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 128MiB 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 32KiB and define numjobs to 4 to
158 fork 4 identical jobs. The result is 4 processes each randomly writing
159 to their own 64MiB 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.
280 int Integer. A whole number value, which may contain an integer prefix
281 and an integer suffix.
282 [integer prefix]number[integer suffix]
284 The optional integer prefix specifies the number's base. The default
285 is decimal. 0x specifies hexadecimal.
287 The optional integer suffix specifies the number's units, and includes
288 an optional unit prefix and an optional unit. For quantities of data,
289 the default unit is bytes. For quantities of time, the default unit
292 With kb_base=1000, fio follows international standards for unit prefixes.
293 To specify power-of-10 decimal values defined in the International
294 System of Units (SI):
295 Ki means kilo (K) or 1000
296 Mi means mega (M) or 1000**2
297 Gi means giga (G) or 1000**3
298 Ti means tera (T) or 1000**4
299 Pi means peta (P) or 1000**5
301 To specify power-of-2 binary values defined in IEC 80000-13:
302 k means kibi (Ki) or 1024
303 M means mebi (Mi) or 1024**2
304 G means gibi (Gi) or 1024**3
305 T means tebi (Ti) or 1024**4
306 P means pebi (Pi) or 1024**5
308 With kb_base=1024 (the default), the unit prefixes are opposite from
309 those specified in the SI and IEC 80000-13 standards to provide
310 compatibility with old scripts. For example, 4k means 4096.
312 For quantities of data, an optional unit of 'B' may be included
313 (e.g., 'kB' is the same as 'k').
315 The integer suffix is not case sensitive (e.g., m/mi mean mebi/mega,
316 not milli). 'b' and 'B' both mean byte, not bit.
318 Examples with kb_base=1000:
319 4 KiB: 4096, 4096b, 4096B, 4ki, 4kib, 4kiB, 4Ki, 4KiB
320 1 MiB: 1048576, 1mi, 1024ki
321 1 MB: 1000000, 1m, 1000k
322 1 TiB: 1073741824, 1ti, 1024mi, 1048576ki
323 1 TB: 1000000000, 1t, 1000m, 1000000k
325 Examples with kb_base=1024 (default):
326 4 KiB: 4096, 4096b, 4096B, 4k, 4kb, 4kB, 4K, 4KB
327 1 MiB: 1048576, 1m, 1024k
328 1 MB: 1000000, 1mi, 1000ki
329 1 TiB: 1073741824, 1t, 1024m, 1048576k
330 1 TB: 1000000000, 1ti, 1000mi, 1000000ki
332 To specify times (units are not case sensitive):
336 s or sec means seconds (default)
337 ms or msec means milliseconds
338 us or usec means microseconds
340 If the option accepts an upper and lower range, use a colon ':' or
341 minus '-' to separate such values. See irange.
343 bool Boolean. Usually parsed as an integer, however only defined for
344 true and false (1 and 0).
345 irange Integer range with suffix. Allows value range to be given, such
346 as 1024-4096. A colon may also be used as the separator, eg
347 1k:4k. If the option allows two sets of ranges, they can be
348 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
350 float_list A list of floating point numbers, separated by a ':' character.
352 With the above in mind, here follows the complete list of fio job
355 name=str ASCII name of the job. This may be used to override the
356 name printed by fio for this job. Otherwise the job
357 name is used. On the command line this parameter has the
358 special purpose of also signaling the start of a new
361 wait_for=str Specifies the name of the already defined job to wait
362 for. Single waitee name only may be specified. If set, the job
363 won't be started until all workers of the waitee job are done.
365 Wait_for operates on the job name basis, so there are a few
366 limitations. First, the waitee must be defined prior to the
367 waiter job (meaning no forward references). Second, if a job
368 is being referenced as a waitee, it must have a unique name
369 (no duplicate waitees).
371 description=str Text description of the job. Doesn't do anything except
372 dump this text description when this job is run. It's
375 directory=str Prefix filenames with this directory. Used to place files
376 in a different location than "./". See the 'filename' option
377 for escaping certain characters.
379 filename=str Fio normally makes up a filename based on the job name,
380 thread number, and file number. If you want to share
381 files between threads in a job or several jobs, specify
382 a filename for each of them to override the default.
383 If the ioengine is file based, you can specify a number of
384 files by separating the names with a ':' colon. So if you
385 wanted a job to open /dev/sda and /dev/sdb as the two working
386 files, you would use filename=/dev/sda:/dev/sdb. On Windows,
387 disk devices are accessed as \\.\PhysicalDrive0 for the first
388 device, \\.\PhysicalDrive1 for the second etc. Note: Windows
389 and FreeBSD prevent write access to areas of the disk
390 containing in-use data (e.g. filesystems).
391 If the wanted filename does need to include a colon, then
392 escape that with a '\' character. For instance, if the filename
393 is "/dev/dsk/foo@3,0:c", then you would use
394 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
395 stdin or stdout. Which of the two depends on the read/write
399 If sharing multiple files between jobs, it is usually necessary
400 to have fio generate the exact names that you want. By default,
401 fio will name a file based on the default file format
402 specification of jobname.jobnumber.filenumber. With this
403 option, that can be customized. Fio will recognize and replace
404 the following keywords in this string:
407 The name of the worker thread or process.
410 The incremental number of the worker thread or
414 The incremental number of the file for that worker
417 To have dependent jobs share a set of files, this option can
418 be set to have fio generate filenames that are shared between
419 the two. For instance, if testfiles.$filenum is specified,
420 file number 4 for any job will be named testfiles.4. The
421 default of $jobname.$jobnum.$filenum will be used if
422 no other format specifier is given.
424 unique_filename=bool To avoid collisions between networked clients, fio
425 defaults to prefixing any generated filenames (with a directory
426 specified) with the source of the client connecting. To disable
427 this behavior, set this option to 0.
429 opendir=str Tell fio to recursively add any file it can find in this
430 directory and down the file system tree.
432 lockfile=str Fio defaults to not locking any files before it does
433 IO to them. If a file or file descriptor is shared, fio
434 can serialize IO to that file to make the end result
435 consistent. This is usual for emulating real workloads that
436 share files. The lock modes are:
438 none No locking. The default.
439 exclusive Only one thread/process may do IO,
440 excluding all others.
441 readwrite Read-write locking on the file. Many
442 readers may access the file at the
443 same time, but writes get exclusive
447 rw=str Type of io pattern. Accepted values are:
449 read Sequential reads
450 write Sequential writes
451 trim Sequential trims
452 randwrite Random writes
453 randread Random reads
454 randtrim Random trims
455 rw,readwrite Sequential mixed reads and writes
456 randrw Random mixed reads and writes
457 trimwrite Sequential trim+write sequences
459 Fio defaults to read if the option is not specified.
460 For the mixed io types, the default is to split them 50/50.
461 For certain types of io the result may still be skewed a bit,
462 since the speed may be different. It is possible to specify
463 a number of IO's to do before getting a new offset, this is
464 done by appending a ':<nr>' to the end of the string given.
465 For a random read, it would look like 'rw=randread:8' for
466 passing in an offset modifier with a value of 8. If the
467 suffix is used with a sequential IO pattern, then the value
468 specified will be added to the generated offset for each IO.
469 For instance, using rw=write:4k will skip 4k for every
470 write. It turns sequential IO into sequential IO with holes.
471 See the 'rw_sequencer' option.
473 rw_sequencer=str If an offset modifier is given by appending a number to
474 the rw=<str> line, then this option controls how that
475 number modifies the IO offset being generated. Accepted
478 sequential Generate sequential offset
479 identical Generate the same offset
481 'sequential' is only useful for random IO, where fio would
482 normally generate a new random offset for every IO. If you
483 append eg 8 to randread, you would get a new random offset for
484 every 8 IO's. The result would be a seek for only every 8
485 IO's, instead of for every IO. Use rw=randread:8 to specify
486 that. As sequential IO is already sequential, setting
487 'sequential' for that would not result in any differences.
488 'identical' behaves in a similar fashion, except it sends
489 the same offset 8 number of times before generating a new
492 kb_base=int Select the interpretation of unit prefixes in input parameters.
493 1000 = Inputs comply with IEC 80000-13 and the International
494 System of Units (SI). Use:
495 - power-of-2 values with IEC prefixes (e.g., KiB)
496 - power-of-10 values with SI prefixes (e.g., kB)
497 1024 = Compatibility mode (default). To avoid breaking
499 - power-of-2 values with SI prefixes
500 - power-of-10 values with IEC prefixes
501 See bs= for more details on input parameters.
503 Outputs always use correct prefixes. Most outputs include
504 both side-by-side, like:
505 bw=2383.3kB/s (2327.4KiB/s)
506 If only one value is reported, then kb_base selects the
511 unified_rw_reporting=bool Fio normally reports statistics on a per
512 data direction basis, meaning that reads, writes, and trims are
513 accounted and reported separately. If this option is set,
514 the fio will sum the results and report them as "mixed"
517 randrepeat=bool For random IO workloads, seed the generator in a predictable
518 way so that results are repeatable across repetitions.
521 randseed=int Seed the random number generators based on this seed value, to
522 be able to control what sequence of output is being generated.
523 If not set, the random sequence depends on the randrepeat
526 fallocate=str Whether pre-allocation is performed when laying down files.
529 none Do not pre-allocate space
530 posix Pre-allocate via posix_fallocate()
531 keep Pre-allocate via fallocate() with
532 FALLOC_FL_KEEP_SIZE set
533 0 Backward-compatible alias for 'none'
534 1 Backward-compatible alias for 'posix'
536 May not be available on all supported platforms. 'keep' is only
537 available on Linux.If using ZFS on Solaris this must be set to
538 'none' because ZFS doesn't support it. Default: 'posix'.
540 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
541 on what IO patterns it is likely to issue. Sometimes you
542 want to test specific IO patterns without telling the
543 kernel about it, in which case you can disable this option.
544 The following options are supported:
546 sequential Use FADV_SEQUENTIAL
547 random Use FADV_RANDOM
548 1 Backwards-compatible hint for basing
549 the hint on the fio workload. Will use
550 FADV_SEQUENTIAL for a sequential
551 workload, and FADV_RANDOM for a random
553 0 Backwards-compatible setting for not
554 issing a fadvise hint.
556 fadvise_stream=int Notify the kernel what write stream ID to place these
557 writes under. Only supported on Linux. Note, this option
558 may change going forward.
560 size=int The total size of file io for this job. Fio will run until
561 this many bytes has been transferred, unless runtime is
562 limited by other options (such as 'runtime', for instance,
563 or increased/decreased by 'io_size'). Unless specific nrfiles
564 and filesize options are given, fio will divide this size
565 between the available files specified by the job. If not set,
566 fio will use the full size of the given files or devices.
567 If the files do not exist, size must be given. It is also
568 possible to give size as a percentage between 1 and 100. If
569 size=20% is given, fio will use 20% of the full size of the
570 given files or devices.
573 io_limit=int Normally fio operates within the region set by 'size', which
574 means that the 'size' option sets both the region and size of
575 IO to be performed. Sometimes that is not what you want. With
576 this option, it is possible to define just the amount of IO
577 that fio should do. For instance, if 'size' is set to 20GiB and
578 'io_size' is set to 5GiB, fio will perform IO within the first
579 20GiB but exit when 5GiB have been done. The opposite is also
580 possible - if 'size' is set to 20GiB, and 'io_size' is set to
581 40GiB, then fio will do 40GiB of IO within the 0..20GiB region.
583 filesize=int Individual file sizes. May be a range, in which case fio
584 will select sizes for files at random within the given range
585 and limited to 'size' in total (if that is given). If not
586 given, each created file is the same size.
588 file_append=bool Perform IO after the end of the file. Normally fio will
589 operate within the size of a file. If this option is set, then
590 fio will append to the file instead. This has identical
591 behavior to setting offset to the size of a file. This option
592 is ignored on non-regular files.
595 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
596 space left on device) as the terminating condition. Only makes
597 sense with sequential write. For a read workload, the mount
598 point will be filled first then IO started on the result. This
599 option doesn't make sense if operating on a raw device node,
600 since the size of that is already known by the file system.
601 Additionally, writing beyond end-of-device will not return
604 blocksize=int[,int][,int]
606 The block size in bytes used for I/O units. Default: 4096.
607 A single value applies to reads, writes, and trims.
608 Comma-separated values may be specified for reads, writes,
609 and trims. A value not terminated in a comma applies to
613 bs=256k means 256k for reads, writes and trims
614 bs=8k,32k means 8k for reads, 32k for writes and trims
615 bs=8k,32k, means 8k for reads, 32k for writes, and
617 bs=,8k means default for reads, 8k for writes and trims
618 bs=,8k, means default for reads, 8k for writes, and
621 blocksize_range=irange[,irange][,irange]
622 bsrange=irange[,irange][,irange]
623 A range of block sizes in bytes for I/O units.
624 The issued I/O unit will always be a multiple of the minimum
625 size, unless blocksize_unaligned is set.
627 Comma-separated ranges may be specified for reads, writes,
628 and trims as described in 'blocksize'.
630 Example: bsrange=1k-4k,2k-8k
632 bssplit=str[,str][,str]
633 Sometimes you want even finer grained control of the
634 block sizes issued, not just an even split between them.
635 This option allows you to weight various block sizes,
636 so that you are able to define a specific amount of
637 block sizes issued. The format for this option is:
639 bssplit=blocksize/percentage:blocksize/percentage
641 for as many block sizes as needed. So if you want to define
642 a workload that has 50% 64k blocks, 10% 4k blocks, and
643 40% 32k blocks, you would write:
645 bssplit=4k/10:64k/50:32k/40
647 Ordering does not matter. If the percentage is left blank,
648 fio will fill in the remaining values evenly. So a bssplit
649 option like this one:
651 bssplit=4k/50:1k/:32k/
653 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
654 always add up to 100, if bssplit is given a range that adds
655 up to more, it will error out.
657 Comma-separated values may be specified for reads, writes,
658 and trims as described in 'blocksize'.
660 If you want a workload that has 50% 2k reads and 50% 4k reads,
661 while having 90% 4k writes and 10% 8k writes, you would
664 bssplit=2k/50:4k/50,4k/90:8k/10
667 bs_unaligned If set, fio will issue I/O units with any size within
668 blocksize_range, not just multiples of the minimum size.
669 This typically won't work with direct I/O, as that normally
670 requires sector alignment.
672 bs_is_seq_rand If this option is set, fio will use the normal read,write
673 blocksize settings as sequential,random blocksize settings
674 instead. Any random read or write will use the WRITE blocksize
675 settings, and any sequential read or write will use the READ
678 blockalign=int[,int][,int]
680 Boundary to which fio will align random I/O units.
681 Default: 'blocksize'.
682 Minimum alignment is typically 512b for using direct IO,
683 though it usually depends on the hardware block size. This
684 option is mutually exclusive with using a random map for
685 files, so it will turn off that option.
686 Comma-separated values may be specified for reads, writes,
687 and trims as described in 'blocksize'.
689 zero_buffers If this option is given, fio will init the IO buffers to
690 all zeroes. The default is to fill them with random data.
692 refill_buffers If this option is given, fio will refill the IO buffers
693 on every submit. The default is to only fill it at init
694 time and reuse that data. Only makes sense if zero_buffers
695 isn't specified, naturally. If data verification is enabled,
696 refill_buffers is also automatically enabled.
698 scramble_buffers=bool If refill_buffers is too costly and the target is
699 using data deduplication, then setting this option will
700 slightly modify the IO buffer contents to defeat normal
701 de-dupe attempts. This is not enough to defeat more clever
702 block compression attempts, but it will stop naive dedupe of
703 blocks. Default: true.
705 buffer_compress_percentage=int If this is set, then fio will attempt to
706 provide IO buffer content (on WRITEs) that compress to
707 the specified level. Fio does this by providing a mix of
708 random data and a fixed pattern. The fixed pattern is either
709 zeroes, or the pattern specified by buffer_pattern. If the
710 pattern option is used, it might skew the compression ratio
711 slightly. Note that this is per block size unit, for file/disk
712 wide compression level that matches this setting, you'll also
713 want to set refill_buffers.
715 buffer_compress_chunk=int See buffer_compress_percentage. This
716 setting allows fio to manage how big the ranges of random
717 data and zeroed data is. Without this set, fio will
718 provide buffer_compress_percentage of blocksize random
719 data, followed by the remaining zeroed. With this set
720 to some chunk size smaller than the block size, fio can
721 alternate random and zeroed data throughout the IO
724 buffer_pattern=str If set, fio will fill the io buffers with this
725 pattern. If not set, the contents of io buffers is defined by
726 the other options related to buffer contents. The setting can
727 be any pattern of bytes, and can be prefixed with 0x for hex
728 values. It may also be a string, where the string must then
729 be wrapped with "", e.g.:
731 buffer_pattern="abcd"
735 buffer_pattern=0xdeadface
737 Also you can combine everything together in any order:
738 buffer_pattern=0xdeadface"abcd"-12
740 dedupe_percentage=int If set, fio will generate this percentage of
741 identical buffers when writing. These buffers will be
742 naturally dedupable. The contents of the buffers depend on
743 what other buffer compression settings have been set. It's
744 possible to have the individual buffers either fully
745 compressible, or not at all. This option only controls the
746 distribution of unique buffers.
748 nrfiles=int Number of files to use for this job. Defaults to 1.
750 openfiles=int Number of files to keep open at the same time. Defaults to
751 the same as nrfiles, can be set smaller to limit the number
754 file_service_type=str Defines how fio decides which file from a job to
755 service next. The following types are defined:
757 random Just choose a file at random.
759 roundrobin Round robin over open files. This
762 sequential Finish one file before moving on to
763 the next. Multiple files can still be
764 open depending on 'openfiles'.
766 zipf Use a zipfian distribution to decide what file
769 pareto Use a pareto distribution to decide what file
772 gauss Use a gaussian (normal) distribution to decide
775 For random, roundrobin, and sequential, a postfix can be
776 appended to tell fio how many I/Os to issue before switching
777 to a new file. For example, specifying
778 'file_service_type=random:8' would cause fio to issue 8 I/Os
779 before selecting a new file at random. For the non-uniform
780 distributions, a floating point postfix can be given to
781 influence how the distribution is skewed. See
782 'random_distribution' for a description of how that would work.
784 ioengine=str Defines how the job issues io to the file. The following
787 sync Basic read(2) or write(2) io. lseek(2) is
788 used to position the io location.
790 psync Basic pread(2) or pwrite(2) io. Default on all
791 supported operating systems except for Windows.
793 vsync Basic readv(2) or writev(2) IO.
795 pvsync Basic preadv(2) or pwritev(2) IO.
797 pvsync2 Basic preadv2(2) or pwritev2(2) IO.
799 libaio Linux native asynchronous io. Note that Linux
800 may only support queued behaviour with
801 non-buffered IO (set direct=1 or buffered=0).
802 This engine defines engine specific options.
804 posixaio glibc posix asynchronous io.
806 solarisaio Solaris native asynchronous io.
808 windowsaio Windows native asynchronous io.
811 mmap File is memory mapped and data copied
812 to/from using memcpy(3).
814 splice splice(2) is used to transfer the data and
815 vmsplice(2) to transfer data from user
818 sg SCSI generic sg v3 io. May either be
819 synchronous using the SG_IO ioctl, or if
820 the target is an sg character device
821 we use read(2) and write(2) for asynchronous
822 io. Requires filename option to specify either
823 block or character devices.
825 null Doesn't transfer any data, just pretends
826 to. This is mainly used to exercise fio
827 itself and for debugging/testing purposes.
829 net Transfer over the network to given host:port.
830 Depending on the protocol used, the hostname,
831 port, listen and filename options are used to
832 specify what sort of connection to make, while
833 the protocol option determines which protocol
835 This engine defines engine specific options.
837 netsplice Like net, but uses splice/vmsplice to
838 map data and send/receive.
839 This engine defines engine specific options.
841 cpuio Doesn't transfer any data, but burns CPU
842 cycles according to the cpuload= and
843 cpuchunks= options. Setting cpuload=85
844 will cause that job to do nothing but burn
845 85% of the CPU. In case of SMP machines,
846 use numjobs=<no_of_cpu> to get desired CPU
847 usage, as the cpuload only loads a single
848 CPU at the desired rate. A job never finishes
849 unless there is at least one non-cpuio job.
851 guasi The GUASI IO engine is the Generic Userspace
852 Asyncronous Syscall Interface approach
855 http://www.xmailserver.org/guasi-lib.html
857 for more info on GUASI.
859 rdma The RDMA I/O engine supports both RDMA
860 memory semantics (RDMA_WRITE/RDMA_READ) and
861 channel semantics (Send/Recv) for the
862 InfiniBand, RoCE and iWARP protocols.
864 falloc IO engine that does regular fallocate to
865 simulate data transfer as fio ioengine.
866 DDIR_READ does fallocate(,mode = keep_size,)
867 DDIR_WRITE does fallocate(,mode = 0)
868 DDIR_TRIM does fallocate(,mode = punch_hole)
870 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
871 ioctls to simulate defragment activity in
872 request to DDIR_WRITE event
874 rbd IO engine supporting direct access to Ceph
875 Rados Block Devices (RBD) via librbd without
876 the need to use the kernel rbd driver. This
877 ioengine defines engine specific options.
879 gfapi Using Glusterfs libgfapi sync interface to
880 direct access to Glusterfs volumes without
883 gfapi_async Using Glusterfs libgfapi async interface
884 to direct access to Glusterfs volumes without
885 having to go through FUSE. This ioengine
886 defines engine specific options.
888 libhdfs Read and write through Hadoop (HDFS).
889 This engine interprets offsets a little
890 differently. In HDFS, files once created
891 cannot be modified. So random writes are not
892 possible. To imitate this, libhdfs engine
893 creates bunch of small files, and engine will
894 pick a file out of those files based on the
895 offset generated by fio backend. Each jobs uses
896 it's own connection to HDFS.
898 mtd Read, write and erase an MTD character device
899 (e.g., /dev/mtd0). Discards are treated as
900 erases. Depending on the underlying device
901 type, the I/O may have to go in a certain
902 pattern, e.g., on NAND, writing sequentially
903 to erase blocks and discarding before
904 overwriting. The writetrim mode works well
907 pmemblk Read and write using filesystem DAX to a file
908 on a filesystem mounted with DAX on a persistent
909 memory device through the NVML libpmemblk library.
911 dev-dax Read and write using device DAX to a persistent
912 memory device (e.g., /dev/dax0.0) through the
913 NVML libpmem library.
915 external Prefix to specify loading an external
916 IO engine object file. Append the engine
917 filename, eg ioengine=external:/tmp/foo.o
918 to load ioengine foo.o in /tmp.
920 iodepth=int This defines how many I/O units to keep in flight against
921 the file. The default is 1 for each file defined in this
922 job, can be overridden with a larger value for higher
923 concurrency. Note that increasing iodepth beyond 1 will not
924 affect synchronous ioengines (except for small degress when
925 verify_async is in use). Even async engines may impose OS
926 restrictions causing the desired depth not to be achieved.
927 This may happen on Linux when using libaio and not setting
928 direct=1, since buffered IO is not async on that OS. Keep an
929 eye on the IO depth distribution in the fio output to verify
930 that the achieved depth is as expected. Default: 1.
932 iodepth_batch_submit=int
933 iodepth_batch=int This defines how many pieces of IO to submit at once.
934 It defaults to 1 which means that we submit each IO
935 as soon as it is available, but can be raised to submit
936 bigger batches of IO at the time. If it is set to 0 the iodepth
939 iodepth_batch_complete_min=int
940 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
941 at once. It defaults to 1 which means that we'll ask
942 for a minimum of 1 IO in the retrieval process from
943 the kernel. The IO retrieval will go on until we
944 hit the limit set by iodepth_low. If this variable is
945 set to 0, then fio will always check for completed
946 events before queuing more IO. This helps reduce
947 IO latency, at the cost of more retrieval system calls.
949 iodepth_batch_complete_max=int This defines maximum pieces of IO to
950 retrieve at once. This variable should be used along with
951 iodepth_batch_complete_min=int variable, specifying the range
952 of min and max amount of IO which should be retrieved. By default
953 it is equal to iodepth_batch_complete_min value.
957 iodepth_batch_complete_min=1
958 iodepth_batch_complete_max=<iodepth>
960 which means that we will retrieve at least 1 IO and up to the
961 whole submitted queue depth. If none of IO has been completed
966 iodepth_batch_complete_min=0
967 iodepth_batch_complete_max=<iodepth>
969 which means that we can retrieve up to the whole submitted
970 queue depth, but if none of IO has been completed yet, we will
971 NOT wait and immediately exit the system call. In this example
972 we simply do polling.
974 iodepth_low=int The low water mark indicating when to start filling
975 the queue again. Defaults to the same as iodepth, meaning
976 that fio will attempt to keep the queue full at all times.
977 If iodepth is set to eg 16 and iodepth_low is set to 4, then
978 after fio has filled the queue of 16 requests, it will let
979 the depth drain down to 4 before starting to fill it again.
981 io_submit_mode=str This option controls how fio submits the IO to
982 the IO engine. The default is 'inline', which means that the
983 fio job threads submit and reap IO directly. If set to
984 'offload', the job threads will offload IO submission to a
985 dedicated pool of IO threads. This requires some coordination
986 and thus has a bit of extra overhead, especially for lower
987 queue depth IO where it can increase latencies. The benefit
988 is that fio can manage submission rates independently of
989 the device completion rates. This avoids skewed latency
990 reporting if IO gets back up on the device side (the
991 coordinated omission problem).
993 direct=bool If value is true, use non-buffered io. This is usually
994 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
995 On Windows the synchronous ioengines don't support direct io.
997 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
998 writes are guaranteed to be stable once acknowledged by
999 the operating system. Only Linux supports O_ATOMIC right
1002 buffered=bool If value is true, use buffered io. This is the opposite
1003 of the 'direct' option. Defaults to true.
1005 offset=int Start io at the given offset in the file. The data before
1006 the given offset will not be touched. This effectively
1007 caps the file size at real_size - offset.
1009 offset_increment=int If this is provided, then the real offset becomes
1010 offset + offset_increment * thread_number, where the thread
1011 number is a counter that starts at 0 and is incremented for
1012 each sub-job (i.e. when numjobs option is specified). This
1013 option is useful if there are several jobs which are intended
1014 to operate on a file in parallel disjoint segments, with
1015 even spacing between the starting points.
1017 number_ios=int Fio will normally perform IOs until it has exhausted the size
1018 of the region set by size=, or if it exhaust the allocated
1019 time (or hits an error condition). With this setting, the
1020 range/size can be set independently of the number of IOs to
1021 perform. When fio reaches this number, it will exit normally
1022 and report status. Note that this does not extend the amount
1023 of IO that will be done, it will only stop fio if this
1024 condition is met before other end-of-job criteria.
1026 fsync=int If writing to a file, issue a sync of the dirty data
1027 for every number of blocks given. For example, if you give
1028 32 as a parameter, fio will sync the file for every 32
1029 writes issued. If fio is using non-buffered io, we may
1030 not sync the file. The exception is the sg io engine, which
1031 synchronizes the disk cache anyway.
1033 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
1035 In FreeBSD and Windows there is no fdatasync(), this falls back
1038 sync_file_range=str:val Use sync_file_range() for every 'val' number of
1039 write operations. Fio will track range of writes that
1040 have happened since the last sync_file_range() call. 'str'
1041 can currently be one or more of:
1043 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
1044 write SYNC_FILE_RANGE_WRITE
1045 wait_after SYNC_FILE_RANGE_WAIT_AFTER
1047 So if you do sync_file_range=wait_before,write:8, fio would
1048 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
1049 every 8 writes. Also see the sync_file_range(2) man page.
1050 This option is Linux specific.
1052 overwrite=bool If true, writes to a file will always overwrite existing
1053 data. If the file doesn't already exist, it will be
1054 created before the write phase begins. If the file exists
1055 and is large enough for the specified write phase, nothing
1058 end_fsync=bool If true, fsync file contents when a write stage has completed.
1060 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
1061 This differs from end_fsync in that it will happen on every
1062 file close, not just at the end of the job.
1064 rwmixread=int How large a percentage of the mix should be reads.
1066 rwmixwrite=int How large a percentage of the mix should be writes. If both
1067 rwmixread and rwmixwrite is given and the values do not add
1068 up to 100%, the latter of the two will be used to override
1069 the first. This may interfere with a given rate setting,
1070 if fio is asked to limit reads or writes to a certain rate.
1071 If that is the case, then the distribution may be skewed.
1073 random_distribution=str:float[,str:float][,str:float]
1074 By default, fio will use a completely uniform
1075 random distribution when asked to perform random IO. Sometimes
1076 it is useful to skew the distribution in specific ways,
1077 ensuring that some parts of the data is more hot than others.
1078 fio includes the following distribution models:
1080 random Uniform random distribution
1081 zipf Zipf distribution
1082 pareto Pareto distribution
1083 gauss Normal (gaussian) distribution
1084 zoned Zoned random distribution
1086 When using a zipf or pareto distribution, an input value
1087 is also needed to define the access pattern. For zipf, this
1088 is the zipf theta. For pareto, it's the pareto power. Fio
1089 includes a test program, genzipf, that can be used visualize
1090 what the given input values will yield in terms of hit rates.
1091 If you wanted to use zipf with a theta of 1.2, you would use
1092 random_distribution=zipf:1.2 as the option. If a non-uniform
1093 model is used, fio will disable use of the random map. For
1094 the gauss distribution, a normal deviation is supplied as
1095 a value between 0 and 100.
1097 For a zoned distribution, fio supports specifying percentages
1098 of IO access that should fall within what range of the file or
1099 device. For example, given a criteria of:
1101 60% of accesses should be to the first 10%
1102 30% of accesses should be to the next 20%
1103 8% of accesses should be to to the next 30%
1104 2% of accesses should be to the next 40%
1106 we can define that through zoning of the random accesses. For
1107 the above example, the user would do:
1109 random_distribution=zoned:60/10:30/20:8/30:2/40
1111 similarly to how bssplit works for setting ranges and
1112 percentages of block sizes. Like bssplit, it's possible to
1113 specify separate zones for reads, writes, and trims. If just
1114 one set is given, it'll apply to all of them.
1116 percentage_random=int[,int][,int]
1117 For a random workload, set how big a percentage should
1118 be random. This defaults to 100%, in which case the workload
1119 is fully random. It can be set from anywhere from 0 to 100.
1120 Setting it to 0 would make the workload fully sequential. Any
1121 setting in between will result in a random mix of sequential
1122 and random IO, at the given percentages.
1123 Comma-separated values may be specified for reads, writes,
1124 and trims as described in 'blocksize'.
1126 norandommap Normally fio will cover every block of the file when doing
1127 random IO. If this option is given, fio will just get a
1128 new random offset without looking at past io history. This
1129 means that some blocks may not be read or written, and that
1130 some blocks may be read/written more than once. If this option
1131 is used with verify= and multiple blocksizes (via bsrange=),
1132 only intact blocks are verified, i.e., partially-overwritten
1135 softrandommap=bool See norandommap. If fio runs with the random block map
1136 enabled and it fails to allocate the map, if this option is
1137 set it will continue without a random block map. As coverage
1138 will not be as complete as with random maps, this option is
1139 disabled by default.
1141 random_generator=str Fio supports the following engines for generating
1142 IO offsets for random IO:
1144 tausworthe Strong 2^88 cycle random number generator
1145 lfsr Linear feedback shift register generator
1146 tausworthe64 Strong 64-bit 2^258 cycle random number
1149 Tausworthe is a strong random number generator, but it
1150 requires tracking on the side if we want to ensure that
1151 blocks are only read or written once. LFSR guarantees
1152 that we never generate the same offset twice, and it's
1153 also less computationally expensive. It's not a true
1154 random generator, however, though for IO purposes it's
1155 typically good enough. LFSR only works with single
1156 block sizes, not with workloads that use multiple block
1157 sizes. If used with such a workload, fio may read or write
1158 some blocks multiple times. The default value is tausworthe,
1159 unless the required space exceeds 2^32 blocks. If it does,
1160 then tausworthe64 is selected automatically.
1162 nice=int Run the job with the given nice value. See man nice(2).
1164 On Windows, values less than -15 set the process class to "High";
1165 -1 through -15 set "Above Normal"; 1 through 15 "Below Normal";
1166 and above 15 "Idle" priority class.
1168 prio=int Set the io priority value of this job. Linux limits us to
1169 a positive value between 0 and 7, with 0 being the highest.
1170 See man ionice(1). Refer to an appropriate manpage for
1171 other operating systems since meaning of priority may differ.
1173 prioclass=int Set the io priority class. See man ionice(1).
1175 thinktime=int Stall the job x microseconds after an io has completed before
1176 issuing the next. May be used to simulate processing being
1177 done by an application. See thinktime_blocks and
1181 Only valid if thinktime is set - pretend to spend CPU time
1182 doing something with the data received, before falling back
1183 to sleeping for the rest of the period specified by
1186 thinktime_blocks=int
1187 Only valid if thinktime is set - control how many blocks
1188 to issue, before waiting 'thinktime' usecs. If not set,
1189 defaults to 1 which will make fio wait 'thinktime' usecs
1190 after every block. This effectively makes any queue depth
1191 setting redundant, since no more than 1 IO will be queued
1192 before we have to complete it and do our thinktime. In
1193 other words, this setting effectively caps the queue depth
1194 if the latter is larger.
1196 rate=int[,int][,int]
1197 Cap the bandwidth used by this job. The number is in bytes/sec,
1198 the normal suffix rules apply.
1199 Comma-separated values may be specified for reads, writes,
1200 and trims as described in 'blocksize'.
1202 rate_min=int[,int][,int]
1203 Tell fio to do whatever it can to maintain at least this
1204 bandwidth. Failing to meet this requirement will cause
1206 Comma-separated values may be specified for reads, writes,
1207 and trims as described in 'blocksize'.
1209 rate_iops=int[,int][,int]
1210 Cap the bandwidth to this number of IOPS. Basically the same
1211 as rate, just specified independently of bandwidth. If the
1212 job is given a block size range instead of a fixed value,
1213 the smallest block size is used as the metric.
1214 Comma-separated values may be specified for reads, writes,
1215 and trims as described in 'blocksize'.
1217 rate_iops_min=int[,int][,int]
1218 If fio doesn't meet this rate of IO, it will cause
1220 Comma-separated values may be specified for reads, writes,
1221 and trims as described in 'blocksize'.
1223 rate_process=str This option controls how fio manages rated IO
1224 submissions. The default is 'linear', which submits IO in a
1225 linear fashion with fixed delays between IOs that gets
1226 adjusted based on IO completion rates. If this is set to
1227 'poisson', fio will submit IO based on a more real world
1228 random request flow, known as the Poisson process
1229 (https://en.wikipedia.org/wiki/Poisson_process). The lambda
1230 will be 10^6 / IOPS for the given workload.
1232 latency_target=int If set, fio will attempt to find the max performance
1233 point that the given workload will run at while maintaining a
1234 latency below this target. The values is given in microseconds.
1235 See latency_window and latency_percentile
1237 latency_window=int Used with latency_target to specify the sample window
1238 that the job is run at varying queue depths to test the
1239 performance. The value is given in microseconds.
1241 latency_percentile=float The percentage of IOs that must fall within the
1242 criteria specified by latency_target and latency_window. If not
1243 set, this defaults to 100.0, meaning that all IOs must be equal
1244 or below to the value set by latency_target.
1246 max_latency=int If set, fio will exit the job if it exceeds this maximum
1247 latency. It will exit with an ETIME error.
1249 rate_cycle=int Average bandwidth for 'rate' and 'rate_min' over this number
1252 cpumask=int Set the CPU affinity of this job. The parameter given is a
1253 bitmask of allowed CPU's the job may run on. So if you want
1254 the allowed CPUs to be 1 and 5, you would pass the decimal
1255 value of (1 << 1 | 1 << 5), or 34. See man
1256 sched_setaffinity(2). This may not work on all supported
1257 operating systems or kernel versions. This option doesn't
1258 work well for a higher CPU count than what you can store in
1259 an integer mask, so it can only control cpus 1-32. For
1260 boxes with larger CPU counts, use cpus_allowed.
1262 cpus_allowed=str Controls the same options as cpumask, but it allows a text
1263 setting of the permitted CPUs instead. So to use CPUs 1 and
1264 5, you would specify cpus_allowed=1,5. This options also
1265 allows a range of CPUs. Say you wanted a binding to CPUs
1266 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
1268 cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
1269 specified by cpus_allowed or cpumask. Two policies are
1272 shared All jobs will share the CPU set specified.
1273 split Each job will get a unique CPU from the CPU set.
1275 'shared' is the default behaviour, if the option isn't
1276 specified. If split is specified, then fio will will assign
1277 one cpu per job. If not enough CPUs are given for the jobs
1278 listed, then fio will roundrobin the CPUs in the set.
1280 numa_cpu_nodes=str Set this job running on specified NUMA nodes' CPUs. The
1281 arguments allow comma delimited list of cpu numbers,
1282 A-B ranges, or 'all'. Note, to enable numa options support,
1283 fio must be built on a system with libnuma-dev(el) installed.
1285 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
1286 nodes. Format of the arguments:
1288 `mode' is one of the following memory policy:
1289 default, prefer, bind, interleave, local
1290 For `default' and `local' memory policy, no node is
1291 needed to be specified.
1292 For `prefer', only one node is allowed.
1293 For `bind' and `interleave', it allow comma delimited
1294 list of numbers, A-B ranges, or 'all'.
1296 startdelay=time Start this job the specified number of seconds after fio
1297 has started. Only useful if the job file contains several
1298 jobs, and you want to delay starting some jobs to a certain
1301 runtime=time Tell fio to terminate processing after the specified number
1302 of seconds. It can be quite hard to determine for how long
1303 a specified job will run, so this parameter is handy to
1304 cap the total runtime to a given time.
1306 time_based If set, fio will run for the duration of the runtime
1307 specified even if the file(s) are completely read or
1308 written. It will simply loop over the same workload
1309 as many times as the runtime allows.
1311 ramp_time=time If set, fio will run the specified workload for this amount
1312 of time before logging any performance numbers. Useful for
1313 letting performance settle before logging results, thus
1314 minimizing the runtime required for stable results. Note
1315 that the ramp_time is considered lead in time for a job,
1316 thus it will increase the total runtime if a special timeout
1317 or runtime is specified.
1319 steadystate=str:float
1320 ss=str:float Define the criterion and limit for assessing steady state
1321 performance. The first parameter designates the criterion
1322 whereas the second parameter sets the threshold. When the
1323 criterion falls below the threshold for the specified duration,
1324 the job will stop. For example, iops_slope:0.1% will direct fio
1325 to terminate the job when the least squares regression slope
1326 falls below 0.1% of the mean IOPS. If group_reporting is
1327 enabled this will apply to all jobs in the group. Below is the
1328 list of available steady state assessment criteria. All
1329 assessments are carried out using only data from the rolling
1330 collection window. Threshold limits can be expressed as a fixed
1331 value or as a percentage of the mean in the collection window.
1332 iops Collect IOPS data. Stop the job if all
1333 individual IOPS measurements are within the
1334 specified limit of the mean IOPS (e.g., iops:2
1335 means that all individual IOPS values must be
1336 within 2 of the mean, whereas iops:0.2% means
1337 that all individual IOPS values must be within
1338 0.2% of the mean IOPS to terminate the job).
1340 Collect IOPS data and calculate the least
1341 squares regression slope. Stop the job if the
1342 slope falls below the specified limit.
1343 bw Collect bandwidth data. Stop the job if all
1344 individual bandwidth measurements are within
1345 the specified limit of the mean bandwidth.
1347 Collect bandwidth data and calculate the least
1348 squares regression slope. Stop the job if the
1349 slope falls below the specified limit.
1351 steadystate_duration=time
1352 ss_dur=time A rolling window of this duration will be used to judge whether
1353 steady state has been reached. Data will be collected once per
1354 second. The default is 0 which disables steady state detection.
1356 steadystate_ramp_time=time
1357 ss_ramp=time Allow the job to run for the specified duration before
1358 beginning data collection for checking the steady state job
1359 termination criterion. The default is 0.
1361 invalidate=bool Invalidate the buffer/page cache parts for this file prior
1362 to starting io. Defaults to true.
1364 sync=bool Use sync io for buffered writes. For the majority of the
1365 io engines, this means using O_SYNC.
1368 mem=str Fio can use various types of memory as the I/O unit buffer.
1369 The allowed values are:
1371 malloc Use memory from malloc(3) as the buffers.
1372 Default memory type.
1374 shm Use shared memory as the buffers. Allocated
1377 shmhuge Same as shm, but use huge pages as backing.
1379 mmap Use mmap to allocate buffers. May either be
1380 anonymous memory, or can be file backed if
1381 a filename is given after the option. The
1382 format is mem=mmap:/path/to/file.
1384 mmaphuge Use a memory mapped huge file as the buffer
1385 backing. Append filename after mmaphuge, ala
1386 mem=mmaphuge:/hugetlbfs/file
1388 mmapshared Same as mmap, but use a MMAP_SHARED
1391 The area allocated is a function of the maximum allowed
1392 bs size for the job, multiplied by the io depth given. Note
1393 that for shmhuge and mmaphuge to work, the system must have
1394 free huge pages allocated. This can normally be checked
1395 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1396 Linux system. Fio assumes a huge page is 4MiB in size. So
1397 to calculate the number of huge pages you need for a given
1398 job file, add up the io depth of all jobs (normally one unless
1399 iodepth= is used) and multiply by the maximum bs set. Then
1400 divide that number by the huge page size. You can see the
1401 size of the huge pages in /proc/meminfo. If no huge pages
1402 are allocated by having a non-zero number in nr_hugepages,
1403 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1405 mmaphuge also needs to have hugetlbfs mounted and the file
1406 location should point there. So if it's mounted in /huge,
1407 you would use mem=mmaphuge:/huge/somefile.
1409 iomem_align=int This indicates the memory alignment of the IO memory buffers.
1410 Note that the given alignment is applied to the first I/O unit
1411 buffer, if using iodepth the alignment of the following buffers
1412 are given by the bs used. In other words, if using a bs that is
1413 a multiple of the page sized in the system, all buffers will
1414 be aligned to this value. If using a bs that is not page
1415 aligned, the alignment of subsequent IO memory buffers is the
1416 sum of the iomem_align and bs used.
1419 Defines the size of a huge page. Must at least be equal
1420 to the system setting, see /proc/meminfo. Defaults to 4MiB.
1421 Should probably always be a multiple of megabytes, so using
1422 hugepage-size=Xm is the preferred way to set this to avoid
1423 setting a non-pow-2 bad value.
1425 exitall When one job finishes, terminate the rest. The default is
1426 to wait for each job to finish, sometimes that is not the
1429 exitall_on_error When one job finishes in error, terminate the rest. The
1430 default is to wait for each job to finish.
1432 bwavgtime=int Average the calculated bandwidth over the given time. Value
1433 is specified in milliseconds. If the job also does bandwidth
1434 logging through 'write_bw_log', then the minimum of this option
1435 and 'log_avg_msec' will be used. Default: 500ms.
1437 iopsavgtime=int Average the calculated IOPS over the given time. Value
1438 is specified in milliseconds. If the job also does IOPS logging
1439 through 'write_iops_log', then the minimum of this option and
1440 'log_avg_msec' will be used. Default: 500ms.
1442 create_serialize=bool If true, serialize the file creation for the jobs.
1443 This may be handy to avoid interleaving of data
1444 files, which may greatly depend on the filesystem
1445 used and even the number of processors in the system.
1447 create_fsync=bool fsync the data file after creation. This is the
1450 create_on_open=bool Don't pre-setup the files for IO, just create open()
1451 when it's time to do IO to that file.
1453 create_only=bool If true, fio will only run the setup phase of the job.
1454 If files need to be laid out or updated on disk, only
1455 that will be done. The actual job contents are not
1458 allow_file_create=bool If true, fio is permitted to create files as part
1459 of its workload. This is the default behavior. If this
1460 option is false, then fio will error out if the files it
1461 needs to use don't already exist. Default: true.
1463 allow_mounted_write=bool If this isn't set, fio will abort jobs that
1464 are destructive (eg that write) to what appears to be a
1465 mounted device or partition. This should help catch creating
1466 inadvertently destructive tests, not realizing that the test
1467 will destroy data on the mounted file system. Default: false.
1469 pre_read=bool If this is given, files will be pre-read into memory before
1470 starting the given IO operation. This will also clear
1471 the 'invalidate' flag, since it is pointless to pre-read
1472 and then drop the cache. This will only work for IO engines
1473 that are seek-able, since they allow you to read the same data
1474 multiple times. Thus it will not work on eg network or splice
1477 unlink=bool Unlink the job files when done. Not the default, as repeated
1478 runs of that job would then waste time recreating the file
1479 set again and again.
1481 unlink_each_loop=bool Unlink job files after each iteration or loop.
1483 loops=int Run the specified number of iterations of this job. Used
1484 to repeat the same workload a given number of times. Defaults
1487 verify_only Do not perform specified workload---only verify data still
1488 matches previous invocation of this workload. This option
1489 allows one to check data multiple times at a later date
1490 without overwriting it. This option makes sense only for
1491 workloads that write data, and does not support workloads
1492 with the time_based option set.
1494 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1495 verify is set. Defaults to 1.
1497 verify=str If writing to a file, fio can verify the file contents
1498 after each iteration of the job. Each verification method also implies
1499 verification of special header, which is written to the beginning of
1500 each block. This header also includes meta information, like offset
1501 of the block, block number, timestamp when block was written, etc.
1502 verify=str can be combined with verify_pattern=str option.
1503 The allowed values are:
1505 md5 Use an md5 sum of the data area and store
1506 it in the header of each block.
1508 crc64 Use an experimental crc64 sum of the data
1509 area and store it in the header of each
1512 crc32c Use a crc32c sum of the data area and store
1513 it in the header of each block.
1515 crc32c-intel Use hardware assisted crc32c calculation
1516 provided on SSE4.2 enabled processors. Falls
1517 back to regular software crc32c, if not
1518 supported by the system.
1520 crc32c-arm64 Use hardware assisted crc32c calculation
1521 provided on CRC enabled ARM 64-bits processors.
1522 Falls back to regular software crc32c, if not
1523 supported by the system.
1525 crc32 Use a crc32 sum of the data area and store
1526 it in the header of each block.
1528 crc16 Use a crc16 sum of the data area and store
1529 it in the header of each block.
1531 crc7 Use a crc7 sum of the data area and store
1532 it in the header of each block.
1534 xxhash Use xxhash as the checksum function. Generally
1535 the fastest software checksum that fio
1538 sha512 Use sha512 as the checksum function.
1540 sha256 Use sha256 as the checksum function.
1542 sha1 Use optimized sha1 as the checksum function.
1544 meta This option is deprecated, since now meta information is
1545 included in generic verification header and meta verification
1546 happens by default. For detailed information see the description
1547 of the verify=str setting. This option is kept because of
1548 compatibility's sake with old configurations. Do not use it.
1550 pattern Verify a strict pattern. Normally fio includes
1551 a header with some basic information and
1552 checksumming, but if this option is set, only
1553 the specific pattern set with 'verify_pattern'
1556 null Only pretend to verify. Useful for testing
1557 internals with ioengine=null, not for much
1560 This option can be used for repeated burn-in tests of a
1561 system to make sure that the written data is also
1562 correctly read back. If the data direction given is
1563 a read or random read, fio will assume that it should
1564 verify a previously written file. If the data direction
1565 includes any form of write, the verify will be of the
1568 verifysort=bool If set, fio will sort written verify blocks when it deems
1569 it faster to read them back in a sorted manner. This is
1570 often the case when overwriting an existing file, since
1571 the blocks are already laid out in the file system. You
1572 can ignore this option unless doing huge amounts of really
1573 fast IO where the red-black tree sorting CPU time becomes
1576 verify_offset=int Swap the verification header with data somewhere else
1577 in the block before writing. Its swapped back before
1580 verify_interval=int Write the verification header at a finer granularity
1581 than the blocksize. It will be written for chunks the
1582 size of header_interval. blocksize should divide this
1585 verify_pattern=str If set, fio will fill the io buffers with this
1586 pattern. Fio defaults to filling with totally random
1587 bytes, but sometimes it's interesting to fill with a known
1588 pattern for io verification purposes. Depending on the
1589 width of the pattern, fio will fill 1/2/3/4 bytes of the
1590 buffer at the time(it can be either a decimal or a hex number).
1591 The verify_pattern if larger than a 32-bit quantity has to
1592 be a hex number that starts with either "0x" or "0X". Use
1593 with verify=str. Also, verify_pattern supports %o format,
1594 which means that for each block offset will be written and
1595 then verified back, e.g.:
1599 Or use combination of everything:
1600 verify_pattern=0xff%o"abcd"-12
1602 verify_fatal=bool Normally fio will keep checking the entire contents
1603 before quitting on a block verification failure. If this
1604 option is set, fio will exit the job on the first observed
1607 verify_dump=bool If set, dump the contents of both the original data
1608 block and the data block we read off disk to files. This
1609 allows later analysis to inspect just what kind of data
1610 corruption occurred. Off by default.
1612 verify_async=int Fio will normally verify IO inline from the submitting
1613 thread. This option takes an integer describing how many
1614 async offload threads to create for IO verification instead,
1615 causing fio to offload the duty of verifying IO contents
1616 to one or more separate threads. If using this offload
1617 option, even sync IO engines can benefit from using an
1618 iodepth setting higher than 1, as it allows them to have
1619 IO in flight while verifies are running.
1621 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1622 async IO verification threads. See cpus_allowed for the
1625 verify_backlog=int Fio will normally verify the written contents of a
1626 job that utilizes verify once that job has completed. In
1627 other words, everything is written then everything is read
1628 back and verified. You may want to verify continually
1629 instead for a variety of reasons. Fio stores the meta data
1630 associated with an IO block in memory, so for large
1631 verify workloads, quite a bit of memory would be used up
1632 holding this meta data. If this option is enabled, fio
1633 will write only N blocks before verifying these blocks.
1635 verify_backlog_batch=int Control how many blocks fio will verify
1636 if verify_backlog is set. If not set, will default to
1637 the value of verify_backlog (meaning the entire queue
1638 is read back and verified). If verify_backlog_batch is
1639 less than verify_backlog then not all blocks will be verified,
1640 if verify_backlog_batch is larger than verify_backlog, some
1641 blocks will be verified more than once.
1643 verify_state_save=bool When a job exits during the write phase of a verify
1644 workload, save its current state. This allows fio to replay
1645 up until that point, if the verify state is loaded for the
1646 verify read phase. The format of the filename is, roughly,
1647 <type>-<jobname>-<jobindex>-verify.state. <type> is "local"
1648 for a local run, "sock" for a client/server socket connection,
1649 and "ip" (192.168.0.1, for instance) for a networked
1650 client/server connection.
1652 verify_state_load=bool If a verify termination trigger was used, fio stores
1653 the current write state of each thread. This can be used at
1654 verification time so that fio knows how far it should verify.
1655 Without this information, fio will run a full verification
1656 pass, according to the settings in the job file used.
1659 wait_for_previous Wait for preceding jobs in the job file to exit, before
1660 starting this one. Can be used to insert serialization
1661 points in the job file. A stone wall also implies starting
1662 a new reporting group.
1664 new_group Start a new reporting group. See: group_reporting.
1666 numjobs=int Create the specified number of clones of this job. May be
1667 used to setup a larger number of threads/processes doing
1668 the same thing. Each thread is reported separately; to see
1669 statistics for all clones as a whole, use group_reporting in
1670 conjunction with new_group.
1672 group_reporting It may sometimes be interesting to display statistics for
1673 groups of jobs as a whole instead of for each individual job.
1674 This is especially true if 'numjobs' is used; looking at
1675 individual thread/process output quickly becomes unwieldy.
1676 To see the final report per-group instead of per-job, use
1677 'group_reporting'. Jobs in a file will be part of the same
1678 reporting group, unless if separated by a stonewall, or by
1681 thread fio defaults to forking jobs, however if this option is
1682 given, fio will use pthread_create(3) to create threads
1685 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1687 zoneskip=int Skip the specified number of bytes when zonesize data has
1688 been read. The two zone options can be used to only do
1689 io on zones of a file.
1691 write_iolog=str Write the issued io patterns to the specified file. See
1692 read_iolog. Specify a separate file for each job, otherwise
1693 the iologs will be interspersed and the file may be corrupt.
1695 read_iolog=str Open an iolog with the specified file name and replay the
1696 io patterns it contains. This can be used to store a
1697 workload and replay it sometime later. The iolog given
1698 may also be a blktrace binary file, which allows fio
1699 to replay a workload captured by blktrace. See blktrace
1700 for how to capture such logging data. For blktrace replay,
1701 the file needs to be turned into a blkparse binary data
1702 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1704 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1705 is to attempt to respect the time stamps within the log and
1706 replay them with the appropriate delay between IOPS. By
1707 setting this variable fio will not respect the timestamps and
1708 attempt to replay them as fast as possible while still
1709 respecting ordering. The result is the same I/O pattern to a
1710 given device, but different timings.
1712 replay_redirect=str While replaying I/O patterns using read_iolog the
1713 default behavior is to replay the IOPS onto the major/minor
1714 device that each IOP was recorded from. This is sometimes
1715 undesirable because on a different machine those major/minor
1716 numbers can map to a different device. Changing hardware on
1717 the same system can also result in a different major/minor
1718 mapping. Replay_redirect causes all IOPS to be replayed onto
1719 the single specified device regardless of the device it was
1720 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1721 IO in the blktrace or iolog to be replayed onto /dev/sdc.
1722 This means multiple devices will be replayed onto a single
1723 device, if the trace contains multiple devices. If you want
1724 multiple devices to be replayed concurrently to multiple
1725 redirected devices you must blkparse your trace into separate
1726 traces and replay them with independent fio invocations.
1727 Unfortunately this also breaks the strict time ordering
1728 between multiple device accesses.
1730 replay_align=int Force alignment of IO offsets and lengths in a trace
1731 to this power of 2 value.
1733 replay_scale=int Scale sector offsets down by this factor when
1736 per_job_logs=bool If set, this generates bw/clat/iops log with per
1737 file private filenames. If not set, jobs with identical names
1738 will share the log filename. Default: true.
1740 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1741 file. Can be used to store data of the bandwidth of the
1742 jobs in their lifetime. The included fio_generate_plots
1743 script uses gnuplot to turn these text files into nice
1744 graphs. See write_lat_log for behaviour of given
1745 filename. For this option, the suffix is _bw.x.log, where
1746 x is the index of the job (1..N, where N is the number of
1747 jobs). If 'per_job_logs' is false, then the filename will not
1748 include the job index. See 'Log File Formats'.
1750 write_lat_log=str Same as write_bw_log, except that this option stores io
1751 submission, completion, and total latencies instead. If no
1752 filename is given with this option, the default filename of
1753 "jobname_type.log" is used. Even if the filename is given,
1754 fio will still append the type of log. So if one specifies
1758 The actual log names will be foo_slat.x.log, foo_clat.x.log,
1759 and foo_lat.x.log, where x is the index of the job (1..N,
1760 where N is the number of jobs). This helps fio_generate_plot
1761 find the logs automatically. If 'per_job_logs' is false, then
1762 the filename will not include the job index. See 'Log File
1765 write_hist_log=str Same as write_lat_log, but writes I/O completion
1766 latency histograms. If no filename is given with this option, the
1767 default filename of "jobname_clat_hist.x.log" is used, where x is
1768 the index of the job (1..N, where N is the number of jobs). Even
1769 if the filename is given, fio will still append the type of log.
1770 If per_job_logs is false, then the filename will not include the
1771 job index. See 'Log File Formats'.
1773 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1774 given with this option, the default filename of
1775 "jobname_type.x.log" is used,where x is the index of the job
1776 (1..N, where N is the number of jobs). Even if the filename
1777 is given, fio will still append the type of log. If
1778 'per_job_logs' is false, then the filename will not include
1779 the job index. See 'Log File Formats'.
1781 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1782 or bw log for every IO that completes. When writing to the
1783 disk log, that can quickly grow to a very large size. Setting
1784 this option makes fio average the each log entry over the
1785 specified period of time, reducing the resolution of the log.
1786 See log_max_value as well. Defaults to 0, logging all entries.
1788 log_hist_msec=int Same as log_avg_msec, but logs entries for completion
1789 latency histograms. Computing latency percentiles from averages of
1790 intervals using log_avg_msec is innacurate. Setting this option makes
1791 fio log histogram entries over the specified period of time, reducing
1792 log sizes for high IOPS devices while retaining percentile accuracy.
1793 See log_hist_coarseness as well. Defaults to 0, meaning histogram
1794 logging is disabled.
1796 log_hist_coarseness=int Integer ranging from 0 to 6, defining the coarseness
1797 of the resolution of the histogram logs enabled with log_hist_msec. For
1798 each increment in coarseness, fio outputs half as many bins. Defaults to
1799 0, for which histogram logs contain 1216 latency bins. See
1802 log_max_value=bool If log_avg_msec is set, fio logs the average over that
1803 window. If you instead want to log the maximum value, set this
1804 option to 1. Defaults to 0, meaning that averaged values are
1807 log_offset=int If this is set, the iolog options will include the byte
1808 offset for the IO entry as well as the other data values.
1810 log_compression=int If this is set, fio will compress the IO logs as
1811 it goes, to keep the memory footprint lower. When a log
1812 reaches the specified size, that chunk is removed and
1813 compressed in the background. Given that IO logs are
1814 fairly highly compressible, this yields a nice memory
1815 savings for longer runs. The downside is that the
1816 compression will consume some background CPU cycles, so
1817 it may impact the run. This, however, is also true if
1818 the logging ends up consuming most of the system memory.
1819 So pick your poison. The IO logs are saved normally at the
1820 end of a run, by decompressing the chunks and storing them
1821 in the specified log file. This feature depends on the
1822 availability of zlib.
1824 log_compression_cpus=str Define the set of CPUs that are allowed to
1825 handle online log compression for the IO jobs. This can
1826 provide better isolation between performance sensitive jobs,
1827 and background compression work.
1829 log_store_compressed=bool If set, fio will store the log files in a
1830 compressed format. They can be decompressed with fio, using
1831 the --inflate-log command line parameter. The files will be
1832 stored with a .fz suffix.
1834 log_unix_epoch=bool If set, fio will log Unix timestamps to the log
1835 files produced by enabling write_type_log for each log type, instead
1836 of the default zero-based timestamps.
1838 block_error_percentiles=bool If set, record errors in trim block-sized
1839 units from writes and trims and output a histogram of
1840 how many trims it took to get to errors, and what kind
1841 of error was encountered.
1843 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1844 potentially be used instead of removing memory or booting
1845 with less memory to simulate a smaller amount of memory.
1846 The amount specified is per worker.
1848 exec_prerun=str Before running this job, issue the command specified
1849 through system(3). Output is redirected in a file called
1852 exec_postrun=str After the job completes, issue the command specified
1853 though system(3). Output is redirected in a file called
1854 jobname.postrun.txt.
1856 ioscheduler=str Attempt to switch the device hosting the file to the specified
1857 io scheduler before running.
1859 disk_util=bool Generate disk utilization statistics, if the platform
1860 supports it. Defaults to on.
1862 disable_lat=bool Disable measurements of total latency numbers. Useful
1863 only for cutting back the number of calls to gettimeofday,
1864 as that does impact performance at really high IOPS rates.
1865 Note that to really get rid of a large amount of these
1866 calls, this option must be used with disable_slat and
1869 disable_clat=bool Disable measurements of completion latency numbers. See
1872 disable_slat=bool Disable measurements of submission latency numbers. See
1875 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1878 clat_percentiles=bool Enable the reporting of percentiles of
1879 completion latencies.
1881 percentile_list=float_list Overwrite the default list of percentiles
1882 for completion latencies and the block error histogram.
1883 Each number is a floating number in the range (0,100],
1884 and the maximum length of the list is 20. Use ':'
1885 to separate the numbers, and list the numbers in ascending
1886 order. For example, --percentile_list=99.5:99.9 will cause
1887 fio to report the values of completion latency below which
1888 99.5% and 99.9% of the observed latencies fell, respectively.
1890 clocksource=str Use the given clocksource as the base of timing. The
1891 supported options are:
1893 gettimeofday gettimeofday(2)
1895 clock_gettime clock_gettime(2)
1897 cpu Internal CPU clock source
1899 cpu is the preferred clocksource if it is reliable, as it
1900 is very fast (and fio is heavy on time calls). Fio will
1901 automatically use this clocksource if it's supported and
1902 considered reliable on the system it is running on, unless
1903 another clocksource is specifically set. For x86/x86-64 CPUs,
1904 this means supporting TSC Invariant.
1906 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1907 (disable_clat, disable_slat, disable_bw) plus reduce
1908 precision of the timeout somewhat to really shrink
1909 the gettimeofday() call count. With this option enabled,
1910 we only do about 0.4% of the gtod() calls we would have
1911 done if all time keeping was enabled.
1913 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1914 execution to just getting the current time. Fio (and
1915 databases, for instance) are very intensive on gettimeofday()
1916 calls. With this option, you can set one CPU aside for
1917 doing nothing but logging current time to a shared memory
1918 location. Then the other threads/processes that run IO
1919 workloads need only copy that segment, instead of entering
1920 the kernel with a gettimeofday() call. The CPU set aside
1921 for doing these time calls will be excluded from other
1922 uses. Fio will manually clear it from the CPU mask of other
1925 continue_on_error=str Normally fio will exit the job on the first observed
1926 failure. If this option is set, fio will continue the job when
1927 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1928 is exceeded or the I/O size specified is completed. If this
1929 option is used, there are two more stats that are appended,
1930 the total error count and the first error. The error field
1931 given in the stats is the first error that was hit during the
1934 The allowed values are:
1936 none Exit on any IO or verify errors.
1938 read Continue on read errors, exit on all others.
1940 write Continue on write errors, exit on all others.
1942 io Continue on any IO error, exit on all others.
1944 verify Continue on verify errors, exit on all others.
1946 all Continue on all errors.
1948 0 Backward-compatible alias for 'none'.
1950 1 Backward-compatible alias for 'all'.
1952 ignore_error=str Sometimes you want to ignore some errors during test
1953 in that case you can specify error list for each error type.
1954 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1955 errors for given error type is separated with ':'. Error
1956 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1958 ignore_error=EAGAIN,ENOSPC:122
1959 This option will ignore EAGAIN from READ, and ENOSPC and
1960 122(EDQUOT) from WRITE.
1962 error_dump=bool If set dump every error even if it is non fatal, true
1963 by default. If disabled only fatal error will be dumped
1965 cgroup=str Add job to this control group. If it doesn't exist, it will
1966 be created. The system must have a mounted cgroup blkio
1967 mount point for this to work. If your system doesn't have it
1968 mounted, you can do so with:
1970 # mount -t cgroup -o blkio none /cgroup
1972 cgroup_weight=int Set the weight of the cgroup to this value. See
1973 the documentation that comes with the kernel, allowed values
1974 are in the range of 100..1000.
1976 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1977 the job completion. To override this behavior and to leave
1978 cgroups around after the job completion, set cgroup_nodelete=1.
1979 This can be useful if one wants to inspect various cgroup
1980 files after job completion. Default: false
1982 uid=int Instead of running as the invoking user, set the user ID to
1983 this value before the thread/process does any work.
1985 gid=int Set group ID, see uid.
1987 flow_id=int The ID of the flow. If not specified, it defaults to being a
1988 global flow. See flow.
1990 flow=int Weight in token-based flow control. If this value is used, then
1991 there is a 'flow counter' which is used to regulate the
1992 proportion of activity between two or more jobs. fio attempts
1993 to keep this flow counter near zero. The 'flow' parameter
1994 stands for how much should be added or subtracted to the flow
1995 counter on each iteration of the main I/O loop. That is, if
1996 one job has flow=8 and another job has flow=-1, then there
1997 will be a roughly 1:8 ratio in how much one runs vs the other.
1999 flow_watermark=int The maximum value that the absolute value of the flow
2000 counter is allowed to reach before the job must wait for a
2001 lower value of the counter.
2003 flow_sleep=int The period of time, in microseconds, to wait after the flow
2004 watermark has been exceeded before retrying operations
2006 In addition, there are some parameters which are only valid when a specific
2007 ioengine is in use. These are used identically to normal parameters, with the
2008 caveat that when used on the command line, they must come after the ioengine
2009 that defines them is selected.
2011 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
2012 the io_getevents system call to reap newly returned events.
2013 With this flag turned on, the AIO ring will be read directly
2014 from user-space to reap events. The reaping mode is only
2015 enabled when polling for a minimum of 0 events (eg when
2016 iodepth_batch_complete=0).
2018 [psyncv2] hipri Set RWF_HIPRI on IO, indicating to the kernel that
2019 it's of higher priority than normal.
2021 [cpuio] cpuload=int Attempt to use the specified percentage of CPU cycles.
2023 [cpuio] cpuchunks=int Split the load into cycles of the given time. In
2026 [cpuio] exit_on_io_done=bool Detect when IO threads are done, then exit.
2028 [netsplice] hostname=str
2029 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
2030 If the job is a TCP listener or UDP reader, the hostname is not
2031 used and must be omitted unless it is a valid UDP multicast
2033 [libhdfs] namenode=str The host name or IP address of a HDFS cluster namenode to contact.
2035 [netsplice] port=int
2036 [net] port=int The TCP or UDP port to bind to or connect to. If this is used
2037 with numjobs to spawn multiple instances of the same job type, then this will
2038 be the starting port number since fio will use a range of ports.
2039 [libhdfs] port=int the listening port of the HFDS cluster namenode.
2041 [netsplice] interface=str
2042 [net] interface=str The IP address of the network interface used to send or
2043 receive UDP multicast
2046 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
2049 [netsplice] nodelay=bool
2050 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
2052 [netsplice] protocol=str
2053 [netsplice] proto=str
2055 [net] proto=str The network protocol to use. Accepted values are:
2057 tcp Transmission control protocol
2058 tcpv6 Transmission control protocol V6
2059 udp User datagram protocol
2060 udpv6 User datagram protocol V6
2061 unix UNIX domain socket
2063 When the protocol is TCP or UDP, the port must also be given,
2064 as well as the hostname if the job is a TCP listener or UDP
2065 reader. For unix sockets, the normal filename option should be
2066 used and the port is invalid.
2068 [net] listen For TCP network connections, tell fio to listen for incoming
2069 connections rather than initiating an outgoing connection. The
2070 hostname must be omitted if this option is used.
2072 [net] pingpong Normally a network writer will just continue writing data, and
2073 a network reader will just consume packages. If pingpong=1
2074 is set, a writer will send its normal payload to the reader,
2075 then wait for the reader to send the same payload back. This
2076 allows fio to measure network latencies. The submission
2077 and completion latencies then measure local time spent
2078 sending or receiving, and the completion latency measures
2079 how long it took for the other end to receive and send back.
2080 For UDP multicast traffic pingpong=1 should only be set for a
2081 single reader when multiple readers are listening to the same
2084 [net] window_size Set the desired socket buffer size for the connection.
2086 [net] mss Set the TCP maximum segment size (TCP_MAXSEG).
2088 [e4defrag] donorname=str
2089 File will be used as a block donor(swap extents between files)
2090 [e4defrag] inplace=int
2091 Configure donor file blocks allocation strategy
2092 0(default): Preallocate donor's file on init
2093 1 : allocate space immediately inside defragment event,
2094 and free right after event
2096 [rbd] clustername=str Specifies the name of the Ceph cluster.
2097 [rbd] rbdname=str Specifies the name of the RBD.
2098 [rbd] pool=str Specifies the name of the Ceph pool containing RBD.
2099 [rbd] clientname=str Specifies the username (without the 'client.' prefix)
2100 used to access the Ceph cluster. If the clustername is
2101 specified, the clientname shall be the full type.id
2102 string. If no type. prefix is given, fio will add
2103 'client.' by default.
2105 [mtd] skip_bad=bool Skip operations against known bad blocks.
2107 [libhdfs] hdfsdirectory libhdfs will create chunk in this HDFS directory
2108 [libhdfs] chunk_size the size of the chunk to use for each file.
2111 6.0 Interpreting the output
2112 ---------------------------
2114 fio spits out a lot of output. While running, fio will display the
2115 status of the jobs created. An example of that would be:
2117 Jobs: 1: [_r] [24.8% done] [r=20992KiB/s,w=24064KiB/s,t=0KiB/s] [r=82,w=94,t=0 iops] [eta 00h:01m:31s]
2119 The characters inside the square brackets denote the current status of
2120 each thread. The possible values (in typical life cycle order) are:
2124 P Thread setup, but not started.
2126 I Thread initialized, waiting or generating necessary data.
2127 p Thread running pre-reading file(s).
2128 R Running, doing sequential reads.
2129 r Running, doing random reads.
2130 W Running, doing sequential writes.
2131 w Running, doing random writes.
2132 M Running, doing mixed sequential reads/writes.
2133 m Running, doing mixed random reads/writes.
2134 F Running, currently waiting for fsync()
2135 f Running, finishing up (writing IO logs, etc)
2136 V Running, doing verification of written data.
2137 E Thread exited, not reaped by main thread yet.
2139 X Thread reaped, exited with an error.
2140 K Thread reaped, exited due to signal.
2142 Fio will condense the thread string as not to take up more space on the
2143 command line as is needed. For instance, if you have 10 readers and 10
2144 writers running, the output would look like this:
2146 Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [r=20992KiB/s,w=24064KiB/s,t=0KiB/s] [r=82,w=94,t=0 iops] [eta 57m:36s]
2148 Fio will still maintain the ordering, though. So the above means that jobs
2149 1..10 are readers, and 11..20 are writers.
2151 The other values are fairly self explanatory - number of threads
2152 currently running and doing io, rate of io since last check (read speed
2153 listed first, then write speed), and the estimated completion percentage
2154 and time for the running group. It's impossible to estimate runtime of
2155 the following groups (if any). Note that the string is displayed in order,
2156 so it's possible to tell which of the jobs are currently doing what. The
2157 first character is the first job defined in the job file, and so forth.
2159 When fio is done (or interrupted by ctrl-c), it will show the data for
2160 each thread, group of threads, and disks in that order. For each data
2161 direction, the output looks like:
2163 Client1 (g=0): err= 0:
2164 write: io= 32MiB, bw= 666KiB/s, iops=89 , runt= 50320msec
2165 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
2166 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
2167 bw (KiB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
2168 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
2169 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
2170 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2171 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
2172 issued r/w: total=0/32768, short=0/0
2173 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
2174 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
2176 The client number is printed, along with the group id and error of that
2177 thread. Below is the io statistics, here for writes. In the order listed,
2180 io= Number of megabytes io performed
2181 bw= Average bandwidth rate
2182 iops= Average IOs performed per second
2183 runt= The runtime of that thread
2184 slat= Submission latency (avg being the average, stdev being the
2185 standard deviation). This is the time it took to submit
2186 the io. For sync io, the slat is really the completion
2187 latency, since queue/complete is one operation there. This
2188 value can be in milliseconds or microseconds, fio will choose
2189 the most appropriate base and print that. In the example
2190 above, milliseconds is the best scale. Note: in --minimal mode
2191 latencies are always expressed in microseconds.
2192 clat= Completion latency. Same names as slat, this denotes the
2193 time from submission to completion of the io pieces. For
2194 sync io, clat will usually be equal (or very close) to 0,
2195 as the time from submit to complete is basically just
2196 CPU time (io has already been done, see slat explanation).
2197 bw= Bandwidth. Same names as the xlat stats, but also includes
2198 an approximate percentage of total aggregate bandwidth
2199 this thread received in this group. This last value is
2200 only really useful if the threads in this group are on the
2201 same disk, since they are then competing for disk access.
2202 cpu= CPU usage. User and system time, along with the number
2203 of context switches this thread went through, usage of
2204 system and user time, and finally the number of major
2205 and minor page faults. The CPU utilization numbers are
2206 averages for the jobs in that reporting group, while the
2207 context and fault counters are summed.
2208 IO depths= The distribution of io depths over the job life time. The
2209 numbers are divided into powers of 2, so for example the
2210 16= entries includes depths up to that value but higher
2211 than the previous entry. In other words, it covers the
2212 range from 16 to 31.
2213 IO submit= How many pieces of IO were submitting in a single submit
2214 call. Each entry denotes that amount and below, until
2215 the previous entry - eg, 8=100% mean that we submitted
2216 anywhere in between 5-8 ios per submit call.
2217 IO complete= Like the above submit number, but for completions instead.
2218 IO issued= The number of read/write requests issued, and how many
2220 IO latencies= The distribution of IO completion latencies. This is the
2221 time from when IO leaves fio and when it gets completed.
2222 The numbers follow the same pattern as the IO depths,
2223 meaning that 2=1.6% means that 1.6% of the IO completed
2224 within 2 msecs, 20=12.8% means that 12.8% of the IO
2225 took more than 10 msecs, but less than (or equal to) 20 msecs.
2227 After each client has been listed, the group statistics are printed. They
2228 will look like this:
2230 Run status group 0 (all jobs):
2231 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
2232 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
2234 For each data direction, it prints:
2236 io= Number of megabytes io performed.
2237 aggrb= Aggregate bandwidth of threads in this group.
2238 minb= The minimum average bandwidth a thread saw.
2239 maxb= The maximum average bandwidth a thread saw.
2240 mint= The smallest runtime of the threads in that group.
2241 maxt= The longest runtime of the threads in that group.
2243 And finally, the disk statistics are printed. They will look like this:
2245 Disk stats (read/write):
2246 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
2248 Each value is printed for both reads and writes, with reads first. The
2251 ios= Number of ios performed by all groups.
2252 merge= Number of merges io the io scheduler.
2253 ticks= Number of ticks we kept the disk busy.
2254 io_queue= Total time spent in the disk queue.
2255 util= The disk utilization. A value of 100% means we kept the disk
2256 busy constantly, 50% would be a disk idling half of the time.
2258 It is also possible to get fio to dump the current output while it is
2259 running, without terminating the job. To do that, send fio the USR1 signal.
2260 You can also get regularly timed dumps by using the --status-interval
2261 parameter, or by creating a file in /tmp named fio-dump-status. If fio
2262 sees this file, it will unlink it and dump the current output status.
2268 For scripted usage where you typically want to generate tables or graphs
2269 of the results, fio can output the results in a semicolon separated format.
2270 The format is one long line of values, such as:
2272 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%
2273 A description of this job goes here.
2275 The job description (if provided) follows on a second line.
2277 To enable terse output, use the --minimal command line option. The first
2278 value is the version of the terse output format. If the output has to
2279 be changed for some reason, this number will be incremented by 1 to
2280 signify that change.
2282 Split up, the format is as follows:
2284 terse version, fio version, jobname, groupid, error
2286 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
2287 Submission latency: min, max, mean, stdev (usec)
2288 Completion latency: min, max, mean, stdev (usec)
2289 Completion latency percentiles: 20 fields (see below)
2290 Total latency: min, max, mean, stdev (usec)
2291 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev
2293 Total IO (KiB), bandwidth (KiB/sec), IOPS, runtime (msec)
2294 Submission latency: min, max, mean, stdev (usec)
2295 Completion latency: min, max, mean, stdev(usec)
2296 Completion latency percentiles: 20 fields (see below)
2297 Total latency: min, max, mean, stdev (usec)
2298 Bw (KiB/s): min, max, aggregate percentage of total, mean, stdev
2299 CPU usage: user, system, context switches, major faults, minor faults
2300 IO depths: <=1, 2, 4, 8, 16, 32, >=64
2301 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
2302 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
2303 Disk utilization: Disk name, Read ios, write ios,
2304 Read merges, write merges,
2305 Read ticks, write ticks,
2306 Time spent in queue, disk utilization percentage
2307 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
2309 Additional Info (dependent on description being set): Text description
2311 Completion latency percentiles can be a grouping of up to 20 sets, so
2312 for the terse output fio writes all of them. Each field will look like this:
2316 which is the Xth percentile, and the usec latency associated with it.
2318 For disk utilization, all disks used by fio are shown. So for each disk
2319 there will be a disk utilization section.
2322 8.0 Trace file format
2323 ---------------------
2324 There are two trace file format that you can encounter. The older (v1) format
2325 is unsupported since version 1.20-rc3 (March 2008). It will still be described
2326 below in case that you get an old trace and want to understand it.
2328 In any case the trace is a simple text file with a single action per line.
2331 8.1 Trace file format v1
2332 ------------------------
2333 Each line represents a single io action in the following format:
2337 where rw=0/1 for read/write, and the offset and length entries being in bytes.
2339 This format is not supported in Fio versions => 1.20-rc3.
2342 8.2 Trace file format v2
2343 ------------------------
2344 The second version of the trace file format was added in Fio version 1.17.
2345 It allows to access more then one file per trace and has a bigger set of
2346 possible file actions.
2348 The first line of the trace file has to be:
2352 Following this can be lines in two different formats, which are described below.
2354 The file management format:
2358 The filename is given as an absolute path. The action can be one of these:
2360 add Add the given filename to the trace
2361 open Open the file with the given filename. The filename has to have
2362 been added with the add action before.
2363 close Close the file with the given filename. The file has to have been
2367 The file io action format:
2369 filename action offset length
2371 The filename is given as an absolute path, and has to have been added and opened
2372 before it can be used with this format. The offset and length are given in
2373 bytes. The action can be one of these:
2375 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
2376 The time is relative to the previous wait statement.
2377 read Read 'length' bytes beginning from 'offset'
2378 write Write 'length' bytes beginning from 'offset'
2379 sync fsync() the file
2380 datasync fdatasync() the file
2381 trim trim the given file from the given 'offset' for 'length' bytes
2384 9.0 CPU idleness profiling
2385 --------------------------
2386 In some cases, we want to understand CPU overhead in a test. For example,
2387 we test patches for the specific goodness of whether they reduce CPU usage.
2388 fio implements a balloon approach to create a thread per CPU that runs at
2389 idle priority, meaning that it only runs when nobody else needs the cpu.
2390 By measuring the amount of work completed by the thread, idleness of each
2391 CPU can be derived accordingly.
2393 An unit work is defined as touching a full page of unsigned characters. Mean
2394 and standard deviation of time to complete an unit work is reported in "unit
2395 work" section. Options can be chosen to report detailed percpu idleness or
2396 overall system idleness by aggregating percpu stats.
2399 10.0 Verification and triggers
2400 ------------------------------
2401 Fio is usually run in one of two ways, when data verification is done. The
2402 first is a normal write job of some sort with verify enabled. When the
2403 write phase has completed, fio switches to reads and verifies everything
2404 it wrote. The second model is running just the write phase, and then later
2405 on running the same job (but with reads instead of writes) to repeat the
2406 same IO patterns and verify the contents. Both of these methods depend
2407 on the write phase being completed, as fio otherwise has no idea how much
2410 With verification triggers, fio supports dumping the current write state
2411 to local files. Then a subsequent read verify workload can load this state
2412 and know exactly where to stop. This is useful for testing cases where
2413 power is cut to a server in a managed fashion, for instance.
2415 A verification trigger consists of two things:
2417 1) Storing the write state of each job
2418 2) Executing a trigger command
2420 The write state is relatively small, on the order of hundreds of bytes
2421 to single kilobytes. It contains information on the number of completions
2422 done, the last X completions, etc.
2424 A trigger is invoked either through creation ('touch') of a specified
2425 file in the system, or through a timeout setting. If fio is run with
2426 --trigger-file=/tmp/trigger-file, then it will continually check for
2427 the existence of /tmp/trigger-file. When it sees this file, it will
2428 fire off the trigger (thus saving state, and executing the trigger
2431 For client/server runs, there's both a local and remote trigger. If
2432 fio is running as a server backend, it will send the job states back
2433 to the client for safe storage, then execute the remote trigger, if
2434 specified. If a local trigger is specified, the server will still send
2435 back the write state, but the client will then execute the trigger.
2437 10.1 Verification trigger example
2438 ---------------------------------
2439 Lets say we want to run a powercut test on the remote machine 'server'.
2440 Our write workload is in write-test.fio. We want to cut power to 'server'
2441 at some point during the run, and we'll run this test from the safety
2442 or our local machine, 'localbox'. On the server, we'll start the fio
2445 server# fio --server
2447 and on the client, we'll fire off the workload:
2449 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
2451 We set /tmp/my-trigger as the trigger file, and we tell fio to execute
2453 echo b > /proc/sysrq-trigger
2455 on the server once it has received the trigger and sent us the write
2456 state. This will work, but it's not _really_ cutting power to the server,
2457 it's merely abruptly rebooting it. If we have a remote way of cutting
2458 power to the server through IPMI or similar, we could do that through
2459 a local trigger command instead. Lets assume we have a script that does
2460 IPMI reboot of a given hostname, ipmi-reboot. On localbox, we could
2461 then have run fio with a local trigger instead:
2463 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
2465 For this case, fio would wait for the server to send us the write state,
2466 then execute 'ipmi-reboot server' when that happened.
2468 10.2 Loading verify state
2469 -------------------------
2470 To load store write state, read verification job file must contain
2471 the verify_state_load option. If that is set, fio will load the previously
2472 stored state. For a local fio run this is done by loading the files directly,
2473 and on a client/server run, the server backend will ask the client to send
2474 the files over and load them from there.
2477 11.0 Log File Formats
2478 ---------------------
2480 Fio supports a variety of log file formats, for logging latencies, bandwidth,
2481 and IOPS. The logs share a common format, which looks like this:
2483 time (msec), value, data direction, offset
2485 Time for the log entry is always in milliseconds. The value logged depends
2486 on the type of log, it will be one of the following:
2488 Latency log Value is latency in usecs
2489 Bandwidth log Value is in KiB/sec
2490 IOPS log Value is IOPS
2492 Data direction is one of the following:
2498 The offset is the offset, in bytes, from the start of the file, for that
2499 particular IO. The logging of the offset can be toggled with 'log_offset'.
2501 If windowed logging is enabled through 'log_avg_msec', then fio doesn't log
2502 individual IOs. Instead of logs the average values over the specified
2503 period of time. Since 'data direction' and 'offset' are per-IO values,
2504 they aren't applicable if windowed logging is enabled. If windowed logging
2505 is enabled and 'log_max_value' is set, then fio logs maximum values in
2506 that window instead of averages.