these text files into nice
e3cedca7
JA
846 graphs. See write_log_log for behaviour of given
847 filename. For this option, the postfix is _bw.log.
71bfa161 848
e3cedca7
JA
849write_lat_log=str Same as write_bw_log, except that this option stores io
850 completion latencies instead. If no filename is given
851 with this option, the default filename of "jobname_type.log"
852 is used. Even if the filename is given, fio will still
853 append the type of log. So if one specifies
854
855 write_lat_log=foo
856
857 The actual log names will be foo_clat.log and foo_slat.log.
858 This helps fio_generate_plot fine the logs automatically.
71bfa161 859
f7fa2653 860lockmem=int Pin down the specified amount of memory with mlock(2). Can
71bfa161
JA
861 potentially be used instead of removing memory or booting
862 with less memory to simulate a smaller amount of memory.
863
864exec_prerun=str Before running this job, issue the command specified
865 through system(3).
866
867exec_postrun=str After the job completes, issue the command specified
868 though system(3).
869
870ioscheduler=str Attempt to switch the device hosting the file to the specified
871 io scheduler before running.
872
873cpuload=int If the job is a CPU cycle eater, attempt to use the specified
874 percentage of CPU cycles.
875
876cpuchunks=int If the job is a CPU cycle eater, split the load into
26eca2db 877 cycles of the given time. In microseconds.
71bfa161 878
0a839f30
JA
879disk_util=bool Generate disk utilization statistics, if the platform
880 supports it. Defaults to on.
881
9520ebb9
JA
882disable_clat=bool Disable measurements of completion latency numbers. Useful
883 only for cutting back the number of calls to gettimeofday,
884 as that does impact performance at really high IOPS rates.
885 Note that to really get rid of a large amount of these
886 calls, this option must be used with disable_slat and
887 disable_bw as well.
888
889disable_slat=bool Disable measurements of submission latency numbers. See
890 disable_clat.
891
892disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
893 disable_clat.
894
993bf48b
JA
895gtod_reduce=bool Enable all of the gettimeofday() reducing options
896 (disable_clat, disable_slat, disable_bw) plus reduce
897 precision of the timeout somewhat to really shrink
898 the gettimeofday() call count. With this option enabled,
899 we only do about 0.4% of the gtod() calls we would have
900 done if all time keeping was enabled.
901
be4ecfdf
JA
902gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
903 execution to just getting the current time. Fio (and
904 databases, for instance) are very intensive on gettimeofday()
905 calls. With this option, you can set one CPU aside for
906 doing nothing but logging current time to a shared memory
907 location. Then the other threads/processes that run IO
908 workloads need only copy that segment, instead of entering
909 the kernel with a gettimeofday() call. The CPU set aside
910 for doing these time calls will be excluded from other
911 uses. Fio will manually clear it from the CPU mask of other
912 jobs.
913
71bfa161
JA
914
9156.0 Interpreting the output
916---------------------------
917
918fio spits out a lot of output. While running, fio will display the
919status of the jobs created. An example of that would be:
920
73c8b082 921Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
71bfa161
JA
922
923The characters inside the square brackets denote the current status of
924each thread. The possible values (in typical life cycle order) are:
925
926Idle Run
927---- ---
928P Thread setup, but not started.
929C Thread created.
930I Thread initialized, waiting.
931 R Running, doing sequential reads.
932 r Running, doing random reads.
933 W Running, doing sequential writes.
934 w Running, doing random writes.
935 M Running, doing mixed sequential reads/writes.
936 m Running, doing mixed random reads/writes.
937 F Running, currently waiting for fsync()
fc6bd43c 938 V Running, doing verification of written data.
71bfa161
JA
939E Thread exited, not reaped by main thread yet.
940_ Thread reaped.
941
942The other values are fairly self explanatory - number of threads
c9f60304
JA
943currently running and doing io, rate of io since last check (read speed
944listed first, then write speed), and the estimated completion percentage
945and time for the running group. It's impossible to estimate runtime of
946the following groups (if any).
71bfa161
JA
947
948When fio is done (or interrupted by ctrl-c), it will show the data for
949each thread, group of threads, and disks in that order. For each data
950direction, the output looks like:
951
952Client1 (g=0): err= 0:
953 write: io= 32MiB, bw= 666KiB/s, runt= 50320msec
6104ddb6
JA
954 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
955 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
956 bw (KiB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
e7823a94 957 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
71619dc2 958 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
959 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
960 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
30061b97 961 issued r/w: total=0/32768, short=0/0
8abdce66
JA
962 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
963 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
71bfa161
JA
964
965The client number is printed, along with the group id and error of that
966thread. Below is the io statistics, here for writes. In the order listed,
967they denote:
968
969io= Number of megabytes io performed
970bw= Average bandwidth rate
971runt= The runtime of that thread
72fbda2a 972 slat= Submission latency (avg being the average, stdev being the
71bfa161
JA
973 standard deviation). This is the time it took to submit
974 the io. For sync io, the slat is really the completion
8a35c71e 975 latency, since queue/complete is one operation there. This
bf9a3edb 976 value can be in milliseconds or microseconds, fio will choose
8a35c71e 977 the most appropriate base and print that. In the example
bf9a3edb 978 above, milliseconds is the best scale.
71bfa161
JA
979 clat= Completion latency. Same names as slat, this denotes the
980 time from submission to completion of the io pieces. For
981 sync io, clat will usually be equal (or very close) to 0,
982 as the time from submit to complete is basically just
983 CPU time (io has already been done, see slat explanation).
984 bw= Bandwidth. Same names as the xlat stats, but also includes
985 an approximate percentage of total aggregate bandwidth
986 this thread received in this group. This last value is
987 only really useful if the threads in this group are on the
988 same disk, since they are then competing for disk access.
989cpu= CPU usage. User and system time, along with the number
e7823a94
JA
990 of context switches this thread went through, usage of
991 system and user time, and finally the number of major
992 and minor page faults.
71619dc2
JA
993IO depths= The distribution of io depths over the job life time. The
994 numbers are divided into powers of 2, so for example the
995 16= entries includes depths up to that value but higher
996 than the previous entry. In other words, it covers the
997 range from 16 to 31.
838bc709
JA
998IO submit= How many pieces of IO were submitting in a single submit
999 call. Each entry denotes that amount and below, until
1000 the previous entry - eg, 8=100% mean that we submitted
1001 anywhere in between 5-8 ios per submit call.
1002IO complete= Like the above submit number, but for completions instead.
30061b97
JA
1003IO issued= The number of read/write requests issued, and how many
1004 of them were short.
ec118304
JA
1005IO latencies= The distribution of IO completion latencies. This is the
1006 time from when IO leaves fio and when it gets completed.
1007 The numbers follow the same pattern as the IO depths,
1008 meaning that 2=1.6% means that 1.6% of the IO completed
8abdce66
JA
1009 within 2 msecs, 20=12.8% means that 12.8% of the IO
1010 took more than 10 msecs, but less than (or equal to) 20 msecs.
71bfa161
JA
1011
1012After each client has been listed, the group statistics are printed. They
1013will look like this:
1014
1015Run status group 0 (all jobs):
1016 READ: io=64MiB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1017 WRITE: io=64MiB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1018
1019For each data direction, it prints:
1020
1021io= Number of megabytes io performed.
1022aggrb= Aggregate bandwidth of threads in this group.
1023minb= The minimum average bandwidth a thread saw.
1024maxb= The maximum average bandwidth a thread saw.
1025mint= The smallest runtime of the threads in that group.
1026maxt= The longest runtime of the threads in that group.
1027
1028And finally, the disk statistics are printed. They will look like this:
1029
1030Disk stats (read/write):
1031 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1032
1033Each value is printed for both reads and writes, with reads first. The
1034numbers denote:
1035
1036ios= Number of ios performed by all groups.
1037merge= Number of merges io the io scheduler.
1038ticks= Number of ticks we kept the disk busy.
1039io_queue= Total time spent in the disk queue.
1040util= The disk utilization. A value of 100% means we kept the disk
1041 busy constantly, 50% would be a disk idling half of the time.
1042
1043
10447.0 Terse output
1045----------------
1046
1047For scripted usage where you typically want to generate tables or graphs
6af019c9 1048of the results, fio can output the results in a semicolon separated format.
71bfa161
JA
1049The format is one long line of values, such as:
1050
6af019c9
JA
1051client1;0;0;1906777;1090804;1790;0;0;0.000000;0.000000;0;0;0.000000;0.000000;929380;1152890;25.510151%;1078276.333333;128948.113404;0;0;0;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000%;0.000000;0.000000;100.000000%;0.000000%;324;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.0%;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%
1052;0.0%;0.0%;0.0%;0.0%;0.0%
71bfa161 1053
6820cb3b
JA
1054To enable terse output, use the --minimal command line option.
1055
71bfa161
JA
1056Split up, the format is as follows:
1057
1058 jobname, groupid, error
1059 READ status:
1060 KiB IO, bandwidth (KiB/sec), runtime (msec)
1061 Submission latency: min, max, mean, deviation
1062 Completion latency: min, max, mean, deviation
6c219763 1063 Bw: min, max, aggregate percentage of total, mean, deviation
71bfa161
JA
1064 WRITE status:
1065 KiB IO, bandwidth (KiB/sec), runtime (msec)
1066 Submission latency: min, max, mean, deviation
1067 Completion latency: min, max, mean, deviation
6c219763 1068 Bw: min, max, aggregate percentage of total, mean, deviation
046ee302 1069 CPU usage: user, system, context switches, major faults, minor faults
2270890c
JA
1070 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1071 IO latencies: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, >=2000
1072 Text description
71bfa161 1073