8 5. Detailed list of parameters
12 9. CPU idleness profiling
13 10. Verification and triggers
17 1.0 Overview and history
18 ------------------------
19 fio was originally written to save me the hassle of writing special test
20 case programs when I wanted to test a specific workload, either for
21 performance reasons or to find/reproduce a bug. The process of writing
22 such a test app can be tiresome, especially if you have to do it often.
23 Hence I needed a tool that would be able to simulate a given io workload
24 without resorting to writing a tailored test case again and again.
26 A test work load is difficult to define, though. There can be any number
27 of processes or threads involved, and they can each be using their own
28 way of generating io. You could have someone dirtying large amounts of
29 memory in an memory mapped file, or maybe several threads issuing
30 reads using asynchronous io. fio needed to be flexible enough to
31 simulate both of these cases, and many more.
35 The first step in getting fio to simulate a desired io workload, is
36 writing a job file describing that specific setup. A job file may contain
37 any number of threads and/or files - the typical contents of the job file
38 is a global section defining shared parameters, and one or more job
39 sections describing the jobs involved. When run, fio parses this file
40 and sets everything up as described. If we break down a job from top to
41 bottom, it contains the following basic parameters:
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.
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
52 IO size How much data are we going to be reading/writing.
54 IO engine How do we issue io? We could be memory mapping the
55 file, we could be using regular read/write, we
56 could be using splice, async io, or even SG
59 IO depth If the io engine is async, how large a queuing
60 depth do we want to maintain?
62 IO type Should we be doing buffered io, or direct/raw io?
64 Num files How many files are we spreading the workload over.
66 Num threads How many threads or processes should we spread
69 The above are the basic parameters defined for a workload, in addition
70 there's a multitude of parameters that modify other aspects of how this
76 See the README file for command line parameters, there are only a few
79 Running fio is normally the easiest part - you just give it the job file
80 (or job files) as parameters:
84 and it will start doing what the job_file tells it to do. You can give
85 more than one job file on the command line, fio will serialize the running
86 of those files. Internally that is the same as using the 'stonewall'
87 parameter described in the parameter section.
89 If the job file contains only one job, you may as well just give the
90 parameters on the command line. The command line parameters are identical
91 to the job parameters, with a few extra that control global parameters
92 (see README). For example, for the job file parameter iodepth=2, the
93 mirror command line option would be --iodepth 2 or --iodepth=2. You can
94 also 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.
96 Command line entries following a --name entry will apply to that job,
97 until there are no more entries or a new --name entry is seen. This is
98 similar to the job file options, where each option applies to the current
99 job until a new [] job entry is seen.
101 fio does not need to run as root, except if the files or devices specified
102 in the job section requires that. Some other options may also be restricted,
103 such as memory locking, io scheduler switching, and decreasing the nice value.
108 As previously described, fio accepts one or more job files describing
109 what it is supposed to do. The job file format is the classic ini file,
110 where the names enclosed in [] brackets define the job name. You are free
111 to use any ascii name you want, except 'global' which has special meaning.
112 A global section sets defaults for the jobs described in that file. A job
113 may override a global section parameter, and a job file may even have
114 several global sections if so desired. A job is only affected by a global
115 section residing above it. If the first character in a line is a ';' or a
116 '#', the entire line is discarded as a comment.
118 So let's look at a really simple job file that defines two processes, each
119 randomly reading from a 128MB file.
121 ; -- start job file --
132 As you can see, the job file sections themselves are empty as all the
133 described parameters are shared. As no filename= option is given, fio
134 makes up a filename for each of the jobs as it sees fit. On the command
135 line, this job would look as follows:
137 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
140 Let's look at an example that has a number of processes writing randomly
143 ; -- start job file --
155 Here we have no global section, as we only have one job defined anyway.
156 We want to use async io here, with a depth of 4 for each file. We also
157 increased the buffer size used to 32KB and define numjobs to 4 to
158 fork 4 identical jobs. The result is 4 processes each randomly writing
159 to their own 64MB file. Instead of using the above job file, you could
160 have given the parameters on the command line. For this case, you would
163 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
165 When fio is utilized as a basis of any reasonably large test suite, it might be
166 desirable to share a set of standardized settings across multiple job files.
167 Instead of copy/pasting such settings, any section may pull in an external
168 .fio file with 'include filename' directive, as in the following example:
170 ; -- start job file including.fio --
174 include glob-include.fio
181 include test-include.fio
182 ; -- end job file including.fio --
184 ; -- start job file glob-include.fio --
187 ; -- end job file glob-include.fio --
189 ; -- start job file test-include.fio --
192 ; -- end job file test-include.fio --
194 Settings pulled into a section apply to that section only (except global
195 section). Include directives may be nested in that any included file may
196 contain further include directive(s). Include files may not contain []
200 4.1 Environment variables
201 -------------------------
203 fio also supports environment variable expansion in job files. Any
204 sub-string of the form "${VARNAME}" as part of an option value (in other
205 words, on the right of the `='), will be expanded to the value of the
206 environment variable called VARNAME. If no such environment variable
207 is defined, or VARNAME is the empty string, the empty string will be
210 As an example, let's look at a sample fio invocation and job file:
212 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
214 ; -- start job file --
221 This will expand to the following equivalent job file at runtime:
223 ; -- start job file --
230 fio ships with a few example job files, you can also look there for
233 4.2 Reserved keywords
234 ---------------------
236 Additionally, fio has a set of reserved keywords that will be replaced
237 internally with the appropriate value. Those keywords are:
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
243 These can be used on the command line or in the job file, and will be
244 automatically substituted with the current system values when the job
245 is run. Simple math is also supported on these keywords, so you can
246 perform actions like:
250 and get that properly expanded to 8 times the size of memory in the
254 5.0 Detailed list of parameters
255 -------------------------------
257 This section describes in details each parameter associated with a job.
258 Some parameters take an option of a given type, such as an integer or
259 a string. Anywhere a numeric value is required, an arithmetic expression
260 may be used, provided it is surrounded by parentheses. Supported operators
270 For time values in expressions, units are microseconds by default. This is
271 different than for time values not in expressions (not enclosed in
272 parentheses). The following types are used:
274 str String. This is a sequence of alpha characters.
275 time Integer with possible time suffix. In seconds unless otherwise
276 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
277 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
278 and 'us' (or 'usec') for microseconds.
279 int 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
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
284 out '4096' or just give 4k. The suffixes signify base 2 values, so
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.
293 bool Boolean. Usually parsed as an integer, however only defined for
294 true and false (1 and 0).
295 irange Integer range with suffix. Allows value range to be given, such
296 as 1024-4096. A colon may also be used as the separator, eg
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
300 float_list A list of floating numbers, separated by a ':' character.
302 With the above in mind, here follows the complete list of fio job
305 name=str ASCII name of the job. This may be used to override the
306 name printed by fio for this job. Otherwise the job
307 name is used. On the command line this parameter has the
308 special purpose of also signaling the start of a new
311 wait_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.
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).
321 description=str Text description of the job. Doesn't do anything except
322 dump this text description when this job is run. It's
325 directory=str Prefix filenames with this directory. Used to place files
326 in a different location than "./". See the 'filename' option
327 for escaping certain characters.
329 filename=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
332 a filename for each of them to override the default.
333 If the ioengine is file based, you can specify a number of
334 files by separating the names with a ':' colon. So if you
335 wanted a job to open /dev/sda and /dev/sdb as the two working
336 files, you would use filename=/dev/sda:/dev/sdb. On Windows,
337 disk devices are accessed as \\.\PhysicalDrive0 for the first
338 device, \\.\PhysicalDrive1 for the second etc. Note: Windows
339 and FreeBSD prevent write access to areas of the disk
340 containing in-use data (e.g. filesystems).
341 If the wanted filename does need to include a colon, then
342 escape that with a '\' character. For instance, if the filename
343 is "/dev/dsk/foo@3,0:c", then you would use
344 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
345 stdin or stdout. Which of the two depends on the read/write
349 If sharing multiple files between jobs, it is usually necessary
350 to have fio generate the exact names that you want. By default,
351 fio will name a file based on the default file format
352 specification of jobname.jobnumber.filenumber. With this
353 option, that can be customized. Fio will recognize and replace
354 the following keywords in this string:
357 The name of the worker thread or process.
360 The incremental number of the worker thread or
364 The incremental number of the file for that worker
367 To have dependent jobs share a set of files, this option can
368 be set to have fio generate filenames that are shared between
369 the two. For instance, if testfiles.$filenum is specified,
370 file number 4 for any job will be named testfiles.4. The
371 default of $jobname.$jobnum.$filenum will be used if
372 no other format specifier is given.
374 unique_filename=bool To avoid collisions between networked clients, fio
375 defaults to prefixing any generated filenames (with a directory
376 specified) with the source of the client connecting. To disable
377 this behavior, set this option to 0.
379 opendir=str Tell fio to recursively add any file it can find in this
380 directory and down the file system tree.
382 lockfile=str Fio defaults to not locking any files before it does
383 IO to them. If a file or file descriptor is shared, fio
384 can serialize IO to that file to make the end result
385 consistent. This is usual for emulating real workloads that
386 share files. The lock modes are:
388 none No locking. The default.
389 exclusive Only one thread/process may do IO,
390 excluding all others.
391 readwrite Read-write locking on the file. Many
392 readers may access the file at the
393 same time, but writes get exclusive
397 rw=str Type of io pattern. Accepted values are:
399 read Sequential reads
400 write Sequential writes
401 randwrite Random writes
402 randread Random reads
403 rw,readwrite Sequential mixed reads and writes
404 randrw Random mixed reads and writes
405 trimwrite Mixed trims and writes. Blocks will be
406 trimmed first, then written to.
408 Fio defaults to read if the option is not specified.
409 For the mixed io types, the default is to split them 50/50.
410 For certain types of io the result may still be skewed a bit,
411 since the speed may be different. It is possible to specify
412 a number of IO's to do before getting a new offset, this is
413 done by appending a ':<nr>' to the end of the string given.
414 For a random read, it would look like 'rw=randread:8' for
415 passing in an offset modifier with a value of 8. If the
416 suffix is used with a sequential IO pattern, then the value
417 specified will be added to the generated offset for each IO.
418 For instance, using rw=write:4k will skip 4k for every
419 write. It turns sequential IO into sequential IO with holes.
420 See the 'rw_sequencer' option.
422 rw_sequencer=str If an offset modifier is given by appending a number to
423 the rw=<str> line, then this option controls how that
424 number modifies the IO offset being generated. Accepted
427 sequential Generate sequential offset
428 identical Generate the same offset
430 'sequential' is only useful for random IO, where fio would
431 normally generate a new random offset for every IO. If you
432 append eg 8 to randread, you would get a new random offset for
433 every 8 IO's. The result would be a seek for only every 8
434 IO's, instead of for every IO. Use rw=randread:8 to specify
435 that. As sequential IO is already sequential, setting
436 'sequential' for that would not result in any differences.
437 'identical' behaves in a similar fashion, except it sends
438 the same offset 8 number of times before generating a new
441 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
442 Storage manufacturers like to use 10^3 or 1000 as a base
443 ten unit instead, for obvious reasons. Allow values are
444 1024 or 1000, with 1024 being the default.
446 unified_rw_reporting=bool Fio normally reports statistics on a per
447 data direction basis, meaning that read, write, and trim are
448 accounted and reported separately. If this option is set,
449 the fio will sum the results and report them as "mixed"
452 randrepeat=bool For random IO workloads, seed the generator in a predictable
453 way so that results are repeatable across repetitions.
456 randseed=int Seed the random number generators based on this seed value, to
457 be able to control what sequence of output is being generated.
458 If not set, the random sequence depends on the randrepeat
461 fallocate=str Whether pre-allocation is performed when laying down files.
464 none Do not pre-allocate space
465 posix Pre-allocate via posix_fallocate()
466 keep Pre-allocate via fallocate() with
467 FALLOC_FL_KEEP_SIZE set
468 0 Backward-compatible alias for 'none'
469 1 Backward-compatible alias for 'posix'
471 May not be available on all supported platforms. 'keep' is only
472 available on Linux.If using ZFS on Solaris this must be set to
473 'none' because ZFS doesn't support it. Default: 'posix'.
475 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
476 on what IO patterns it is likely to issue. Sometimes you
477 want to test specific IO patterns without telling the
478 kernel about it, in which case you can disable this option.
479 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
480 IO and POSIX_FADV_RANDOM for random IO.
482 fadvise_stream=int Notify the kernel what write stream ID to place these
483 writes under. Only supported on Linux. Note, this option
484 may change going forward.
486 size=int The total size of file io for this job. Fio will run until
487 this many bytes has been transferred, unless runtime is
488 limited by other options (such as 'runtime', for instance,
489 or increased/decreased by 'io_size'). Unless specific nrfiles
490 and filesize options are given, fio will divide this size
491 between the available files specified by the job. If not set,
492 fio will use the full size of the given files or devices.
493 If the files do not exist, size must be given. It is also
494 possible to give size as a percentage between 1 and 100. If
495 size=20% is given, fio will use 20% of the full size of the
496 given files or devices.
499 io_limit=int Normally fio operates within the region set by 'size', which
500 means that the 'size' option sets both the region and size of
501 IO to be performed. Sometimes that is not what you want. With
502 this option, it is possible to define just the amount of IO
503 that fio should do. For instance, if 'size' is set to 20G and
504 'io_size' is set to 5G, fio will perform IO within the first
505 20G but exit when 5G have been done. The opposite is also
506 possible - if 'size' is set to 20G, and 'io_size' is set to
507 40G, then fio will do 40G of IO within the 0..20G region.
509 filesize=int Individual file sizes. May be a range, in which case fio
510 will select sizes for files at random within the given range
511 and limited to 'size' in total (if that is given). If not
512 given, each created file is the same size.
514 file_append=bool Perform IO after the end of the file. Normally fio will
515 operate within the size of a file. If this option is set, then
516 fio will append to the file instead. This has identical
517 behavior to setting offset to the size of a file. This option
518 is ignored on non-regular files.
521 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
522 space left on device) as the terminating condition. Only makes
523 sense with sequential write. For a read workload, the mount
524 point will be filled first then IO started on the result. This
525 option doesn't make sense if operating on a raw device node,
526 since the size of that is already known by the file system.
527 Additionally, writing beyond end-of-device will not return
531 bs=int The block size used for the io units. Defaults to 4k. Values
532 can be given for both read and writes. If a single int is
533 given, it will apply to both. If a second int is specified
534 after a comma, it will apply to writes only. In other words,
535 the format is either bs=read_and_write or bs=read,write,trim.
536 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
537 writes, and 8k for trims. You can terminate the list with
538 a trailing comma. bs=4k,8k, would use the default value for
539 trims.. If you only wish to set the write size, you
540 can do so by passing an empty read size - bs=,8k will set
541 8k for writes and leave the read default value.
544 ba=int At what boundary to align random IO offsets. Defaults to
545 the same as 'blocksize' the minimum blocksize given.
546 Minimum alignment is typically 512b for using direct IO,
547 though it usually depends on the hardware block size. This
548 option is mutually exclusive with using a random map for
549 files, so it will turn off that option.
551 blocksize_range=irange
552 bsrange=irange Instead of giving a single block size, specify a range
553 and fio will mix the issued io block sizes. The issued
554 io unit will always be a multiple of the minimum value
555 given (also see bs_unaligned). Applies to both reads and
556 writes, however a second range can be given after a comma.
559 bssplit=str Sometimes you want even finer grained control of the
560 block sizes issued, not just an even split between them.
561 This option allows you to weight various block sizes,
562 so that you are able to define a specific amount of
563 block sizes issued. The format for this option is:
565 bssplit=blocksize/percentage:blocksize/percentage
567 for as many block sizes as needed. So if you want to define
568 a workload that has 50% 64k blocks, 10% 4k blocks, and
569 40% 32k blocks, you would write:
571 bssplit=4k/10:64k/50:32k/40
573 Ordering does not matter. If the percentage is left blank,
574 fio will fill in the remaining values evenly. So a bssplit
575 option like this one:
577 bssplit=4k/50:1k/:32k/
579 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
580 always add up to 100, if bssplit is given a range that adds
581 up to more, it will error out.
583 bssplit also supports giving separate splits to reads and
584 writes. The format is identical to what bs= accepts. You
585 have to separate the read and write parts with a comma. So
586 if you want a workload that has 50% 2k reads and 50% 4k reads,
587 while having 90% 4k writes and 10% 8k writes, you would
590 bssplit=2k/50:4k/50,4k/90:8k/10
593 bs_unaligned If this option is given, any byte size value within bsrange
594 may be used as a block range. This typically wont work with
595 direct IO, as that normally requires sector alignment.
597 bs_is_seq_rand If this option is set, fio will use the normal read,write
598 blocksize settings as sequential,random instead. Any random
599 read or write will use the WRITE blocksize settings, and any
600 sequential read or write will use the READ blocksize setting.
602 zero_buffers If this option is given, fio will init the IO buffers to
603 all zeroes. The default is to fill them with random data.
605 refill_buffers If this option is given, fio will refill the IO buffers
606 on every submit. The default is to only fill it at init
607 time and reuse that data. Only makes sense if zero_buffers
608 isn't specified, naturally. If data verification is enabled,
609 refill_buffers is also automatically enabled.
611 scramble_buffers=bool If refill_buffers is too costly and the target is
612 using data deduplication, then setting this option will
613 slightly modify the IO buffer contents to defeat normal
614 de-dupe attempts. This is not enough to defeat more clever
615 block compression attempts, but it will stop naive dedupe of
616 blocks. Default: true.
618 buffer_compress_percentage=int If this is set, then fio will attempt to
619 provide IO buffer content (on WRITEs) that compress to
620 the specified level. Fio does this by providing a mix of
621 random data and a fixed pattern. The fixed pattern is either
622 zeroes, or the pattern specified by buffer_pattern. If the
623 pattern option is used, it might skew the compression ratio
624 slightly. Note that this is per block size unit, for file/disk
625 wide compression level that matches this setting, you'll also
626 want to set refill_buffers.
628 buffer_compress_chunk=int See buffer_compress_percentage. This
629 setting allows fio to manage how big the ranges of random
630 data and zeroed data is. Without this set, fio will
631 provide buffer_compress_percentage of blocksize random
632 data, followed by the remaining zeroed. With this set
633 to some chunk size smaller than the block size, fio can
634 alternate random and zeroed data throughout the IO
637 buffer_pattern=str If set, fio will fill the io buffers with this
638 pattern. If not set, the contents of io buffers is defined by
639 the other options related to buffer contents. The setting can
640 be any pattern of bytes, and can be prefixed with 0x for hex
641 values. It may also be a string, where the string must then
642 be wrapped with "", e.g.:
644 buffer_pattern="abcd"
648 buffer_pattern=0xdeadface
650 Also you can combine everything together in any order:
651 buffer_pattern=0xdeadface"abcd"-12
653 dedupe_percentage=int If set, fio will generate this percentage of
654 identical buffers when writing. These buffers will be
655 naturally dedupable. The contents of the buffers depend on
656 what other buffer compression settings have been set. It's
657 possible to have the individual buffers either fully
658 compressible, or not at all. This option only controls the
659 distribution of unique buffers.
661 nrfiles=int Number of files to use for this job. Defaults to 1.
663 openfiles=int Number of files to keep open at the same time. Defaults to
664 the same as nrfiles, can be set smaller to limit the number
667 file_service_type=str Defines how fio decides which file from a job to
668 service next. The following types are defined:
670 random Just choose a file at random.
672 roundrobin Round robin over open files. This
675 sequential Finish one file before moving on to
676 the next. Multiple files can still be
677 open depending on 'openfiles'.
679 zipf Use a zipfian distribution to decide what file
682 pareto Use a pareto distribution to decide what file
685 gauss Use a gaussian (normal) distribution to decide
688 For random, roundrobin, and sequential, a postfix can be
689 appended to tell fio how many I/Os to issue before switching
690 to a new file. For example, specifying
691 'file_service_type=random:8' would cause fio to issue 8 I/Os
692 before selecting a new file at random. For the non-uniform
693 distributions, a floating point postfix can be given to
694 influence how the distribution is skewed. See
695 'random_distribution' for a description of how that would work.
697 ioengine=str Defines how the job issues io to the file. The following
700 sync Basic read(2) or write(2) io. lseek(2) is
701 used to position the io location.
703 psync Basic pread(2) or pwrite(2) io. Default on all
704 supported operating systems except for Windows.
706 vsync Basic readv(2) or writev(2) IO.
708 pvsync Basic preadv(2) or pwritev(2) IO.
710 pvsync2 Basic preadv2(2) or pwritev2(2) IO.
712 libaio Linux native asynchronous io. Note that Linux
713 may only support queued behaviour with
714 non-buffered IO (set direct=1 or buffered=0).
715 This engine defines engine specific options.
717 posixaio glibc posix asynchronous io.
719 solarisaio Solaris native asynchronous io.
721 windowsaio Windows native asynchronous io.
724 mmap File is memory mapped and data copied
725 to/from using memcpy(3).
727 splice splice(2) is used to transfer the data and
728 vmsplice(2) to transfer data from user
731 sg SCSI generic sg v3 io. May either be
732 synchronous using the SG_IO ioctl, or if
733 the target is an sg character device
734 we use read(2) and write(2) for asynchronous
737 null Doesn't transfer any data, just pretends
738 to. This is mainly used to exercise fio
739 itself and for debugging/testing purposes.
741 net Transfer over the network to given host:port.
742 Depending on the protocol used, the hostname,
743 port, listen and filename options are used to
744 specify what sort of connection to make, while
745 the protocol option determines which protocol
747 This engine defines engine specific options.
749 netsplice Like net, but uses splice/vmsplice to
750 map data and send/receive.
751 This engine defines engine specific options.
753 cpuio Doesn't transfer any data, but burns CPU
754 cycles according to the cpuload= and
755 cpuchunks= options. Setting cpuload=85
756 will cause that job to do nothing but burn
757 85% of the CPU. In case of SMP machines,
758 use numjobs=<no_of_cpu> to get desired CPU
759 usage, as the cpuload only loads a single
760 CPU at the desired rate. A job never finishes
761 unless there is at least one non-cpuio job.
763 guasi The GUASI IO engine is the Generic Userspace
764 Asyncronous Syscall Interface approach
767 http://www.xmailserver.org/guasi-lib.html
769 for more info on GUASI.
771 rdma The RDMA I/O engine supports both RDMA
772 memory semantics (RDMA_WRITE/RDMA_READ) and
773 channel semantics (Send/Recv) for the
774 InfiniBand, RoCE and iWARP protocols.
776 falloc IO engine that does regular fallocate to
777 simulate data transfer as fio ioengine.
778 DDIR_READ does fallocate(,mode = keep_size,)
779 DDIR_WRITE does fallocate(,mode = 0)
780 DDIR_TRIM does fallocate(,mode = punch_hole)
782 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
783 ioctls to simulate defragment activity in
784 request to DDIR_WRITE event
786 rbd IO engine supporting direct access to Ceph
787 Rados Block Devices (RBD) via librbd without
788 the need to use the kernel rbd driver. This
789 ioengine defines engine specific options.
791 gfapi Using Glusterfs libgfapi sync interface to
792 direct access to Glusterfs volumes without
795 gfapi_async Using Glusterfs libgfapi async interface
796 to direct access to Glusterfs volumes without
797 having to go through FUSE. This ioengine
798 defines engine specific options.
800 libhdfs Read and write through Hadoop (HDFS).
801 This engine interprets offsets a little
802 differently. In HDFS, files once created
803 cannot be modified. So random writes are not
804 possible. To imitate this, libhdfs engine
805 creates bunch of small files, and engine will
806 pick a file out of those files based on the
807 offset generated by fio backend. Each jobs uses
808 it's own connection to HDFS.
810 mtd Read, write and erase an MTD character device
811 (e.g., /dev/mtd0). Discards are treated as
812 erases. Depending on the underlying device
813 type, the I/O may have to go in a certain
814 pattern, e.g., on NAND, writing sequentially
815 to erase blocks and discarding before
816 overwriting. The writetrim mode works well
819 pmemblk Read and write through the NVML libpmemblk
822 external Prefix to specify loading an external
823 IO engine object file. Append the engine
824 filename, eg ioengine=external:/tmp/foo.o
825 to load ioengine foo.o in /tmp.
827 iodepth=int This defines how many io units to keep in flight against
828 the file. The default is 1 for each file defined in this
829 job, can be overridden with a larger value for higher
830 concurrency. Note that increasing iodepth beyond 1 will not
831 affect synchronous ioengines (except for small degress when
832 verify_async is in use). Even async engines may impose OS
833 restrictions causing the desired depth not to be achieved.
834 This may happen on Linux when using libaio and not setting
835 direct=1, since buffered IO is not async on that OS. Keep an
836 eye on the IO depth distribution in the fio output to verify
837 that the achieved depth is as expected. Default: 1.
839 iodepth_batch_submit=int
840 iodepth_batch=int This defines how many pieces of IO to submit at once.
841 It defaults to 1 which means that we submit each IO
842 as soon as it is available, but can be raised to submit
843 bigger batches of IO at the time. If it is set to 0 the iodepth
846 iodepth_batch_complete_min=int
847 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
848 at once. It defaults to 1 which means that we'll ask
849 for a minimum of 1 IO in the retrieval process from
850 the kernel. The IO retrieval will go on until we
851 hit the limit set by iodepth_low. If this variable is
852 set to 0, then fio will always check for completed
853 events before queuing more IO. This helps reduce
854 IO latency, at the cost of more retrieval system calls.
856 iodepth_batch_complete_max=int This defines maximum pieces of IO to
857 retrieve at once. This variable should be used along with
858 iodepth_batch_complete_min=int variable, specifying the range
859 of min and max amount of IO which should be retrieved. By default
860 it is equal to iodepth_batch_complete_min value.
864 iodepth_batch_complete_min=1
865 iodepth_batch_complete_max=<iodepth>
867 which means that we will retrieve at least 1 IO and up to the
868 whole submitted queue depth. If none of IO has been completed
873 iodepth_batch_complete_min=0
874 iodepth_batch_complete_max=<iodepth>
876 which means that we can retrieve up to the whole submitted
877 queue depth, but if none of IO has been completed yet, we will
878 NOT wait and immediately exit the system call. In this example
879 we simply do polling.
881 iodepth_low=int The low water mark indicating when to start filling
882 the queue again. Defaults to the same as iodepth, meaning
883 that fio will attempt to keep the queue full at all times.
884 If iodepth is set to eg 16 and iodepth_low is set to 4, then
885 after fio has filled the queue of 16 requests, it will let
886 the depth drain down to 4 before starting to fill it again.
888 io_submit_mode=str This option controls how fio submits the IO to
889 the IO engine. The default is 'inline', which means that the
890 fio job threads submit and reap IO directly. If set to
891 'offload', the job threads will offload IO submission to a
892 dedicated pool of IO threads. This requires some coordination
893 and thus has a bit of extra overhead, especially for lower
894 queue depth IO where it can increase latencies. The benefit
895 is that fio can manage submission rates independently of
896 the device completion rates. This avoids skewed latency
897 reporting if IO gets back up on the device side (the
898 coordinated omission problem).
900 direct=bool If value is true, use non-buffered io. This is usually
901 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
902 On Windows the synchronous ioengines don't support direct io.
904 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
905 writes are guaranteed to be stable once acknowledged by
906 the operating system. Only Linux supports O_ATOMIC right
909 buffered=bool If value is true, use buffered io. This is the opposite
910 of the 'direct' option. Defaults to true.
912 offset=int Start io at the given offset in the file. The data before
913 the given offset will not be touched. This effectively
914 caps the file size at real_size - offset.
916 offset_increment=int If this is provided, then the real offset becomes
917 offset + offset_increment * thread_number, where the thread
918 number is a counter that starts at 0 and is incremented for
919 each sub-job (i.e. when numjobs option is specified). This
920 option is useful if there are several jobs which are intended
921 to operate on a file in parallel disjoint segments, with
922 even spacing between the starting points.
924 number_ios=int Fio will normally perform IOs until it has exhausted the size
925 of the region set by size=, or if it exhaust the allocated
926 time (or hits an error condition). With this setting, the
927 range/size can be set independently of the number of IOs to
928 perform. When fio reaches this number, it will exit normally
929 and report status. Note that this does not extend the amount
930 of IO that will be done, it will only stop fio if this
931 condition is met before other end-of-job criteria.
933 fsync=int If writing to a file, issue a sync of the dirty data
934 for every number of blocks given. For example, if you give
935 32 as a parameter, fio will sync the file for every 32
936 writes issued. If fio is using non-buffered io, we may
937 not sync the file. The exception is the sg io engine, which
938 synchronizes the disk cache anyway.
940 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
942 In FreeBSD and Windows there is no fdatasync(), this falls back
945 sync_file_range=str:val Use sync_file_range() for every 'val' number of
946 write operations. Fio will track range of writes that
947 have happened since the last sync_file_range() call. 'str'
948 can currently be one or more of:
950 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
951 write SYNC_FILE_RANGE_WRITE
952 wait_after SYNC_FILE_RANGE_WAIT_AFTER
954 So if you do sync_file_range=wait_before,write:8, fio would
955 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
956 every 8 writes. Also see the sync_file_range(2) man page.
957 This option is Linux specific.
959 overwrite=bool If true, writes to a file will always overwrite existing
960 data. If the file doesn't already exist, it will be
961 created before the write phase begins. If the file exists
962 and is large enough for the specified write phase, nothing
965 end_fsync=bool If true, fsync file contents when a write stage has completed.
967 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
968 This differs from end_fsync in that it will happen on every
969 file close, not just at the end of the job.
971 rwmixread=int How large a percentage of the mix should be reads.
973 rwmixwrite=int How large a percentage of the mix should be writes. If both
974 rwmixread and rwmixwrite is given and the values do not add
975 up to 100%, the latter of the two will be used to override
976 the first. This may interfere with a given rate setting,
977 if fio is asked to limit reads or writes to a certain rate.
978 If that is the case, then the distribution may be skewed.
980 random_distribution=str:float By default, fio will use a completely uniform
981 random distribution when asked to perform random IO. Sometimes
982 it is useful to skew the distribution in specific ways,
983 ensuring that some parts of the data is more hot than others.
984 fio includes the following distribution models:
986 random Uniform random distribution
987 zipf Zipf distribution
988 pareto Pareto distribution
989 gauss Normal (gaussian) distribution
990 zoned Zoned random distribution
992 When using a zipf or pareto distribution, an input value
993 is also needed to define the access pattern. For zipf, this
994 is the zipf theta. For pareto, it's the pareto power. Fio
995 includes a test program, genzipf, that can be used visualize
996 what the given input values will yield in terms of hit rates.
997 If you wanted to use zipf with a theta of 1.2, you would use
998 random_distribution=zipf:1.2 as the option. If a non-uniform
999 model is used, fio will disable use of the random map. For
1000 the gauss distribution, a normal deviation is supplied as
1001 a value between 0 and 100.
1003 For a zoned distribution, fio supports specifying percentages
1004 of IO access that should fall within what range of the file or
1005 device. For example, given a criteria of:
1007 60% of accesses should be to the first 10%
1008 30% of accesses should be to the next 20%
1009 8% of accesses should be to to the next 30%
1010 2% of accesses should be to the next 40%
1012 we can define that through zoning of the random accesses. For
1013 the above example, the user would do:
1015 random_distribution=zoned:60/10:30/20:8/30:2/40
1017 similarly to how bssplit works for setting ranges and
1018 percentages of block sizes. Like bssplit, it's possible to
1019 specify separate zones for reads, writes, and trims. If just
1020 one set is given, it'll apply to all of them.
1022 percentage_random=int For a random workload, set how big a percentage should
1023 be random. This defaults to 100%, in which case the workload
1024 is fully random. It can be set from anywhere from 0 to 100.
1025 Setting it to 0 would make the workload fully sequential. Any
1026 setting in between will result in a random mix of sequential
1027 and random IO, at the given percentages. It is possible to
1028 set different values for reads, writes, and trim. To do so,
1029 simply use a comma separated list. See blocksize.
1031 norandommap Normally fio will cover every block of the file when doing
1032 random IO. If this option is given, fio will just get a
1033 new random offset without looking at past io history. This
1034 means that some blocks may not be read or written, and that
1035 some blocks may be read/written more than once. If this option
1036 is used with verify= and multiple blocksizes (via bsrange=),
1037 only intact blocks are verified, i.e., partially-overwritten
1040 softrandommap=bool See norandommap. If fio runs with the random block map
1041 enabled and it fails to allocate the map, if this option is
1042 set it will continue without a random block map. As coverage
1043 will not be as complete as with random maps, this option is
1044 disabled by default.
1046 random_generator=str Fio supports the following engines for generating
1047 IO offsets for random IO:
1049 tausworthe Strong 2^88 cycle random number generator
1050 lfsr Linear feedback shift register generator
1051 tausworthe64 Strong 64-bit 2^258 cycle random number
1054 Tausworthe is a strong random number generator, but it
1055 requires tracking on the side if we want to ensure that
1056 blocks are only read or written once. LFSR guarantees
1057 that we never generate the same offset twice, and it's
1058 also less computationally expensive. It's not a true
1059 random generator, however, though for IO purposes it's
1060 typically good enough. LFSR only works with single
1061 block sizes, not with workloads that use multiple block
1062 sizes. If used with such a workload, fio may read or write
1063 some blocks multiple times. The default value is tausworthe,
1064 unless the required space exceeds 2^32 blocks. If it does,
1065 then tausworthe64 is selected automatically.
1067 nice=int Run the job with the given nice value. See man nice(2).
1069 On Windows, values less than -15 set the process class to "High";
1070 -1 through -15 set "Above Normal"; 1 through 15 "Below Normal";
1071 and above 15 "Idle" priority class.
1073 prio=int Set the io priority value of this job. Linux limits us to
1074 a positive value between 0 and 7, with 0 being the highest.
1075 See man ionice(1). Refer to an appropriate manpage for
1076 other operating systems since meaning of priority may differ.
1078 prioclass=int Set the io priority class. See man ionice(1).
1080 thinktime=int Stall the job x microseconds after an io has completed before
1081 issuing the next. May be used to simulate processing being
1082 done by an application. See thinktime_blocks and
1086 Only valid if thinktime is set - pretend to spend CPU time
1087 doing something with the data received, before falling back
1088 to sleeping for the rest of the period specified by
1091 thinktime_blocks=int
1092 Only valid if thinktime is set - control how many blocks
1093 to issue, before waiting 'thinktime' usecs. If not set,
1094 defaults to 1 which will make fio wait 'thinktime' usecs
1095 after every block. This effectively makes any queue depth
1096 setting redundant, since no more than 1 IO will be queued
1097 before we have to complete it and do our thinktime. In
1098 other words, this setting effectively caps the queue depth
1099 if the latter is larger.
1101 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
1102 the normal suffix rules apply. You can use rate=500k to limit
1103 reads and writes to 500k each, or you can specify read and
1104 writes separately. Using rate=1m,500k would limit reads to
1105 1MB/sec and writes to 500KB/sec. Capping only reads or
1106 writes can be done with rate=,500k or rate=500k,. The former
1107 will only limit writes (to 500KB/sec), the latter will only
1110 rate_min=int Tell fio to do whatever it can to maintain at least this
1111 bandwidth. Failing to meet this requirement, will cause
1112 the job to exit. The same format as rate is used for
1113 read vs write separation.
1115 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
1116 as rate, just specified independently of bandwidth. If the
1117 job is given a block size range instead of a fixed value,
1118 the smallest block size is used as the metric. The same format
1119 as rate is used for read vs write separation.
1121 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
1122 the job to exit. The same format as rate is used for read vs
1125 rate_process=str This option controls how fio manages rated IO
1126 submissions. The default is 'linear', which submits IO in a
1127 linear fashion with fixed delays between IOs that gets
1128 adjusted based on IO completion rates. If this is set to
1129 'poisson', fio will submit IO based on a more real world
1130 random request flow, known as the Poisson process
1131 (https://en.wikipedia.org/wiki/Poisson_process). The lambda
1132 will be 10^6 / IOPS for the given workload.
1134 latency_target=int If set, fio will attempt to find the max performance
1135 point that the given workload will run at while maintaining a
1136 latency below this target. The values is given in microseconds.
1137 See latency_window and latency_percentile
1139 latency_window=int Used with latency_target to specify the sample window
1140 that the job is run at varying queue depths to test the
1141 performance. The value is given in microseconds.
1143 latency_percentile=float The percentage of IOs that must fall within the
1144 criteria specified by latency_target and latency_window. If not
1145 set, this defaults to 100.0, meaning that all IOs must be equal
1146 or below to the value set by latency_target.
1148 max_latency=int If set, fio will exit the job if it exceeds this maximum
1149 latency. It will exit with an ETIME error.
1151 rate_cycle=int Average bandwidth for 'rate' and 'rate_min' over this number
1154 cpumask=int Set the CPU affinity of this job. The parameter given is a
1155 bitmask of allowed CPU's the job may run on. So if you want
1156 the allowed CPUs to be 1 and 5, you would pass the decimal
1157 value of (1 << 1 | 1 << 5), or 34. See man
1158 sched_setaffinity(2). This may not work on all supported
1159 operating systems or kernel versions. This option doesn't
1160 work well for a higher CPU count than what you can store in
1161 an integer mask, so it can only control cpus 1-32. For
1162 boxes with larger CPU counts, use cpus_allowed.
1164 cpus_allowed=str Controls the same options as cpumask, but it allows a text
1165 setting of the permitted CPUs instead. So to use CPUs 1 and
1166 5, you would specify cpus_allowed=1,5. This options also
1167 allows a range of CPUs. Say you wanted a binding to CPUs
1168 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
1170 cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
1171 specified by cpus_allowed or cpumask. Two policies are
1174 shared All jobs will share the CPU set specified.
1175 split Each job will get a unique CPU from the CPU set.
1177 'shared' is the default behaviour, if the option isn't
1178 specified. If split is specified, then fio will will assign
1179 one cpu per job. If not enough CPUs are given for the jobs
1180 listed, then fio will roundrobin the CPUs in the set.
1182 numa_cpu_nodes=str Set this job running on specified NUMA nodes' CPUs. The
1183 arguments allow comma delimited list of cpu numbers,
1184 A-B ranges, or 'all'. Note, to enable numa options support,
1185 fio must be built on a system with libnuma-dev(el) installed.
1187 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
1188 nodes. Format of the arguments:
1190 `mode' is one of the following memory policy:
1191 default, prefer, bind, interleave, local
1192 For `default' and `local' memory policy, no node is
1193 needed to be specified.
1194 For `prefer', only one node is allowed.
1195 For `bind' and `interleave', it allow comma delimited
1196 list of numbers, A-B ranges, or 'all'.
1198 startdelay=time Start this job the specified number of seconds after fio
1199 has started. Only useful if the job file contains several
1200 jobs, and you want to delay starting some jobs to a certain
1203 runtime=time Tell fio to terminate processing after the specified number
1204 of seconds. It can be quite hard to determine for how long
1205 a specified job will run, so this parameter is handy to
1206 cap the total runtime to a given time.
1208 time_based If set, fio will run for the duration of the runtime
1209 specified even if the file(s) are completely read or
1210 written. It will simply loop over the same workload
1211 as many times as the runtime allows.
1213 ramp_time=time If set, fio will run the specified workload for this amount
1214 of time before logging any performance numbers. Useful for
1215 letting performance settle before logging results, thus
1216 minimizing the runtime required for stable results. Note
1217 that the ramp_time is considered lead in time for a job,
1218 thus it will increase the total runtime if a special timeout
1219 or runtime is specified.
1221 invalidate=bool Invalidate the buffer/page cache parts for this file prior
1222 to starting io. Defaults to true.
1224 sync=bool Use sync io for buffered writes. For the majority of the
1225 io engines, this means using O_SYNC.
1228 mem=str Fio can use various types of memory as the io unit buffer.
1229 The allowed values are:
1231 malloc Use memory from malloc(3) as the buffers.
1232 Default memory type.
1234 shm Use shared memory as the buffers. Allocated
1237 shmhuge Same as shm, but use huge pages as backing.
1239 mmap Use mmap to allocate buffers. May either be
1240 anonymous memory, or can be file backed if
1241 a filename is given after the option. The
1242 format is mem=mmap:/path/to/file.
1244 mmaphuge Use a memory mapped huge file as the buffer
1245 backing. Append filename after mmaphuge, ala
1246 mem=mmaphuge:/hugetlbfs/file
1248 mmapshared Same as mmap, but use a MMAP_SHARED
1251 The area allocated is a function of the maximum allowed
1252 bs size for the job, multiplied by the io depth given. Note
1253 that for shmhuge and mmaphuge to work, the system must have
1254 free huge pages allocated. This can normally be checked
1255 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1256 Linux system. Fio assumes a huge page is 4MB in size. So
1257 to calculate the number of huge pages you need for a given
1258 job file, add up the io depth of all jobs (normally one unless
1259 iodepth= is used) and multiply by the maximum bs set. Then
1260 divide that number by the huge page size. You can see the
1261 size of the huge pages in /proc/meminfo. If no huge pages
1262 are allocated by having a non-zero number in nr_hugepages,
1263 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1265 mmaphuge also needs to have hugetlbfs mounted and the file
1266 location should point there. So if it's mounted in /huge,
1267 you would use mem=mmaphuge:/huge/somefile.
1269 iomem_align=int This indicates the memory alignment of the IO memory buffers.
1270 Note that the given alignment is applied to the first IO unit
1271 buffer, if using iodepth the alignment of the following buffers
1272 are given by the bs used. In other words, if using a bs that is
1273 a multiple of the page sized in the system, all buffers will
1274 be aligned to this value. If using a bs that is not page
1275 aligned, the alignment of subsequent IO memory buffers is the
1276 sum of the iomem_align and bs used.
1279 Defines the size of a huge page. Must at least be equal
1280 to the system setting, see /proc/meminfo. Defaults to 4MB.
1281 Should probably always be a multiple of megabytes, so using
1282 hugepage-size=Xm is the preferred way to set this to avoid
1283 setting a non-pow-2 bad value.
1285 exitall When one job finishes, terminate the rest. The default is
1286 to wait for each job to finish, sometimes that is not the
1289 exitall_on_error When one job finishes in error, terminate the rest. The
1290 default is to wait for each job to finish.
1292 bwavgtime=int Average the calculated bandwidth over the given time. Value
1293 is specified in milliseconds. If the job also does bandwidth
1294 logging through 'write_bw_log', then the minimum of this option
1295 and 'log_avg_msec' will be used. Default: 500ms.
1297 iopsavgtime=int Average the calculated IOPS over the given time. Value
1298 is specified in milliseconds. If the job also does IOPS logging
1299 through 'write_iops_log', then the minimum of this option and
1300 'log_avg_msec' will be used. Default: 500ms.
1302 create_serialize=bool If true, serialize the file creation for the jobs.
1303 This may be handy to avoid interleaving of data
1304 files, which may greatly depend on the filesystem
1305 used and even the number of processors in the system.
1307 create_fsync=bool fsync the data file after creation. This is the
1310 create_on_open=bool Don't pre-setup the files for IO, just create open()
1311 when it's time to do IO to that file.
1313 create_only=bool If true, fio will only run the setup phase of the job.
1314 If files need to be laid out or updated on disk, only
1315 that will be done. The actual job contents are not
1318 allow_file_create=bool If true, fio is permitted to create files as part
1319 of its workload. This is the default behavior. If this
1320 option is false, then fio will error out if the files it
1321 needs to use don't already exist. Default: true.
1323 allow_mounted_write=bool If this isn't set, fio will abort jobs that
1324 are destructive (eg that write) to what appears to be a
1325 mounted device or partition. This should help catch creating
1326 inadvertently destructive tests, not realizing that the test
1327 will destroy data on the mounted file system. Default: false.
1329 pre_read=bool If this is given, files will be pre-read into memory before
1330 starting the given IO operation. This will also clear
1331 the 'invalidate' flag, since it is pointless to pre-read
1332 and then drop the cache. This will only work for IO engines
1333 that are seek-able, since they allow you to read the same data
1334 multiple times. Thus it will not work on eg network or splice
1337 unlink=bool Unlink the job files when done. Not the default, as repeated
1338 runs of that job would then waste time recreating the file
1339 set again and again.
1341 unlink_each_loop=bool Unlink job files after each iteration or loop.
1343 loops=int Run the specified number of iterations of this job. Used
1344 to repeat the same workload a given number of times. Defaults
1347 verify_only Do not perform specified workload---only verify data still
1348 matches previous invocation of this workload. This option
1349 allows one to check data multiple times at a later date
1350 without overwriting it. This option makes sense only for
1351 workloads that write data, and does not support workloads
1352 with the time_based option set.
1354 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1355 verify is set. Defaults to 1.
1357 verify=str If writing to a file, fio can verify the file contents
1358 after each iteration of the job. Each verification method also implies
1359 verification of special header, which is written to the beginning of
1360 each block. This header also includes meta information, like offset
1361 of the block, block number, timestamp when block was written, etc.
1362 verify=str can be combined with verify_pattern=str option.
1363 The allowed values are:
1365 md5 Use an md5 sum of the data area and store
1366 it in the header of each block.
1368 crc64 Use an experimental crc64 sum of the data
1369 area and store it in the header of each
1372 crc32c Use a crc32c sum of the data area and store
1373 it in the header of each block.
1375 crc32c-intel Use hardware assisted crc32c calculation
1376 provided on SSE4.2 enabled processors. Falls
1377 back to regular software crc32c, if not
1378 supported by the system.
1380 crc32 Use a crc32 sum of the data area and store
1381 it in the header of each block.
1383 crc16 Use a crc16 sum of the data area and store
1384 it in the header of each block.
1386 crc7 Use a crc7 sum of the data area and store
1387 it in the header of each block.
1389 xxhash Use xxhash as the checksum function. Generally
1390 the fastest software checksum that fio
1393 sha512 Use sha512 as the checksum function.
1395 sha256 Use sha256 as the checksum function.
1397 sha1 Use optimized sha1 as the checksum function.
1399 meta This option is deprecated, since now meta information is
1400 included in generic verification header and meta verification
1401 happens by default. For detailed information see the description
1402 of the verify=str setting. This option is kept because of
1403 compatibility's sake with old configurations. Do not use it.
1405 pattern Verify a strict pattern. Normally fio includes
1406 a header with some basic information and
1407 checksumming, but if this option is set, only
1408 the specific pattern set with 'verify_pattern'
1411 null Only pretend to verify. Useful for testing
1412 internals with ioengine=null, not for much
1415 This option can be used for repeated burn-in tests of a
1416 system to make sure that the written data is also
1417 correctly read back. If the data direction given is
1418 a read or random read, fio will assume that it should
1419 verify a previously written file. If the data direction
1420 includes any form of write, the verify will be of the
1423 verifysort=bool If set, fio will sort written verify blocks when it deems
1424 it faster to read them back in a sorted manner. This is
1425 often the case when overwriting an existing file, since
1426 the blocks are already laid out in the file system. You
1427 can ignore this option unless doing huge amounts of really
1428 fast IO where the red-black tree sorting CPU time becomes
1431 verify_offset=int Swap the verification header with data somewhere else
1432 in the block before writing. Its swapped back before
1435 verify_interval=int Write the verification header at a finer granularity
1436 than the blocksize. It will be written for chunks the
1437 size of header_interval. blocksize should divide this
1440 verify_pattern=str If set, fio will fill the io buffers with this
1441 pattern. Fio defaults to filling with totally random
1442 bytes, but sometimes it's interesting to fill with a known
1443 pattern for io verification purposes. Depending on the
1444 width of the pattern, fio will fill 1/2/3/4 bytes of the
1445 buffer at the time(it can be either a decimal or a hex number).
1446 The verify_pattern if larger than a 32-bit quantity has to
1447 be a hex number that starts with either "0x" or "0X". Use
1448 with verify=str. Also, verify_pattern supports %o format,
1449 which means that for each block offset will be written and
1450 then verified back, e.g.:
1454 Or use combination of everything:
1455 verify_pattern=0xff%o"abcd"-12
1457 verify_fatal=bool Normally fio will keep checking the entire contents
1458 before quitting on a block verification failure. If this
1459 option is set, fio will exit the job on the first observed
1462 verify_dump=bool If set, dump the contents of both the original data
1463 block and the data block we read off disk to files. This
1464 allows later analysis to inspect just what kind of data
1465 corruption occurred. Off by default.
1467 verify_async=int Fio will normally verify IO inline from the submitting
1468 thread. This option takes an integer describing how many
1469 async offload threads to create for IO verification instead,
1470 causing fio to offload the duty of verifying IO contents
1471 to one or more separate threads. If using this offload
1472 option, even sync IO engines can benefit from using an
1473 iodepth setting higher than 1, as it allows them to have
1474 IO in flight while verifies are running.
1476 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1477 async IO verification threads. See cpus_allowed for the
1480 verify_backlog=int Fio will normally verify the written contents of a
1481 job that utilizes verify once that job has completed. In
1482 other words, everything is written then everything is read
1483 back and verified. You may want to verify continually
1484 instead for a variety of reasons. Fio stores the meta data
1485 associated with an IO block in memory, so for large
1486 verify workloads, quite a bit of memory would be used up
1487 holding this meta data. If this option is enabled, fio
1488 will write only N blocks before verifying these blocks.
1490 verify_backlog_batch=int Control how many blocks fio will verify
1491 if verify_backlog is set. If not set, will default to
1492 the value of verify_backlog (meaning the entire queue
1493 is read back and verified). If verify_backlog_batch is
1494 less than verify_backlog then not all blocks will be verified,
1495 if verify_backlog_batch is larger than verify_backlog, some
1496 blocks will be verified more than once.
1498 verify_state_save=bool When a job exits during the write phase of a verify
1499 workload, save its current state. This allows fio to replay
1500 up until that point, if the verify state is loaded for the
1501 verify read phase. The format of the filename is, roughly,
1502 <type>-<jobname>-<jobindex>-verify.state. <type> is "local"
1503 for a local run, "sock" for a client/server socket connection,
1504 and "ip" (192.168.0.1, for instance) for a networked
1505 client/server connection.
1507 verify_state_load=bool If a verify termination trigger was used, fio stores
1508 the current write state of each thread. This can be used at
1509 verification time so that fio knows how far it should verify.
1510 Without this information, fio will run a full verification
1511 pass, according to the settings in the job file used.
1514 wait_for_previous Wait for preceding jobs in the job file to exit, before
1515 starting this one. Can be used to insert serialization
1516 points in the job file. A stone wall also implies starting
1517 a new reporting group.
1519 new_group Start a new reporting group. See: group_reporting.
1521 numjobs=int Create the specified number of clones of this job. May be
1522 used to setup a larger number of threads/processes doing
1523 the same thing. Each thread is reported separately; to see
1524 statistics for all clones as a whole, use group_reporting in
1525 conjunction with new_group.
1527 group_reporting It may sometimes be interesting to display statistics for
1528 groups of jobs as a whole instead of for each individual job.
1529 This is especially true if 'numjobs' is used; looking at
1530 individual thread/process output quickly becomes unwieldy.
1531 To see the final report per-group instead of per-job, use
1532 'group_reporting'. Jobs in a file will be part of the same
1533 reporting group, unless if separated by a stonewall, or by
1536 thread fio defaults to forking jobs, however if this option is
1537 given, fio will use pthread_create(3) to create threads
1540 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1542 zoneskip=int Skip the specified number of bytes when zonesize data has
1543 been read. The two zone options can be used to only do
1544 io on zones of a file.
1546 write_iolog=str Write the issued io patterns to the specified file. See
1547 read_iolog. Specify a separate file for each job, otherwise
1548 the iologs will be interspersed and the file may be corrupt.
1550 read_iolog=str Open an iolog with the specified file name and replay the
1551 io patterns it contains. This can be used to store a
1552 workload and replay it sometime later. The iolog given
1553 may also be a blktrace binary file, which allows fio
1554 to replay a workload captured by blktrace. See blktrace
1555 for how to capture such logging data. For blktrace replay,
1556 the file needs to be turned into a blkparse binary data
1557 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1559 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1560 is to attempt to respect the time stamps within the log and
1561 replay them with the appropriate delay between IOPS. By
1562 setting this variable fio will not respect the timestamps and
1563 attempt to replay them as fast as possible while still
1564 respecting ordering. The result is the same I/O pattern to a
1565 given device, but different timings.
1567 replay_redirect=str While replaying I/O patterns using read_iolog the
1568 default behavior is to replay the IOPS onto the major/minor
1569 device that each IOP was recorded from. This is sometimes
1570 undesirable because on a different machine those major/minor
1571 numbers can map to a different device. Changing hardware on
1572 the same system can also result in a different major/minor
1573 mapping. Replay_redirect causes all IOPS to be replayed onto
1574 the single specified device regardless of the device it was
1575 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1576 IO in the blktrace or iolog to be replayed onto /dev/sdc.
1577 This means multiple devices will be replayed onto a single
1578 device, if the trace contains multiple devices. If you want
1579 multiple devices to be replayed concurrently to multiple
1580 redirected devices you must blkparse your trace into separate
1581 traces and replay them with independent fio invocations.
1582 Unfortunately this also breaks the strict time ordering
1583 between multiple device accesses.
1585 replay_align=int Force alignment of IO offsets and lengths in a trace
1586 to this power of 2 value.
1588 replay_scale=int Scale sector offsets down by this factor when
1591 per_job_logs=bool If set, this generates bw/clat/iops log with per
1592 file private filenames. If not set, jobs with identical names
1593 will share the log filename. Default: true.
1595 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1596 file. Can be used to store data of the bandwidth of the
1597 jobs in their lifetime. The included fio_generate_plots
1598 script uses gnuplot to turn these text files into nice
1599 graphs. See write_lat_log for behaviour of given
1600 filename. For this option, the suffix is _bw.x.log, where
1601 x is the index of the job (1..N, where N is the number of
1602 jobs). If 'per_job_logs' is false, then the filename will not
1603 include the job index. See 'Log File Formats'.
1605 write_lat_log=str Same as write_bw_log, except that this option stores io
1606 submission, completion, and total latencies instead. If no
1607 filename is given with this option, the default filename of
1608 "jobname_type.log" is used. Even if the filename is given,
1609 fio will still append the type of log. So if one specifies
1613 The actual log names will be foo_slat.x.log, foo_clat.x.log,
1614 and foo_lat.x.log, where x is the index of the job (1..N,
1615 where N is the number of jobs). This helps fio_generate_plot
1616 find the logs automatically. If 'per_job_logs' is false, then
1617 the filename will not include the job index. See 'Log File
1620 write_hist_log=str Same as write_lat_log, but writes I/O completion
1621 latency histograms. If no filename is given with this option, the
1622 default filename of "jobname_clat_hist.x.log" is used, where x is
1623 the index of the job (1..N, where N is the number of jobs). Even
1624 if the filename is given, fio will still append the type of log.
1625 If per_job_logs is false, then the filename will not include the
1626 job index. See 'Log File Formats'.
1628 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1629 given with this option, the default filename of
1630 "jobname_type.x.log" is used,where x is the index of the job
1631 (1..N, where N is the number of jobs). Even if the filename
1632 is given, fio will still append the type of log. If
1633 'per_job_logs' is false, then the filename will not include
1634 the job index. See 'Log File Formats'.
1636 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1637 or bw log for every IO that completes. When writing to the
1638 disk log, that can quickly grow to a very large size. Setting
1639 this option makes fio average the each log entry over the
1640 specified period of time, reducing the resolution of the log.
1641 See log_max_value as well. Defaults to 0, logging all entries.
1643 log_hist_msec=int Same as log_avg_msec, but logs entries for completion
1644 latency histograms. Computing latency percentiles from averages of
1645 intervals using log_avg_msec is innacurate. Setting this option makes
1646 fio log histogram entries over the specified period of time, reducing
1647 log sizes for high IOPS devices while retaining percentile accuracy.
1648 See log_hist_coarseness as well. Defaults to 0, meaning histogram
1649 logging is disabled.
1651 log_hist_coarseness=int Integer ranging from 0 to 6, defining the coarseness
1652 of the resolution of the histogram logs enabled with log_hist_msec. For
1653 each increment in coarseness, fio outputs half as many bins. Defaults to
1654 0, for which histogram logs contain 1216 latency bins. See
1657 log_max_value=bool If log_avg_msec is set, fio logs the average over that
1658 window. If you instead want to log the maximum value, set this
1659 option to 1. Defaults to 0, meaning that averaged values are
1662 log_offset=int If this is set, the iolog options will include the byte
1663 offset for the IO entry as well as the other data values.
1665 log_compression=int If this is set, fio will compress the IO logs as
1666 it goes, to keep the memory footprint lower. When a log
1667 reaches the specified size, that chunk is removed and
1668 compressed in the background. Given that IO logs are
1669 fairly highly compressible, this yields a nice memory
1670 savings for longer runs. The downside is that the
1671 compression will consume some background CPU cycles, so
1672 it may impact the run. This, however, is also true if
1673 the logging ends up consuming most of the system memory.
1674 So pick your poison. The IO logs are saved normally at the
1675 end of a run, by decompressing the chunks and storing them
1676 in the specified log file. This feature depends on the
1677 availability of zlib.
1679 log_compression_cpus=str Define the set of CPUs that are allowed to
1680 handle online log compression for the IO jobs. This can
1681 provide better isolation between performance sensitive jobs,
1682 and background compression work.
1684 log_store_compressed=bool If set, fio will store the log files in a
1685 compressed format. They can be decompressed with fio, using
1686 the --inflate-log command line parameter. The files will be
1687 stored with a .fz suffix.
1689 log_unix_epoch=bool If set, fio will log Unix timestamps to the log
1690 files produced by enabling write_type_log for each log type, instead
1691 of the default zero-based timestamps.
1693 block_error_percentiles=bool If set, record errors in trim block-sized
1694 units from writes and trims and output a histogram of
1695 how many trims it took to get to errors, and what kind
1696 of error was encountered.
1698 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1699 potentially be used instead of removing memory or booting
1700 with less memory to simulate a smaller amount of memory.
1701 The amount specified is per worker.
1703 exec_prerun=str Before running this job, issue the command specified
1704 through system(3). Output is redirected in a file called
1707 exec_postrun=str After the job completes, issue the command specified
1708 though system(3). Output is redirected in a file called
1709 jobname.postrun.txt.
1711 ioscheduler=str Attempt to switch the device hosting the file to the specified
1712 io scheduler before running.
1714 disk_util=bool Generate disk utilization statistics, if the platform
1715 supports it. Defaults to on.
1717 disable_lat=bool Disable measurements of total latency numbers. Useful
1718 only for cutting back the number of calls to gettimeofday,
1719 as that does impact performance at really high IOPS rates.
1720 Note that to really get rid of a large amount of these
1721 calls, this option must be used with disable_slat and
1724 disable_clat=bool Disable measurements of completion latency numbers. See
1727 disable_slat=bool Disable measurements of submission latency numbers. See
1730 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1733 clat_percentiles=bool Enable the reporting of percentiles of
1734 completion latencies.
1736 percentile_list=float_list Overwrite the default list of percentiles
1737 for completion latencies and the block error histogram.
1738 Each number is a floating number in the range (0,100],
1739 and the maximum length of the list is 20. Use ':'
1740 to separate the numbers, and list the numbers in ascending
1741 order. For example, --percentile_list=99.5:99.9 will cause
1742 fio to report the values of completion latency below which
1743 99.5% and 99.9% of the observed latencies fell, respectively.
1745 clocksource=str Use the given clocksource as the base of timing. The
1746 supported options are:
1748 gettimeofday gettimeofday(2)
1750 clock_gettime clock_gettime(2)
1752 cpu Internal CPU clock source
1754 cpu is the preferred clocksource if it is reliable, as it
1755 is very fast (and fio is heavy on time calls). Fio will
1756 automatically use this clocksource if it's supported and
1757 considered reliable on the system it is running on, unless
1758 another clocksource is specifically set. For x86/x86-64 CPUs,
1759 this means supporting TSC Invariant.
1761 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1762 (disable_clat, disable_slat, disable_bw) plus reduce
1763 precision of the timeout somewhat to really shrink
1764 the gettimeofday() call count. With this option enabled,
1765 we only do about 0.4% of the gtod() calls we would have
1766 done if all time keeping was enabled.
1768 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1769 execution to just getting the current time. Fio (and
1770 databases, for instance) are very intensive on gettimeofday()
1771 calls. With this option, you can set one CPU aside for
1772 doing nothing but logging current time to a shared memory
1773 location. Then the other threads/processes that run IO
1774 workloads need only copy that segment, instead of entering
1775 the kernel with a gettimeofday() call. The CPU set aside
1776 for doing these time calls will be excluded from other
1777 uses. Fio will manually clear it from the CPU mask of other
1780 continue_on_error=str Normally fio will exit the job on the first observed
1781 failure. If this option is set, fio will continue the job when
1782 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1783 is exceeded or the I/O size specified is completed. If this
1784 option is used, there are two more stats that are appended,
1785 the total error count and the first error. The error field
1786 given in the stats is the first error that was hit during the
1789 The allowed values are:
1791 none Exit on any IO or verify errors.
1793 read Continue on read errors, exit on all others.
1795 write Continue on write errors, exit on all others.
1797 io Continue on any IO error, exit on all others.
1799 verify Continue on verify errors, exit on all others.
1801 all Continue on all errors.
1803 0 Backward-compatible alias for 'none'.
1805 1 Backward-compatible alias for 'all'.
1807 ignore_error=str Sometimes you want to ignore some errors during test
1808 in that case you can specify error list for each error type.
1809 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1810 errors for given error type is separated with ':'. Error
1811 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1813 ignore_error=EAGAIN,ENOSPC:122
1814 This option will ignore EAGAIN from READ, and ENOSPC and
1815 122(EDQUOT) from WRITE.
1817 error_dump=bool If set dump every error even if it is non fatal, true
1818 by default. If disabled only fatal error will be dumped
1820 cgroup=str Add job to this control group. If it doesn't exist, it will
1821 be created. The system must have a mounted cgroup blkio
1822 mount point for this to work. If your system doesn't have it
1823 mounted, you can do so with:
1825 # mount -t cgroup -o blkio none /cgroup
1827 cgroup_weight=int Set the weight of the cgroup to this value. See
1828 the documentation that comes with the kernel, allowed values
1829 are in the range of 100..1000.
1831 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1832 the job completion. To override this behavior and to leave
1833 cgroups around after the job completion, set cgroup_nodelete=1.
1834 This can be useful if one wants to inspect various cgroup
1835 files after job completion. Default: false
1837 uid=int Instead of running as the invoking user, set the user ID to
1838 this value before the thread/process does any work.
1840 gid=int Set group ID, see uid.
1842 flow_id=int The ID of the flow. If not specified, it defaults to being a
1843 global flow. See flow.
1845 flow=int Weight in token-based flow control. If this value is used, then
1846 there is a 'flow counter' which is used to regulate the
1847 proportion of activity between two or more jobs. fio attempts
1848 to keep this flow counter near zero. The 'flow' parameter
1849 stands for how much should be added or subtracted to the flow
1850 counter on each iteration of the main I/O loop. That is, if
1851 one job has flow=8 and another job has flow=-1, then there
1852 will be a roughly 1:8 ratio in how much one runs vs the other.
1854 flow_watermark=int The maximum value that the absolute value of the flow
1855 counter is allowed to reach before the job must wait for a
1856 lower value of the counter.
1858 flow_sleep=int The period of time, in microseconds, to wait after the flow
1859 watermark has been exceeded before retrying operations
1861 In addition, there are some parameters which are only valid when a specific
1862 ioengine is in use. These are used identically to normal parameters, with the
1863 caveat that when used on the command line, they must come after the ioengine
1864 that defines them is selected.
1866 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1867 the io_getevents system call to reap newly returned events.
1868 With this flag turned on, the AIO ring will be read directly
1869 from user-space to reap events. The reaping mode is only
1870 enabled when polling for a minimum of 0 events (eg when
1871 iodepth_batch_complete=0).
1873 [psyncv2] hipri Set RWF_HIPRI on IO, indicating to the kernel that
1874 it's of higher priority than normal.
1876 [cpuio] cpuload=int Attempt to use the specified percentage of CPU cycles.
1878 [cpuio] cpuchunks=int Split the load into cycles of the given time. In
1881 [cpuio] exit_on_io_done=bool Detect when IO threads are done, then exit.
1883 [netsplice] hostname=str
1884 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1885 If the job is a TCP listener or UDP reader, the hostname is not
1886 used and must be omitted unless it is a valid UDP multicast
1888 [libhdfs] namenode=str The host name or IP address of a HDFS cluster namenode to contact.
1890 [netsplice] port=int
1891 [net] port=int The TCP or UDP port to bind to or connect to. If this is used
1892 with numjobs to spawn multiple instances of the same job type, then this will
1893 be the starting port number since fio will use a range of ports.
1894 [libhdfs] port=int the listening port of the HFDS cluster namenode.
1896 [netsplice] interface=str
1897 [net] interface=str The IP address of the network interface used to send or
1898 receive UDP multicast
1901 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1904 [netsplice] nodelay=bool
1905 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1907 [netsplice] protocol=str
1908 [netsplice] proto=str
1910 [net] proto=str The network protocol to use. Accepted values are:
1912 tcp Transmission control protocol
1913 tcpv6 Transmission control protocol V6
1914 udp User datagram protocol
1915 udpv6 User datagram protocol V6
1916 unix UNIX domain socket
1918 When the protocol is TCP or UDP, the port must also be given,
1919 as well as the hostname if the job is a TCP listener or UDP
1920 reader. For unix sockets, the normal filename option should be
1921 used and the port is invalid.
1923 [net] listen For TCP network connections, tell fio to listen for incoming
1924 connections rather than initiating an outgoing connection. The
1925 hostname must be omitted if this option is used.
1927 [net] pingpong Normally a network writer will just continue writing data, and
1928 a network reader will just consume packages. If pingpong=1
1929 is set, a writer will send its normal payload to the reader,
1930 then wait for the reader to send the same payload back. This
1931 allows fio to measure network latencies. The submission
1932 and completion latencies then measure local time spent
1933 sending or receiving, and the completion latency measures
1934 how long it took for the other end to receive and send back.
1935 For UDP multicast traffic pingpong=1 should only be set for a
1936 single reader when multiple readers are listening to the same
1939 [net] window_size Set the desired socket buffer size for the connection.
1941 [net] mss Set the TCP maximum segment size (TCP_MAXSEG).
1943 [e4defrag] donorname=str
1944 File will be used as a block donor(swap extents between files)
1945 [e4defrag] inplace=int
1946 Configure donor file blocks allocation strategy
1947 0(default): Preallocate donor's file on init
1948 1 : allocate space immediately inside defragment event,
1949 and free right after event
1951 [rbd] clustername=str Specifies the name of the Ceph cluster.
1952 [rbd] rbdname=str Specifies the name of the RBD.
1953 [rbd] pool=str Specifies the name of the Ceph pool containing RBD.
1954 [rbd] clientname=str Specifies the username (without the 'client.' prefix)
1955 used to access the Ceph cluster. If the clustername is
1956 specified, the clientname shall be the full type.id
1957 string. If no type. prefix is given, fio will add
1958 'client.' by default.
1960 [mtd] skip_bad=bool Skip operations against known bad blocks.
1962 [libhdfs] hdfsdirectory libhdfs will create chunk in this HDFS directory
1963 [libhdfs] chunk_size the size of the chunk to use for each file.
1966 6.0 Interpreting the output
1967 ---------------------------
1969 fio spits out a lot of output. While running, fio will display the
1970 status of the jobs created. An example of that would be:
1972 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1974 The characters inside the square brackets denote the current status of
1975 each thread. The possible values (in typical life cycle order) are:
1979 P Thread setup, but not started.
1981 I Thread initialized, waiting or generating necessary data.
1982 p Thread running pre-reading file(s).
1983 R Running, doing sequential reads.
1984 r Running, doing random reads.
1985 W Running, doing sequential writes.
1986 w Running, doing random writes.
1987 M Running, doing mixed sequential reads/writes.
1988 m Running, doing mixed random reads/writes.
1989 F Running, currently waiting for fsync()
1990 f Running, finishing up (writing IO logs, etc)
1991 V Running, doing verification of written data.
1992 E Thread exited, not reaped by main thread yet.
1994 X Thread reaped, exited with an error.
1995 K Thread reaped, exited due to signal.
1997 Fio will condense the thread string as not to take up more space on the
1998 command line as is needed. For instance, if you have 10 readers and 10
1999 writers running, the output would look like this:
2001 Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s]
2003 Fio will still maintain the ordering, though. So the above means that jobs
2004 1..10 are readers, and 11..20 are writers.
2006 The other values are fairly self explanatory - number of threads
2007 currently running and doing io, rate of io since last check (read speed
2008 listed first, then write speed), and the estimated completion percentage
2009 and time for the running group. It's impossible to estimate runtime of
2010 the following groups (if any). Note that the string is displayed in order,
2011 so it's possible to tell which of the jobs are currently doing what. The
2012 first character is the first job defined in the job file, and so forth.
2014 When fio is done (or interrupted by ctrl-c), it will show the data for
2015 each thread, group of threads, and disks in that order. For each data
2016 direction, the output looks like:
2018 Client1 (g=0): err= 0:
2019 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
2020 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
2021 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
2022 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
2023 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
2024 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
2025 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2026 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2027 issued r/w: total=0/32768, short=0/0
2028 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
2029 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
2031 The client number is printed, along with the group id and error of that
2032 thread. Below is the io statistics, here for writes. In the order listed,
2035 io= Number of megabytes io performed
2036 bw= Average bandwidth rate
2037 iops= Average IOs performed per second
2038 runt= The runtime of that thread
2039 slat= Submission latency (avg being the average, stdev being the
2040 standard deviation). This is the time it took to submit
2041 the io. For sync io, the slat is really the completion
2042 latency, since queue/complete is one operation there. This
2043 value can be in milliseconds or microseconds, fio will choose
2044 the most appropriate base and print that. In the example
2045 above, milliseconds is the best scale. Note: in --minimal mode
2046 latencies are always expressed in microseconds.
2047 clat= Completion latency. Same names as slat, this denotes the
2048 time from submission to completion of the io pieces. For
2049 sync io, clat will usually be equal (or very close) to 0,
2050 as the time from submit to complete is basically just
2051 CPU time (io has already been done, see slat explanation).
2052 bw= Bandwidth. Same names as the xlat stats, but also includes
2053 an approximate percentage of total aggregate bandwidth
2054 this thread received in this group. This last value is
2055 only really useful if the threads in this group are on the
2056 same disk, since they are then competing for disk access.
2057 cpu= CPU usage. User and system time, along with the number
2058 of context switches this thread went through, usage of
2059 system and user time, and finally the number of major
2060 and minor page faults. The CPU utilization numbers are
2061 averages for the jobs in that reporting group, while the
2062 context and fault counters are summed.
2063 IO depths= The distribution of io depths over the job life time. The
2064 numbers are divided into powers of 2, so for example the
2065 16= entries includes depths up to that value but higher
2066 than the previous entry. In other words, it covers the
2067 range from 16 to 31.
2068 IO submit= How many pieces of IO were submitting in a single submit
2069 call. Each entry denotes that amount and below, until
2070 the previous entry - eg, 8=100% mean that we submitted
2071 anywhere in between 5-8 ios per submit call.
2072 IO complete= Like the above submit number, but for completions instead.
2073 IO issued= The number of read/write requests issued, and how many
2075 IO latencies= The distribution of IO completion latencies. This is the
2076 time from when IO leaves fio and when it gets completed.
2077 The numbers follow the same pattern as the IO depths,
2078 meaning that 2=1.6% means that 1.6% of the IO completed
2079 within 2 msecs, 20=12.8% means that 12.8% of the IO
2080 took more than 10 msecs, but less than (or equal to) 20 msecs.
2082 After each client has been listed, the group statistics are printed. They
2083 will look like this:
2085 Run status group 0 (all jobs):
2086 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
2087 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
2089 For each data direction, it prints:
2091 io= Number of megabytes io performed.
2092 aggrb= Aggregate bandwidth of threads in this group.
2093 minb= The minimum average bandwidth a thread saw.
2094 maxb= The maximum average bandwidth a thread saw.
2095 mint= The smallest runtime of the threads in that group.
2096 maxt= The longest runtime of the threads in that group.
2098 And finally, the disk statistics are printed. They will look like this:
2100 Disk stats (read/write):
2101 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
2103 Each value is printed for both reads and writes, with reads first. The
2106 ios= Number of ios performed by all groups.
2107 merge= Number of merges io the io scheduler.
2108 ticks= Number of ticks we kept the disk busy.
2109 io_queue= Total time spent in the disk queue.
2110 util= The disk utilization. A value of 100% means we kept the disk
2111 busy constantly, 50% would be a disk idling half of the time.
2113 It is also possible to get fio to dump the current output while it is
2114 running, without terminating the job. To do that, send fio the USR1 signal.
2115 You can also get regularly timed dumps by using the --status-interval
2116 parameter, or by creating a file in /tmp named fio-dump-status. If fio
2117 sees this file, it will unlink it and dump the current output status.
2123 For scripted usage where you typically want to generate tables or graphs
2124 of the results, fio can output the results in a semicolon separated format.
2125 The format is one long line of values, such as:
2127 2;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%
2128 A description of this job goes here.
2130 The job description (if provided) follows on a second line.
2132 To enable terse output, use the --minimal command line option. The first
2133 value is the version of the terse output format. If the output has to
2134 be changed for some reason, this number will be incremented by 1 to
2135 signify that change.
2137 Split up, the format is as follows:
2139 terse version, fio version, jobname, groupid, error
2141 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
2142 Submission latency: min, max, mean, stdev (usec)
2143 Completion latency: min, max, mean, stdev (usec)
2144 Completion latency percentiles: 20 fields (see below)
2145 Total latency: min, max, mean, stdev (usec)
2146 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
2148 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
2149 Submission latency: min, max, mean, stdev (usec)
2150 Completion latency: min, max, mean, stdev(usec)
2151 Completion latency percentiles: 20 fields (see below)
2152 Total latency: min, max, mean, stdev (usec)
2153 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
2154 CPU usage: user, system, context switches, major faults, minor faults
2155 IO depths: <=1, 2, 4, 8, 16, 32, >=64
2156 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
2157 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
2158 Disk utilization: Disk name, Read ios, write ios,
2159 Read merges, write merges,
2160 Read ticks, write ticks,
2161 Time spent in queue, disk utilization percentage
2162 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
2164 Additional Info (dependent on description being set): Text description
2166 Completion latency percentiles can be a grouping of up to 20 sets, so
2167 for the terse output fio writes all of them. Each field will look like this:
2171 which is the Xth percentile, and the usec latency associated with it.
2173 For disk utilization, all disks used by fio are shown. So for each disk
2174 there will be a disk utilization section.
2177 8.0 Trace file format
2178 ---------------------
2179 There are two trace file format that you can encounter. The older (v1) format
2180 is unsupported since version 1.20-rc3 (March 2008). It will still be described
2181 below in case that you get an old trace and want to understand it.
2183 In any case the trace is a simple text file with a single action per line.
2186 8.1 Trace file format v1
2187 ------------------------
2188 Each line represents a single io action in the following format:
2192 where rw=0/1 for read/write, and the offset and length entries being in bytes.
2194 This format is not supported in Fio versions => 1.20-rc3.
2197 8.2 Trace file format v2
2198 ------------------------
2199 The second version of the trace file format was added in Fio version 1.17.
2200 It allows to access more then one file per trace and has a bigger set of
2201 possible file actions.
2203 The first line of the trace file has to be:
2207 Following this can be lines in two different formats, which are described below.
2209 The file management format:
2213 The filename is given as an absolute path. The action can be one of these:
2215 add Add the given filename to the trace
2216 open Open the file with the given filename. The filename has to have
2217 been added with the add action before.
2218 close Close the file with the given filename. The file has to have been
2222 The file io action format:
2224 filename action offset length
2226 The filename is given as an absolute path, and has to have been added and opened
2227 before it can be used with this format. The offset and length are given in
2228 bytes. The action can be one of these:
2230 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
2231 The time is relative to the previous wait statement.
2232 read Read 'length' bytes beginning from 'offset'
2233 write Write 'length' bytes beginning from 'offset'
2234 sync fsync() the file
2235 datasync fdatasync() the file
2236 trim trim the given file from the given 'offset' for 'length' bytes
2239 9.0 CPU idleness profiling
2240 --------------------------
2241 In some cases, we want to understand CPU overhead in a test. For example,
2242 we test patches for the specific goodness of whether they reduce CPU usage.
2243 fio implements a balloon approach to create a thread per CPU that runs at
2244 idle priority, meaning that it only runs when nobody else needs the cpu.
2245 By measuring the amount of work completed by the thread, idleness of each
2246 CPU can be derived accordingly.
2248 An unit work is defined as touching a full page of unsigned characters. Mean
2249 and standard deviation of time to complete an unit work is reported in "unit
2250 work" section. Options can be chosen to report detailed percpu idleness or
2251 overall system idleness by aggregating percpu stats.
2254 10.0 Verification and triggers
2255 ------------------------------
2256 Fio is usually run in one of two ways, when data verification is done. The
2257 first is a normal write job of some sort with verify enabled. When the
2258 write phase has completed, fio switches to reads and verifies everything
2259 it wrote. The second model is running just the write phase, and then later
2260 on running the same job (but with reads instead of writes) to repeat the
2261 same IO patterns and verify the contents. Both of these methods depend
2262 on the write phase being completed, as fio otherwise has no idea how much
2265 With verification triggers, fio supports dumping the current write state
2266 to local files. Then a subsequent read verify workload can load this state
2267 and know exactly where to stop. This is useful for testing cases where
2268 power is cut to a server in a managed fashion, for instance.
2270 A verification trigger consists of two things:
2272 1) Storing the write state of each job
2273 2) Executing a trigger command
2275 The write state is relatively small, on the order of hundreds of bytes
2276 to single kilobytes. It contains information on the number of completions
2277 done, the last X completions, etc.
2279 A trigger is invoked either through creation ('touch') of a specified
2280 file in the system, or through a timeout setting. If fio is run with
2281 --trigger-file=/tmp/trigger-file, then it will continually check for
2282 the existence of /tmp/trigger-file. When it sees this file, it will
2283 fire off the trigger (thus saving state, and executing the trigger
2286 For client/server runs, there's both a local and remote trigger. If
2287 fio is running as a server backend, it will send the job states back
2288 to the client for safe storage, then execute the remote trigger, if
2289 specified. If a local trigger is specified, the server will still send
2290 back the write state, but the client will then execute the trigger.
2292 10.1 Verification trigger example
2293 ---------------------------------
2294 Lets say we want to run a powercut test on the remote machine 'server'.
2295 Our write workload is in write-test.fio. We want to cut power to 'server'
2296 at some point during the run, and we'll run this test from the safety
2297 or our local machine, 'localbox'. On the server, we'll start the fio
2300 server# fio --server
2302 and on the client, we'll fire off the workload:
2304 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
2306 We set /tmp/my-trigger as the trigger file, and we tell fio to execute
2308 echo b > /proc/sysrq-trigger
2310 on the server once it has received the trigger and sent us the write
2311 state. This will work, but it's not _really_ cutting power to the server,
2312 it's merely abruptly rebooting it. If we have a remote way of cutting
2313 power to the server through IPMI or similar, we could do that through
2314 a local trigger command instead. Lets assume we have a script that does
2315 IPMI reboot of a given hostname, ipmi-reboot. On localbox, we could
2316 then have run fio with a local trigger instead:
2318 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
2320 For this case, fio would wait for the server to send us the write state,
2321 then execute 'ipmi-reboot server' when that happened.
2323 10.2 Loading verify state
2324 -------------------------
2325 To load store write state, read verification job file must contain
2326 the verify_state_load option. If that is set, fio will load the previously
2327 stored state. For a local fio run this is done by loading the files directly,
2328 and on a client/server run, the server backend will ask the client to send
2329 the files over and load them from there.
2332 11.0 Log File Formats
2333 ---------------------
2335 Fio supports a variety of log file formats, for logging latencies, bandwidth,
2336 and IOPS. The logs share a common format, which looks like this:
2338 time (msec), value, data direction, offset
2340 Time for the log entry is always in milliseconds. The value logged depends
2341 on the type of log, it will be one of the following:
2343 Latency log Value is latency in usecs
2344 Bandwidth log Value is in KB/sec
2345 IOPS log Value is IOPS
2347 Data direction is one of the following:
2353 The offset is the offset, in bytes, from the start of the file, for that
2354 particular IO. The logging of the offset can be toggled with 'log_offset'.
2356 If windowed logging is enabled through 'log_avg_msec', then fio doesn't log
2357 individual IOs. Instead of logs the average values over the specified
2358 period of time. Since 'data direction' and 'offset' are per-IO values,
2359 they aren't applicable if windowed logging is enabled. If windowed logging
2360 is enabled and 'log_max_value' is set, then fio logs maximum values in
2361 that window instead of averages.