8 5. Detailed list of parameters
12 9. CPU idleness profiling
14 1.0 Overview and history
15 ------------------------
16 fio was originally written to save me the hassle of writing special test
17 case programs when I wanted to test a specific workload, either for
18 performance reasons or to find/reproduce a bug. The process of writing
19 such a test app can be tiresome, especially if you have to do it often.
20 Hence I needed a tool that would be able to simulate a given io workload
21 without resorting to writing a tailored test case again and again.
23 A test work load is difficult to define, though. There can be any number
24 of processes or threads involved, and they can each be using their own
25 way of generating io. You could have someone dirtying large amounts of
26 memory in an memory mapped file, or maybe several threads issuing
27 reads using asynchronous io. fio needed to be flexible enough to
28 simulate both of these cases, and many more.
32 The first step in getting fio to simulate a desired io workload, is
33 writing a job file describing that specific setup. A job file may contain
34 any number of threads and/or files - the typical contents of the job file
35 is a global section defining shared parameters, and one or more job
36 sections describing the jobs involved. When run, fio parses this file
37 and sets everything up as described. If we break down a job from top to
38 bottom, it contains the following basic parameters:
40 IO type Defines the io pattern issued to the file(s).
41 We may only be reading sequentially from this
42 file(s), or we may be writing randomly. Or even
43 mixing reads and writes, sequentially or randomly.
45 Block size In how large chunks are we issuing io? This may be
46 a single value, or it may describe a range of
49 IO size How much data are we going to be reading/writing.
51 IO engine How do we issue io? We could be memory mapping the
52 file, we could be using regular read/write, we
53 could be using splice, async io, syslet, or even
56 IO depth If the io engine is async, how large a queuing
57 depth do we want to maintain?
59 IO type Should we be doing buffered io, or direct/raw io?
61 Num files How many files are we spreading the workload over.
63 Num threads How many threads or processes should we spread
66 The above are the basic parameters defined for a workload, in addition
67 there's a multitude of parameters that modify other aspects of how this
73 See the README file for command line parameters, there are only a few
76 Running fio is normally the easiest part - you just give it the job file
77 (or job files) as parameters:
81 and it will start doing what the job_file tells it to do. You can give
82 more than one job file on the command line, fio will serialize the running
83 of those files. Internally that is the same as using the 'stonewall'
84 parameter described in the parameter section.
86 If the job file contains only one job, you may as well just give the
87 parameters on the command line. The command line parameters are identical
88 to the job parameters, with a few extra that control global parameters
89 (see README). For example, for the job file parameter iodepth=2, the
90 mirror command line option would be --iodepth 2 or --iodepth=2. You can
91 also use the command line for giving more than one job entry. For each
92 --name option that fio sees, it will start a new job with that name.
93 Command line entries following a --name entry will apply to that job,
94 until there are no more entries or a new --name entry is seen. This is
95 similar to the job file options, where each option applies to the current
96 job until a new [] job entry is seen.
98 fio does not need to run as root, except if the files or devices specified
99 in the job section requires that. Some other options may also be restricted,
100 such as memory locking, io scheduler switching, and decreasing the nice value.
105 As previously described, fio accepts one or more job files describing
106 what it is supposed to do. The job file format is the classic ini file,
107 where the names enclosed in [] brackets define the job name. You are free
108 to use any ascii name you want, except 'global' which has special meaning.
109 A global section sets defaults for the jobs described in that file. A job
110 may override a global section parameter, and a job file may even have
111 several global sections if so desired. A job is only affected by a global
112 section residing above it. If the first character in a line is a ';' or a
113 '#', the entire line is discarded as a comment.
115 So let's look at a really simple job file that defines two processes, each
116 randomly reading from a 128MB file.
118 ; -- start job file --
129 As you can see, the job file sections themselves are empty as all the
130 described parameters are shared. As no filename= option is given, fio
131 makes up a filename for each of the jobs as it sees fit. On the command
132 line, this job would look as follows:
134 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
137 Let's look at an example that has a number of processes writing randomly
140 ; -- start job file --
152 Here we have no global section, as we only have one job defined anyway.
153 We want to use async io here, with a depth of 4 for each file. We also
154 increased the buffer size used to 32KB and define numjobs to 4 to
155 fork 4 identical jobs. The result is 4 processes each randomly writing
156 to their own 64MB file. Instead of using the above job file, you could
157 have given the parameters on the command line. For this case, you would
160 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
162 When fio is utilized as a basis of any reasonably large test suite, it might be
163 desirable to share a set of standardized settings across multiple job files.
164 Instead of copy/pasting such settings, any section may pull in an external
165 .fio file with 'include filename' directive, as in the following example:
167 ; -- start job file including.fio --
171 include glob-include.fio
178 include test-include.fio
179 ; -- end job file including.fio --
181 ; -- start job file glob-include.fio --
184 ; -- end job file glob-include.fio --
186 ; -- start job file test-include.fio --
189 ; -- end job file test-include.fio --
191 Settings pulled into a section apply to that section only (except global
192 section). Include directives may be nested in that any included file may
193 contain further include directive(s). Include files may not contain []
197 4.1 Environment variables
198 -------------------------
200 fio also supports environment variable expansion in job files. Any
201 sub-string of the form "${VARNAME}" as part of an option value (in other
202 words, on the right of the `='), will be expanded to the value of the
203 environment variable called VARNAME. If no such environment variable
204 is defined, or VARNAME is the empty string, the empty string will be
207 As an example, let's look at a sample fio invocation and job file:
209 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
211 ; -- start job file --
218 This will expand to the following equivalent job file at runtime:
220 ; -- start job file --
227 fio ships with a few example job files, you can also look there for
230 4.2 Reserved keywords
231 ---------------------
233 Additionally, fio has a set of reserved keywords that will be replaced
234 internally with the appropriate value. Those keywords are:
236 $pagesize The architecture page size of the running system
237 $mb_memory Megabytes of total memory in the system
238 $ncpus Number of online available CPUs
240 These can be used on the command line or in the job file, and will be
241 automatically substituted with the current system values when the job
242 is run. Simple math is also supported on these keywords, so you can
243 perform actions like:
247 and get that properly expanded to 8 times the size of memory in the
251 5.0 Detailed list of parameters
252 -------------------------------
254 This section describes in details each parameter associated with a job.
255 Some parameters take an option of a given type, such as an integer or
256 a string. Anywhere a numeric value is required, an arithmetic expression
257 may be used, provided it is surrounded by parentheses. Supported operators
267 For time values in expressions, units are microseconds by default. This is
268 different than for time values not in expressions (not enclosed in
269 parentheses). The following types are used:
271 str String. This is a sequence of alpha characters.
272 time Integer with possible time suffix. In seconds unless otherwise
273 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
274 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
275 and 'us' (or 'usec') for microseconds.
276 int SI integer. A whole number value, which may contain a suffix
277 describing the base of the number. Accepted suffixes are k/m/g/t/p,
278 meaning kilo, mega, giga, tera, and peta. The suffix is not case
279 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
280 as 'k'). So if you want to specify 4096, you could either write
281 out '4096' or just give 4k. The suffixes signify base 2 values, so
282 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
283 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
284 case, then 1000 is used as the multiplier. This can be handy for
285 disks, since manufacturers generally use base 10 values when listing
286 the capacity of a drive. If the option accepts an upper and lower
287 range, use a colon ':' or minus '-' to separate such values. May also
288 include a prefix to indicate numbers base. If 0x is used, the number
289 is assumed to be hexadecimal. See irange.
290 bool Boolean. Usually parsed as an integer, however only defined for
291 true and false (1 and 0).
292 irange Integer range with suffix. Allows value range to be given, such
293 as 1024-4096. A colon may also be used as the separator, eg
294 1k:4k. If the option allows two sets of ranges, they can be
295 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
297 float_list A list of floating numbers, separated by a ':' character.
299 With the above in mind, here follows the complete list of fio job
302 name=str ASCII name of the job. This may be used to override the
303 name printed by fio for this job. Otherwise the job
304 name is used. On the command line this parameter has the
305 special purpose of also signaling the start of a new
308 wait_for=str Specifies the name of the already defined job to wait
309 for. Single waitee name only may be specified. If set, the job
310 won't be started until all workers of the waitee job are done.
312 Wait_for operates on the job name basis, so there are a few
313 limitations. First, the waitee must be defined prior to the
314 waiter job (meaning no forward references). Second, if a job
315 is being referenced as a waitee, it must have a unique name
316 (no duplicate waitees).
318 description=str Text description of the job. Doesn't do anything except
319 dump this text description when this job is run. It's
322 directory=str Prefix filenames with this directory. Used to place files
323 in a different location than "./". See the 'filename' option
324 for escaping certain characters.
326 filename=str Fio normally makes up a filename based on the job name,
327 thread number, and file number. If you want to share
328 files between threads in a job or several jobs, specify
329 a filename for each of them to override the default. If
330 the ioengine used is 'net', the filename is the host, port,
331 and protocol to use in the format of =host,port,protocol.
332 See ioengine=net for more. If the ioengine is file based, you
333 can specify a number of files by separating the names with a
334 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
335 as the two working files, you would use
336 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
337 accessed as \\.\PhysicalDrive0 for the first device,
338 \\.\PhysicalDrive1 for the second etc. Note: Windows and
339 FreeBSD prevent write access to areas of the disk containing
340 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 opendir=str Tell fio to recursively add any file it can find in this
375 directory and down the file system tree.
377 lockfile=str Fio defaults to not locking any files before it does
378 IO to them. If a file or file descriptor is shared, fio
379 can serialize IO to that file to make the end result
380 consistent. This is usual for emulating real workloads that
381 share files. The lock modes are:
383 none No locking. The default.
384 exclusive Only one thread/process may do IO,
385 excluding all others.
386 readwrite Read-write locking on the file. Many
387 readers may access the file at the
388 same time, but writes get exclusive
392 rw=str Type of io pattern. Accepted values are:
394 read Sequential reads
395 write Sequential writes
396 randwrite Random writes
397 randread Random reads
398 rw,readwrite Sequential mixed reads and writes
399 randrw Random mixed reads and writes
400 trimwrite Mixed trims and writes. Blocks will be
401 trimmed first, then written to.
403 For the mixed io types, the default is to split them 50/50.
404 For certain types of io the result may still be skewed a bit,
405 since the speed may be different. It is possible to specify
406 a number of IO's to do before getting a new offset, this is
407 done by appending a ':<nr>' to the end of the string given.
408 For a random read, it would look like 'rw=randread:8' for
409 passing in an offset modifier with a value of 8. If the
410 suffix is used with a sequential IO pattern, then the value
411 specified will be added to the generated offset for each IO.
412 For instance, using rw=write:4k will skip 4k for every
413 write. It turns sequential IO into sequential IO with holes.
414 See the 'rw_sequencer' option.
416 rw_sequencer=str If an offset modifier is given by appending a number to
417 the rw=<str> line, then this option controls how that
418 number modifies the IO offset being generated. Accepted
421 sequential Generate sequential offset
422 identical Generate the same offset
424 'sequential' is only useful for random IO, where fio would
425 normally generate a new random offset for every IO. If you
426 append eg 8 to randread, you would get a new random offset for
427 every 8 IO's. The result would be a seek for only every 8
428 IO's, instead of for every IO. Use rw=randread:8 to specify
429 that. As sequential IO is already sequential, setting
430 'sequential' for that would not result in any differences.
431 'identical' behaves in a similar fashion, except it sends
432 the same offset 8 number of times before generating a new
435 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
436 Storage manufacturers like to use 10^3 or 1000 as a base
437 ten unit instead, for obvious reasons. Allow values are
438 1024 or 1000, with 1024 being the default.
440 unified_rw_reporting=bool Fio normally reports statistics on a per
441 data direction basis, meaning that read, write, and trim are
442 accounted and reported separately. If this option is set,
443 the fio will sum the results and report them as "mixed"
446 randrepeat=bool For random IO workloads, seed the generator in a predictable
447 way so that results are repeatable across repetitions.
450 randseed=int Seed the random number generators based on this seed value, to
451 be able to control what sequence of output is being generated.
452 If not set, the random sequence depends on the randrepeat
455 fallocate=str Whether pre-allocation is performed when laying down files.
458 none Do not pre-allocate space
459 posix Pre-allocate via posix_fallocate()
460 keep Pre-allocate via fallocate() with
461 FALLOC_FL_KEEP_SIZE set
462 0 Backward-compatible alias for 'none'
463 1 Backward-compatible alias for 'posix'
465 May not be available on all supported platforms. 'keep' is only
466 available on Linux.If using ZFS on Solaris this must be set to
467 'none' because ZFS doesn't support it. Default: 'posix'.
469 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
470 on what IO patterns it is likely to issue. Sometimes you
471 want to test specific IO patterns without telling the
472 kernel about it, in which case you can disable this option.
473 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
474 IO and POSIX_FADV_RANDOM for random IO.
476 fadvise_stream=int Notify the kernel what write stream ID to place these
477 writes under. Only supported on Linux. Note, this option
478 may change going forward.
480 size=int The total size of file io for this job. Fio will run until
481 this many bytes has been transferred, unless runtime is
482 limited by other options (such as 'runtime', for instance,
483 or increased/decreased by 'io_size'). Unless specific nrfiles
484 and filesize options are given, fio will divide this size
485 between the available files specified by the job. If not set,
486 fio will use the full size of the given files or devices.
487 If the files do not exist, size must be given. It is also
488 possible to give size as a percentage between 1 and 100. If
489 size=20% is given, fio will use 20% of the full size of the
490 given files or devices.
493 io_limit=int Normally fio operates within the region set by 'size', which
494 means that the 'size' option sets both the region and size of
495 IO to be performed. Sometimes that is not what you want. With
496 this option, it is possible to define just the amount of IO
497 that fio should do. For instance, if 'size' is set to 20G and
498 'io_size' is set to 5G, fio will perform IO within the first
499 20G but exit when 5G have been done. The opposite is also
500 possible - if 'size' is set to 20G, and 'io_size' is set to
501 40G, then fio will do 40G of IO within the 0..20G region.
503 filesize=int Individual file sizes. May be a range, in which case fio
504 will select sizes for files at random within the given range
505 and limited to 'size' in total (if that is given). If not
506 given, each created file is the same size.
508 file_append=bool Perform IO after the end of the file. Normally fio will
509 operate within the size of a file. If this option is set, then
510 fio will append to the file instead. This has identical
511 behavior to setting offset to the size of a file. This option
512 is ignored on non-regular files.
515 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
516 space left on device) as the terminating condition. Only makes
517 sense with sequential write. For a read workload, the mount
518 point will be filled first then IO started on the result. This
519 option doesn't make sense if operating on a raw device node,
520 since the size of that is already known by the file system.
521 Additionally, writing beyond end-of-device will not return
525 bs=int The block size used for the io units. Defaults to 4k. Values
526 can be given for both read and writes. If a single int is
527 given, it will apply to both. If a second int is specified
528 after a comma, it will apply to writes only. In other words,
529 the format is either bs=read_and_write or bs=read,write,trim.
530 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
531 writes, and 8k for trims. You can terminate the list with
532 a trailing comma. bs=4k,8k, would use the default value for
533 trims.. If you only wish to set the write size, you
534 can do so by passing an empty read size - bs=,8k will set
535 8k for writes and leave the read default value.
538 ba=int At what boundary to align random IO offsets. Defaults to
539 the same as 'blocksize' the minimum blocksize given.
540 Minimum alignment is typically 512b for using direct IO,
541 though it usually depends on the hardware block size. This
542 option is mutually exclusive with using a random map for
543 files, so it will turn off that option.
545 blocksize_range=irange
546 bsrange=irange Instead of giving a single block size, specify a range
547 and fio will mix the issued io block sizes. The issued
548 io unit will always be a multiple of the minimum value
549 given (also see bs_unaligned). Applies to both reads and
550 writes, however a second range can be given after a comma.
553 bssplit=str Sometimes you want even finer grained control of the
554 block sizes issued, not just an even split between them.
555 This option allows you to weight various block sizes,
556 so that you are able to define a specific amount of
557 block sizes issued. The format for this option is:
559 bssplit=blocksize/percentage:blocksize/percentage
561 for as many block sizes as needed. So if you want to define
562 a workload that has 50% 64k blocks, 10% 4k blocks, and
563 40% 32k blocks, you would write:
565 bssplit=4k/10:64k/50:32k/40
567 Ordering does not matter. If the percentage is left blank,
568 fio will fill in the remaining values evenly. So a bssplit
569 option like this one:
571 bssplit=4k/50:1k/:32k/
573 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
574 always add up to 100, if bssplit is given a range that adds
575 up to more, it will error out.
577 bssplit also supports giving separate splits to reads and
578 writes. The format is identical to what bs= accepts. You
579 have to separate the read and write parts with a comma. So
580 if you want a workload that has 50% 2k reads and 50% 4k reads,
581 while having 90% 4k writes and 10% 8k writes, you would
584 bssplit=2k/50:4k/50,4k/90:8k/10
587 bs_unaligned If this option is given, any byte size value within bsrange
588 may be used as a block range. This typically wont work with
589 direct IO, as that normally requires sector alignment.
591 bs_is_seq_rand If this option is set, fio will use the normal read,write
592 blocksize settings as sequential,random instead. Any random
593 read or write will use the WRITE blocksize settings, and any
594 sequential read or write will use the READ blocksize setting.
596 zero_buffers If this option is given, fio will init the IO buffers to
597 all zeroes. The default is to fill them with random data.
599 refill_buffers If this option is given, fio will refill the IO buffers
600 on every submit. The default is to only fill it at init
601 time and reuse that data. Only makes sense if zero_buffers
602 isn't specified, naturally. If data verification is enabled,
603 refill_buffers is also automatically enabled.
605 scramble_buffers=bool If refill_buffers is too costly and the target is
606 using data deduplication, then setting this option will
607 slightly modify the IO buffer contents to defeat normal
608 de-dupe attempts. This is not enough to defeat more clever
609 block compression attempts, but it will stop naive dedupe of
610 blocks. Default: true.
612 buffer_compress_percentage=int If this is set, then fio will attempt to
613 provide IO buffer content (on WRITEs) that compress to
614 the specified level. Fio does this by providing a mix of
615 random data and a fixed pattern. The fixed pattern is either
616 zeroes, or the pattern specified by buffer_pattern. If the
617 pattern option is used, it might skew the compression ratio
618 slightly. Note that this is per block size unit, for file/disk
619 wide compression level that matches this setting, you'll also
620 want to set refill_buffers.
622 buffer_compress_chunk=int See buffer_compress_percentage. This
623 setting allows fio to manage how big the ranges of random
624 data and zeroed data is. Without this set, fio will
625 provide buffer_compress_percentage of blocksize random
626 data, followed by the remaining zeroed. With this set
627 to some chunk size smaller than the block size, fio can
628 alternate random and zeroed data throughout the IO
631 buffer_pattern=str If set, fio will fill the io buffers with this
632 pattern. If not set, the contents of io buffers is defined by
633 the other options related to buffer contents. The setting can
634 be any pattern of bytes, and can be prefixed with 0x for hex
635 values. It may also be a string, where the string must then
636 be wrapped with "", e.g.:
638 buffer_pattern="abcd"
642 buffer_pattern=0xdeadface
644 Also you can combine everything together in any order:
645 buffer_pattern=0xdeadface"abcd"-12
647 dedupe_percentage=int If set, fio will generate this percentage of
648 identical buffers when writing. These buffers will be
649 naturally dedupable. The contents of the buffers depend on
650 what other buffer compression settings have been set. It's
651 possible to have the individual buffers either fully
652 compressible, or not at all. This option only controls the
653 distribution of unique buffers.
655 nrfiles=int Number of files to use for this job. Defaults to 1.
657 openfiles=int Number of files to keep open at the same time. Defaults to
658 the same as nrfiles, can be set smaller to limit the number
661 file_service_type=str Defines how fio decides which file from a job to
662 service next. The following types are defined:
664 random Just choose a file at random.
666 roundrobin Round robin over open files. This
669 sequential Finish one file before moving on to
670 the next. Multiple files can still be
671 open depending on 'openfiles'.
673 The string can have a number appended, indicating how
674 often to switch to a new file. So if option random:4 is
675 given, fio will switch to a new random file after 4 ios
678 ioengine=str Defines how the job issues io to the file. The following
681 sync Basic read(2) or write(2) io. lseek(2) is
682 used to position the io location.
684 psync Basic pread(2) or pwrite(2) io.
686 vsync Basic readv(2) or writev(2) IO.
688 psyncv Basic preadv(2) or pwritev(2) IO.
690 libaio Linux native asynchronous io. Note that Linux
691 may only support queued behaviour with
692 non-buffered IO (set direct=1 or buffered=0).
693 This engine defines engine specific options.
695 posixaio glibc posix asynchronous io.
697 solarisaio Solaris native asynchronous io.
699 windowsaio Windows native asynchronous io.
701 mmap File is memory mapped and data copied
702 to/from using memcpy(3).
704 splice splice(2) is used to transfer the data and
705 vmsplice(2) to transfer data from user
708 syslet-rw Use the syslet system calls to make
709 regular read/write async.
711 sg SCSI generic sg v3 io. May either be
712 synchronous using the SG_IO ioctl, or if
713 the target is an sg character device
714 we use read(2) and write(2) for asynchronous
717 null Doesn't transfer any data, just pretends
718 to. This is mainly used to exercise fio
719 itself and for debugging/testing purposes.
721 net Transfer over the network to given host:port.
722 Depending on the protocol used, the hostname,
723 port, listen and filename options are used to
724 specify what sort of connection to make, while
725 the protocol option determines which protocol
727 This engine defines engine specific options.
729 netsplice Like net, but uses splice/vmsplice to
730 map data and send/receive.
731 This engine defines engine specific options.
733 cpuio Doesn't transfer any data, but burns CPU
734 cycles according to the cpuload= and
735 cpucycle= options. Setting cpuload=85
736 will cause that job to do nothing but burn
737 85% of the CPU. In case of SMP machines,
738 use numjobs=<no_of_cpu> to get desired CPU
739 usage, as the cpuload only loads a single
740 CPU at the desired rate.
742 guasi The GUASI IO engine is the Generic Userspace
743 Asyncronous Syscall Interface approach
746 http://www.xmailserver.org/guasi-lib.html
748 for more info on GUASI.
750 rdma The RDMA I/O engine supports both RDMA
751 memory semantics (RDMA_WRITE/RDMA_READ) and
752 channel semantics (Send/Recv) for the
753 InfiniBand, RoCE and iWARP protocols.
755 falloc IO engine that does regular fallocate to
756 simulate data transfer as fio ioengine.
757 DDIR_READ does fallocate(,mode = keep_size,)
758 DDIR_WRITE does fallocate(,mode = 0)
759 DDIR_TRIM does fallocate(,mode = punch_hole)
761 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
762 ioctls to simulate defragment activity in
763 request to DDIR_WRITE event
765 rbd IO engine supporting direct access to Ceph
766 Rados Block Devices (RBD) via librbd without
767 the need to use the kernel rbd driver. This
768 ioengine defines engine specific options.
770 gfapi Using Glusterfs libgfapi sync interface to
771 direct access to Glusterfs volumes without
774 gfapi_async Using Glusterfs libgfapi async interface
775 to direct access to Glusterfs volumes without
776 having to go through FUSE. This ioengine
777 defines engine specific options.
779 libhdfs Read and write through Hadoop (HDFS).
780 This engine interprets offsets a little
781 differently. In HDFS, files once created
782 cannot be modified. So random writes are not
783 possible. To imitate this, libhdfs engine
784 creates bunch of small files, and engine will
785 pick a file out of those files based on the
786 offset enerated by fio backend. Each jobs uses
787 it's own connection to HDFS.
789 mtd Read, write and erase an MTD character device
790 (e.g., /dev/mtd0). Discards are treated as
791 erases. Depending on the underlying device
792 type, the I/O may have to go in a certain
793 pattern, e.g., on NAND, writing sequentially
794 to erase blocks and discarding before
795 overwriting. The writetrim mode works well
798 external Prefix to specify loading an external
799 IO engine object file. Append the engine
800 filename, eg ioengine=external:/tmp/foo.o
801 to load ioengine foo.o in /tmp.
803 iodepth=int This defines how many io units to keep in flight against
804 the file. The default is 1 for each file defined in this
805 job, can be overridden with a larger value for higher
806 concurrency. Note that increasing iodepth beyond 1 will not
807 affect synchronous ioengines (except for small degress when
808 verify_async is in use). Even async engines may impose OS
809 restrictions causing the desired depth not to be achieved.
810 This may happen on Linux when using libaio and not setting
811 direct=1, since buffered IO is not async on that OS. Keep an
812 eye on the IO depth distribution in the fio output to verify
813 that the achieved depth is as expected. Default: 1.
815 iodepth_batch_submit=int
816 iodepth_batch=int This defines how many pieces of IO to submit at once.
817 It defaults to 1 which means that we submit each IO
818 as soon as it is available, but can be raised to submit
819 bigger batches of IO at the time. If it is set to 0 the iodepth
822 iodepth_batch_complete_min=int
823 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
824 at once. It defaults to 1 which means that we'll ask
825 for a minimum of 1 IO in the retrieval process from
826 the kernel. The IO retrieval will go on until we
827 hit the limit set by iodepth_low. If this variable is
828 set to 0, then fio will always check for completed
829 events before queuing more IO. This helps reduce
830 IO latency, at the cost of more retrieval system calls.
832 iodepth_batch_complete_max=int This defines maximum pieces of IO to
833 retrieve at once. This variable should be used along with
834 iodepth_batch_complete_min=int variable, specifying the range
835 of min and max amount of IO which should be retrieved. By default
836 it is equal to iodepth_batch_complete_min value.
840 iodepth_batch_complete_min=1
841 iodepth_batch_complete_max=<iodepth>
843 which means that we will retrieve at leat 1 IO and up to the
844 whole submitted queue depth. If none of IO has been completed
849 iodepth_batch_complete_min=0
850 iodepth_batch_complete_max=<iodepth>
852 which means that we can retrieve up to the whole submitted
853 queue depth, but if none of IO has been completed yet, we will
854 NOT wait and immediately exit the system call. In this example
855 we simply do polling.
857 iodepth_low=int The low water mark indicating when to start filling
858 the queue again. Defaults to the same as iodepth, meaning
859 that fio will attempt to keep the queue full at all times.
860 If iodepth is set to eg 16 and iodepth_low is set to 4, then
861 after fio has filled the queue of 16 requests, it will let
862 the depth drain down to 4 before starting to fill it again.
864 io_submit_mode=str This option controls how fio submits the IO to
865 the IO engine. The default is 'inline', which means that the
866 fio job threads submit and reap IO directly. If set to
867 'offload', the job threads will offload IO submission to a
868 dedicated pool of IO threads. This requires some coordination
869 and thus has a bit of extra overhead, especially for lower
870 queue depth IO where it can increase latencies. The benefit
871 is that fio can manage submission rates independently of
872 the device completion rates. This avoids skewed latency
873 reporting if IO gets back up on the device side (the
874 coordinated omission problem).
876 direct=bool If value is true, use non-buffered io. This is usually
877 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
878 On Windows the synchronous ioengines don't support direct io.
880 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
881 writes are guaranteed to be stable once acknowledged by
882 the operating system. Only Linux supports O_ATOMIC right
885 buffered=bool If value is true, use buffered io. This is the opposite
886 of the 'direct' option. Defaults to true.
888 offset=int Start io at the given offset in the file. The data before
889 the given offset will not be touched. This effectively
890 caps the file size at real_size - offset.
892 offset_increment=int If this is provided, then the real offset becomes
893 offset + offset_increment * thread_number, where the thread
894 number is a counter that starts at 0 and is incremented for
895 each sub-job (i.e. when numjobs option is specified). This
896 option is useful if there are several jobs which are intended
897 to operate on a file in parallel disjoint segments, with
898 even spacing between the starting points.
900 number_ios=int Fio will normally perform IOs until it has exhausted the size
901 of the region set by size=, or if it exhaust the allocated
902 time (or hits an error condition). With this setting, the
903 range/size can be set independently of the number of IOs to
904 perform. When fio reaches this number, it will exit normally
905 and report status. Note that this does not extend the amount
906 of IO that will be done, it will only stop fio if this
907 condition is met before other end-of-job criteria.
909 fsync=int If writing to a file, issue a sync of the dirty data
910 for every number of blocks given. For example, if you give
911 32 as a parameter, fio will sync the file for every 32
912 writes issued. If fio is using non-buffered io, we may
913 not sync the file. The exception is the sg io engine, which
914 synchronizes the disk cache anyway.
916 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
918 In FreeBSD and Windows there is no fdatasync(), this falls back
921 sync_file_range=str:val Use sync_file_range() for every 'val' number of
922 write operations. Fio will track range of writes that
923 have happened since the last sync_file_range() call. 'str'
924 can currently be one or more of:
926 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
927 write SYNC_FILE_RANGE_WRITE
928 wait_after SYNC_FILE_RANGE_WAIT_AFTER
930 So if you do sync_file_range=wait_before,write:8, fio would
931 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
932 every 8 writes. Also see the sync_file_range(2) man page.
933 This option is Linux specific.
935 overwrite=bool If true, writes to a file will always overwrite existing
936 data. If the file doesn't already exist, it will be
937 created before the write phase begins. If the file exists
938 and is large enough for the specified write phase, nothing
941 end_fsync=bool If true, fsync file contents when a write stage has completed.
943 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
944 This differs from end_fsync in that it will happen on every
945 file close, not just at the end of the job.
947 rwmixread=int How large a percentage of the mix should be reads.
949 rwmixwrite=int How large a percentage of the mix should be writes. If both
950 rwmixread and rwmixwrite is given and the values do not add
951 up to 100%, the latter of the two will be used to override
952 the first. This may interfere with a given rate setting,
953 if fio is asked to limit reads or writes to a certain rate.
954 If that is the case, then the distribution may be skewed.
956 random_distribution=str:float By default, fio will use a completely uniform
957 random distribution when asked to perform random IO. Sometimes
958 it is useful to skew the distribution in specific ways,
959 ensuring that some parts of the data is more hot than others.
960 fio includes the following distribution models:
962 random Uniform random distribution
963 zipf Zipf distribution
964 pareto Pareto distribution
966 When using a zipf or pareto distribution, an input value
967 is also needed to define the access pattern. For zipf, this
968 is the zipf theta. For pareto, it's the pareto power. Fio
969 includes a test program, genzipf, that can be used visualize
970 what the given input values will yield in terms of hit rates.
971 If you wanted to use zipf with a theta of 1.2, you would use
972 random_distribution=zipf:1.2 as the option. If a non-uniform
973 model is used, fio will disable use of the random map.
975 percentage_random=int For a random workload, set how big a percentage should
976 be random. This defaults to 100%, in which case the workload
977 is fully random. It can be set from anywhere from 0 to 100.
978 Setting it to 0 would make the workload fully sequential. Any
979 setting in between will result in a random mix of sequential
980 and random IO, at the given percentages. It is possible to
981 set different values for reads, writes, and trim. To do so,
982 simply use a comma separated list. See blocksize.
984 norandommap Normally fio will cover every block of the file when doing
985 random IO. If this option is given, fio will just get a
986 new random offset without looking at past io history. This
987 means that some blocks may not be read or written, and that
988 some blocks may be read/written more than once. If this option
989 is used with verify= and multiple blocksizes (via bsrange=),
990 only intact blocks are verified, i.e., partially-overwritten
993 softrandommap=bool See norandommap. If fio runs with the random block map
994 enabled and it fails to allocate the map, if this option is
995 set it will continue without a random block map. As coverage
996 will not be as complete as with random maps, this option is
999 random_generator=str Fio supports the following engines for generating
1000 IO offsets for random IO:
1002 tausworthe Strong 2^88 cycle random number generator
1003 lfsr Linear feedback shift register generator
1004 tausworthe64 Strong 64-bit 2^258 cycle random number
1007 Tausworthe is a strong random number generator, but it
1008 requires tracking on the side if we want to ensure that
1009 blocks are only read or written once. LFSR guarantees
1010 that we never generate the same offset twice, and it's
1011 also less computationally expensive. It's not a true
1012 random generator, however, though for IO purposes it's
1013 typically good enough. LFSR only works with single
1014 block sizes, not with workloads that use multiple block
1015 sizes. If used with such a workload, fio may read or write
1016 some blocks multiple times. The default value is tausworthe,
1017 unless the required space exceeds 2^32 blocks. If it does,
1018 then tausworthe64 is selected automatically.
1020 nice=int Run the job with the given nice value. See man nice(2).
1022 prio=int Set the io priority value of this job. Linux limits us to
1023 a positive value between 0 and 7, with 0 being the highest.
1026 prioclass=int Set the io priority class. See man ionice(1).
1028 thinktime=int Stall the job x microseconds after an io has completed before
1029 issuing the next. May be used to simulate processing being
1030 done by an application. See thinktime_blocks and
1034 Only valid if thinktime is set - pretend to spend CPU time
1035 doing something with the data received, before falling back
1036 to sleeping for the rest of the period specified by
1039 thinktime_blocks=int
1040 Only valid if thinktime is set - control how many blocks
1041 to issue, before waiting 'thinktime' usecs. If not set,
1042 defaults to 1 which will make fio wait 'thinktime' usecs
1043 after every block. This effectively makes any queue depth
1044 setting redundant, since no more than 1 IO will be queued
1045 before we have to complete it and do our thinktime. In
1046 other words, this setting effectively caps the queue depth
1047 if the latter is larger.
1049 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
1050 the normal suffix rules apply. You can use rate=500k to limit
1051 reads and writes to 500k each, or you can specify read and
1052 writes separately. Using rate=1m,500k would limit reads to
1053 1MB/sec and writes to 500KB/sec. Capping only reads or
1054 writes can be done with rate=,500k or rate=500k,. The former
1055 will only limit writes (to 500KB/sec), the latter will only
1058 rate_min=int Tell fio to do whatever it can to maintain at least this
1059 bandwidth. Failing to meet this requirement, will cause
1060 the job to exit. The same format as rate is used for
1061 read vs write separation.
1063 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
1064 as rate, just specified independently of bandwidth. If the
1065 job is given a block size range instead of a fixed value,
1066 the smallest block size is used as the metric. The same format
1067 as rate is used for read vs write separation.
1069 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
1070 the job to exit. The same format as rate is used for read vs
1073 rate_process=str This option controls how fio manages rated IO
1074 submissions. The default is 'linear', which submits IO in a
1075 linear fashion with fixed delays between IOs that gets
1076 adjusted based on IO completion rates. If this is set to
1077 'poisson', fio will submit IO based on a more real world
1078 random request flow, known as the Poisson process
1079 (https://en.wikipedia.org/wiki/Poisson_process). The lambda
1080 will be 10^6 / IOPS for the given workload.
1082 latency_target=int If set, fio will attempt to find the max performance
1083 point that the given workload will run at while maintaining a
1084 latency below this target. The values is given in microseconds.
1085 See latency_window and latency_percentile
1087 latency_window=int Used with latency_target to specify the sample window
1088 that the job is run at varying queue depths to test the
1089 performance. The value is given in microseconds.
1091 latency_percentile=float The percentage of IOs that must fall within the
1092 criteria specified by latency_target and latency_window. If not
1093 set, this defaults to 100.0, meaning that all IOs must be equal
1094 or below to the value set by latency_target.
1096 max_latency=int If set, fio will exit the job if it exceeds this maximum
1097 latency. It will exit with an ETIME error.
1099 rate_cycle=int Average bandwidth for 'rate' and 'rate_min' over this number
1102 cpumask=int Set the CPU affinity of this job. The parameter given is a
1103 bitmask of allowed CPU's the job may run on. So if you want
1104 the allowed CPUs to be 1 and 5, you would pass the decimal
1105 value of (1 << 1 | 1 << 5), or 34. See man
1106 sched_setaffinity(2). This may not work on all supported
1107 operating systems or kernel versions. This option doesn't
1108 work well for a higher CPU count than what you can store in
1109 an integer mask, so it can only control cpus 1-32. For
1110 boxes with larger CPU counts, use cpus_allowed.
1112 cpus_allowed=str Controls the same options as cpumask, but it allows a text
1113 setting of the permitted CPUs instead. So to use CPUs 1 and
1114 5, you would specify cpus_allowed=1,5. This options also
1115 allows a range of CPUs. Say you wanted a binding to CPUs
1116 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
1118 cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
1119 specified by cpus_allowed or cpumask. Two policies are
1122 shared All jobs will share the CPU set specified.
1123 split Each job will get a unique CPU from the CPU set.
1125 'shared' is the default behaviour, if the option isn't
1126 specified. If split is specified, then fio will will assign
1127 one cpu per job. If not enough CPUs are given for the jobs
1128 listed, then fio will roundrobin the CPUs in the set.
1130 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
1131 arguments allow comma delimited list of cpu numbers,
1132 A-B ranges, or 'all'. Note, to enable numa options support,
1133 fio must be built on a system with libnuma-dev(el) installed.
1135 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
1136 nodes. Format of the argements:
1138 `mode' is one of the following memory policy:
1139 default, prefer, bind, interleave, local
1140 For `default' and `local' memory policy, no node is
1141 needed to be specified.
1142 For `prefer', only one node is allowed.
1143 For `bind' and `interleave', it allow comma delimited
1144 list of numbers, A-B ranges, or 'all'.
1146 startdelay=time Start this job the specified number of seconds after fio
1147 has started. Only useful if the job file contains several
1148 jobs, and you want to delay starting some jobs to a certain
1151 runtime=time Tell fio to terminate processing after the specified number
1152 of seconds. It can be quite hard to determine for how long
1153 a specified job will run, so this parameter is handy to
1154 cap the total runtime to a given time.
1156 time_based If set, fio will run for the duration of the runtime
1157 specified even if the file(s) are completely read or
1158 written. It will simply loop over the same workload
1159 as many times as the runtime allows.
1161 ramp_time=time If set, fio will run the specified workload for this amount
1162 of time before logging any performance numbers. Useful for
1163 letting performance settle before logging results, thus
1164 minimizing the runtime required for stable results. Note
1165 that the ramp_time is considered lead in time for a job,
1166 thus it will increase the total runtime if a special timeout
1167 or runtime is specified.
1169 invalidate=bool Invalidate the buffer/page cache parts for this file prior
1170 to starting io. Defaults to true.
1172 sync=bool Use sync io for buffered writes. For the majority of the
1173 io engines, this means using O_SYNC.
1176 mem=str Fio can use various types of memory as the io unit buffer.
1177 The allowed values are:
1179 malloc Use memory from malloc(3) as the buffers.
1181 shm Use shared memory as the buffers. Allocated
1184 shmhuge Same as shm, but use huge pages as backing.
1186 mmap Use mmap to allocate buffers. May either be
1187 anonymous memory, or can be file backed if
1188 a filename is given after the option. The
1189 format is mem=mmap:/path/to/file.
1191 mmaphuge Use a memory mapped huge file as the buffer
1192 backing. Append filename after mmaphuge, ala
1193 mem=mmaphuge:/hugetlbfs/file
1195 mmapshared Same as mmap, but use a MMAP_SHARED
1198 The area allocated is a function of the maximum allowed
1199 bs size for the job, multiplied by the io depth given. Note
1200 that for shmhuge and mmaphuge to work, the system must have
1201 free huge pages allocated. This can normally be checked
1202 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1203 Linux system. Fio assumes a huge page is 4MB in size. So
1204 to calculate the number of huge pages you need for a given
1205 job file, add up the io depth of all jobs (normally one unless
1206 iodepth= is used) and multiply by the maximum bs set. Then
1207 divide that number by the huge page size. You can see the
1208 size of the huge pages in /proc/meminfo. If no huge pages
1209 are allocated by having a non-zero number in nr_hugepages,
1210 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1212 mmaphuge also needs to have hugetlbfs mounted and the file
1213 location should point there. So if it's mounted in /huge,
1214 you would use mem=mmaphuge:/huge/somefile.
1216 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1217 Note that the given alignment is applied to the first IO unit
1218 buffer, if using iodepth the alignment of the following buffers
1219 are given by the bs used. In other words, if using a bs that is
1220 a multiple of the page sized in the system, all buffers will
1221 be aligned to this value. If using a bs that is not page
1222 aligned, the alignment of subsequent IO memory buffers is the
1223 sum of the iomem_align and bs used.
1226 Defines the size of a huge page. Must at least be equal
1227 to the system setting, see /proc/meminfo. Defaults to 4MB.
1228 Should probably always be a multiple of megabytes, so using
1229 hugepage-size=Xm is the preferred way to set this to avoid
1230 setting a non-pow-2 bad value.
1232 exitall When one job finishes, terminate the rest. The default is
1233 to wait for each job to finish, sometimes that is not the
1236 exitall_on_error When one job finishes in error, terminate the rest. The
1237 default is to wait for each job to finish.
1239 bwavgtime=int Average the calculated bandwidth over the given time. Value
1240 is specified in milliseconds.
1242 iopsavgtime=int Average the calculated IOPS over the given time. Value
1243 is specified in milliseconds.
1245 create_serialize=bool If true, serialize the file creating for the jobs.
1246 This may be handy to avoid interleaving of data
1247 files, which may greatly depend on the filesystem
1248 used and even the number of processors in the system.
1250 create_fsync=bool fsync the data file after creation. This is the
1253 create_on_open=bool Don't pre-setup the files for IO, just create open()
1254 when it's time to do IO to that file.
1256 create_only=bool If true, fio will only run the setup phase of the job.
1257 If files need to be laid out or updated on disk, only
1258 that will be done. The actual job contents are not
1261 allow_file_create=bool If true, fio is permitted to create files as part
1262 of its workload. This is the default behavior. If this
1263 option is false, then fio will error out if the files it
1264 needs to use don't already exist. Default: true.
1266 allow_mounted_write=bool If this isn't set, fio will abort jobs that
1267 are destructive (eg that write) to what appears to be a
1268 mounted device or partition. This should help catch creating
1269 inadvertently destructive tests, not realizing that the test
1270 will destroy data on the mounted file system. Default: false.
1272 pre_read=bool If this is given, files will be pre-read into memory before
1273 starting the given IO operation. This will also clear
1274 the 'invalidate' flag, since it is pointless to pre-read
1275 and then drop the cache. This will only work for IO engines
1276 that are seekable, since they allow you to read the same data
1277 multiple times. Thus it will not work on eg network or splice
1280 unlink=bool Unlink the job files when done. Not the default, as repeated
1281 runs of that job would then waste time recreating the file
1282 set again and again.
1284 loops=int Run the specified number of iterations of this job. Used
1285 to repeat the same workload a given number of times. Defaults
1288 verify_only Do not perform specified workload---only verify data still
1289 matches previous invocation of this workload. This option
1290 allows one to check data multiple times at a later date
1291 without overwriting it. This option makes sense only for
1292 workloads that write data, and does not support workloads
1293 with the time_based option set.
1295 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1296 verify is set. Defaults to 1.
1298 verify=str If writing to a file, fio can verify the file contents
1299 after each iteration of the job. Each verification method also implies
1300 verification of special header, which is written to the beginning of
1301 each block. This header also includes meta information, like offset
1302 of the block, block number, timestamp when block was written, etc.
1303 verify=str can be combined with verify_pattern=str option.
1304 The allowed values are:
1306 md5 Use an md5 sum of the data area and store
1307 it in the header of each block.
1309 crc64 Use an experimental crc64 sum of the data
1310 area and store it in the header of each
1313 crc32c Use a crc32c sum of the data area and store
1314 it in the header of each block.
1316 crc32c-intel Use hardware assisted crc32c calcuation
1317 provided on SSE4.2 enabled processors. Falls
1318 back to regular software crc32c, if not
1319 supported by the system.
1321 crc32 Use a crc32 sum of the data area and store
1322 it in the header of each block.
1324 crc16 Use a crc16 sum of the data area and store
1325 it in the header of each block.
1327 crc7 Use a crc7 sum of the data area and store
1328 it in the header of each block.
1330 xxhash Use xxhash as the checksum function. Generally
1331 the fastest software checksum that fio
1334 sha512 Use sha512 as the checksum function.
1336 sha256 Use sha256 as the checksum function.
1338 sha1 Use optimized sha1 as the checksum function.
1340 meta This option is deprecated, since now meta information is
1341 included in generic verification header and meta verification
1342 happens by default. For detailed information see the description
1343 of the verify=str setting. This option is kept because of
1344 compatibility's sake with old configurations. Do not use it.
1346 pattern Verify a strict pattern. Normally fio includes
1347 a header with some basic information and
1348 checksumming, but if this option is set, only
1349 the specific pattern set with 'verify_pattern'
1352 null Only pretend to verify. Useful for testing
1353 internals with ioengine=null, not for much
1356 This option can be used for repeated burn-in tests of a
1357 system to make sure that the written data is also
1358 correctly read back. If the data direction given is
1359 a read or random read, fio will assume that it should
1360 verify a previously written file. If the data direction
1361 includes any form of write, the verify will be of the
1364 verifysort=bool If set, fio will sort written verify blocks when it deems
1365 it faster to read them back in a sorted manner. This is
1366 often the case when overwriting an existing file, since
1367 the blocks are already laid out in the file system. You
1368 can ignore this option unless doing huge amounts of really
1369 fast IO where the red-black tree sorting CPU time becomes
1372 verify_offset=int Swap the verification header with data somewhere else
1373 in the block before writing. Its swapped back before
1376 verify_interval=int Write the verification header at a finer granularity
1377 than the blocksize. It will be written for chunks the
1378 size of header_interval. blocksize should divide this
1381 verify_pattern=str If set, fio will fill the io buffers with this
1382 pattern. Fio defaults to filling with totally random
1383 bytes, but sometimes it's interesting to fill with a known
1384 pattern for io verification purposes. Depending on the
1385 width of the pattern, fio will fill 1/2/3/4 bytes of the
1386 buffer at the time(it can be either a decimal or a hex number).
1387 The verify_pattern if larger than a 32-bit quantity has to
1388 be a hex number that starts with either "0x" or "0X". Use
1389 with verify=str. Also, verify_pattern supports %o format,
1390 which means that for each block offset will be written and
1391 then verifyied back, e.g.:
1395 Or use combination of everything:
1396 verify_pattern=0xff%o"abcd"-12
1398 verify_fatal=bool Normally fio will keep checking the entire contents
1399 before quitting on a block verification failure. If this
1400 option is set, fio will exit the job on the first observed
1403 verify_dump=bool If set, dump the contents of both the original data
1404 block and the data block we read off disk to files. This
1405 allows later analysis to inspect just what kind of data
1406 corruption occurred. Off by default.
1408 verify_async=int Fio will normally verify IO inline from the submitting
1409 thread. This option takes an integer describing how many
1410 async offload threads to create for IO verification instead,
1411 causing fio to offload the duty of verifying IO contents
1412 to one or more separate threads. If using this offload
1413 option, even sync IO engines can benefit from using an
1414 iodepth setting higher than 1, as it allows them to have
1415 IO in flight while verifies are running.
1417 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1418 async IO verification threads. See cpus_allowed for the
1421 verify_backlog=int Fio will normally verify the written contents of a
1422 job that utilizes verify once that job has completed. In
1423 other words, everything is written then everything is read
1424 back and verified. You may want to verify continually
1425 instead for a variety of reasons. Fio stores the meta data
1426 associated with an IO block in memory, so for large
1427 verify workloads, quite a bit of memory would be used up
1428 holding this meta data. If this option is enabled, fio
1429 will write only N blocks before verifying these blocks.
1431 verify_backlog_batch=int Control how many blocks fio will verify
1432 if verify_backlog is set. If not set, will default to
1433 the value of verify_backlog (meaning the entire queue
1434 is read back and verified). If verify_backlog_batch is
1435 less than verify_backlog then not all blocks will be verified,
1436 if verify_backlog_batch is larger than verify_backlog, some
1437 blocks will be verified more than once.
1439 verify_state_save=bool When a job exits during the write phase of a verify
1440 workload, save its current state. This allows fio to replay
1441 up until that point, if the verify state is loaded for the
1442 verify read phase. The format of the filename is, roughly,
1443 <type>-<jobname>-<jobindex>-verify.state. <type> is "local"
1444 for a local run, "sock" for a client/server socket connection,
1445 and "ip" (192.168.0.1, for instance) for a networked
1446 client/server connection.
1448 verify_state_load=bool If a verify termination trigger was used, fio stores
1449 the current write state of each thread. This can be used at
1450 verification time so that fio knows how far it should verify.
1451 Without this information, fio will run a full verification
1452 pass, according to the settings in the job file used.
1455 wait_for_previous Wait for preceding jobs in the job file to exit, before
1456 starting this one. Can be used to insert serialization
1457 points in the job file. A stone wall also implies starting
1458 a new reporting group.
1460 new_group Start a new reporting group. See: group_reporting.
1462 numjobs=int Create the specified number of clones of this job. May be
1463 used to setup a larger number of threads/processes doing
1464 the same thing. Each thread is reported separately; to see
1465 statistics for all clones as a whole, use group_reporting in
1466 conjunction with new_group.
1468 group_reporting It may sometimes be interesting to display statistics for
1469 groups of jobs as a whole instead of for each individual job.
1470 This is especially true if 'numjobs' is used; looking at
1471 individual thread/process output quickly becomes unwieldy.
1472 To see the final report per-group instead of per-job, use
1473 'group_reporting'. Jobs in a file will be part of the same
1474 reporting group, unless if separated by a stonewall, or by
1477 thread fio defaults to forking jobs, however if this option is
1478 given, fio will use pthread_create(3) to create threads
1481 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1483 zoneskip=int Skip the specified number of bytes when zonesize data has
1484 been read. The two zone options can be used to only do
1485 io on zones of a file.
1487 write_iolog=str Write the issued io patterns to the specified file. See
1488 read_iolog. Specify a separate file for each job, otherwise
1489 the iologs will be interspersed and the file may be corrupt.
1491 read_iolog=str Open an iolog with the specified file name and replay the
1492 io patterns it contains. This can be used to store a
1493 workload and replay it sometime later. The iolog given
1494 may also be a blktrace binary file, which allows fio
1495 to replay a workload captured by blktrace. See blktrace
1496 for how to capture such logging data. For blktrace replay,
1497 the file needs to be turned into a blkparse binary data
1498 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1500 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1501 is to attempt to respect the time stamps within the log and
1502 replay them with the appropriate delay between IOPS. By
1503 setting this variable fio will not respect the timestamps and
1504 attempt to replay them as fast as possible while still
1505 respecting ordering. The result is the same I/O pattern to a
1506 given device, but different timings.
1508 replay_redirect=str While replaying I/O patterns using read_iolog the
1509 default behavior is to replay the IOPS onto the major/minor
1510 device that each IOP was recorded from. This is sometimes
1511 undesirable because on a different machine those major/minor
1512 numbers can map to a different device. Changing hardware on
1513 the same system can also result in a different major/minor
1514 mapping. Replay_redirect causes all IOPS to be replayed onto
1515 the single specified device regardless of the device it was
1516 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1517 IO in the blktrace to be replayed onto /dev/sdc. This means
1518 multiple devices will be replayed onto a single, if the trace
1519 contains multiple devices. If you want multiple devices to be
1520 replayed concurrently to multiple redirected devices you must
1521 blkparse your trace into separate traces and replay them with
1522 independent fio invocations. Unfortuantely this also breaks
1523 the strict time ordering between multiple device accesses.
1525 replay_align=int Force alignment of IO offsets and lengths in a trace
1526 to this power of 2 value.
1528 replay_scale=int Scale sector offsets down by this factor when
1531 per_job_logs=bool If set, this generates bw/clat/iops log with per
1532 file private filenames. If not set, jobs with identical names
1533 will share the log filename. Default: true.
1535 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1536 file. Can be used to store data of the bandwidth of the
1537 jobs in their lifetime. The included fio_generate_plots
1538 script uses gnuplot to turn these text files into nice
1539 graphs. See write_lat_log for behaviour of given
1540 filename. For this option, the suffix is _bw.x.log, where
1541 x is the index of the job (1..N, where N is the number of
1542 jobs). If 'per_job_logs' is false, then the filename will not
1543 include the job index.
1545 write_lat_log=str Same as write_bw_log, except that this option stores io
1546 submission, completion, and total latencies instead. If no
1547 filename is given with this option, the default filename of
1548 "jobname_type.log" is used. Even if the filename is given,
1549 fio will still append the type of log. So if one specifies
1553 The actual log names will be foo_slat.x.log, foo_clat.x.log,
1554 and foo_lat.x.log, where x is the index of the job (1..N,
1555 where N is the number of jobs). This helps fio_generate_plot
1556 fine the logs automatically. If 'per_job_logs' is false, then
1557 the filename will not include the job index.
1560 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1561 given with this option, the default filename of
1562 "jobname_type.x.log" is used,where x is the index of the job
1563 (1..N, where N is the number of jobs). Even if the filename
1564 is given, fio will still append the type of log. If
1565 'per_job_logs' is false, then the filename will not include
1568 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1569 or bw log for every IO that completes. When writing to the
1570 disk log, that can quickly grow to a very large size. Setting
1571 this option makes fio average the each log entry over the
1572 specified period of time, reducing the resolution of the log.
1575 log_offset=int If this is set, the iolog options will include the byte
1576 offset for the IO entry as well as the other data values.
1578 log_compression=int If this is set, fio will compress the IO logs as
1579 it goes, to keep the memory footprint lower. When a log
1580 reaches the specified size, that chunk is removed and
1581 compressed in the background. Given that IO logs are
1582 fairly highly compressible, this yields a nice memory
1583 savings for longer runs. The downside is that the
1584 compression will consume some background CPU cycles, so
1585 it may impact the run. This, however, is also true if
1586 the logging ends up consuming most of the system memory.
1587 So pick your poison. The IO logs are saved normally at the
1588 end of a run, by decompressing the chunks and storing them
1589 in the specified log file. This feature depends on the
1590 availability of zlib.
1592 log_compression_cpus=str Define the set of CPUs that are allowed to
1593 handle online log compression for the IO jobs. This can
1594 provide better isolation between performance sensitive jobs,
1595 and background compression work.
1597 log_store_compressed=bool If set, fio will store the log files in a
1598 compressed format. They can be decompressed with fio, using
1599 the --inflate-log command line parameter. The files will be
1600 stored with a .fz suffix.
1602 block_error_percentiles=bool If set, record errors in trim block-sized
1603 units from writes and trims and output a histogram of
1604 how many trims it took to get to errors, and what kind
1605 of error was encountered.
1607 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1608 potentially be used instead of removing memory or booting
1609 with less memory to simulate a smaller amount of memory.
1610 The amount specified is per worker.
1612 exec_prerun=str Before running this job, issue the command specified
1613 through system(3). Output is redirected in a file called
1616 exec_postrun=str After the job completes, issue the command specified
1617 though system(3). Output is redirected in a file called
1618 jobname.postrun.txt.
1620 ioscheduler=str Attempt to switch the device hosting the file to the specified
1621 io scheduler before running.
1623 disk_util=bool Generate disk utilization statistics, if the platform
1624 supports it. Defaults to on.
1626 disable_lat=bool Disable measurements of total latency numbers. Useful
1627 only for cutting back the number of calls to gettimeofday,
1628 as that does impact performance at really high IOPS rates.
1629 Note that to really get rid of a large amount of these
1630 calls, this option must be used with disable_slat and
1633 disable_clat=bool Disable measurements of completion latency numbers. See
1636 disable_slat=bool Disable measurements of submission latency numbers. See
1639 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1642 clat_percentiles=bool Enable the reporting of percentiles of
1643 completion latencies.
1645 percentile_list=float_list Overwrite the default list of percentiles
1646 for completion latencies and the block error histogram.
1647 Each number is a floating number in the range (0,100],
1648 and the maximum length of the list is 20. Use ':'
1649 to separate the numbers, and list the numbers in ascending
1650 order. For example, --percentile_list=99.5:99.9 will cause
1651 fio to report the values of completion latency below which
1652 99.5% and 99.9% of the observed latencies fell, respectively.
1654 clocksource=str Use the given clocksource as the base of timing. The
1655 supported options are:
1657 gettimeofday gettimeofday(2)
1659 clock_gettime clock_gettime(2)
1661 cpu Internal CPU clock source
1663 cpu is the preferred clocksource if it is reliable, as it
1664 is very fast (and fio is heavy on time calls). Fio will
1665 automatically use this clocksource if it's supported and
1666 considered reliable on the system it is running on, unless
1667 another clocksource is specifically set. For x86/x86-64 CPUs,
1668 this means supporting TSC Invariant.
1670 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1671 (disable_clat, disable_slat, disable_bw) plus reduce
1672 precision of the timeout somewhat to really shrink
1673 the gettimeofday() call count. With this option enabled,
1674 we only do about 0.4% of the gtod() calls we would have
1675 done if all time keeping was enabled.
1677 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1678 execution to just getting the current time. Fio (and
1679 databases, for instance) are very intensive on gettimeofday()
1680 calls. With this option, you can set one CPU aside for
1681 doing nothing but logging current time to a shared memory
1682 location. Then the other threads/processes that run IO
1683 workloads need only copy that segment, instead of entering
1684 the kernel with a gettimeofday() call. The CPU set aside
1685 for doing these time calls will be excluded from other
1686 uses. Fio will manually clear it from the CPU mask of other
1689 continue_on_error=str Normally fio will exit the job on the first observed
1690 failure. If this option is set, fio will continue the job when
1691 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1692 is exceeded or the I/O size specified is completed. If this
1693 option is used, there are two more stats that are appended,
1694 the total error count and the first error. The error field
1695 given in the stats is the first error that was hit during the
1698 The allowed values are:
1700 none Exit on any IO or verify errors.
1702 read Continue on read errors, exit on all others.
1704 write Continue on write errors, exit on all others.
1706 io Continue on any IO error, exit on all others.
1708 verify Continue on verify errors, exit on all others.
1710 all Continue on all errors.
1712 0 Backward-compatible alias for 'none'.
1714 1 Backward-compatible alias for 'all'.
1716 ignore_error=str Sometimes you want to ignore some errors during test
1717 in that case you can specify error list for each error type.
1718 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1719 errors for given error type is separated with ':'. Error
1720 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1722 ignore_error=EAGAIN,ENOSPC:122
1723 This option will ignore EAGAIN from READ, and ENOSPC and
1724 122(EDQUOT) from WRITE.
1726 error_dump=bool If set dump every error even if it is non fatal, true
1727 by default. If disabled only fatal error will be dumped
1729 cgroup=str Add job to this control group. If it doesn't exist, it will
1730 be created. The system must have a mounted cgroup blkio
1731 mount point for this to work. If your system doesn't have it
1732 mounted, you can do so with:
1734 # mount -t cgroup -o blkio none /cgroup
1736 cgroup_weight=int Set the weight of the cgroup to this value. See
1737 the documentation that comes with the kernel, allowed values
1738 are in the range of 100..1000.
1740 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1741 the job completion. To override this behavior and to leave
1742 cgroups around after the job completion, set cgroup_nodelete=1.
1743 This can be useful if one wants to inspect various cgroup
1744 files after job completion. Default: false
1746 uid=int Instead of running as the invoking user, set the user ID to
1747 this value before the thread/process does any work.
1749 gid=int Set group ID, see uid.
1751 flow_id=int The ID of the flow. If not specified, it defaults to being a
1752 global flow. See flow.
1754 flow=int Weight in token-based flow control. If this value is used, then
1755 there is a 'flow counter' which is used to regulate the
1756 proportion of activity between two or more jobs. fio attempts
1757 to keep this flow counter near zero. The 'flow' parameter
1758 stands for how much should be added or subtracted to the flow
1759 counter on each iteration of the main I/O loop. That is, if
1760 one job has flow=8 and another job has flow=-1, then there
1761 will be a roughly 1:8 ratio in how much one runs vs the other.
1763 flow_watermark=int The maximum value that the absolute value of the flow
1764 counter is allowed to reach before the job must wait for a
1765 lower value of the counter.
1767 flow_sleep=int The period of time, in microseconds, to wait after the flow
1768 watermark has been exceeded before retrying operations
1770 In addition, there are some parameters which are only valid when a specific
1771 ioengine is in use. These are used identically to normal parameters, with the
1772 caveat that when used on the command line, they must come after the ioengine
1773 that defines them is selected.
1775 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1776 the io_getevents system call to reap newly returned events.
1777 With this flag turned on, the AIO ring will be read directly
1778 from user-space to reap events. The reaping mode is only
1779 enabled when polling for a minimum of 0 events (eg when
1780 iodepth_batch_complete=0).
1782 [cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1784 [cpu] cpuchunks=int Split the load into cycles of the given time. In
1787 [cpu] exit_on_io_done=bool Detect when IO threads are done, then exit.
1789 [netsplice] hostname=str
1790 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1791 If the job is a TCP listener or UDP reader, the hostname is not
1792 used and must be omitted unless it is a valid UDP multicast
1794 [libhdfs] namenode=str The host name or IP address of a HDFS cluster namenode to contact.
1796 [netsplice] port=int
1797 [net] port=int The TCP or UDP port to bind to or connect to. If this is used
1798 with numjobs to spawn multiple instances of the same job type, then this will
1799 be the starting port number since fio will use a range of ports.
1800 [libhdfs] port=int the listening port of the HFDS cluster namenode.
1802 [netsplice] interface=str
1803 [net] interface=str The IP address of the network interface used to send or
1804 receive UDP multicast
1807 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1810 [netsplice] nodelay=bool
1811 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1813 [netsplice] protocol=str
1814 [netsplice] proto=str
1816 [net] proto=str The network protocol to use. Accepted values are:
1818 tcp Transmission control protocol
1819 tcpv6 Transmission control protocol V6
1820 udp User datagram protocol
1821 udpv6 User datagram protocol V6
1822 unix UNIX domain socket
1824 When the protocol is TCP or UDP, the port must also be given,
1825 as well as the hostname if the job is a TCP listener or UDP
1826 reader. For unix sockets, the normal filename option should be
1827 used and the port is invalid.
1829 [net] listen For TCP network connections, tell fio to listen for incoming
1830 connections rather than initiating an outgoing connection. The
1831 hostname must be omitted if this option is used.
1833 [net] pingpong Normaly a network writer will just continue writing data, and
1834 a network reader will just consume packages. If pingpong=1
1835 is set, a writer will send its normal payload to the reader,
1836 then wait for the reader to send the same payload back. This
1837 allows fio to measure network latencies. The submission
1838 and completion latencies then measure local time spent
1839 sending or receiving, and the completion latency measures
1840 how long it took for the other end to receive and send back.
1841 For UDP multicast traffic pingpong=1 should only be set for a
1842 single reader when multiple readers are listening to the same
1845 [net] window_size Set the desired socket buffer size for the connection.
1847 [net] mss Set the TCP maximum segment size (TCP_MAXSEG).
1849 [e4defrag] donorname=str
1850 File will be used as a block donor(swap extents between files)
1851 [e4defrag] inplace=int
1852 Configure donor file blocks allocation strategy
1853 0(default): Preallocate donor's file on init
1854 1 : allocate space immidietly inside defragment event,
1855 and free right after event
1857 [mtd] skip_bad=bool Skip operations against known bad blocks.
1859 [libhdfs] hdfsdirectory libhdfs will create chunk in this HDFS directory
1860 [libhdfs] chunck_size the size of the chunck to use for each file.
1863 6.0 Interpreting the output
1864 ---------------------------
1866 fio spits out a lot of output. While running, fio will display the
1867 status of the jobs created. An example of that would be:
1869 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1871 The characters inside the square brackets denote the current status of
1872 each thread. The possible values (in typical life cycle order) are:
1876 P Thread setup, but not started.
1878 I Thread initialized, waiting or generating necessary data.
1879 p Thread running pre-reading file(s).
1880 R Running, doing sequential reads.
1881 r Running, doing random reads.
1882 W Running, doing sequential writes.
1883 w Running, doing random writes.
1884 M Running, doing mixed sequential reads/writes.
1885 m Running, doing mixed random reads/writes.
1886 F Running, currently waiting for fsync()
1887 f Running, finishing up (writing IO logs, etc)
1888 V Running, doing verification of written data.
1889 E Thread exited, not reaped by main thread yet.
1891 X Thread reaped, exited with an error.
1892 K Thread reaped, exited due to signal.
1894 Fio will condense the thread string as not to take up more space on the
1895 command line as is needed. For instance, if you have 10 readers and 10
1896 writers running, the output would look like this:
1898 Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s]
1900 Fio will still maintain the ordering, though. So the above means that jobs
1901 1..10 are readers, and 11..20 are writers.
1903 The other values are fairly self explanatory - number of threads
1904 currently running and doing io, rate of io since last check (read speed
1905 listed first, then write speed), and the estimated completion percentage
1906 and time for the running group. It's impossible to estimate runtime of
1907 the following groups (if any). Note that the string is displayed in order,
1908 so it's possible to tell which of the jobs are currently doing what. The
1909 first character is the first job defined in the job file, and so forth.
1911 When fio is done (or interrupted by ctrl-c), it will show the data for
1912 each thread, group of threads, and disks in that order. For each data
1913 direction, the output looks like:
1915 Client1 (g=0): err= 0:
1916 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1917 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1918 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1919 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1920 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1921 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1922 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1923 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1924 issued r/w: total=0/32768, short=0/0
1925 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1926 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1928 The client number is printed, along with the group id and error of that
1929 thread. Below is the io statistics, here for writes. In the order listed,
1932 io= Number of megabytes io performed
1933 bw= Average bandwidth rate
1934 iops= Average IOs performed per second
1935 runt= The runtime of that thread
1936 slat= Submission latency (avg being the average, stdev being the
1937 standard deviation). This is the time it took to submit
1938 the io. For sync io, the slat is really the completion
1939 latency, since queue/complete is one operation there. This
1940 value can be in milliseconds or microseconds, fio will choose
1941 the most appropriate base and print that. In the example
1942 above, milliseconds is the best scale. Note: in --minimal mode
1943 latencies are always expressed in microseconds.
1944 clat= Completion latency. Same names as slat, this denotes the
1945 time from submission to completion of the io pieces. For
1946 sync io, clat will usually be equal (or very close) to 0,
1947 as the time from submit to complete is basically just
1948 CPU time (io has already been done, see slat explanation).
1949 bw= Bandwidth. Same names as the xlat stats, but also includes
1950 an approximate percentage of total aggregate bandwidth
1951 this thread received in this group. This last value is
1952 only really useful if the threads in this group are on the
1953 same disk, since they are then competing for disk access.
1954 cpu= CPU usage. User and system time, along with the number
1955 of context switches this thread went through, usage of
1956 system and user time, and finally the number of major
1957 and minor page faults.
1958 IO depths= The distribution of io depths over the job life time. The
1959 numbers are divided into powers of 2, so for example the
1960 16= entries includes depths up to that value but higher
1961 than the previous entry. In other words, it covers the
1962 range from 16 to 31.
1963 IO submit= How many pieces of IO were submitting in a single submit
1964 call. Each entry denotes that amount and below, until
1965 the previous entry - eg, 8=100% mean that we submitted
1966 anywhere in between 5-8 ios per submit call.
1967 IO complete= Like the above submit number, but for completions instead.
1968 IO issued= The number of read/write requests issued, and how many
1970 IO latencies= The distribution of IO completion latencies. This is the
1971 time from when IO leaves fio and when it gets completed.
1972 The numbers follow the same pattern as the IO depths,
1973 meaning that 2=1.6% means that 1.6% of the IO completed
1974 within 2 msecs, 20=12.8% means that 12.8% of the IO
1975 took more than 10 msecs, but less than (or equal to) 20 msecs.
1977 After each client has been listed, the group statistics are printed. They
1978 will look like this:
1980 Run status group 0 (all jobs):
1981 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1982 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1984 For each data direction, it prints:
1986 io= Number of megabytes io performed.
1987 aggrb= Aggregate bandwidth of threads in this group.
1988 minb= The minimum average bandwidth a thread saw.
1989 maxb= The maximum average bandwidth a thread saw.
1990 mint= The smallest runtime of the threads in that group.
1991 maxt= The longest runtime of the threads in that group.
1993 And finally, the disk statistics are printed. They will look like this:
1995 Disk stats (read/write):
1996 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1998 Each value is printed for both reads and writes, with reads first. The
2001 ios= Number of ios performed by all groups.
2002 merge= Number of merges io the io scheduler.
2003 ticks= Number of ticks we kept the disk busy.
2004 io_queue= Total time spent in the disk queue.
2005 util= The disk utilization. A value of 100% means we kept the disk
2006 busy constantly, 50% would be a disk idling half of the time.
2008 It is also possible to get fio to dump the current output while it is
2009 running, without terminating the job. To do that, send fio the USR1 signal.
2010 You can also get regularly timed dumps by using the --status-interval
2011 parameter, or by creating a file in /tmp named fio-dump-status. If fio
2012 sees this file, it will unlink it and dump the current output status.
2018 For scripted usage where you typically want to generate tables or graphs
2019 of the results, fio can output the results in a semicolon separated format.
2020 The format is one long line of values, such as:
2022 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%
2023 A description of this job goes here.
2025 The job description (if provided) follows on a second line.
2027 To enable terse output, use the --minimal command line option. The first
2028 value is the version of the terse output format. If the output has to
2029 be changed for some reason, this number will be incremented by 1 to
2030 signify that change.
2032 Split up, the format is as follows:
2034 terse version, fio version, jobname, groupid, error
2036 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
2037 Submission latency: min, max, mean, stdev (usec)
2038 Completion latency: min, max, mean, stdev (usec)
2039 Completion latency percentiles: 20 fields (see below)
2040 Total latency: min, max, mean, stdev (usec)
2041 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
2043 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
2044 Submission latency: min, max, mean, stdev (usec)
2045 Completion latency: min, max, mean, stdev(usec)
2046 Completion latency percentiles: 20 fields (see below)
2047 Total latency: min, max, mean, stdev (usec)
2048 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
2049 CPU usage: user, system, context switches, major faults, minor faults
2050 IO depths: <=1, 2, 4, 8, 16, 32, >=64
2051 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
2052 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
2053 Disk utilization: Disk name, Read ios, write ios,
2054 Read merges, write merges,
2055 Read ticks, write ticks,
2056 Time spent in queue, disk utilization percentage
2057 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
2059 Additional Info (dependent on description being set): Text description
2061 Completion latency percentiles can be a grouping of up to 20 sets, so
2062 for the terse output fio writes all of them. Each field will look like this:
2066 which is the Xth percentile, and the usec latency associated with it.
2068 For disk utilization, all disks used by fio are shown. So for each disk
2069 there will be a disk utilization section.
2072 8.0 Trace file format
2073 ---------------------
2074 There are two trace file format that you can encounter. The older (v1) format
2075 is unsupported since version 1.20-rc3 (March 2008). It will still be described
2076 below in case that you get an old trace and want to understand it.
2078 In any case the trace is a simple text file with a single action per line.
2081 8.1 Trace file format v1
2082 ------------------------
2083 Each line represents a single io action in the following format:
2087 where rw=0/1 for read/write, and the offset and length entries being in bytes.
2089 This format is not supported in Fio versions => 1.20-rc3.
2092 8.2 Trace file format v2
2093 ------------------------
2094 The second version of the trace file format was added in Fio version 1.17.
2095 It allows to access more then one file per trace and has a bigger set of
2096 possible file actions.
2098 The first line of the trace file has to be:
2102 Following this can be lines in two different formats, which are described below.
2104 The file management format:
2108 The filename is given as an absolute path. The action can be one of these:
2110 add Add the given filename to the trace
2111 open Open the file with the given filename. The filename has to have
2112 been added with the add action before.
2113 close Close the file with the given filename. The file has to have been
2117 The file io action format:
2119 filename action offset length
2121 The filename is given as an absolute path, and has to have been added and opened
2122 before it can be used with this format. The offset and length are given in
2123 bytes. The action can be one of these:
2125 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
2126 The time is relative to the previous wait statement.
2127 read Read 'length' bytes beginning from 'offset'
2128 write Write 'length' bytes beginning from 'offset'
2129 sync fsync() the file
2130 datasync fdatasync() the file
2131 trim trim the given file from the given 'offset' for 'length' bytes
2134 9.0 CPU idleness profiling
2135 --------------------------
2136 In some cases, we want to understand CPU overhead in a test. For example,
2137 we test patches for the specific goodness of whether they reduce CPU usage.
2138 fio implements a balloon approach to create a thread per CPU that runs at
2139 idle priority, meaning that it only runs when nobody else needs the cpu.
2140 By measuring the amount of work completed by the thread, idleness of each
2141 CPU can be derived accordingly.
2143 An unit work is defined as touching a full page of unsigned characters. Mean
2144 and standard deviation of time to complete an unit work is reported in "unit
2145 work" section. Options can be chosen to report detailed percpu idleness or
2146 overall system idleness by aggregating percpu stats.
2149 10.0 Verification and triggers
2150 ------------------------------
2151 Fio is usually run in one of two ways, when data verification is done. The
2152 first is a normal write job of some sort with verify enabled. When the
2153 write phase has completed, fio switches to reads and verifies everything
2154 it wrote. The second model is running just the write phase, and then later
2155 on running the same job (but with reads instead of writes) to repeat the
2156 same IO patterns and verify the contents. Both of these methods depend
2157 on the write phase being completed, as fio otherwise has no idea how much
2160 With verification triggers, fio supports dumping the current write state
2161 to local files. Then a subsequent read verify workload can load this state
2162 and know exactly where to stop. This is useful for testing cases where
2163 power is cut to a server in a managed fashion, for instance.
2165 A verification trigger consists of two things:
2167 1) Storing the write state of each job
2168 2) Executing a trigger command
2170 The write state is relatively small, on the order of hundreds of bytes
2171 to single kilobytes. It contains information on the number of completions
2172 done, the last X completions, etc.
2174 A trigger is invoked either through creation ('touch') of a specified
2175 file in the system, or through a timeout setting. If fio is run with
2176 --trigger-file=/tmp/trigger-file, then it will continually check for
2177 the existence of /tmp/trigger-file. When it sees this file, it will
2178 fire off the trigger (thus saving state, and executing the trigger
2181 For client/server runs, there's both a local and remote trigger. If
2182 fio is running as a server backend, it will send the job states back
2183 to the client for safe storage, then execute the remote trigger, if
2184 specified. If a local trigger is specified, the server will still send
2185 back the write state, but the client will then execute the trigger.
2187 10.1 Verification trigger example
2188 ---------------------------------
2189 Lets say we want to run a powercut test on the remote machine 'server'.
2190 Our write workload is in write-test.fio. We want to cut power to 'server'
2191 at some point during the run, and we'll run this test from the safety
2192 or our local machine, 'localbox'. On the server, we'll start the fio
2195 server# fio --server
2197 and on the client, we'll fire off the workload:
2199 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
2201 We set /tmp/my-trigger as the trigger file, and we tell fio to execute
2203 echo b > /proc/sysrq-trigger
2205 on the server once it has received the trigger and sent us the write
2206 state. This will work, but it's not _really_ cutting power to the server,
2207 it's merely abruptly rebooting it. If we have a remote way of cutting
2208 power to the server through IPMI or similar, we could do that through
2209 a local trigger command instead. Lets assume we have a script that does
2210 IPMI reboot of a given hostname, ipmi-reboot. On localbox, we could
2211 then have run fio with a local trigger instead:
2213 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
2215 For this case, fio would wait for the server to send us the write state,
2216 then execute 'ipmi-reboot server' when that happened.
2218 10.1 Loading verify state
2219 -------------------------
2220 To load store write state, read verification job file must contain
2221 the verify_state_load option. If that is set, fio will load the previously
2222 stored state. For a local fio run this is done by loading the files directly,
2223 and on a client/server run, the server backend will ask the client to send
2224 the files over and load them from there.