8 5. Detailed list of parameters
12 9. CPU idleness profiling
13 10. Verification and triggers
17 1.0 Overview and history
18 ------------------------
19 fio was originally written to save me the hassle of writing special test
20 case programs when I wanted to test a specific workload, either for
21 performance reasons or to find/reproduce a bug. The process of writing
22 such a test app can be tiresome, especially if you have to do it often.
23 Hence I needed a tool that would be able to simulate a given io workload
24 without resorting to writing a tailored test case again and again.
26 A test work load is difficult to define, though. There can be any number
27 of processes or threads involved, and they can each be using their own
28 way of generating io. You could have someone dirtying large amounts of
29 memory in an memory mapped file, or maybe several threads issuing
30 reads using asynchronous io. fio needed to be flexible enough to
31 simulate both of these cases, and many more.
35 The first step in getting fio to simulate a desired io workload, is
36 writing a job file describing that specific setup. A job file may contain
37 any number of threads and/or files - the typical contents of the job file
38 is a global section defining shared parameters, and one or more job
39 sections describing the jobs involved. When run, fio parses this file
40 and sets everything up as described. If we break down a job from top to
41 bottom, it contains the following basic parameters:
43 IO type Defines the io pattern issued to the file(s).
44 We may only be reading sequentially from this
45 file(s), or we may be writing randomly. Or even
46 mixing reads and writes, sequentially or randomly.
48 Block size In how large chunks are we issuing io? This may be
49 a single value, or it may describe a range of
52 IO size How much data are we going to be reading/writing.
54 IO engine How do we issue io? We could be memory mapping the
55 file, we could be using regular read/write, we
56 could be using splice, async io, or even SG
59 IO depth If the io engine is async, how large a queuing
60 depth do we want to maintain?
62 IO type Should we be doing buffered io, or direct/raw io?
64 Num files How many files are we spreading the workload over.
66 Num threads How many threads or processes should we spread
69 The above are the basic parameters defined for a workload, in addition
70 there's a multitude of parameters that modify other aspects of how this
76 See the README file for command line parameters, there are only a few
79 Running fio is normally the easiest part - you just give it the job file
80 (or job files) as parameters:
84 and it will start doing what the job_file tells it to do. You can give
85 more than one job file on the command line, fio will serialize the running
86 of those files. Internally that is the same as using the 'stonewall'
87 parameter described in the parameter section.
89 If the job file contains only one job, you may as well just give the
90 parameters on the command line. The command line parameters are identical
91 to the job parameters, with a few extra that control global parameters
92 (see README). For example, for the job file parameter iodepth=2, the
93 mirror command line option would be --iodepth 2 or --iodepth=2. You can
94 also use the command line for giving more than one job entry. For each
95 --name option that fio sees, it will start a new job with that name.
96 Command line entries following a --name entry will apply to that job,
97 until there are no more entries or a new --name entry is seen. This is
98 similar to the job file options, where each option applies to the current
99 job until a new [] job entry is seen.
101 fio does not need to run as root, except if the files or devices specified
102 in the job section requires that. Some other options may also be restricted,
103 such as memory locking, io scheduler switching, and decreasing the nice value.
108 As previously described, fio accepts one or more job files describing
109 what it is supposed to do. The job file format is the classic ini file,
110 where the names enclosed in [] brackets define the job name. You are free
111 to use any ascii name you want, except 'global' which has special meaning.
112 A global section sets defaults for the jobs described in that file. A job
113 may override a global section parameter, and a job file may even have
114 several global sections if so desired. A job is only affected by a global
115 section residing above it. If the first character in a line is a ';' or a
116 '#', the entire line is discarded as a comment.
118 So let's look at a really simple job file that defines two processes, each
119 randomly reading from a 128MB file.
121 ; -- start job file --
132 As you can see, the job file sections themselves are empty as all the
133 described parameters are shared. As no filename= option is given, fio
134 makes up a filename for each of the jobs as it sees fit. On the command
135 line, this job would look as follows:
137 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
140 Let's look at an example that has a number of processes writing randomly
143 ; -- start job file --
155 Here we have no global section, as we only have one job defined anyway.
156 We want to use async io here, with a depth of 4 for each file. We also
157 increased the buffer size used to 32KB and define numjobs to 4 to
158 fork 4 identical jobs. The result is 4 processes each randomly writing
159 to their own 64MB file. Instead of using the above job file, you could
160 have given the parameters on the command line. For this case, you would
163 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
165 When fio is utilized as a basis of any reasonably large test suite, it might be
166 desirable to share a set of standardized settings across multiple job files.
167 Instead of copy/pasting such settings, any section may pull in an external
168 .fio file with 'include filename' directive, as in the following example:
170 ; -- start job file including.fio --
174 include glob-include.fio
181 include test-include.fio
182 ; -- end job file including.fio --
184 ; -- start job file glob-include.fio --
187 ; -- end job file glob-include.fio --
189 ; -- start job file test-include.fio --
192 ; -- end job file test-include.fio --
194 Settings pulled into a section apply to that section only (except global
195 section). Include directives may be nested in that any included file may
196 contain further include directive(s). Include files may not contain []
200 4.1 Environment variables
201 -------------------------
203 fio also supports environment variable expansion in job files. Any
204 sub-string of the form "${VARNAME}" as part of an option value (in other
205 words, on the right of the `='), will be expanded to the value of the
206 environment variable called VARNAME. If no such environment variable
207 is defined, or VARNAME is the empty string, the empty string will be
210 As an example, let's look at a sample fio invocation and job file:
212 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
214 ; -- start job file --
221 This will expand to the following equivalent job file at runtime:
223 ; -- start job file --
230 fio ships with a few example job files, you can also look there for
233 4.2 Reserved keywords
234 ---------------------
236 Additionally, fio has a set of reserved keywords that will be replaced
237 internally with the appropriate value. Those keywords are:
239 $pagesize The architecture page size of the running system
240 $mb_memory Megabytes of total memory in the system
241 $ncpus Number of online available CPUs
243 These can be used on the command line or in the job file, and will be
244 automatically substituted with the current system values when the job
245 is run. Simple math is also supported on these keywords, so you can
246 perform actions like:
250 and get that properly expanded to 8 times the size of memory in the
254 5.0 Detailed list of parameters
255 -------------------------------
257 This section describes in details each parameter associated with a job.
258 Some parameters take an option of a given type, such as an integer or
259 a string. Anywhere a numeric value is required, an arithmetic expression
260 may be used, provided it is surrounded by parentheses. Supported operators
270 For time values in expressions, units are microseconds by default. This is
271 different than for time values not in expressions (not enclosed in
272 parentheses). The following types are used:
274 str String. This is a sequence of alpha characters.
275 time Integer with possible time suffix. In seconds unless otherwise
276 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
277 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
278 and 'us' (or 'usec') for microseconds.
279 int SI integer. A whole number value, which may contain a suffix
280 describing the base of the number. Accepted suffixes are k/m/g/t/p,
281 meaning kilo, mega, giga, tera, and peta. The suffix is not case
282 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
283 as 'k'). So if you want to specify 4096, you could either write
284 out '4096' or just give 4k. The suffixes signify base 2 values, so
285 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
286 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
287 case, then 1000 is used as the multiplier. This can be handy for
288 disks, since manufacturers generally use base 10 values when listing
289 the capacity of a drive. If the option accepts an upper and lower
290 range, use a colon ':' or minus '-' to separate such values. May also
291 include a prefix to indicate numbers base. If 0x is used, the number
292 is assumed to be hexadecimal. See irange.
293 bool Boolean. Usually parsed as an integer, however only defined for
294 true and false (1 and 0).
295 irange Integer range with suffix. Allows value range to be given, such
296 as 1024-4096. A colon may also be used as the separator, eg
297 1k:4k. If the option allows two sets of ranges, they can be
298 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
300 float_list A list of floating point numbers, separated by a ':' character.
302 With the above in mind, here follows the complete list of fio job
305 name=str ASCII name of the job. This may be used to override the
306 name printed by fio for this job. Otherwise the job
307 name is used. On the command line this parameter has the
308 special purpose of also signaling the start of a new
311 wait_for=str Specifies the name of the already defined job to wait
312 for. Single waitee name only may be specified. If set, the job
313 won't be started until all workers of the waitee job are done.
315 Wait_for operates on the job name basis, so there are a few
316 limitations. First, the waitee must be defined prior to the
317 waiter job (meaning no forward references). Second, if a job
318 is being referenced as a waitee, it must have a unique name
319 (no duplicate waitees).
321 description=str Text description of the job. Doesn't do anything except
322 dump this text description when this job is run. It's
325 directory=str Prefix filenames with this directory. Used to place files
326 in a different location than "./". See the 'filename' option
327 for escaping certain characters.
329 filename=str Fio normally makes up a filename based on the job name,
330 thread number, and file number. If you want to share
331 files between threads in a job or several jobs, specify
332 a filename for each of them to override the default.
333 If the ioengine is file based, you can specify a number of
334 files by separating the names with a ':' colon. So if you
335 wanted a job to open /dev/sda and /dev/sdb as the two working
336 files, you would use filename=/dev/sda:/dev/sdb. On Windows,
337 disk devices are accessed as \\.\PhysicalDrive0 for the first
338 device, \\.\PhysicalDrive1 for the second etc. Note: Windows
339 and FreeBSD prevent write access to areas of the disk
340 containing in-use data (e.g. filesystems).
341 If the wanted filename does need to include a colon, then
342 escape that with a '\' character. For instance, if the filename
343 is "/dev/dsk/foo@3,0:c", then you would use
344 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
345 stdin or stdout. Which of the two depends on the read/write
349 If sharing multiple files between jobs, it is usually necessary
350 to have fio generate the exact names that you want. By default,
351 fio will name a file based on the default file format
352 specification of jobname.jobnumber.filenumber. With this
353 option, that can be customized. Fio will recognize and replace
354 the following keywords in this string:
357 The name of the worker thread or process.
360 The incremental number of the worker thread or
364 The incremental number of the file for that worker
367 To have dependent jobs share a set of files, this option can
368 be set to have fio generate filenames that are shared between
369 the two. For instance, if testfiles.$filenum is specified,
370 file number 4 for any job will be named testfiles.4. The
371 default of $jobname.$jobnum.$filenum will be used if
372 no other format specifier is given.
374 unique_filename=bool To avoid collisions between networked clients, fio
375 defaults to prefixing any generated filenames (with a directory
376 specified) with the source of the client connecting. To disable
377 this behavior, set this option to 0.
379 opendir=str Tell fio to recursively add any file it can find in this
380 directory and down the file system tree.
382 lockfile=str Fio defaults to not locking any files before it does
383 IO to them. If a file or file descriptor is shared, fio
384 can serialize IO to that file to make the end result
385 consistent. This is usual for emulating real workloads that
386 share files. The lock modes are:
388 none No locking. The default.
389 exclusive Only one thread/process may do IO,
390 excluding all others.
391 readwrite Read-write locking on the file. Many
392 readers may access the file at the
393 same time, but writes get exclusive
397 rw=str Type of io pattern. Accepted values are:
399 read Sequential reads
400 write Sequential writes
401 trim Sequential trims
402 randwrite Random writes
403 randread Random reads
404 randtrim Random trims
405 rw,readwrite Sequential mixed reads and writes
406 randrw Random mixed reads and writes
407 trimwrite Sequential trim+write sequences
409 Fio defaults to read if the option is not specified.
410 For the mixed io types, the default is to split them 50/50.
411 For certain types of io the result may still be skewed a bit,
412 since the speed may be different. It is possible to specify
413 a number of IO's to do before getting a new offset, this is
414 done by appending a ':<nr>' to the end of the string given.
415 For a random read, it would look like 'rw=randread:8' for
416 passing in an offset modifier with a value of 8. If the
417 suffix is used with a sequential IO pattern, then the value
418 specified will be added to the generated offset for each IO.
419 For instance, using rw=write:4k will skip 4k for every
420 write. It turns sequential IO into sequential IO with holes.
421 See the 'rw_sequencer' option.
423 rw_sequencer=str If an offset modifier is given by appending a number to
424 the rw=<str> line, then this option controls how that
425 number modifies the IO offset being generated. Accepted
428 sequential Generate sequential offset
429 identical Generate the same offset
431 'sequential' is only useful for random IO, where fio would
432 normally generate a new random offset for every IO. If you
433 append eg 8 to randread, you would get a new random offset for
434 every 8 IO's. The result would be a seek for only every 8
435 IO's, instead of for every IO. Use rw=randread:8 to specify
436 that. As sequential IO is already sequential, setting
437 'sequential' for that would not result in any differences.
438 'identical' behaves in a similar fashion, except it sends
439 the same offset 8 number of times before generating a new
442 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
443 Storage manufacturers like to use 10^3 or 1000 as a base
444 ten unit instead, for obvious reasons. Allow values are
445 1024 or 1000, with 1024 being the default.
447 unified_rw_reporting=bool Fio normally reports statistics on a per
448 data direction basis, meaning that reads, writes, and trims are
449 accounted and reported separately. If this option is set,
450 the fio will sum the results and report them as "mixed"
453 randrepeat=bool For random IO workloads, seed the generator in a predictable
454 way so that results are repeatable across repetitions.
457 randseed=int Seed the random number generators based on this seed value, to
458 be able to control what sequence of output is being generated.
459 If not set, the random sequence depends on the randrepeat
462 fallocate=str Whether pre-allocation is performed when laying down files.
465 none Do not pre-allocate space
466 posix Pre-allocate via posix_fallocate()
467 keep Pre-allocate via fallocate() with
468 FALLOC_FL_KEEP_SIZE set
469 0 Backward-compatible alias for 'none'
470 1 Backward-compatible alias for 'posix'
472 May not be available on all supported platforms. 'keep' is only
473 available on Linux.If using ZFS on Solaris this must be set to
474 'none' because ZFS doesn't support it. Default: 'posix'.
476 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
477 on what IO patterns it is likely to issue. Sometimes you
478 want to test specific IO patterns without telling the
479 kernel about it, in which case you can disable this option.
480 The following options are supported:
482 sequential Use FADV_SEQUENTIAL
483 random Use FADV_RANDOM
484 1 Backwards-compatible hint for basing
485 the hint on the fio workload. Will use
486 FADV_SEQUENTIAL for a sequential
487 workload, and FADV_RANDOM for a random
489 0 Backwards-compatible setting for not
490 issing a fadvise hint.
492 fadvise_stream=int Notify the kernel what write stream ID to place these
493 writes under. Only supported on Linux. Note, this option
494 may change going forward.
496 size=int The total size of file io for this job. Fio will run until
497 this many bytes has been transferred, unless runtime is
498 limited by other options (such as 'runtime', for instance,
499 or increased/decreased by 'io_size'). Unless specific nrfiles
500 and filesize options are given, fio will divide this size
501 between the available files specified by the job. If not set,
502 fio will use the full size of the given files or devices.
503 If the files do not exist, size must be given. It is also
504 possible to give size as a percentage between 1 and 100. If
505 size=20% is given, fio will use 20% of the full size of the
506 given files or devices.
509 io_limit=int Normally fio operates within the region set by 'size', which
510 means that the 'size' option sets both the region and size of
511 IO to be performed. Sometimes that is not what you want. With
512 this option, it is possible to define just the amount of IO
513 that fio should do. For instance, if 'size' is set to 20G and
514 'io_size' is set to 5G, fio will perform IO within the first
515 20G but exit when 5G have been done. The opposite is also
516 possible - if 'size' is set to 20G, and 'io_size' is set to
517 40G, then fio will do 40G of IO within the 0..20G region.
519 filesize=int Individual file sizes. May be a range, in which case fio
520 will select sizes for files at random within the given range
521 and limited to 'size' in total (if that is given). If not
522 given, each created file is the same size.
524 file_append=bool Perform IO after the end of the file. Normally fio will
525 operate within the size of a file. If this option is set, then
526 fio will append to the file instead. This has identical
527 behavior to setting offset to the size of a file. This option
528 is ignored on non-regular files.
531 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
532 space left on device) as the terminating condition. Only makes
533 sense with sequential write. For a read workload, the mount
534 point will be filled first then IO started on the result. This
535 option doesn't make sense if operating on a raw device node,
536 since the size of that is already known by the file system.
537 Additionally, writing beyond end-of-device will not return
541 bs=int The block size used for the io units. Defaults to 4k. Values
542 can be given for both read and writes. If a single int is
543 given, it will apply to both. If a second int is specified
544 after a comma, it will apply to writes only. In other words,
545 the format is either bs=read_and_write or bs=read,write,trim.
546 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
547 writes, and 8k for trims. You can terminate the list with
548 a trailing comma. bs=4k,8k, would use the default value for
549 trims.. If you only wish to set the write size, you
550 can do so by passing an empty read size - bs=,8k will set
551 8k for writes and leave the read default value.
554 ba=int At what boundary to align random IO offsets. Defaults to
555 the same as 'blocksize' the minimum blocksize given.
556 Minimum alignment is typically 512b for using direct IO,
557 though it usually depends on the hardware block size. This
558 option is mutually exclusive with using a random map for
559 files, so it will turn off that option.
561 blocksize_range=irange
562 bsrange=irange Instead of giving a single block size, specify a range
563 and fio will mix the issued io block sizes. The issued
564 io unit will always be a multiple of the minimum value
565 given (also see bs_unaligned). Applies to both reads and
566 writes, however a second range can be given after a comma.
569 bssplit=str Sometimes you want even finer grained control of the
570 block sizes issued, not just an even split between them.
571 This option allows you to weight various block sizes,
572 so that you are able to define a specific amount of
573 block sizes issued. The format for this option is:
575 bssplit=blocksize/percentage:blocksize/percentage
577 for as many block sizes as needed. So if you want to define
578 a workload that has 50% 64k blocks, 10% 4k blocks, and
579 40% 32k blocks, you would write:
581 bssplit=4k/10:64k/50:32k/40
583 Ordering does not matter. If the percentage is left blank,
584 fio will fill in the remaining values evenly. So a bssplit
585 option like this one:
587 bssplit=4k/50:1k/:32k/
589 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
590 always add up to 100, if bssplit is given a range that adds
591 up to more, it will error out.
593 bssplit also supports giving separate splits to reads and
594 writes. The format is identical to what bs= accepts. You
595 have to separate the read and write parts with a comma. So
596 if you want a workload that has 50% 2k reads and 50% 4k reads,
597 while having 90% 4k writes and 10% 8k writes, you would
600 bssplit=2k/50:4k/50,4k/90:8k/10
603 bs_unaligned If this option is given, any byte size value within bsrange
604 may be used as a block range. This typically wont work with
605 direct IO, as that normally requires sector alignment.
607 bs_is_seq_rand If this option is set, fio will use the normal read,write
608 blocksize settings as sequential,random instead. Any random
609 read or write will use the WRITE blocksize settings, and any
610 sequential read or write will use the READ blocksize setting.
612 zero_buffers If this option is given, fio will init the IO buffers to
613 all zeroes. The default is to fill them with random data.
615 refill_buffers If this option is given, fio will refill the IO buffers
616 on every submit. The default is to only fill it at init
617 time and reuse that data. Only makes sense if zero_buffers
618 isn't specified, naturally. If data verification is enabled,
619 refill_buffers is also automatically enabled.
621 scramble_buffers=bool If refill_buffers is too costly and the target is
622 using data deduplication, then setting this option will
623 slightly modify the IO buffer contents to defeat normal
624 de-dupe attempts. This is not enough to defeat more clever
625 block compression attempts, but it will stop naive dedupe of
626 blocks. Default: true.
628 buffer_compress_percentage=int If this is set, then fio will attempt to
629 provide IO buffer content (on WRITEs) that compress to
630 the specified level. Fio does this by providing a mix of
631 random data and a fixed pattern. The fixed pattern is either
632 zeroes, or the pattern specified by buffer_pattern. If the
633 pattern option is used, it might skew the compression ratio
634 slightly. Note that this is per block size unit, for file/disk
635 wide compression level that matches this setting, you'll also
636 want to set refill_buffers.
638 buffer_compress_chunk=int See buffer_compress_percentage. This
639 setting allows fio to manage how big the ranges of random
640 data and zeroed data is. Without this set, fio will
641 provide buffer_compress_percentage of blocksize random
642 data, followed by the remaining zeroed. With this set
643 to some chunk size smaller than the block size, fio can
644 alternate random and zeroed data throughout the IO
647 buffer_pattern=str If set, fio will fill the io buffers with this
648 pattern. If not set, the contents of io buffers is defined by
649 the other options related to buffer contents. The setting can
650 be any pattern of bytes, and can be prefixed with 0x for hex
651 values. It may also be a string, where the string must then
652 be wrapped with "", e.g.:
654 buffer_pattern="abcd"
658 buffer_pattern=0xdeadface
660 Also you can combine everything together in any order:
661 buffer_pattern=0xdeadface"abcd"-12
663 dedupe_percentage=int If set, fio will generate this percentage of
664 identical buffers when writing. These buffers will be
665 naturally dedupable. The contents of the buffers depend on
666 what other buffer compression settings have been set. It's
667 possible to have the individual buffers either fully
668 compressible, or not at all. This option only controls the
669 distribution of unique buffers.
671 nrfiles=int Number of files to use for this job. Defaults to 1.
673 openfiles=int Number of files to keep open at the same time. Defaults to
674 the same as nrfiles, can be set smaller to limit the number
677 file_service_type=str Defines how fio decides which file from a job to
678 service next. The following types are defined:
680 random Just choose a file at random.
682 roundrobin Round robin over open files. This
685 sequential Finish one file before moving on to
686 the next. Multiple files can still be
687 open depending on 'openfiles'.
689 zipf Use a zipfian distribution to decide what file
692 pareto Use a pareto distribution to decide what file
695 gauss Use a gaussian (normal) distribution to decide
698 For random, roundrobin, and sequential, a postfix can be
699 appended to tell fio how many I/Os to issue before switching
700 to a new file. For example, specifying
701 'file_service_type=random:8' would cause fio to issue 8 I/Os
702 before selecting a new file at random. For the non-uniform
703 distributions, a floating point postfix can be given to
704 influence how the distribution is skewed. See
705 'random_distribution' for a description of how that would work.
707 ioengine=str Defines how the job issues io to the file. The following
710 sync Basic read(2) or write(2) io. lseek(2) is
711 used to position the io location.
713 psync Basic pread(2) or pwrite(2) io. Default on all
714 supported operating systems except for Windows.
716 vsync Basic readv(2) or writev(2) IO.
718 pvsync Basic preadv(2) or pwritev(2) IO.
720 pvsync2 Basic preadv2(2) or pwritev2(2) IO.
722 libaio Linux native asynchronous io. Note that Linux
723 may only support queued behaviour with
724 non-buffered IO (set direct=1 or buffered=0).
725 This engine defines engine specific options.
727 posixaio glibc posix asynchronous io.
729 solarisaio Solaris native asynchronous io.
731 windowsaio Windows native asynchronous io.
734 mmap File is memory mapped and data copied
735 to/from using memcpy(3).
737 splice splice(2) is used to transfer the data and
738 vmsplice(2) to transfer data from user
741 sg SCSI generic sg v3 io. May either be
742 synchronous using the SG_IO ioctl, or if
743 the target is an sg character device
744 we use read(2) and write(2) for asynchronous
747 null Doesn't transfer any data, just pretends
748 to. This is mainly used to exercise fio
749 itself and for debugging/testing purposes.
751 net Transfer over the network to given host:port.
752 Depending on the protocol used, the hostname,
753 port, listen and filename options are used to
754 specify what sort of connection to make, while
755 the protocol option determines which protocol
757 This engine defines engine specific options.
759 netsplice Like net, but uses splice/vmsplice to
760 map data and send/receive.
761 This engine defines engine specific options.
763 cpuio Doesn't transfer any data, but burns CPU
764 cycles according to the cpuload= and
765 cpuchunks= options. Setting cpuload=85
766 will cause that job to do nothing but burn
767 85% of the CPU. In case of SMP machines,
768 use numjobs=<no_of_cpu> to get desired CPU
769 usage, as the cpuload only loads a single
770 CPU at the desired rate. A job never finishes
771 unless there is at least one non-cpuio job.
773 guasi The GUASI IO engine is the Generic Userspace
774 Asyncronous Syscall Interface approach
777 http://www.xmailserver.org/guasi-lib.html
779 for more info on GUASI.
781 rdma The RDMA I/O engine supports both RDMA
782 memory semantics (RDMA_WRITE/RDMA_READ) and
783 channel semantics (Send/Recv) for the
784 InfiniBand, RoCE and iWARP protocols.
786 falloc IO engine that does regular fallocate to
787 simulate data transfer as fio ioengine.
788 DDIR_READ does fallocate(,mode = keep_size,)
789 DDIR_WRITE does fallocate(,mode = 0)
790 DDIR_TRIM does fallocate(,mode = punch_hole)
792 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
793 ioctls to simulate defragment activity in
794 request to DDIR_WRITE event
796 rbd IO engine supporting direct access to Ceph
797 Rados Block Devices (RBD) via librbd without
798 the need to use the kernel rbd driver. This
799 ioengine defines engine specific options.
801 gfapi Using Glusterfs libgfapi sync interface to
802 direct access to Glusterfs volumes without
805 gfapi_async Using Glusterfs libgfapi async interface
806 to direct access to Glusterfs volumes without
807 having to go through FUSE. This ioengine
808 defines engine specific options.
810 libhdfs Read and write through Hadoop (HDFS).
811 This engine interprets offsets a little
812 differently. In HDFS, files once created
813 cannot be modified. So random writes are not
814 possible. To imitate this, libhdfs engine
815 creates bunch of small files, and engine will
816 pick a file out of those files based on the
817 offset generated by fio backend. Each jobs uses
818 it's own connection to HDFS.
820 mtd Read, write and erase an MTD character device
821 (e.g., /dev/mtd0). Discards are treated as
822 erases. Depending on the underlying device
823 type, the I/O may have to go in a certain
824 pattern, e.g., on NAND, writing sequentially
825 to erase blocks and discarding before
826 overwriting. The writetrim mode works well
829 pmemblk Read and write through the NVML libpmemblk
832 dev-dax Read and write through a DAX device exposed
833 from persistent memory.
835 external Prefix to specify loading an external
836 IO engine object file. Append the engine
837 filename, eg ioengine=external:/tmp/foo.o
838 to load ioengine foo.o in /tmp.
840 iodepth=int This defines how many io units to keep in flight against
841 the file. The default is 1 for each file defined in this
842 job, can be overridden with a larger value for higher
843 concurrency. Note that increasing iodepth beyond 1 will not
844 affect synchronous ioengines (except for small degress when
845 verify_async is in use). Even async engines may impose OS
846 restrictions causing the desired depth not to be achieved.
847 This may happen on Linux when using libaio and not setting
848 direct=1, since buffered IO is not async on that OS. Keep an
849 eye on the IO depth distribution in the fio output to verify
850 that the achieved depth is as expected. Default: 1.
852 iodepth_batch_submit=int
853 iodepth_batch=int This defines how many pieces of IO to submit at once.
854 It defaults to 1 which means that we submit each IO
855 as soon as it is available, but can be raised to submit
856 bigger batches of IO at the time. If it is set to 0 the iodepth
859 iodepth_batch_complete_min=int
860 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
861 at once. It defaults to 1 which means that we'll ask
862 for a minimum of 1 IO in the retrieval process from
863 the kernel. The IO retrieval will go on until we
864 hit the limit set by iodepth_low. If this variable is
865 set to 0, then fio will always check for completed
866 events before queuing more IO. This helps reduce
867 IO latency, at the cost of more retrieval system calls.
869 iodepth_batch_complete_max=int This defines maximum pieces of IO to
870 retrieve at once. This variable should be used along with
871 iodepth_batch_complete_min=int variable, specifying the range
872 of min and max amount of IO which should be retrieved. By default
873 it is equal to iodepth_batch_complete_min value.
877 iodepth_batch_complete_min=1
878 iodepth_batch_complete_max=<iodepth>
880 which means that we will retrieve at least 1 IO and up to the
881 whole submitted queue depth. If none of IO has been completed
886 iodepth_batch_complete_min=0
887 iodepth_batch_complete_max=<iodepth>
889 which means that we can retrieve up to the whole submitted
890 queue depth, but if none of IO has been completed yet, we will
891 NOT wait and immediately exit the system call. In this example
892 we simply do polling.
894 iodepth_low=int The low water mark indicating when to start filling
895 the queue again. Defaults to the same as iodepth, meaning
896 that fio will attempt to keep the queue full at all times.
897 If iodepth is set to eg 16 and iodepth_low is set to 4, then
898 after fio has filled the queue of 16 requests, it will let
899 the depth drain down to 4 before starting to fill it again.
901 io_submit_mode=str This option controls how fio submits the IO to
902 the IO engine. The default is 'inline', which means that the
903 fio job threads submit and reap IO directly. If set to
904 'offload', the job threads will offload IO submission to a
905 dedicated pool of IO threads. This requires some coordination
906 and thus has a bit of extra overhead, especially for lower
907 queue depth IO where it can increase latencies. The benefit
908 is that fio can manage submission rates independently of
909 the device completion rates. This avoids skewed latency
910 reporting if IO gets back up on the device side (the
911 coordinated omission problem).
913 direct=bool If value is true, use non-buffered io. This is usually
914 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
915 On Windows the synchronous ioengines don't support direct io.
917 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
918 writes are guaranteed to be stable once acknowledged by
919 the operating system. Only Linux supports O_ATOMIC right
922 buffered=bool If value is true, use buffered io. This is the opposite
923 of the 'direct' option. Defaults to true.
925 offset=int Start io at the given offset in the file. The data before
926 the given offset will not be touched. This effectively
927 caps the file size at real_size - offset.
929 offset_increment=int If this is provided, then the real offset becomes
930 offset + offset_increment * thread_number, where the thread
931 number is a counter that starts at 0 and is incremented for
932 each sub-job (i.e. when numjobs option is specified). This
933 option is useful if there are several jobs which are intended
934 to operate on a file in parallel disjoint segments, with
935 even spacing between the starting points.
937 number_ios=int Fio will normally perform IOs until it has exhausted the size
938 of the region set by size=, or if it exhaust the allocated
939 time (or hits an error condition). With this setting, the
940 range/size can be set independently of the number of IOs to
941 perform. When fio reaches this number, it will exit normally
942 and report status. Note that this does not extend the amount
943 of IO that will be done, it will only stop fio if this
944 condition is met before other end-of-job criteria.
946 fsync=int If writing to a file, issue a sync of the dirty data
947 for every number of blocks given. For example, if you give
948 32 as a parameter, fio will sync the file for every 32
949 writes issued. If fio is using non-buffered io, we may
950 not sync the file. The exception is the sg io engine, which
951 synchronizes the disk cache anyway.
953 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
955 In FreeBSD and Windows there is no fdatasync(), this falls back
958 sync_file_range=str:val Use sync_file_range() for every 'val' number of
959 write operations. Fio will track range of writes that
960 have happened since the last sync_file_range() call. 'str'
961 can currently be one or more of:
963 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
964 write SYNC_FILE_RANGE_WRITE
965 wait_after SYNC_FILE_RANGE_WAIT_AFTER
967 So if you do sync_file_range=wait_before,write:8, fio would
968 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
969 every 8 writes. Also see the sync_file_range(2) man page.
970 This option is Linux specific.
972 overwrite=bool If true, writes to a file will always overwrite existing
973 data. If the file doesn't already exist, it will be
974 created before the write phase begins. If the file exists
975 and is large enough for the specified write phase, nothing
978 end_fsync=bool If true, fsync file contents when a write stage has completed.
980 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
981 This differs from end_fsync in that it will happen on every
982 file close, not just at the end of the job.
984 rwmixread=int How large a percentage of the mix should be reads.
986 rwmixwrite=int How large a percentage of the mix should be writes. If both
987 rwmixread and rwmixwrite is given and the values do not add
988 up to 100%, the latter of the two will be used to override
989 the first. This may interfere with a given rate setting,
990 if fio is asked to limit reads or writes to a certain rate.
991 If that is the case, then the distribution may be skewed.
993 random_distribution=str:float By default, fio will use a completely uniform
994 random distribution when asked to perform random IO. Sometimes
995 it is useful to skew the distribution in specific ways,
996 ensuring that some parts of the data is more hot than others.
997 fio includes the following distribution models:
999 random Uniform random distribution
1000 zipf Zipf distribution
1001 pareto Pareto distribution
1002 gauss Normal (gaussian) distribution
1003 zoned Zoned random distribution
1005 When using a zipf or pareto distribution, an input value
1006 is also needed to define the access pattern. For zipf, this
1007 is the zipf theta. For pareto, it's the pareto power. Fio
1008 includes a test program, genzipf, that can be used visualize
1009 what the given input values will yield in terms of hit rates.
1010 If you wanted to use zipf with a theta of 1.2, you would use
1011 random_distribution=zipf:1.2 as the option. If a non-uniform
1012 model is used, fio will disable use of the random map. For
1013 the gauss distribution, a normal deviation is supplied as
1014 a value between 0 and 100.
1016 For a zoned distribution, fio supports specifying percentages
1017 of IO access that should fall within what range of the file or
1018 device. For example, given a criteria of:
1020 60% of accesses should be to the first 10%
1021 30% of accesses should be to the next 20%
1022 8% of accesses should be to to the next 30%
1023 2% of accesses should be to the next 40%
1025 we can define that through zoning of the random accesses. For
1026 the above example, the user would do:
1028 random_distribution=zoned:60/10:30/20:8/30:2/40
1030 similarly to how bssplit works for setting ranges and
1031 percentages of block sizes. Like bssplit, it's possible to
1032 specify separate zones for reads, writes, and trims. If just
1033 one set is given, it'll apply to all of them.
1035 percentage_random=int For a random workload, set how big a percentage should
1036 be random. This defaults to 100%, in which case the workload
1037 is fully random. It can be set from anywhere from 0 to 100.
1038 Setting it to 0 would make the workload fully sequential. Any
1039 setting in between will result in a random mix of sequential
1040 and random IO, at the given percentages. It is possible to
1041 set different values for reads, writes, and trim. To do so,
1042 simply use a comma separated list. See blocksize.
1044 norandommap Normally fio will cover every block of the file when doing
1045 random IO. If this option is given, fio will just get a
1046 new random offset without looking at past io history. This
1047 means that some blocks may not be read or written, and that
1048 some blocks may be read/written more than once. If this option
1049 is used with verify= and multiple blocksizes (via bsrange=),
1050 only intact blocks are verified, i.e., partially-overwritten
1053 softrandommap=bool See norandommap. If fio runs with the random block map
1054 enabled and it fails to allocate the map, if this option is
1055 set it will continue without a random block map. As coverage
1056 will not be as complete as with random maps, this option is
1057 disabled by default.
1059 random_generator=str Fio supports the following engines for generating
1060 IO offsets for random IO:
1062 tausworthe Strong 2^88 cycle random number generator
1063 lfsr Linear feedback shift register generator
1064 tausworthe64 Strong 64-bit 2^258 cycle random number
1067 Tausworthe is a strong random number generator, but it
1068 requires tracking on the side if we want to ensure that
1069 blocks are only read or written once. LFSR guarantees
1070 that we never generate the same offset twice, and it's
1071 also less computationally expensive. It's not a true
1072 random generator, however, though for IO purposes it's
1073 typically good enough. LFSR only works with single
1074 block sizes, not with workloads that use multiple block
1075 sizes. If used with such a workload, fio may read or write
1076 some blocks multiple times. The default value is tausworthe,
1077 unless the required space exceeds 2^32 blocks. If it does,
1078 then tausworthe64 is selected automatically.
1080 nice=int Run the job with the given nice value. See man nice(2).
1082 On Windows, values less than -15 set the process class to "High";
1083 -1 through -15 set "Above Normal"; 1 through 15 "Below Normal";
1084 and above 15 "Idle" priority class.
1086 prio=int Set the io priority value of this job. Linux limits us to
1087 a positive value between 0 and 7, with 0 being the highest.
1088 See man ionice(1). Refer to an appropriate manpage for
1089 other operating systems since meaning of priority may differ.
1091 prioclass=int Set the io priority class. See man ionice(1).
1093 thinktime=int Stall the job x microseconds after an io has completed before
1094 issuing the next. May be used to simulate processing being
1095 done by an application. See thinktime_blocks and
1099 Only valid if thinktime is set - pretend to spend CPU time
1100 doing something with the data received, before falling back
1101 to sleeping for the rest of the period specified by
1104 thinktime_blocks=int
1105 Only valid if thinktime is set - control how many blocks
1106 to issue, before waiting 'thinktime' usecs. If not set,
1107 defaults to 1 which will make fio wait 'thinktime' usecs
1108 after every block. This effectively makes any queue depth
1109 setting redundant, since no more than 1 IO will be queued
1110 before we have to complete it and do our thinktime. In
1111 other words, this setting effectively caps the queue depth
1112 if the latter is larger.
1114 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
1115 the normal suffix rules apply. You can use rate=500k to limit
1116 reads and writes to 500k each, or you can specify read and
1117 writes separately. Using rate=1m,500k would limit reads to
1118 1MB/sec and writes to 500KB/sec. Capping only reads or
1119 writes can be done with rate=,500k or rate=500k,. The former
1120 will only limit writes (to 500KB/sec), the latter will only
1123 rate_min=int Tell fio to do whatever it can to maintain at least this
1124 bandwidth. Failing to meet this requirement, will cause
1125 the job to exit. The same format as rate is used for
1126 read vs write separation.
1128 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
1129 as rate, just specified independently of bandwidth. If the
1130 job is given a block size range instead of a fixed value,
1131 the smallest block size is used as the metric. The same format
1132 as rate is used for read vs write separation.
1134 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
1135 the job to exit. The same format as rate is used for read vs
1138 rate_process=str This option controls how fio manages rated IO
1139 submissions. The default is 'linear', which submits IO in a
1140 linear fashion with fixed delays between IOs that gets
1141 adjusted based on IO completion rates. If this is set to
1142 'poisson', fio will submit IO based on a more real world
1143 random request flow, known as the Poisson process
1144 (https://en.wikipedia.org/wiki/Poisson_process). The lambda
1145 will be 10^6 / IOPS for the given workload.
1147 latency_target=int If set, fio will attempt to find the max performance
1148 point that the given workload will run at while maintaining a
1149 latency below this target. The values is given in microseconds.
1150 See latency_window and latency_percentile
1152 latency_window=int Used with latency_target to specify the sample window
1153 that the job is run at varying queue depths to test the
1154 performance. The value is given in microseconds.
1156 latency_percentile=float The percentage of IOs that must fall within the
1157 criteria specified by latency_target and latency_window. If not
1158 set, this defaults to 100.0, meaning that all IOs must be equal
1159 or below to the value set by latency_target.
1161 max_latency=int If set, fio will exit the job if it exceeds this maximum
1162 latency. It will exit with an ETIME error.
1164 rate_cycle=int Average bandwidth for 'rate' and 'rate_min' over this number
1167 cpumask=int Set the CPU affinity of this job. The parameter given is a
1168 bitmask of allowed CPU's the job may run on. So if you want
1169 the allowed CPUs to be 1 and 5, you would pass the decimal
1170 value of (1 << 1 | 1 << 5), or 34. See man
1171 sched_setaffinity(2). This may not work on all supported
1172 operating systems or kernel versions. This option doesn't
1173 work well for a higher CPU count than what you can store in
1174 an integer mask, so it can only control cpus 1-32. For
1175 boxes with larger CPU counts, use cpus_allowed.
1177 cpus_allowed=str Controls the same options as cpumask, but it allows a text
1178 setting of the permitted CPUs instead. So to use CPUs 1 and
1179 5, you would specify cpus_allowed=1,5. This options also
1180 allows a range of CPUs. Say you wanted a binding to CPUs
1181 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
1183 cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
1184 specified by cpus_allowed or cpumask. Two policies are
1187 shared All jobs will share the CPU set specified.
1188 split Each job will get a unique CPU from the CPU set.
1190 'shared' is the default behaviour, if the option isn't
1191 specified. If split is specified, then fio will will assign
1192 one cpu per job. If not enough CPUs are given for the jobs
1193 listed, then fio will roundrobin the CPUs in the set.
1195 numa_cpu_nodes=str Set this job running on specified NUMA nodes' CPUs. The
1196 arguments allow comma delimited list of cpu numbers,
1197 A-B ranges, or 'all'. Note, to enable numa options support,
1198 fio must be built on a system with libnuma-dev(el) installed.
1200 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
1201 nodes. Format of the arguments:
1203 `mode' is one of the following memory policy:
1204 default, prefer, bind, interleave, local
1205 For `default' and `local' memory policy, no node is
1206 needed to be specified.
1207 For `prefer', only one node is allowed.
1208 For `bind' and `interleave', it allow comma delimited
1209 list of numbers, A-B ranges, or 'all'.
1211 startdelay=time Start this job the specified number of seconds after fio
1212 has started. Only useful if the job file contains several
1213 jobs, and you want to delay starting some jobs to a certain
1216 runtime=time Tell fio to terminate processing after the specified number
1217 of seconds. It can be quite hard to determine for how long
1218 a specified job will run, so this parameter is handy to
1219 cap the total runtime to a given time.
1221 time_based If set, fio will run for the duration of the runtime
1222 specified even if the file(s) are completely read or
1223 written. It will simply loop over the same workload
1224 as many times as the runtime allows.
1226 ramp_time=time If set, fio will run the specified workload for this amount
1227 of time before logging any performance numbers. Useful for
1228 letting performance settle before logging results, thus
1229 minimizing the runtime required for stable results. Note
1230 that the ramp_time is considered lead in time for a job,
1231 thus it will increase the total runtime if a special timeout
1232 or runtime is specified.
1234 steadystate=str:float
1235 ss=str:float Define the criterion and limit for assessing steady state
1236 performance. The first parameter designates the criterion
1237 whereas the second parameter sets the threshold. When the
1238 criterion falls below the threshold for the specified duration,
1239 the job will stop. For example, iops_slope:0.1% will direct fio
1240 to terminate the job when the least squares regression slope
1241 falls below 0.1% of the mean IOPS. If group_reporting is
1242 enabled this will apply to all jobs in the group. Below is the
1243 list of available steady state assessment criteria. All
1244 assessments are carried out using only data from the rolling
1245 collection window. Threshold limits can be expressed as a fixed
1246 value or as a percentage of the mean in the collection window.
1247 iops Collect IOPS data. Stop the job if all
1248 individual IOPS measurements are within the
1249 specified limit of the mean IOPS (e.g., iops:2
1250 means that all individual IOPS values must be
1251 within 2 of the mean, whereas iops:0.2% means
1252 that all individual IOPS values must be within
1253 0.2% of the mean IOPS to terminate the job).
1255 Collect IOPS data and calculate the least
1256 squares regression slope. Stop the job if the
1257 slope falls below the specified limit.
1258 bw Collect bandwidth data. Stop the job if all
1259 individual bandwidth measurements are within
1260 the specified limit of the mean bandwidth.
1262 Collect bandwidth data and calculate the least
1263 squares regression slope. Stop the job if the
1264 slope falls below the specified limit.
1266 steadystate_duration=time
1267 ss_dur=time A rolling window of this duration will be used to judge whether
1268 steady state has been reached. Data will be collected once per
1269 second. The default is 0 which disables steady state detection.
1271 steadystate_ramp_time=time
1272 ss_ramp=time Allow the job to run for the specified duration before
1273 beginning data collection for checking the steady state job
1274 termination criterion. The default is 0.
1276 invalidate=bool Invalidate the buffer/page cache parts for this file prior
1277 to starting io. Defaults to true.
1279 sync=bool Use sync io for buffered writes. For the majority of the
1280 io engines, this means using O_SYNC.
1283 mem=str Fio can use various types of memory as the io unit buffer.
1284 The allowed values are:
1286 malloc Use memory from malloc(3) as the buffers.
1287 Default memory type.
1289 shm Use shared memory as the buffers. Allocated
1292 shmhuge Same as shm, but use huge pages as backing.
1294 mmap Use mmap to allocate buffers. May either be
1295 anonymous memory, or can be file backed if
1296 a filename is given after the option. The
1297 format is mem=mmap:/path/to/file.
1299 mmaphuge Use a memory mapped huge file as the buffer
1300 backing. Append filename after mmaphuge, ala
1301 mem=mmaphuge:/hugetlbfs/file
1303 mmapshared Same as mmap, but use a MMAP_SHARED
1306 The area allocated is a function of the maximum allowed
1307 bs size for the job, multiplied by the io depth given. Note
1308 that for shmhuge and mmaphuge to work, the system must have
1309 free huge pages allocated. This can normally be checked
1310 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1311 Linux system. Fio assumes a huge page is 4MB in size. So
1312 to calculate the number of huge pages you need for a given
1313 job file, add up the io depth of all jobs (normally one unless
1314 iodepth= is used) and multiply by the maximum bs set. Then
1315 divide that number by the huge page size. You can see the
1316 size of the huge pages in /proc/meminfo. If no huge pages
1317 are allocated by having a non-zero number in nr_hugepages,
1318 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1320 mmaphuge also needs to have hugetlbfs mounted and the file
1321 location should point there. So if it's mounted in /huge,
1322 you would use mem=mmaphuge:/huge/somefile.
1324 iomem_align=int This indicates the memory alignment of the IO memory buffers.
1325 Note that the given alignment is applied to the first IO unit
1326 buffer, if using iodepth the alignment of the following buffers
1327 are given by the bs used. In other words, if using a bs that is
1328 a multiple of the page sized in the system, all buffers will
1329 be aligned to this value. If using a bs that is not page
1330 aligned, the alignment of subsequent IO memory buffers is the
1331 sum of the iomem_align and bs used.
1334 Defines the size of a huge page. Must at least be equal
1335 to the system setting, see /proc/meminfo. Defaults to 4MB.
1336 Should probably always be a multiple of megabytes, so using
1337 hugepage-size=Xm is the preferred way to set this to avoid
1338 setting a non-pow-2 bad value.
1340 exitall When one job finishes, terminate the rest. The default is
1341 to wait for each job to finish, sometimes that is not the
1344 exitall_on_error When one job finishes in error, terminate the rest. The
1345 default is to wait for each job to finish.
1347 bwavgtime=int Average the calculated bandwidth over the given time. Value
1348 is specified in milliseconds. If the job also does bandwidth
1349 logging through 'write_bw_log', then the minimum of this option
1350 and 'log_avg_msec' will be used. Default: 500ms.
1352 iopsavgtime=int Average the calculated IOPS over the given time. Value
1353 is specified in milliseconds. If the job also does IOPS logging
1354 through 'write_iops_log', then the minimum of this option and
1355 'log_avg_msec' will be used. Default: 500ms.
1357 create_serialize=bool If true, serialize the file creation for the jobs.
1358 This may be handy to avoid interleaving of data
1359 files, which may greatly depend on the filesystem
1360 used and even the number of processors in the system.
1362 create_fsync=bool fsync the data file after creation. This is the
1365 create_on_open=bool Don't pre-setup the files for IO, just create open()
1366 when it's time to do IO to that file.
1368 create_only=bool If true, fio will only run the setup phase of the job.
1369 If files need to be laid out or updated on disk, only
1370 that will be done. The actual job contents are not
1373 allow_file_create=bool If true, fio is permitted to create files as part
1374 of its workload. This is the default behavior. If this
1375 option is false, then fio will error out if the files it
1376 needs to use don't already exist. Default: true.
1378 allow_mounted_write=bool If this isn't set, fio will abort jobs that
1379 are destructive (eg that write) to what appears to be a
1380 mounted device or partition. This should help catch creating
1381 inadvertently destructive tests, not realizing that the test
1382 will destroy data on the mounted file system. Default: false.
1384 pre_read=bool If this is given, files will be pre-read into memory before
1385 starting the given IO operation. This will also clear
1386 the 'invalidate' flag, since it is pointless to pre-read
1387 and then drop the cache. This will only work for IO engines
1388 that are seek-able, since they allow you to read the same data
1389 multiple times. Thus it will not work on eg network or splice
1392 unlink=bool Unlink the job files when done. Not the default, as repeated
1393 runs of that job would then waste time recreating the file
1394 set again and again.
1396 unlink_each_loop=bool Unlink job files after each iteration or loop.
1398 loops=int Run the specified number of iterations of this job. Used
1399 to repeat the same workload a given number of times. Defaults
1402 verify_only Do not perform specified workload---only verify data still
1403 matches previous invocation of this workload. This option
1404 allows one to check data multiple times at a later date
1405 without overwriting it. This option makes sense only for
1406 workloads that write data, and does not support workloads
1407 with the time_based option set.
1409 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1410 verify is set. Defaults to 1.
1412 verify=str If writing to a file, fio can verify the file contents
1413 after each iteration of the job. Each verification method also implies
1414 verification of special header, which is written to the beginning of
1415 each block. This header also includes meta information, like offset
1416 of the block, block number, timestamp when block was written, etc.
1417 verify=str can be combined with verify_pattern=str option.
1418 The allowed values are:
1420 md5 Use an md5 sum of the data area and store
1421 it in the header of each block.
1423 crc64 Use an experimental crc64 sum of the data
1424 area and store it in the header of each
1427 crc32c Use a crc32c sum of the data area and store
1428 it in the header of each block.
1430 crc32c-intel Use hardware assisted crc32c calculation
1431 provided on SSE4.2 enabled processors. Falls
1432 back to regular software crc32c, if not
1433 supported by the system.
1435 crc32 Use a crc32 sum of the data area and store
1436 it in the header of each block.
1438 crc16 Use a crc16 sum of the data area and store
1439 it in the header of each block.
1441 crc7 Use a crc7 sum of the data area and store
1442 it in the header of each block.
1444 xxhash Use xxhash as the checksum function. Generally
1445 the fastest software checksum that fio
1448 sha512 Use sha512 as the checksum function.
1450 sha256 Use sha256 as the checksum function.
1452 sha1 Use optimized sha1 as the checksum function.
1454 meta This option is deprecated, since now meta information is
1455 included in generic verification header and meta verification
1456 happens by default. For detailed information see the description
1457 of the verify=str setting. This option is kept because of
1458 compatibility's sake with old configurations. Do not use it.
1460 pattern Verify a strict pattern. Normally fio includes
1461 a header with some basic information and
1462 checksumming, but if this option is set, only
1463 the specific pattern set with 'verify_pattern'
1466 null Only pretend to verify. Useful for testing
1467 internals with ioengine=null, not for much
1470 This option can be used for repeated burn-in tests of a
1471 system to make sure that the written data is also
1472 correctly read back. If the data direction given is
1473 a read or random read, fio will assume that it should
1474 verify a previously written file. If the data direction
1475 includes any form of write, the verify will be of the
1478 verifysort=bool If set, fio will sort written verify blocks when it deems
1479 it faster to read them back in a sorted manner. This is
1480 often the case when overwriting an existing file, since
1481 the blocks are already laid out in the file system. You
1482 can ignore this option unless doing huge amounts of really
1483 fast IO where the red-black tree sorting CPU time becomes
1486 verify_offset=int Swap the verification header with data somewhere else
1487 in the block before writing. Its swapped back before
1490 verify_interval=int Write the verification header at a finer granularity
1491 than the blocksize. It will be written for chunks the
1492 size of header_interval. blocksize should divide this
1495 verify_pattern=str If set, fio will fill the io buffers with this
1496 pattern. Fio defaults to filling with totally random
1497 bytes, but sometimes it's interesting to fill with a known
1498 pattern for io verification purposes. Depending on the
1499 width of the pattern, fio will fill 1/2/3/4 bytes of the
1500 buffer at the time(it can be either a decimal or a hex number).
1501 The verify_pattern if larger than a 32-bit quantity has to
1502 be a hex number that starts with either "0x" or "0X". Use
1503 with verify=str. Also, verify_pattern supports %o format,
1504 which means that for each block offset will be written and
1505 then verified back, e.g.:
1509 Or use combination of everything:
1510 verify_pattern=0xff%o"abcd"-12
1512 verify_fatal=bool Normally fio will keep checking the entire contents
1513 before quitting on a block verification failure. If this
1514 option is set, fio will exit the job on the first observed
1517 verify_dump=bool If set, dump the contents of both the original data
1518 block and the data block we read off disk to files. This
1519 allows later analysis to inspect just what kind of data
1520 corruption occurred. Off by default.
1522 verify_async=int Fio will normally verify IO inline from the submitting
1523 thread. This option takes an integer describing how many
1524 async offload threads to create for IO verification instead,
1525 causing fio to offload the duty of verifying IO contents
1526 to one or more separate threads. If using this offload
1527 option, even sync IO engines can benefit from using an
1528 iodepth setting higher than 1, as it allows them to have
1529 IO in flight while verifies are running.
1531 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1532 async IO verification threads. See cpus_allowed for the
1535 verify_backlog=int Fio will normally verify the written contents of a
1536 job that utilizes verify once that job has completed. In
1537 other words, everything is written then everything is read
1538 back and verified. You may want to verify continually
1539 instead for a variety of reasons. Fio stores the meta data
1540 associated with an IO block in memory, so for large
1541 verify workloads, quite a bit of memory would be used up
1542 holding this meta data. If this option is enabled, fio
1543 will write only N blocks before verifying these blocks.
1545 verify_backlog_batch=int Control how many blocks fio will verify
1546 if verify_backlog is set. If not set, will default to
1547 the value of verify_backlog (meaning the entire queue
1548 is read back and verified). If verify_backlog_batch is
1549 less than verify_backlog then not all blocks will be verified,
1550 if verify_backlog_batch is larger than verify_backlog, some
1551 blocks will be verified more than once.
1553 verify_state_save=bool When a job exits during the write phase of a verify
1554 workload, save its current state. This allows fio to replay
1555 up until that point, if the verify state is loaded for the
1556 verify read phase. The format of the filename is, roughly,
1557 <type>-<jobname>-<jobindex>-verify.state. <type> is "local"
1558 for a local run, "sock" for a client/server socket connection,
1559 and "ip" (192.168.0.1, for instance) for a networked
1560 client/server connection.
1562 verify_state_load=bool If a verify termination trigger was used, fio stores
1563 the current write state of each thread. This can be used at
1564 verification time so that fio knows how far it should verify.
1565 Without this information, fio will run a full verification
1566 pass, according to the settings in the job file used.
1569 wait_for_previous Wait for preceding jobs in the job file to exit, before
1570 starting this one. Can be used to insert serialization
1571 points in the job file. A stone wall also implies starting
1572 a new reporting group.
1574 new_group Start a new reporting group. See: group_reporting.
1576 numjobs=int Create the specified number of clones of this job. May be
1577 used to setup a larger number of threads/processes doing
1578 the same thing. Each thread is reported separately; to see
1579 statistics for all clones as a whole, use group_reporting in
1580 conjunction with new_group.
1582 group_reporting It may sometimes be interesting to display statistics for
1583 groups of jobs as a whole instead of for each individual job.
1584 This is especially true if 'numjobs' is used; looking at
1585 individual thread/process output quickly becomes unwieldy.
1586 To see the final report per-group instead of per-job, use
1587 'group_reporting'. Jobs in a file will be part of the same
1588 reporting group, unless if separated by a stonewall, or by
1591 thread fio defaults to forking jobs, however if this option is
1592 given, fio will use pthread_create(3) to create threads
1595 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1597 zoneskip=int Skip the specified number of bytes when zonesize data has
1598 been read. The two zone options can be used to only do
1599 io on zones of a file.
1601 write_iolog=str Write the issued io patterns to the specified file. See
1602 read_iolog. Specify a separate file for each job, otherwise
1603 the iologs will be interspersed and the file may be corrupt.
1605 read_iolog=str Open an iolog with the specified file name and replay the
1606 io patterns it contains. This can be used to store a
1607 workload and replay it sometime later. The iolog given
1608 may also be a blktrace binary file, which allows fio
1609 to replay a workload captured by blktrace. See blktrace
1610 for how to capture such logging data. For blktrace replay,
1611 the file needs to be turned into a blkparse binary data
1612 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1614 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1615 is to attempt to respect the time stamps within the log and
1616 replay them with the appropriate delay between IOPS. By
1617 setting this variable fio will not respect the timestamps and
1618 attempt to replay them as fast as possible while still
1619 respecting ordering. The result is the same I/O pattern to a
1620 given device, but different timings.
1622 replay_redirect=str While replaying I/O patterns using read_iolog the
1623 default behavior is to replay the IOPS onto the major/minor
1624 device that each IOP was recorded from. This is sometimes
1625 undesirable because on a different machine those major/minor
1626 numbers can map to a different device. Changing hardware on
1627 the same system can also result in a different major/minor
1628 mapping. Replay_redirect causes all IOPS to be replayed onto
1629 the single specified device regardless of the device it was
1630 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1631 IO in the blktrace or iolog to be replayed onto /dev/sdc.
1632 This means multiple devices will be replayed onto a single
1633 device, if the trace contains multiple devices. If you want
1634 multiple devices to be replayed concurrently to multiple
1635 redirected devices you must blkparse your trace into separate
1636 traces and replay them with independent fio invocations.
1637 Unfortunately this also breaks the strict time ordering
1638 between multiple device accesses.
1640 replay_align=int Force alignment of IO offsets and lengths in a trace
1641 to this power of 2 value.
1643 replay_scale=int Scale sector offsets down by this factor when
1646 per_job_logs=bool If set, this generates bw/clat/iops log with per
1647 file private filenames. If not set, jobs with identical names
1648 will share the log filename. Default: true.
1650 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1651 file. Can be used to store data of the bandwidth of the
1652 jobs in their lifetime. The included fio_generate_plots
1653 script uses gnuplot to turn these text files into nice
1654 graphs. See write_lat_log for behaviour of given
1655 filename. For this option, the suffix is _bw.x.log, where
1656 x is the index of the job (1..N, where N is the number of
1657 jobs). If 'per_job_logs' is false, then the filename will not
1658 include the job index. See 'Log File Formats'.
1660 write_lat_log=str Same as write_bw_log, except that this option stores io
1661 submission, completion, and total latencies instead. If no
1662 filename is given with this option, the default filename of
1663 "jobname_type.log" is used. Even if the filename is given,
1664 fio will still append the type of log. So if one specifies
1668 The actual log names will be foo_slat.x.log, foo_clat.x.log,
1669 and foo_lat.x.log, where x is the index of the job (1..N,
1670 where N is the number of jobs). This helps fio_generate_plot
1671 find the logs automatically. If 'per_job_logs' is false, then
1672 the filename will not include the job index. See 'Log File
1675 write_hist_log=str Same as write_lat_log, but writes I/O completion
1676 latency histograms. If no filename is given with this option, the
1677 default filename of "jobname_clat_hist.x.log" is used, where x is
1678 the index of the job (1..N, where N is the number of jobs). Even
1679 if the filename is given, fio will still append the type of log.
1680 If per_job_logs is false, then the filename will not include the
1681 job index. See 'Log File Formats'.
1683 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1684 given with this option, the default filename of
1685 "jobname_type.x.log" is used,where x is the index of the job
1686 (1..N, where N is the number of jobs). Even if the filename
1687 is given, fio will still append the type of log. If
1688 'per_job_logs' is false, then the filename will not include
1689 the job index. See 'Log File Formats'.
1691 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1692 or bw log for every IO that completes. When writing to the
1693 disk log, that can quickly grow to a very large size. Setting
1694 this option makes fio average the each log entry over the
1695 specified period of time, reducing the resolution of the log.
1696 See log_max_value as well. Defaults to 0, logging all entries.
1698 log_hist_msec=int Same as log_avg_msec, but logs entries for completion
1699 latency histograms. Computing latency percentiles from averages of
1700 intervals using log_avg_msec is innacurate. Setting this option makes
1701 fio log histogram entries over the specified period of time, reducing
1702 log sizes for high IOPS devices while retaining percentile accuracy.
1703 See log_hist_coarseness as well. Defaults to 0, meaning histogram
1704 logging is disabled.
1706 log_hist_coarseness=int Integer ranging from 0 to 6, defining the coarseness
1707 of the resolution of the histogram logs enabled with log_hist_msec. For
1708 each increment in coarseness, fio outputs half as many bins. Defaults to
1709 0, for which histogram logs contain 1216 latency bins. See
1712 log_max_value=bool If log_avg_msec is set, fio logs the average over that
1713 window. If you instead want to log the maximum value, set this
1714 option to 1. Defaults to 0, meaning that averaged values are
1717 log_offset=int If this is set, the iolog options will include the byte
1718 offset for the IO entry as well as the other data values.
1720 log_compression=int If this is set, fio will compress the IO logs as
1721 it goes, to keep the memory footprint lower. When a log
1722 reaches the specified size, that chunk is removed and
1723 compressed in the background. Given that IO logs are
1724 fairly highly compressible, this yields a nice memory
1725 savings for longer runs. The downside is that the
1726 compression will consume some background CPU cycles, so
1727 it may impact the run. This, however, is also true if
1728 the logging ends up consuming most of the system memory.
1729 So pick your poison. The IO logs are saved normally at the
1730 end of a run, by decompressing the chunks and storing them
1731 in the specified log file. This feature depends on the
1732 availability of zlib.
1734 log_compression_cpus=str Define the set of CPUs that are allowed to
1735 handle online log compression for the IO jobs. This can
1736 provide better isolation between performance sensitive jobs,
1737 and background compression work.
1739 log_store_compressed=bool If set, fio will store the log files in a
1740 compressed format. They can be decompressed with fio, using
1741 the --inflate-log command line parameter. The files will be
1742 stored with a .fz suffix.
1744 log_unix_epoch=bool If set, fio will log Unix timestamps to the log
1745 files produced by enabling write_type_log for each log type, instead
1746 of the default zero-based timestamps.
1748 block_error_percentiles=bool If set, record errors in trim block-sized
1749 units from writes and trims and output a histogram of
1750 how many trims it took to get to errors, and what kind
1751 of error was encountered.
1753 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1754 potentially be used instead of removing memory or booting
1755 with less memory to simulate a smaller amount of memory.
1756 The amount specified is per worker.
1758 exec_prerun=str Before running this job, issue the command specified
1759 through system(3). Output is redirected in a file called
1762 exec_postrun=str After the job completes, issue the command specified
1763 though system(3). Output is redirected in a file called
1764 jobname.postrun.txt.
1766 ioscheduler=str Attempt to switch the device hosting the file to the specified
1767 io scheduler before running.
1769 disk_util=bool Generate disk utilization statistics, if the platform
1770 supports it. Defaults to on.
1772 disable_lat=bool Disable measurements of total latency numbers. Useful
1773 only for cutting back the number of calls to gettimeofday,
1774 as that does impact performance at really high IOPS rates.
1775 Note that to really get rid of a large amount of these
1776 calls, this option must be used with disable_slat and
1779 disable_clat=bool Disable measurements of completion latency numbers. See
1782 disable_slat=bool Disable measurements of submission latency numbers. See
1785 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1788 clat_percentiles=bool Enable the reporting of percentiles of
1789 completion latencies.
1791 percentile_list=float_list Overwrite the default list of percentiles
1792 for completion latencies and the block error histogram.
1793 Each number is a floating number in the range (0,100],
1794 and the maximum length of the list is 20. Use ':'
1795 to separate the numbers, and list the numbers in ascending
1796 order. For example, --percentile_list=99.5:99.9 will cause
1797 fio to report the values of completion latency below which
1798 99.5% and 99.9% of the observed latencies fell, respectively.
1800 clocksource=str Use the given clocksource as the base of timing. The
1801 supported options are:
1803 gettimeofday gettimeofday(2)
1805 clock_gettime clock_gettime(2)
1807 cpu Internal CPU clock source
1809 cpu is the preferred clocksource if it is reliable, as it
1810 is very fast (and fio is heavy on time calls). Fio will
1811 automatically use this clocksource if it's supported and
1812 considered reliable on the system it is running on, unless
1813 another clocksource is specifically set. For x86/x86-64 CPUs,
1814 this means supporting TSC Invariant.
1816 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1817 (disable_clat, disable_slat, disable_bw) plus reduce
1818 precision of the timeout somewhat to really shrink
1819 the gettimeofday() call count. With this option enabled,
1820 we only do about 0.4% of the gtod() calls we would have
1821 done if all time keeping was enabled.
1823 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1824 execution to just getting the current time. Fio (and
1825 databases, for instance) are very intensive on gettimeofday()
1826 calls. With this option, you can set one CPU aside for
1827 doing nothing but logging current time to a shared memory
1828 location. Then the other threads/processes that run IO
1829 workloads need only copy that segment, instead of entering
1830 the kernel with a gettimeofday() call. The CPU set aside
1831 for doing these time calls will be excluded from other
1832 uses. Fio will manually clear it from the CPU mask of other
1835 continue_on_error=str Normally fio will exit the job on the first observed
1836 failure. If this option is set, fio will continue the job when
1837 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1838 is exceeded or the I/O size specified is completed. If this
1839 option is used, there are two more stats that are appended,
1840 the total error count and the first error. The error field
1841 given in the stats is the first error that was hit during the
1844 The allowed values are:
1846 none Exit on any IO or verify errors.
1848 read Continue on read errors, exit on all others.
1850 write Continue on write errors, exit on all others.
1852 io Continue on any IO error, exit on all others.
1854 verify Continue on verify errors, exit on all others.
1856 all Continue on all errors.
1858 0 Backward-compatible alias for 'none'.
1860 1 Backward-compatible alias for 'all'.
1862 ignore_error=str Sometimes you want to ignore some errors during test
1863 in that case you can specify error list for each error type.
1864 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1865 errors for given error type is separated with ':'. Error
1866 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1868 ignore_error=EAGAIN,ENOSPC:122
1869 This option will ignore EAGAIN from READ, and ENOSPC and
1870 122(EDQUOT) from WRITE.
1872 error_dump=bool If set dump every error even if it is non fatal, true
1873 by default. If disabled only fatal error will be dumped
1875 cgroup=str Add job to this control group. If it doesn't exist, it will
1876 be created. The system must have a mounted cgroup blkio
1877 mount point for this to work. If your system doesn't have it
1878 mounted, you can do so with:
1880 # mount -t cgroup -o blkio none /cgroup
1882 cgroup_weight=int Set the weight of the cgroup to this value. See
1883 the documentation that comes with the kernel, allowed values
1884 are in the range of 100..1000.
1886 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1887 the job completion. To override this behavior and to leave
1888 cgroups around after the job completion, set cgroup_nodelete=1.
1889 This can be useful if one wants to inspect various cgroup
1890 files after job completion. Default: false
1892 uid=int Instead of running as the invoking user, set the user ID to
1893 this value before the thread/process does any work.
1895 gid=int Set group ID, see uid.
1897 flow_id=int The ID of the flow. If not specified, it defaults to being a
1898 global flow. See flow.
1900 flow=int Weight in token-based flow control. If this value is used, then
1901 there is a 'flow counter' which is used to regulate the
1902 proportion of activity between two or more jobs. fio attempts
1903 to keep this flow counter near zero. The 'flow' parameter
1904 stands for how much should be added or subtracted to the flow
1905 counter on each iteration of the main I/O loop. That is, if
1906 one job has flow=8 and another job has flow=-1, then there
1907 will be a roughly 1:8 ratio in how much one runs vs the other.
1909 flow_watermark=int The maximum value that the absolute value of the flow
1910 counter is allowed to reach before the job must wait for a
1911 lower value of the counter.
1913 flow_sleep=int The period of time, in microseconds, to wait after the flow
1914 watermark has been exceeded before retrying operations
1916 In addition, there are some parameters which are only valid when a specific
1917 ioengine is in use. These are used identically to normal parameters, with the
1918 caveat that when used on the command line, they must come after the ioengine
1919 that defines them is selected.
1921 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1922 the io_getevents system call to reap newly returned events.
1923 With this flag turned on, the AIO ring will be read directly
1924 from user-space to reap events. The reaping mode is only
1925 enabled when polling for a minimum of 0 events (eg when
1926 iodepth_batch_complete=0).
1928 [psyncv2] hipri Set RWF_HIPRI on IO, indicating to the kernel that
1929 it's of higher priority than normal.
1931 [cpuio] cpuload=int Attempt to use the specified percentage of CPU cycles.
1933 [cpuio] cpuchunks=int Split the load into cycles of the given time. In
1936 [cpuio] exit_on_io_done=bool Detect when IO threads are done, then exit.
1938 [netsplice] hostname=str
1939 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1940 If the job is a TCP listener or UDP reader, the hostname is not
1941 used and must be omitted unless it is a valid UDP multicast
1943 [libhdfs] namenode=str The host name or IP address of a HDFS cluster namenode to contact.
1945 [netsplice] port=int
1946 [net] port=int The TCP or UDP port to bind to or connect to. If this is used
1947 with numjobs to spawn multiple instances of the same job type, then this will
1948 be the starting port number since fio will use a range of ports.
1949 [libhdfs] port=int the listening port of the HFDS cluster namenode.
1951 [netsplice] interface=str
1952 [net] interface=str The IP address of the network interface used to send or
1953 receive UDP multicast
1956 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1959 [netsplice] nodelay=bool
1960 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1962 [netsplice] protocol=str
1963 [netsplice] proto=str
1965 [net] proto=str The network protocol to use. Accepted values are:
1967 tcp Transmission control protocol
1968 tcpv6 Transmission control protocol V6
1969 udp User datagram protocol
1970 udpv6 User datagram protocol V6
1971 unix UNIX domain socket
1973 When the protocol is TCP or UDP, the port must also be given,
1974 as well as the hostname if the job is a TCP listener or UDP
1975 reader. For unix sockets, the normal filename option should be
1976 used and the port is invalid.
1978 [net] listen For TCP network connections, tell fio to listen for incoming
1979 connections rather than initiating an outgoing connection. The
1980 hostname must be omitted if this option is used.
1982 [net] pingpong Normally a network writer will just continue writing data, and
1983 a network reader will just consume packages. If pingpong=1
1984 is set, a writer will send its normal payload to the reader,
1985 then wait for the reader to send the same payload back. This
1986 allows fio to measure network latencies. The submission
1987 and completion latencies then measure local time spent
1988 sending or receiving, and the completion latency measures
1989 how long it took for the other end to receive and send back.
1990 For UDP multicast traffic pingpong=1 should only be set for a
1991 single reader when multiple readers are listening to the same
1994 [net] window_size Set the desired socket buffer size for the connection.
1996 [net] mss Set the TCP maximum segment size (TCP_MAXSEG).
1998 [e4defrag] donorname=str
1999 File will be used as a block donor(swap extents between files)
2000 [e4defrag] inplace=int
2001 Configure donor file blocks allocation strategy
2002 0(default): Preallocate donor's file on init
2003 1 : allocate space immediately inside defragment event,
2004 and free right after event
2006 [rbd] clustername=str Specifies the name of the Ceph cluster.
2007 [rbd] rbdname=str Specifies the name of the RBD.
2008 [rbd] pool=str Specifies the name of the Ceph pool containing RBD.
2009 [rbd] clientname=str Specifies the username (without the 'client.' prefix)
2010 used to access the Ceph cluster. If the clustername is
2011 specified, the clientname shall be the full type.id
2012 string. If no type. prefix is given, fio will add
2013 'client.' by default.
2015 [mtd] skip_bad=bool Skip operations against known bad blocks.
2017 [libhdfs] hdfsdirectory libhdfs will create chunk in this HDFS directory
2018 [libhdfs] chunk_size the size of the chunk to use for each file.
2021 6.0 Interpreting the output
2022 ---------------------------
2024 fio spits out a lot of output. While running, fio will display the
2025 status of the jobs created. An example of that would be:
2027 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
2029 The characters inside the square brackets denote the current status of
2030 each thread. The possible values (in typical life cycle order) are:
2034 P Thread setup, but not started.
2036 I Thread initialized, waiting or generating necessary data.
2037 p Thread running pre-reading file(s).
2038 R Running, doing sequential reads.
2039 r Running, doing random reads.
2040 W Running, doing sequential writes.
2041 w Running, doing random writes.
2042 M Running, doing mixed sequential reads/writes.
2043 m Running, doing mixed random reads/writes.
2044 F Running, currently waiting for fsync()
2045 f Running, finishing up (writing IO logs, etc)
2046 V Running, doing verification of written data.
2047 E Thread exited, not reaped by main thread yet.
2049 X Thread reaped, exited with an error.
2050 K Thread reaped, exited due to signal.
2052 Fio will condense the thread string as not to take up more space on the
2053 command line as is needed. For instance, if you have 10 readers and 10
2054 writers running, the output would look like this:
2056 Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s]
2058 Fio will still maintain the ordering, though. So the above means that jobs
2059 1..10 are readers, and 11..20 are writers.
2061 The other values are fairly self explanatory - number of threads
2062 currently running and doing io, rate of io since last check (read speed
2063 listed first, then write speed), and the estimated completion percentage
2064 and time for the running group. It's impossible to estimate runtime of
2065 the following groups (if any). Note that the string is displayed in order,
2066 so it's possible to tell which of the jobs are currently doing what. The
2067 first character is the first job defined in the job file, and so forth.
2069 When fio is done (or interrupted by ctrl-c), it will show the data for
2070 each thread, group of threads, and disks in that order. For each data
2071 direction, the output looks like:
2073 Client1 (g=0): err= 0:
2074 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
2075 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
2076 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
2077 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
2078 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
2079 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
2080 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2081 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2082 issued r/w: total=0/32768, short=0/0
2083 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
2084 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
2086 The client number is printed, along with the group id and error of that
2087 thread. Below is the io statistics, here for writes. In the order listed,
2090 io= Number of megabytes io performed
2091 bw= Average bandwidth rate
2092 iops= Average IOs performed per second
2093 runt= The runtime of that thread
2094 slat= Submission latency (avg being the average, stdev being the
2095 standard deviation). This is the time it took to submit
2096 the io. For sync io, the slat is really the completion
2097 latency, since queue/complete is one operation there. This
2098 value can be in milliseconds or microseconds, fio will choose
2099 the most appropriate base and print that. In the example
2100 above, milliseconds is the best scale. Note: in --minimal mode
2101 latencies are always expressed in microseconds.
2102 clat= Completion latency. Same names as slat, this denotes the
2103 time from submission to completion of the io pieces. For
2104 sync io, clat will usually be equal (or very close) to 0,
2105 as the time from submit to complete is basically just
2106 CPU time (io has already been done, see slat explanation).
2107 bw= Bandwidth. Same names as the xlat stats, but also includes
2108 an approximate percentage of total aggregate bandwidth
2109 this thread received in this group. This last value is
2110 only really useful if the threads in this group are on the
2111 same disk, since they are then competing for disk access.
2112 cpu= CPU usage. User and system time, along with the number
2113 of context switches this thread went through, usage of
2114 system and user time, and finally the number of major
2115 and minor page faults. The CPU utilization numbers are
2116 averages for the jobs in that reporting group, while the
2117 context and fault counters are summed.
2118 IO depths= The distribution of io depths over the job life time. The
2119 numbers are divided into powers of 2, so for example the
2120 16= entries includes depths up to that value but higher
2121 than the previous entry. In other words, it covers the
2122 range from 16 to 31.
2123 IO submit= How many pieces of IO were submitting in a single submit
2124 call. Each entry denotes that amount and below, until
2125 the previous entry - eg, 8=100% mean that we submitted
2126 anywhere in between 5-8 ios per submit call.
2127 IO complete= Like the above submit number, but for completions instead.
2128 IO issued= The number of read/write requests issued, and how many
2130 IO latencies= The distribution of IO completion latencies. This is the
2131 time from when IO leaves fio and when it gets completed.
2132 The numbers follow the same pattern as the IO depths,
2133 meaning that 2=1.6% means that 1.6% of the IO completed
2134 within 2 msecs, 20=12.8% means that 12.8% of the IO
2135 took more than 10 msecs, but less than (or equal to) 20 msecs.
2137 After each client has been listed, the group statistics are printed. They
2138 will look like this:
2140 Run status group 0 (all jobs):
2141 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
2142 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
2144 For each data direction, it prints:
2146 io= Number of megabytes io performed.
2147 aggrb= Aggregate bandwidth of threads in this group.
2148 minb= The minimum average bandwidth a thread saw.
2149 maxb= The maximum average bandwidth a thread saw.
2150 mint= The smallest runtime of the threads in that group.
2151 maxt= The longest runtime of the threads in that group.
2153 And finally, the disk statistics are printed. They will look like this:
2155 Disk stats (read/write):
2156 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
2158 Each value is printed for both reads and writes, with reads first. The
2161 ios= Number of ios performed by all groups.
2162 merge= Number of merges io the io scheduler.
2163 ticks= Number of ticks we kept the disk busy.
2164 io_queue= Total time spent in the disk queue.
2165 util= The disk utilization. A value of 100% means we kept the disk
2166 busy constantly, 50% would be a disk idling half of the time.
2168 It is also possible to get fio to dump the current output while it is
2169 running, without terminating the job. To do that, send fio the USR1 signal.
2170 You can also get regularly timed dumps by using the --status-interval
2171 parameter, or by creating a file in /tmp named fio-dump-status. If fio
2172 sees this file, it will unlink it and dump the current output status.
2178 For scripted usage where you typically want to generate tables or graphs
2179 of the results, fio can output the results in a semicolon separated format.
2180 The format is one long line of values, such as:
2182 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%
2183 A description of this job goes here.
2185 The job description (if provided) follows on a second line.
2187 To enable terse output, use the --minimal command line option. The first
2188 value is the version of the terse output format. If the output has to
2189 be changed for some reason, this number will be incremented by 1 to
2190 signify that change.
2192 Split up, the format is as follows:
2194 terse version, fio version, jobname, groupid, error
2196 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
2197 Submission latency: min, max, mean, stdev (usec)
2198 Completion latency: min, max, mean, stdev (usec)
2199 Completion latency percentiles: 20 fields (see below)
2200 Total latency: min, max, mean, stdev (usec)
2201 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
2203 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
2204 Submission latency: min, max, mean, stdev (usec)
2205 Completion latency: min, max, mean, stdev(usec)
2206 Completion latency percentiles: 20 fields (see below)
2207 Total latency: min, max, mean, stdev (usec)
2208 Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
2209 CPU usage: user, system, context switches, major faults, minor faults
2210 IO depths: <=1, 2, 4, 8, 16, 32, >=64
2211 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
2212 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
2213 Disk utilization: Disk name, Read ios, write ios,
2214 Read merges, write merges,
2215 Read ticks, write ticks,
2216 Time spent in queue, disk utilization percentage
2217 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
2219 Additional Info (dependent on description being set): Text description
2221 Completion latency percentiles can be a grouping of up to 20 sets, so
2222 for the terse output fio writes all of them. Each field will look like this:
2226 which is the Xth percentile, and the usec latency associated with it.
2228 For disk utilization, all disks used by fio are shown. So for each disk
2229 there will be a disk utilization section.
2232 8.0 Trace file format
2233 ---------------------
2234 There are two trace file format that you can encounter. The older (v1) format
2235 is unsupported since version 1.20-rc3 (March 2008). It will still be described
2236 below in case that you get an old trace and want to understand it.
2238 In any case the trace is a simple text file with a single action per line.
2241 8.1 Trace file format v1
2242 ------------------------
2243 Each line represents a single io action in the following format:
2247 where rw=0/1 for read/write, and the offset and length entries being in bytes.
2249 This format is not supported in Fio versions => 1.20-rc3.
2252 8.2 Trace file format v2
2253 ------------------------
2254 The second version of the trace file format was added in Fio version 1.17.
2255 It allows to access more then one file per trace and has a bigger set of
2256 possible file actions.
2258 The first line of the trace file has to be:
2262 Following this can be lines in two different formats, which are described below.
2264 The file management format:
2268 The filename is given as an absolute path. The action can be one of these:
2270 add Add the given filename to the trace
2271 open Open the file with the given filename. The filename has to have
2272 been added with the add action before.
2273 close Close the file with the given filename. The file has to have been
2277 The file io action format:
2279 filename action offset length
2281 The filename is given as an absolute path, and has to have been added and opened
2282 before it can be used with this format. The offset and length are given in
2283 bytes. The action can be one of these:
2285 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
2286 The time is relative to the previous wait statement.
2287 read Read 'length' bytes beginning from 'offset'
2288 write Write 'length' bytes beginning from 'offset'
2289 sync fsync() the file
2290 datasync fdatasync() the file
2291 trim trim the given file from the given 'offset' for 'length' bytes
2294 9.0 CPU idleness profiling
2295 --------------------------
2296 In some cases, we want to understand CPU overhead in a test. For example,
2297 we test patches for the specific goodness of whether they reduce CPU usage.
2298 fio implements a balloon approach to create a thread per CPU that runs at
2299 idle priority, meaning that it only runs when nobody else needs the cpu.
2300 By measuring the amount of work completed by the thread, idleness of each
2301 CPU can be derived accordingly.
2303 An unit work is defined as touching a full page of unsigned characters. Mean
2304 and standard deviation of time to complete an unit work is reported in "unit
2305 work" section. Options can be chosen to report detailed percpu idleness or
2306 overall system idleness by aggregating percpu stats.
2309 10.0 Verification and triggers
2310 ------------------------------
2311 Fio is usually run in one of two ways, when data verification is done. The
2312 first is a normal write job of some sort with verify enabled. When the
2313 write phase has completed, fio switches to reads and verifies everything
2314 it wrote. The second model is running just the write phase, and then later
2315 on running the same job (but with reads instead of writes) to repeat the
2316 same IO patterns and verify the contents. Both of these methods depend
2317 on the write phase being completed, as fio otherwise has no idea how much
2320 With verification triggers, fio supports dumping the current write state
2321 to local files. Then a subsequent read verify workload can load this state
2322 and know exactly where to stop. This is useful for testing cases where
2323 power is cut to a server in a managed fashion, for instance.
2325 A verification trigger consists of two things:
2327 1) Storing the write state of each job
2328 2) Executing a trigger command
2330 The write state is relatively small, on the order of hundreds of bytes
2331 to single kilobytes. It contains information on the number of completions
2332 done, the last X completions, etc.
2334 A trigger is invoked either through creation ('touch') of a specified
2335 file in the system, or through a timeout setting. If fio is run with
2336 --trigger-file=/tmp/trigger-file, then it will continually check for
2337 the existence of /tmp/trigger-file. When it sees this file, it will
2338 fire off the trigger (thus saving state, and executing the trigger
2341 For client/server runs, there's both a local and remote trigger. If
2342 fio is running as a server backend, it will send the job states back
2343 to the client for safe storage, then execute the remote trigger, if
2344 specified. If a local trigger is specified, the server will still send
2345 back the write state, but the client will then execute the trigger.
2347 10.1 Verification trigger example
2348 ---------------------------------
2349 Lets say we want to run a powercut test on the remote machine 'server'.
2350 Our write workload is in write-test.fio. We want to cut power to 'server'
2351 at some point during the run, and we'll run this test from the safety
2352 or our local machine, 'localbox'. On the server, we'll start the fio
2355 server# fio --server
2357 and on the client, we'll fire off the workload:
2359 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
2361 We set /tmp/my-trigger as the trigger file, and we tell fio to execute
2363 echo b > /proc/sysrq-trigger
2365 on the server once it has received the trigger and sent us the write
2366 state. This will work, but it's not _really_ cutting power to the server,
2367 it's merely abruptly rebooting it. If we have a remote way of cutting
2368 power to the server through IPMI or similar, we could do that through
2369 a local trigger command instead. Lets assume we have a script that does
2370 IPMI reboot of a given hostname, ipmi-reboot. On localbox, we could
2371 then have run fio with a local trigger instead:
2373 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
2375 For this case, fio would wait for the server to send us the write state,
2376 then execute 'ipmi-reboot server' when that happened.
2378 10.2 Loading verify state
2379 -------------------------
2380 To load store write state, read verification job file must contain
2381 the verify_state_load option. If that is set, fio will load the previously
2382 stored state. For a local fio run this is done by loading the files directly,
2383 and on a client/server run, the server backend will ask the client to send
2384 the files over and load them from there.
2387 11.0 Log File Formats
2388 ---------------------
2390 Fio supports a variety of log file formats, for logging latencies, bandwidth,
2391 and IOPS. The logs share a common format, which looks like this:
2393 time (msec), value, data direction, offset
2395 Time for the log entry is always in milliseconds. The value logged depends
2396 on the type of log, it will be one of the following:
2398 Latency log Value is latency in usecs
2399 Bandwidth log Value is in KB/sec
2400 IOPS log Value is IOPS
2402 Data direction is one of the following:
2408 The offset is the offset, in bytes, from the start of the file, for that
2409 particular IO. The logging of the offset can be toggled with 'log_offset'.
2411 If windowed logging is enabled through 'log_avg_msec', then fio doesn't log
2412 individual IOs. Instead of logs the average values over the specified
2413 period of time. Since 'data direction' and 'offset' are per-IO values,
2414 they aren't applicable if windowed logging is enabled. If windowed logging
2415 is enabled and 'log_max_value' is set, then fio logs maximum values in
2416 that window instead of averages.
projects / fio.git / blob