8 5. Detailed list of parameters
12 9. CPU idleness profiling
14 1.0 Overview and history
15 ------------------------
16 fio was originally written to save me the hassle of writing special test
17 case programs when I wanted to test a specific workload, either for
18 performance reasons or to find/reproduce a bug. The process of writing
19 such a test app can be tiresome, especially if you have to do it often.
20 Hence I needed a tool that would be able to simulate a given io workload
21 without resorting to writing a tailored test case again and again.
23 A test work load is difficult to define, though. There can be any number
24 of processes or threads involved, and they can each be using their own
25 way of generating io. You could have someone dirtying large amounts of
26 memory in an memory mapped file, or maybe several threads issuing
27 reads using asynchronous io. fio needed to be flexible enough to
28 simulate both of these cases, and many more.
32 The first step in getting fio to simulate a desired io workload, is
33 writing a job file describing that specific setup. A job file may contain
34 any number of threads and/or files - the typical contents of the job file
35 is a global section defining shared parameters, and one or more job
36 sections describing the jobs involved. When run, fio parses this file
37 and sets everything up as described. If we break down a job from top to
38 bottom, it contains the following basic parameters:
40 IO type Defines the io pattern issued to the file(s).
41 We may only be reading sequentially from this
42 file(s), or we may be writing randomly. Or even
43 mixing reads and writes, sequentially or randomly.
45 Block size In how large chunks are we issuing io? This may be
46 a single value, or it may describe a range of
49 IO size How much data are we going to be reading/writing.
51 IO engine How do we issue io? We could be memory mapping the
52 file, we could be using regular read/write, we
53 could be using splice, async io, syslet, or even
56 IO depth If the io engine is async, how large a queuing
57 depth do we want to maintain?
59 IO type Should we be doing buffered io, or direct/raw io?
61 Num files How many files are we spreading the workload over.
63 Num threads How many threads or processes should we spread
66 The above are the basic parameters defined for a workload, in addition
67 there's a multitude of parameters that modify other aspects of how this
73 See the README file for command line parameters, there are only a few
76 Running fio is normally the easiest part - you just give it the job file
77 (or job files) as parameters:
81 and it will start doing what the job_file tells it to do. You can give
82 more than one job file on the command line, fio will serialize the running
83 of those files. Internally that is the same as using the 'stonewall'
84 parameter described the the parameter section.
86 If the job file contains only one job, you may as well just give the
87 parameters on the command line. The command line parameters are identical
88 to the job parameters, with a few extra that control global parameters
89 (see README). For example, for the job file parameter iodepth=2, the
90 mirror command line option would be --iodepth 2 or --iodepth=2. You can
91 also use the command line for giving more than one job entry. For each
92 --name option that fio sees, it will start a new job with that name.
93 Command line entries following a --name entry will apply to that job,
94 until there are no more entries or a new --name entry is seen. This is
95 similar to the job file options, where each option applies to the current
96 job until a new [] job entry is seen.
98 fio does not need to run as root, except if the files or devices specified
99 in the job section requires that. Some other options may also be restricted,
100 such as memory locking, io scheduler switching, and decreasing the nice value.
105 As previously described, fio accepts one or more job files describing
106 what it is supposed to do. The job file format is the classic ini file,
107 where the names enclosed in [] brackets define the job name. You are free
108 to use any ascii name you want, except 'global' which has special meaning.
109 A global section sets defaults for the jobs described in that file. A job
110 may override a global section parameter, and a job file may even have
111 several global sections if so desired. A job is only affected by a global
112 section residing above it. If the first character in a line is a ';' or a
113 '#', the entire line is discarded as a comment.
115 So let's look at a really simple job file that defines two processes, each
116 randomly reading from a 128MB file.
118 ; -- start job file --
129 As you can see, the job file sections themselves are empty as all the
130 described parameters are shared. As no filename= option is given, fio
131 makes up a filename for each of the jobs as it sees fit. On the command
132 line, this job would look as follows:
134 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
137 Let's look at an example that has a number of processes writing randomly
140 ; -- start job file --
152 Here we have no global section, as we only have one job defined anyway.
153 We want to use async io here, with a depth of 4 for each file. We also
154 increased the buffer size used to 32KB and define numjobs to 4 to
155 fork 4 identical jobs. The result is 4 processes each randomly writing
156 to their own 64MB file. Instead of using the above job file, you could
157 have given the parameters on the command line. For this case, you would
160 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
162 4.1 Environment variables
163 -------------------------
165 fio also supports environment variable expansion in job files. Any
166 substring of the form "${VARNAME}" as part of an option value (in other
167 words, on the right of the `='), will be expanded to the value of the
168 environment variable called VARNAME. If no such environment variable
169 is defined, or VARNAME is the empty string, the empty string will be
172 As an example, let's look at a sample fio invocation and job file:
174 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
176 ; -- start job file --
183 This will expand to the following equivalent job file at runtime:
185 ; -- start job file --
192 fio ships with a few example job files, you can also look there for
195 4.2 Reserved keywords
196 ---------------------
198 Additionally, fio has a set of reserved keywords that will be replaced
199 internally with the appropriate value. Those keywords are:
201 $pagesize The architecture page size of the running system
202 $mb_memory Megabytes of total memory in the system
203 $ncpus Number of online available CPUs
205 These can be used on the command line or in the job file, and will be
206 automatically substituted with the current system values when the job
207 is run. Simple math is also supported on these keywords, so you can
208 perform actions like:
212 and get that properly expanded to 8 times the size of memory in the
216 5.0 Detailed list of parameters
217 -------------------------------
219 This section describes in details each parameter associated with a job.
220 Some parameters take an option of a given type, such as an integer or
221 a string. The following types are used:
223 str String. This is a sequence of alpha characters.
224 time Integer with possible time suffix. In seconds unless otherwise
225 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
226 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
227 and 'us' (or 'usec') for microseconds.
228 int SI integer. A whole number value, which may contain a suffix
229 describing the base of the number. Accepted suffixes are k/m/g/t/p,
230 meaning kilo, mega, giga, tera, and peta. The suffix is not case
231 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
232 as 'k'). So if you want to specify 4096, you could either write
233 out '4096' or just give 4k. The suffixes signify base 2 values, so
234 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
235 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
236 case, then 1000 is used as the multiplier. This can be handy for
237 disks, since manufacturers generally use base 10 values when listing
238 the capacity of a drive. If the option accepts an upper and lower
239 range, use a colon ':' or minus '-' to separate such values. May also
240 include a prefix to indicate numbers base. If 0x is used, the number
241 is assumed to be hexadecimal. See irange.
242 bool Boolean. Usually parsed as an integer, however only defined for
243 true and false (1 and 0).
244 irange Integer range with suffix. Allows value range to be given, such
245 as 1024-4096. A colon may also be used as the separator, eg
246 1k:4k. If the option allows two sets of ranges, they can be
247 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
249 float_list A list of floating numbers, separated by a ':' character.
251 With the above in mind, here follows the complete list of fio job
254 name=str ASCII name of the job. This may be used to override the
255 name printed by fio for this job. Otherwise the job
256 name is used. On the command line this parameter has the
257 special purpose of also signaling the start of a new
260 description=str Text description of the job. Doesn't do anything except
261 dump this text description when this job is run. It's
264 directory=str Prefix filenames with this directory. Used to place files
265 in a different location than "./". See the 'filename' option
266 for escaping certain characters.
268 filename=str Fio normally makes up a filename based on the job name,
269 thread number, and file number. If you want to share
270 files between threads in a job or several jobs, specify
271 a filename for each of them to override the default. If
272 the ioengine used is 'net', the filename is the host, port,
273 and protocol to use in the format of =host,port,protocol.
274 See ioengine=net for more. If the ioengine is file based, you
275 can specify a number of files by separating the names with a
276 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
277 as the two working files, you would use
278 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
279 accessed as \\.\PhysicalDrive0 for the first device,
280 \\.\PhysicalDrive1 for the second etc. Note: Windows and
281 FreeBSD prevent write access to areas of the disk containing
282 in-use data (e.g. filesystems).
283 If the wanted filename does need to include a colon, then
284 escape that with a '\' character. For instance, if the filename
285 is "/dev/dsk/foo@3,0:c", then you would use
286 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
287 stdin or stdout. Which of the two depends on the read/write
291 If sharing multiple files between jobs, it is usually necessary
292 to have fio generate the exact names that you want. By default,
293 fio will name a file based on the default file format
294 specification of jobname.jobnumber.filenumber. With this
295 option, that can be customized. Fio will recognize and replace
296 the following keywords in this string:
299 The name of the worker thread or process.
302 The incremental number of the worker thread or
306 The incremental number of the file for that worker
309 To have dependent jobs share a set of files, this option can
310 be set to have fio generate filenames that are shared between
311 the two. For instance, if testfiles.$filenum is specified,
312 file number 4 for any job will be named testfiles.4. The
313 default of $jobname.$jobnum.$filenum will be used if
314 no other format specifier is given.
316 opendir=str Tell fio to recursively add any file it can find in this
317 directory and down the file system tree.
319 lockfile=str Fio defaults to not locking any files before it does
320 IO to them. If a file or file descriptor is shared, fio
321 can serialize IO to that file to make the end result
322 consistent. This is usual for emulating real workloads that
323 share files. The lock modes are:
325 none No locking. The default.
326 exclusive Only one thread/process may do IO,
327 excluding all others.
328 readwrite Read-write locking on the file. Many
329 readers may access the file at the
330 same time, but writes get exclusive
334 rw=str Type of io pattern. Accepted values are:
336 read Sequential reads
337 write Sequential writes
338 randwrite Random writes
339 randread Random reads
340 rw,readwrite Sequential mixed reads and writes
341 randrw Random mixed reads and writes
343 For the mixed io types, the default is to split them 50/50.
344 For certain types of io the result may still be skewed a bit,
345 since the speed may be different. It is possible to specify
346 a number of IO's to do before getting a new offset, this is
347 one by appending a ':<nr>' to the end of the string given.
348 For a random read, it would look like 'rw=randread:8' for
349 passing in an offset modifier with a value of 8. If the
350 suffix is used with a sequential IO pattern, then the value
351 specified will be added to the generated offset for each IO.
352 For instance, using rw=write:4k will skip 4k for every
353 write. It turns sequential IO into sequential IO with holes.
354 See the 'rw_sequencer' option.
356 rw_sequencer=str If an offset modifier is given by appending a number to
357 the rw=<str> line, then this option controls how that
358 number modifies the IO offset being generated. Accepted
361 sequential Generate sequential offset
362 identical Generate the same offset
364 'sequential' is only useful for random IO, where fio would
365 normally generate a new random offset for every IO. If you
366 append eg 8 to randread, you would get a new random offset for
367 every 8 IO's. The result would be a seek for only every 8
368 IO's, instead of for every IO. Use rw=randread:8 to specify
369 that. As sequential IO is already sequential, setting
370 'sequential' for that would not result in any differences.
371 'identical' behaves in a similar fashion, except it sends
372 the same offset 8 number of times before generating a new
375 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
376 Storage manufacturers like to use 10^3 or 1000 as a base
377 ten unit instead, for obvious reasons. Allow values are
378 1024 or 1000, with 1024 being the default.
380 unified_rw_reporting=bool Fio normally reports statistics on a per
381 data direction basis, meaning that read, write, and trim are
382 accounted and reported separately. If this option is set,
383 the fio will sum the results and report them as "mixed"
386 randrepeat=bool For random IO workloads, seed the generator in a predictable
387 way so that results are repeatable across repetitions.
389 randseed=int Seed the random number generators based on this seed value, to
390 be able to control what sequence of output is being generated.
391 If not set, the random sequence depends on the randrepeat
394 use_os_rand=bool Fio can either use the random generator supplied by the OS
395 to generator random offsets, or it can use it's own internal
396 generator (based on Tausworthe). Default is to use the
397 internal generator, which is often of better quality and
400 fallocate=str Whether pre-allocation is performed when laying down files.
403 none Do not pre-allocate space
404 posix Pre-allocate via posix_fallocate()
405 keep Pre-allocate via fallocate() with
406 FALLOC_FL_KEEP_SIZE set
407 0 Backward-compatible alias for 'none'
408 1 Backward-compatible alias for 'posix'
410 May not be available on all supported platforms. 'keep' is only
411 available on Linux.If using ZFS on Solaris this must be set to
412 'none' because ZFS doesn't support it. Default: 'posix'.
414 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
415 on what IO patterns it is likely to issue. Sometimes you
416 want to test specific IO patterns without telling the
417 kernel about it, in which case you can disable this option.
418 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
419 IO and POSIX_FADV_RANDOM for random IO.
421 size=int The total size of file io for this job. Fio will run until
422 this many bytes has been transferred, unless runtime is
423 limited by other options (such as 'runtime', for instance).
424 Unless specific nrfiles and filesize options are given,
425 fio will divide this size between the available files
426 specified by the job. If not set, fio will use the full
427 size of the given files or devices. If the the files
428 do not exist, size must be given. It is also possible to
429 give size as a percentage between 1 and 100. If size=20%
430 is given, fio will use 20% of the full size of the given
433 filesize=int Individual file sizes. May be a range, in which case fio
434 will select sizes for files at random within the given range
435 and limited to 'size' in total (if that is given). If not
436 given, each created file is the same size.
439 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
440 space left on device) as the terminating condition. Only makes
441 sense with sequential write. For a read workload, the mount
442 point will be filled first then IO started on the result. This
443 option doesn't make sense if operating on a raw device node,
444 since the size of that is already known by the file system.
445 Additionally, writing beyond end-of-device will not return
449 bs=int The block size used for the io units. Defaults to 4k. Values
450 can be given for both read and writes. If a single int is
451 given, it will apply to both. If a second int is specified
452 after a comma, it will apply to writes only. In other words,
453 the format is either bs=read_and_write or bs=read,write,trim.
454 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
455 writes, and 8k for trims. You can terminate the list with
456 a trailing comma. bs=4k,8k, would use the default value for
457 trims.. If you only wish to set the write size, you
458 can do so by passing an empty read size - bs=,8k will set
459 8k for writes and leave the read default value.
462 ba=int At what boundary to align random IO offsets. Defaults to
463 the same as 'blocksize' the minimum blocksize given.
464 Minimum alignment is typically 512b for using direct IO,
465 though it usually depends on the hardware block size. This
466 option is mutually exclusive with using a random map for
467 files, so it will turn off that option.
469 blocksize_range=irange
470 bsrange=irange Instead of giving a single block size, specify a range
471 and fio will mix the issued io block sizes. The issued
472 io unit will always be a multiple of the minimum value
473 given (also see bs_unaligned). Applies to both reads and
474 writes, however a second range can be given after a comma.
477 bssplit=str Sometimes you want even finer grained control of the
478 block sizes issued, not just an even split between them.
479 This option allows you to weight various block sizes,
480 so that you are able to define a specific amount of
481 block sizes issued. The format for this option is:
483 bssplit=blocksize/percentage:blocksize/percentage
485 for as many block sizes as needed. So if you want to define
486 a workload that has 50% 64k blocks, 10% 4k blocks, and
487 40% 32k blocks, you would write:
489 bssplit=4k/10:64k/50:32k/40
491 Ordering does not matter. If the percentage is left blank,
492 fio will fill in the remaining values evenly. So a bssplit
493 option like this one:
495 bssplit=4k/50:1k/:32k/
497 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
498 always add up to 100, if bssplit is given a range that adds
499 up to more, it will error out.
501 bssplit also supports giving separate splits to reads and
502 writes. The format is identical to what bs= accepts. You
503 have to separate the read and write parts with a comma. So
504 if you want a workload that has 50% 2k reads and 50% 4k reads,
505 while having 90% 4k writes and 10% 8k writes, you would
508 bssplit=2k/50:4k/50,4k/90,8k/10
511 bs_unaligned If this option is given, any byte size value within bsrange
512 may be used as a block range. This typically wont work with
513 direct IO, as that normally requires sector alignment.
515 bs_is_seq_rand If this option is set, fio will use the normal read,write
516 blocksize settings as sequential,random instead. Any random
517 read or write will use the WRITE blocksize settings, and any
518 sequential read or write will use the READ blocksize setting.
520 zero_buffers If this option is given, fio will init the IO buffers to
521 all zeroes. The default is to fill them with random data.
522 The resulting IO buffers will not be completely zeroed,
523 unless scramble_buffers is also turned off.
525 refill_buffers If this option is given, fio will refill the IO buffers
526 on every submit. The default is to only fill it at init
527 time and reuse that data. Only makes sense if zero_buffers
528 isn't specified, naturally. If data verification is enabled,
529 refill_buffers is also automatically enabled.
531 scramble_buffers=bool If refill_buffers is too costly and the target is
532 using data deduplication, then setting this option will
533 slightly modify the IO buffer contents to defeat normal
534 de-dupe attempts. This is not enough to defeat more clever
535 block compression attempts, but it will stop naive dedupe of
536 blocks. Default: true.
538 buffer_compress_percentage=int If this is set, then fio will attempt to
539 provide IO buffer content (on WRITEs) that compress to
540 the specified level. Fio does this by providing a mix of
541 random data and zeroes. Note that this is per block size
542 unit, for file/disk wide compression level that matches
543 this setting, you'll also want to set refill_buffers.
545 buffer_compress_chunk=int See buffer_compress_percentage. This
546 setting allows fio to manage how big the ranges of random
547 data and zeroed data is. Without this set, fio will
548 provide buffer_compress_percentage of blocksize random
549 data, followed by the remaining zeroed. With this set
550 to some chunk size smaller than the block size, fio can
551 alternate random and zeroed data throughout the IO
554 buffer_pattern=str If set, fio will fill the io buffers with this pattern.
555 If not set, the contents of io buffers is defined by the other
556 options related to buffer contents. The setting can be any
557 pattern of bytes, and can be prefixed with 0x for hex values.
559 nrfiles=int Number of files to use for this job. Defaults to 1.
561 openfiles=int Number of files to keep open at the same time. Defaults to
562 the same as nrfiles, can be set smaller to limit the number
565 file_service_type=str Defines how fio decides which file from a job to
566 service next. The following types are defined:
568 random Just choose a file at random.
570 roundrobin Round robin over open files. This
573 sequential Finish one file before moving on to
574 the next. Multiple files can still be
575 open depending on 'openfiles'.
577 The string can have a number appended, indicating how
578 often to switch to a new file. So if option random:4 is
579 given, fio will switch to a new random file after 4 ios
582 ioengine=str Defines how the job issues io to the file. The following
585 sync Basic read(2) or write(2) io. lseek(2) is
586 used to position the io location.
588 psync Basic pread(2) or pwrite(2) io.
590 vsync Basic readv(2) or writev(2) IO.
592 psyncv Basic preadv(2) or pwritev(2) IO.
594 libaio Linux native asynchronous io. Note that Linux
595 may only support queued behaviour with
596 non-buffered IO (set direct=1 or buffered=0).
597 This engine defines engine specific options.
599 posixaio glibc posix asynchronous io.
601 solarisaio Solaris native asynchronous io.
603 windowsaio Windows native asynchronous io.
605 mmap File is memory mapped and data copied
606 to/from using memcpy(3).
608 splice splice(2) is used to transfer the data and
609 vmsplice(2) to transfer data from user
612 syslet-rw Use the syslet system calls to make
613 regular read/write async.
615 sg SCSI generic sg v3 io. May either be
616 synchronous using the SG_IO ioctl, or if
617 the target is an sg character device
618 we use read(2) and write(2) for asynchronous
621 null Doesn't transfer any data, just pretends
622 to. This is mainly used to exercise fio
623 itself and for debugging/testing purposes.
625 net Transfer over the network to given host:port.
626 Depending on the protocol used, the hostname,
627 port, listen and filename options are used to
628 specify what sort of connection to make, while
629 the protocol option determines which protocol
631 This engine defines engine specific options.
633 netsplice Like net, but uses splice/vmsplice to
634 map data and send/receive.
635 This engine defines engine specific options.
637 cpuio Doesn't transfer any data, but burns CPU
638 cycles according to the cpuload= and
639 cpucycle= options. Setting cpuload=85
640 will cause that job to do nothing but burn
641 85% of the CPU. In case of SMP machines,
642 use numjobs=<no_of_cpu> to get desired CPU
643 usage, as the cpuload only loads a single
644 CPU at the desired rate.
646 guasi The GUASI IO engine is the Generic Userspace
647 Asyncronous Syscall Interface approach
650 http://www.xmailserver.org/guasi-lib.html
652 for more info on GUASI.
654 rdma The RDMA I/O engine supports both RDMA
655 memory semantics (RDMA_WRITE/RDMA_READ) and
656 channel semantics (Send/Recv) for the
657 InfiniBand, RoCE and iWARP protocols.
659 falloc IO engine that does regular fallocate to
660 simulate data transfer as fio ioengine.
661 DDIR_READ does fallocate(,mode = keep_size,)
662 DDIR_WRITE does fallocate(,mode = 0)
663 DDIR_TRIM does fallocate(,mode = punch_hole)
665 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
666 ioctls to simulate defragment activity in
667 request to DDIR_WRITE event
669 external Prefix to specify loading an external
670 IO engine object file. Append the engine
671 filename, eg ioengine=external:/tmp/foo.o
672 to load ioengine foo.o in /tmp.
674 iodepth=int This defines how many io units to keep in flight against
675 the file. The default is 1 for each file defined in this
676 job, can be overridden with a larger value for higher
677 concurrency. Note that increasing iodepth beyond 1 will not
678 affect synchronous ioengines (except for small degress when
679 verify_async is in use). Even async engines may impose OS
680 restrictions causing the desired depth not to be achieved.
681 This may happen on Linux when using libaio and not setting
682 direct=1, since buffered IO is not async on that OS. Keep an
683 eye on the IO depth distribution in the fio output to verify
684 that the achieved depth is as expected. Default: 1.
686 iodepth_batch_submit=int
687 iodepth_batch=int This defines how many pieces of IO to submit at once.
688 It defaults to 1 which means that we submit each IO
689 as soon as it is available, but can be raised to submit
690 bigger batches of IO at the time.
692 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
693 at once. It defaults to 1 which means that we'll ask
694 for a minimum of 1 IO in the retrieval process from
695 the kernel. The IO retrieval will go on until we
696 hit the limit set by iodepth_low. If this variable is
697 set to 0, then fio will always check for completed
698 events before queuing more IO. This helps reduce
699 IO latency, at the cost of more retrieval system calls.
701 iodepth_low=int The low water mark indicating when to start filling
702 the queue again. Defaults to the same as iodepth, meaning
703 that fio will attempt to keep the queue full at all times.
704 If iodepth is set to eg 16 and iodepth_low is set to 4, then
705 after fio has filled the queue of 16 requests, it will let
706 the depth drain down to 4 before starting to fill it again.
708 direct=bool If value is true, use non-buffered io. This is usually
709 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
710 On Windows the synchronous ioengines don't support direct io.
712 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
713 writes are guaranteed to be stable once acknowledged by
714 the operating system. Only Linux supports O_ATOMIC right
717 buffered=bool If value is true, use buffered io. This is the opposite
718 of the 'direct' option. Defaults to true.
720 offset=int Start io at the given offset in the file. The data before
721 the given offset will not be touched. This effectively
722 caps the file size at real_size - offset.
724 offset_increment=int If this is provided, then the real offset becomes
725 the offset + offset_increment * thread_number, where the
726 thread number is a counter that starts at 0 and is incremented
727 for each job. This option is useful if there are several jobs
728 which are intended to operate on a file in parallel in disjoint
729 segments, with even spacing between the starting points.
731 number_ios=int Fio will normally perform IOs until it has exhausted the size
732 of the region set by size=, or if it exhaust the allocated
733 time (or hits an error condition). With this setting, the
734 range/size can be set independently of the number of IOs to
735 perform. When fio reaches this number, it will exit normally
738 fsync=int If writing to a file, issue a sync of the dirty data
739 for every number of blocks given. For example, if you give
740 32 as a parameter, fio will sync the file for every 32
741 writes issued. If fio is using non-buffered io, we may
742 not sync the file. The exception is the sg io engine, which
743 synchronizes the disk cache anyway.
745 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
747 In FreeBSD and Windows there is no fdatasync(), this falls back to
750 sync_file_range=str:val Use sync_file_range() for every 'val' number of
751 write operations. Fio will track range of writes that
752 have happened since the last sync_file_range() call. 'str'
753 can currently be one or more of:
755 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
756 write SYNC_FILE_RANGE_WRITE
757 wait_after SYNC_FILE_RANGE_WAIT_AFTER
759 So if you do sync_file_range=wait_before,write:8, fio would
760 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
761 every 8 writes. Also see the sync_file_range(2) man page.
762 This option is Linux specific.
764 overwrite=bool If true, writes to a file will always overwrite existing
765 data. If the file doesn't already exist, it will be
766 created before the write phase begins. If the file exists
767 and is large enough for the specified write phase, nothing
770 end_fsync=bool If true, fsync file contents when a write stage has completed.
772 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
773 This differs from end_fsync in that it will happen on every
774 file close, not just at the end of the job.
776 rwmixread=int How large a percentage of the mix should be reads.
778 rwmixwrite=int How large a percentage of the mix should be writes. If both
779 rwmixread and rwmixwrite is given and the values do not add
780 up to 100%, the latter of the two will be used to override
781 the first. This may interfere with a given rate setting,
782 if fio is asked to limit reads or writes to a certain rate.
783 If that is the case, then the distribution may be skewed.
785 random_distribution=str:float By default, fio will use a completely uniform
786 random distribution when asked to perform random IO. Sometimes
787 it is useful to skew the distribution in specific ways,
788 ensuring that some parts of the data is more hot than others.
789 fio includes the following distribution models:
791 random Uniform random distribution
792 zipf Zipf distribution
793 pareto Pareto distribution
795 When using a zipf or pareto distribution, an input value
796 is also needed to define the access pattern. For zipf, this
797 is the zipf theta. For pareto, it's the pareto power. Fio
798 includes a test program, genzipf, that can be used visualize
799 what the given input values will yield in terms of hit rates.
800 If you wanted to use zipf with a theta of 1.2, you would use
801 random_distribution=zipf:1.2 as the option. If a non-uniform
802 model is used, fio will disable use of the random map.
804 percentage_random=int For a random workload, set how big a percentage should
805 be random. This defaults to 100%, in which case the workload
806 is fully random. It can be set from anywhere from 0 to 100.
807 Setting it to 0 would make the workload fully sequential. Any
808 setting in between will result in a random mix of sequential
809 and random IO, at the given percentages. It is possible to
810 set different values for reads, writes, and trim. To do so,
811 simply use a comma separated list. See blocksize.
813 norandommap Normally fio will cover every block of the file when doing
814 random IO. If this option is given, fio will just get a
815 new random offset without looking at past io history. This
816 means that some blocks may not be read or written, and that
817 some blocks may be read/written more than once. This option
818 is mutually exclusive with verify= if and only if multiple
819 blocksizes (via bsrange=) are used, since fio only tracks
820 complete rewrites of blocks.
822 softrandommap=bool See norandommap. If fio runs with the random block map
823 enabled and it fails to allocate the map, if this option is
824 set it will continue without a random block map. As coverage
825 will not be as complete as with random maps, this option is
828 random_generator=str Fio supports the following engines for generating
829 IO offsets for random IO:
831 tausworthe Strong 2^88 cycle random number generator
832 lfsr Linear feedback shift register generator
834 Tausworthe is a strong random number generator, but it
835 requires tracking on the side if we want to ensure that
836 blocks are only read or written once. LFSR guarantees
837 that we never generate the same offset twice, and it's
838 also less computationally expensive. It's not a true
839 random generator, however, though for IO purposes it's
840 typically good enough. LFSR only works with single
841 block sizes, not with workloads that use multiple block
842 sizes. If used with such a workload, fio may read or write
843 some blocks multiple times.
845 nice=int Run the job with the given nice value. See man nice(2).
847 prio=int Set the io priority value of this job. Linux limits us to
848 a positive value between 0 and 7, with 0 being the highest.
851 prioclass=int Set the io priority class. See man ionice(1).
853 thinktime=int Stall the job x microseconds after an io has completed before
854 issuing the next. May be used to simulate processing being
855 done by an application. See thinktime_blocks and
859 Only valid if thinktime is set - pretend to spend CPU time
860 doing something with the data received, before falling back
861 to sleeping for the rest of the period specified by
865 Only valid if thinktime is set - control how many blocks
866 to issue, before waiting 'thinktime' usecs. If not set,
867 defaults to 1 which will make fio wait 'thinktime' usecs
868 after every block. This effectively makes any queue depth
869 setting redundant, since no more than 1 IO will be queued
870 before we have to complete it and do our thinktime. In
871 other words, this setting effectively caps the queue depth
872 if the latter is larger.
874 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
875 the normal suffix rules apply. You can use rate=500k to limit
876 reads and writes to 500k each, or you can specify read and
877 writes separately. Using rate=1m,500k would limit reads to
878 1MB/sec and writes to 500KB/sec. Capping only reads or
879 writes can be done with rate=,500k or rate=500k,. The former
880 will only limit writes (to 500KB/sec), the latter will only
883 ratemin=int Tell fio to do whatever it can to maintain at least this
884 bandwidth. Failing to meet this requirement, will cause
885 the job to exit. The same format as rate is used for
886 read vs write separation.
888 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
889 as rate, just specified independently of bandwidth. If the
890 job is given a block size range instead of a fixed value,
891 the smallest block size is used as the metric. The same format
892 as rate is used for read vs write separation.
894 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
895 the job to exit. The same format as rate is used for read vs
898 latency_target=int If set, fio will attempt to find the max performance
899 point that the given workload will run at while maintaining a
900 latency below this target. The values is given in microseconds.
901 See latency_window and latency_percentile
903 latency_window=int Used with latency_target to specify the sample window
904 that the job is run at varying queue depths to test the
905 performance. The value is given in microseconds.
907 latency_percentile=float The percentage of IOs that must fall within the
908 criteria specified by latency_target and latency_window. If not
909 set, this defaults to 100.0, meaning that all IOs must be equal
910 or below to the value set by latency_target.
912 max_latency=int If set, fio will exit the job if it exceeds this maximum
913 latency. It will exit with an ETIME error.
915 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
918 cpumask=int Set the CPU affinity of this job. The parameter given is a
919 bitmask of allowed CPU's the job may run on. So if you want
920 the allowed CPUs to be 1 and 5, you would pass the decimal
921 value of (1 << 1 | 1 << 5), or 34. See man
922 sched_setaffinity(2). This may not work on all supported
923 operating systems or kernel versions. This option doesn't
924 work well for a higher CPU count than what you can store in
925 an integer mask, so it can only control cpus 1-32. For
926 boxes with larger CPU counts, use cpus_allowed.
928 cpus_allowed=str Controls the same options as cpumask, but it allows a text
929 setting of the permitted CPUs instead. So to use CPUs 1 and
930 5, you would specify cpus_allowed=1,5. This options also
931 allows a range of CPUs. Say you wanted a binding to CPUs
932 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
934 cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
935 specified by cpus_allowed or cpumask. Two policies are
938 shared All jobs will share the CPU set specified.
939 split Each job will get a unique CPU from the CPU set.
941 'shared' is the default behaviour, if the option isn't
942 specified. If split is specified, then fio will will assign
943 one cpu per job. If not enough CPUs are given for the jobs
944 listed, then fio will roundrobin the CPUs in the set.
946 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
947 arguments allow comma delimited list of cpu numbers,
948 A-B ranges, or 'all'. Note, to enable numa options support,
949 fio must be built on a system with libnuma-dev(el) installed.
951 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
952 nodes. Format of the argements:
954 `mode' is one of the following memory policy:
955 default, prefer, bind, interleave, local
956 For `default' and `local' memory policy, no node is
957 needed to be specified.
958 For `prefer', only one node is allowed.
959 For `bind' and `interleave', it allow comma delimited
960 list of numbers, A-B ranges, or 'all'.
962 startdelay=time Start this job the specified number of seconds after fio
963 has started. Only useful if the job file contains several
964 jobs, and you want to delay starting some jobs to a certain
967 runtime=time Tell fio to terminate processing after the specified number
968 of seconds. It can be quite hard to determine for how long
969 a specified job will run, so this parameter is handy to
970 cap the total runtime to a given time.
972 time_based If set, fio will run for the duration of the runtime
973 specified even if the file(s) are completely read or
974 written. It will simply loop over the same workload
975 as many times as the runtime allows.
977 ramp_time=time If set, fio will run the specified workload for this amount
978 of time before logging any performance numbers. Useful for
979 letting performance settle before logging results, thus
980 minimizing the runtime required for stable results. Note
981 that the ramp_time is considered lead in time for a job,
982 thus it will increase the total runtime if a special timeout
983 or runtime is specified.
985 invalidate=bool Invalidate the buffer/page cache parts for this file prior
986 to starting io. Defaults to true.
988 sync=bool Use sync io for buffered writes. For the majority of the
989 io engines, this means using O_SYNC.
992 mem=str Fio can use various types of memory as the io unit buffer.
993 The allowed values are:
995 malloc Use memory from malloc(3) as the buffers.
997 shm Use shared memory as the buffers. Allocated
1000 shmhuge Same as shm, but use huge pages as backing.
1002 mmap Use mmap to allocate buffers. May either be
1003 anonymous memory, or can be file backed if
1004 a filename is given after the option. The
1005 format is mem=mmap:/path/to/file.
1007 mmaphuge Use a memory mapped huge file as the buffer
1008 backing. Append filename after mmaphuge, ala
1009 mem=mmaphuge:/hugetlbfs/file
1011 The area allocated is a function of the maximum allowed
1012 bs size for the job, multiplied by the io depth given. Note
1013 that for shmhuge and mmaphuge to work, the system must have
1014 free huge pages allocated. This can normally be checked
1015 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1016 Linux system. Fio assumes a huge page is 4MB in size. So
1017 to calculate the number of huge pages you need for a given
1018 job file, add up the io depth of all jobs (normally one unless
1019 iodepth= is used) and multiply by the maximum bs set. Then
1020 divide that number by the huge page size. You can see the
1021 size of the huge pages in /proc/meminfo. If no huge pages
1022 are allocated by having a non-zero number in nr_hugepages,
1023 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1025 mmaphuge also needs to have hugetlbfs mounted and the file
1026 location should point there. So if it's mounted in /huge,
1027 you would use mem=mmaphuge:/huge/somefile.
1029 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1030 Note that the given alignment is applied to the first IO unit
1031 buffer, if using iodepth the alignment of the following buffers
1032 are given by the bs used. In other words, if using a bs that is
1033 a multiple of the page sized in the system, all buffers will
1034 be aligned to this value. If using a bs that is not page
1035 aligned, the alignment of subsequent IO memory buffers is the
1036 sum of the iomem_align and bs used.
1039 Defines the size of a huge page. Must at least be equal
1040 to the system setting, see /proc/meminfo. Defaults to 4MB.
1041 Should probably always be a multiple of megabytes, so using
1042 hugepage-size=Xm is the preferred way to set this to avoid
1043 setting a non-pow-2 bad value.
1045 exitall When one job finishes, terminate the rest. The default is
1046 to wait for each job to finish, sometimes that is not the
1049 bwavgtime=int Average the calculated bandwidth over the given time. Value
1050 is specified in milliseconds.
1052 iopsavgtime=int Average the calculated IOPS over the given time. Value
1053 is specified in milliseconds.
1055 create_serialize=bool If true, serialize the file creating for the jobs.
1056 This may be handy to avoid interleaving of data
1057 files, which may greatly depend on the filesystem
1058 used and even the number of processors in the system.
1060 create_fsync=bool fsync the data file after creation. This is the
1063 create_on_open=bool Don't pre-setup the files for IO, just create open()
1064 when it's time to do IO to that file.
1066 create_only=bool If true, fio will only run the setup phase of the job.
1067 If files need to be laid out or updated on disk, only
1068 that will be done. The actual job contents are not
1071 pre_read=bool If this is given, files will be pre-read into memory before
1072 starting the given IO operation. This will also clear
1073 the 'invalidate' flag, since it is pointless to pre-read
1074 and then drop the cache. This will only work for IO engines
1075 that are seekable, since they allow you to read the same data
1076 multiple times. Thus it will not work on eg network or splice
1079 unlink=bool Unlink the job files when done. Not the default, as repeated
1080 runs of that job would then waste time recreating the file
1081 set again and again.
1083 loops=int Run the specified number of iterations of this job. Used
1084 to repeat the same workload a given number of times. Defaults
1087 verify_only Do not perform specified workload---only verify data still
1088 matches previous invocation of this workload. This option
1089 allows one to check data multiple times at a later date
1090 without overwriting it. This option makes sense only for
1091 workloads that write data, and does not support workloads
1092 with the time_based option set.
1094 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1095 verify is set. Defaults to 1.
1097 verify=str If writing to a file, fio can verify the file contents
1098 after each iteration of the job. The allowed values are:
1100 md5 Use an md5 sum of the data area and store
1101 it in the header of each block.
1103 crc64 Use an experimental crc64 sum of the data
1104 area and store it in the header of each
1107 crc32c Use a crc32c sum of the data area and store
1108 it in the header of each block.
1110 crc32c-intel Use hardware assisted crc32c calcuation
1111 provided on SSE4.2 enabled processors. Falls
1112 back to regular software crc32c, if not
1113 supported by the system.
1115 crc32 Use a crc32 sum of the data area and store
1116 it in the header of each block.
1118 crc16 Use a crc16 sum of the data area and store
1119 it in the header of each block.
1121 crc7 Use a crc7 sum of the data area and store
1122 it in the header of each block.
1124 xxhash Use xxhash as the checksum function. Generally
1125 the fastest software checksum that fio
1128 sha512 Use sha512 as the checksum function.
1130 sha256 Use sha256 as the checksum function.
1132 sha1 Use optimized sha1 as the checksum function.
1134 meta Write extra information about each io
1135 (timestamp, block number etc.). The block
1136 number is verified. The io sequence number is
1137 verified for workloads that write data.
1138 See also verify_pattern.
1140 null Only pretend to verify. Useful for testing
1141 internals with ioengine=null, not for much
1144 This option can be used for repeated burn-in tests of a
1145 system to make sure that the written data is also
1146 correctly read back. If the data direction given is
1147 a read or random read, fio will assume that it should
1148 verify a previously written file. If the data direction
1149 includes any form of write, the verify will be of the
1152 verifysort=bool If set, fio will sort written verify blocks when it deems
1153 it faster to read them back in a sorted manner. This is
1154 often the case when overwriting an existing file, since
1155 the blocks are already laid out in the file system. You
1156 can ignore this option unless doing huge amounts of really
1157 fast IO where the red-black tree sorting CPU time becomes
1160 verify_offset=int Swap the verification header with data somewhere else
1161 in the block before writing. Its swapped back before
1164 verify_interval=int Write the verification header at a finer granularity
1165 than the blocksize. It will be written for chunks the
1166 size of header_interval. blocksize should divide this
1169 verify_pattern=str If set, fio will fill the io buffers with this
1170 pattern. Fio defaults to filling with totally random
1171 bytes, but sometimes it's interesting to fill with a known
1172 pattern for io verification purposes. Depending on the
1173 width of the pattern, fio will fill 1/2/3/4 bytes of the
1174 buffer at the time(it can be either a decimal or a hex number).
1175 The verify_pattern if larger than a 32-bit quantity has to
1176 be a hex number that starts with either "0x" or "0X". Use
1179 verify_fatal=bool Normally fio will keep checking the entire contents
1180 before quitting on a block verification failure. If this
1181 option is set, fio will exit the job on the first observed
1184 verify_dump=bool If set, dump the contents of both the original data
1185 block and the data block we read off disk to files. This
1186 allows later analysis to inspect just what kind of data
1187 corruption occurred. Off by default.
1189 verify_async=int Fio will normally verify IO inline from the submitting
1190 thread. This option takes an integer describing how many
1191 async offload threads to create for IO verification instead,
1192 causing fio to offload the duty of verifying IO contents
1193 to one or more separate threads. If using this offload
1194 option, even sync IO engines can benefit from using an
1195 iodepth setting higher than 1, as it allows them to have
1196 IO in flight while verifies are running.
1198 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1199 async IO verification threads. See cpus_allowed for the
1202 verify_backlog=int Fio will normally verify the written contents of a
1203 job that utilizes verify once that job has completed. In
1204 other words, everything is written then everything is read
1205 back and verified. You may want to verify continually
1206 instead for a variety of reasons. Fio stores the meta data
1207 associated with an IO block in memory, so for large
1208 verify workloads, quite a bit of memory would be used up
1209 holding this meta data. If this option is enabled, fio
1210 will write only N blocks before verifying these blocks.
1212 verify_backlog_batch=int Control how many blocks fio will verify
1213 if verify_backlog is set. If not set, will default to
1214 the value of verify_backlog (meaning the entire queue
1215 is read back and verified). If verify_backlog_batch is
1216 less than verify_backlog then not all blocks will be verified,
1217 if verify_backlog_batch is larger than verify_backlog, some
1218 blocks will be verified more than once.
1221 wait_for_previous Wait for preceding jobs in the job file to exit, before
1222 starting this one. Can be used to insert serialization
1223 points in the job file. A stone wall also implies starting
1224 a new reporting group.
1226 new_group Start a new reporting group. See: group_reporting.
1228 numjobs=int Create the specified number of clones of this job. May be
1229 used to setup a larger number of threads/processes doing
1230 the same thing. Each thread is reported separately; to see
1231 statistics for all clones as a whole, use group_reporting in
1232 conjunction with new_group.
1234 group_reporting It may sometimes be interesting to display statistics for
1235 groups of jobs as a whole instead of for each individual job.
1236 This is especially true if 'numjobs' is used; looking at
1237 individual thread/process output quickly becomes unwieldy.
1238 To see the final report per-group instead of per-job, use
1239 'group_reporting'. Jobs in a file will be part of the same
1240 reporting group, unless if separated by a stonewall, or by
1243 thread fio defaults to forking jobs, however if this option is
1244 given, fio will use pthread_create(3) to create threads
1247 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1249 zoneskip=int Skip the specified number of bytes when zonesize data has
1250 been read. The two zone options can be used to only do
1251 io on zones of a file.
1253 write_iolog=str Write the issued io patterns to the specified file. See
1254 read_iolog. Specify a separate file for each job, otherwise
1255 the iologs will be interspersed and the file may be corrupt.
1257 read_iolog=str Open an iolog with the specified file name and replay the
1258 io patterns it contains. This can be used to store a
1259 workload and replay it sometime later. The iolog given
1260 may also be a blktrace binary file, which allows fio
1261 to replay a workload captured by blktrace. See blktrace
1262 for how to capture such logging data. For blktrace replay,
1263 the file needs to be turned into a blkparse binary data
1264 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1266 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1267 is to attempt to respect the time stamps within the log and
1268 replay them with the appropriate delay between IOPS. By
1269 setting this variable fio will not respect the timestamps and
1270 attempt to replay them as fast as possible while still
1271 respecting ordering. The result is the same I/O pattern to a
1272 given device, but different timings.
1274 replay_redirect=str While replaying I/O patterns using read_iolog the
1275 default behavior is to replay the IOPS onto the major/minor
1276 device that each IOP was recorded from. This is sometimes
1277 undesirable because on a different machine those major/minor
1278 numbers can map to a different device. Changing hardware on
1279 the same system can also result in a different major/minor
1280 mapping. Replay_redirect causes all IOPS to be replayed onto
1281 the single specified device regardless of the device it was
1282 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1283 IO in the blktrace to be replayed onto /dev/sdc. This means
1284 multiple devices will be replayed onto a single, if the trace
1285 contains multiple devices. If you want multiple devices to be
1286 replayed concurrently to multiple redirected devices you must
1287 blkparse your trace into separate traces and replay them with
1288 independent fio invocations. Unfortuantely this also breaks
1289 the strict time ordering between multiple device accesses.
1291 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1292 file. Can be used to store data of the bandwidth of the
1293 jobs in their lifetime. The included fio_generate_plots
1294 script uses gnuplot to turn these text files into nice
1295 graphs. See write_lat_log for behaviour of given
1296 filename. For this option, the suffix is _bw.log.
1298 write_lat_log=str Same as write_bw_log, except that this option stores io
1299 submission, completion, and total latencies instead. If no
1300 filename is given with this option, the default filename of
1301 "jobname_type.log" is used. Even if the filename is given,
1302 fio will still append the type of log. So if one specifies
1306 The actual log names will be foo_slat.log, foo_clat.log,
1307 and foo_lat.log. This helps fio_generate_plot fine the logs
1310 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1311 given with this option, the default filename of
1312 "jobname_type.log" is used. Even if the filename is given,
1313 fio will still append the type of log.
1315 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1316 or bw log for every IO that completes. When writing to the
1317 disk log, that can quickly grow to a very large size. Setting
1318 this option makes fio average the each log entry over the
1319 specified period of time, reducing the resolution of the log.
1322 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1323 potentially be used instead of removing memory or booting
1324 with less memory to simulate a smaller amount of memory.
1325 The amount specified is per worker.
1327 exec_prerun=str Before running this job, issue the command specified
1328 through system(3). Output is redirected in a file called
1331 exec_postrun=str After the job completes, issue the command specified
1332 though system(3). Output is redirected in a file called
1333 jobname.postrun.txt.
1335 ioscheduler=str Attempt to switch the device hosting the file to the specified
1336 io scheduler before running.
1338 disk_util=bool Generate disk utilization statistics, if the platform
1339 supports it. Defaults to on.
1341 disable_lat=bool Disable measurements of total latency numbers. Useful
1342 only for cutting back the number of calls to gettimeofday,
1343 as that does impact performance at really high IOPS rates.
1344 Note that to really get rid of a large amount of these
1345 calls, this option must be used with disable_slat and
1348 disable_clat=bool Disable measurements of completion latency numbers. See
1351 disable_slat=bool Disable measurements of submission latency numbers. See
1354 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1357 clat_percentiles=bool Enable the reporting of percentiles of
1358 completion latencies.
1360 percentile_list=float_list Overwrite the default list of percentiles
1361 for completion latencies. Each number is a floating
1362 number in the range (0,100], and the maximum length of
1363 the list is 20. Use ':' to separate the numbers, and
1364 list the numbers in ascending order. For example,
1365 --percentile_list=99.5:99.9 will cause fio to report
1366 the values of completion latency below which 99.5% and
1367 99.9% of the observed latencies fell, respectively.
1369 clocksource=str Use the given clocksource as the base of timing. The
1370 supported options are:
1372 gettimeofday gettimeofday(2)
1374 clock_gettime clock_gettime(2)
1376 cpu Internal CPU clock source
1378 cpu is the preferred clocksource if it is reliable, as it
1379 is very fast (and fio is heavy on time calls). Fio will
1380 automatically use this clocksource if it's supported and
1381 considered reliable on the system it is running on, unless
1382 another clocksource is specifically set. For x86/x86-64 CPUs,
1383 this means supporting TSC Invariant.
1385 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1386 (disable_clat, disable_slat, disable_bw) plus reduce
1387 precision of the timeout somewhat to really shrink
1388 the gettimeofday() call count. With this option enabled,
1389 we only do about 0.4% of the gtod() calls we would have
1390 done if all time keeping was enabled.
1392 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1393 execution to just getting the current time. Fio (and
1394 databases, for instance) are very intensive on gettimeofday()
1395 calls. With this option, you can set one CPU aside for
1396 doing nothing but logging current time to a shared memory
1397 location. Then the other threads/processes that run IO
1398 workloads need only copy that segment, instead of entering
1399 the kernel with a gettimeofday() call. The CPU set aside
1400 for doing these time calls will be excluded from other
1401 uses. Fio will manually clear it from the CPU mask of other
1404 continue_on_error=str Normally fio will exit the job on the first observed
1405 failure. If this option is set, fio will continue the job when
1406 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1407 is exceeded or the I/O size specified is completed. If this
1408 option is used, there are two more stats that are appended,
1409 the total error count and the first error. The error field
1410 given in the stats is the first error that was hit during the
1413 The allowed values are:
1415 none Exit on any IO or verify errors.
1417 read Continue on read errors, exit on all others.
1419 write Continue on write errors, exit on all others.
1421 io Continue on any IO error, exit on all others.
1423 verify Continue on verify errors, exit on all others.
1425 all Continue on all errors.
1427 0 Backward-compatible alias for 'none'.
1429 1 Backward-compatible alias for 'all'.
1431 ignore_error=str Sometimes you want to ignore some errors during test
1432 in that case you can specify error list for each error type.
1433 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1434 errors for given error type is separated with ':'. Error
1435 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1437 ignore_error=EAGAIN,ENOSPC:122
1438 This option will ignore EAGAIN from READ, and ENOSPC and
1439 122(EDQUOT) from WRITE.
1441 error_dump=bool If set dump every error even if it is non fatal, true
1442 by default. If disabled only fatal error will be dumped
1444 cgroup=str Add job to this control group. If it doesn't exist, it will
1445 be created. The system must have a mounted cgroup blkio
1446 mount point for this to work. If your system doesn't have it
1447 mounted, you can do so with:
1449 # mount -t cgroup -o blkio none /cgroup
1451 cgroup_weight=int Set the weight of the cgroup to this value. See
1452 the documentation that comes with the kernel, allowed values
1453 are in the range of 100..1000.
1455 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1456 the job completion. To override this behavior and to leave
1457 cgroups around after the job completion, set cgroup_nodelete=1.
1458 This can be useful if one wants to inspect various cgroup
1459 files after job completion. Default: false
1461 uid=int Instead of running as the invoking user, set the user ID to
1462 this value before the thread/process does any work.
1464 gid=int Set group ID, see uid.
1466 flow_id=int The ID of the flow. If not specified, it defaults to being a
1467 global flow. See flow.
1469 flow=int Weight in token-based flow control. If this value is used, then
1470 there is a 'flow counter' which is used to regulate the
1471 proportion of activity between two or more jobs. fio attempts
1472 to keep this flow counter near zero. The 'flow' parameter
1473 stands for how much should be added or subtracted to the flow
1474 counter on each iteration of the main I/O loop. That is, if
1475 one job has flow=8 and another job has flow=-1, then there
1476 will be a roughly 1:8 ratio in how much one runs vs the other.
1478 flow_watermark=int The maximum value that the absolute value of the flow
1479 counter is allowed to reach before the job must wait for a
1480 lower value of the counter.
1482 flow_sleep=int The period of time, in microseconds, to wait after the flow
1483 watermark has been exceeded before retrying operations
1485 In addition, there are some parameters which are only valid when a specific
1486 ioengine is in use. These are used identically to normal parameters, with the
1487 caveat that when used on the command line, they must come after the ioengine
1488 that defines them is selected.
1490 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1491 the io_getevents system call to reap newly returned events.
1492 With this flag turned on, the AIO ring will be read directly
1493 from user-space to reap events. The reaping mode is only
1494 enabled when polling for a minimum of 0 events (eg when
1495 iodepth_batch_complete=0).
1497 [cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1499 [cpu] cpuchunks=int Split the load into cycles of the given time. In
1502 [netsplice] hostname=str
1503 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1504 If the job is a TCP listener or UDP reader, the hostname is not
1505 used and must be omitted unless it is a valid UDP multicast
1508 [netsplice] port=int
1509 [net] port=int The TCP or UDP port to bind to or connect to.
1511 [netsplice] interface=str
1512 [net] interface=str The IP address of the network interface used to send or
1513 receive UDP multicast
1516 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1519 [netsplice] nodelay=bool
1520 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1522 [netsplice] protocol=str
1523 [netsplice] proto=str
1525 [net] proto=str The network protocol to use. Accepted values are:
1527 tcp Transmission control protocol
1528 tcpv6 Transmission control protocol V6
1529 udp User datagram protocol
1530 udpv6 User datagram protocol V6
1531 unix UNIX domain socket
1533 When the protocol is TCP or UDP, the port must also be given,
1534 as well as the hostname if the job is a TCP listener or UDP
1535 reader. For unix sockets, the normal filename option should be
1536 used and the port is invalid.
1538 [net] listen For TCP network connections, tell fio to listen for incoming
1539 connections rather than initiating an outgoing connection. The
1540 hostname must be omitted if this option is used.
1541 [net] pingpong Normaly a network writer will just continue writing data, and
1542 a network reader will just consume packages. If pingpong=1
1543 is set, a writer will send its normal payload to the reader,
1544 then wait for the reader to send the same payload back. This
1545 allows fio to measure network latencies. The submission
1546 and completion latencies then measure local time spent
1547 sending or receiving, and the completion latency measures
1548 how long it took for the other end to receive and send back.
1549 For UDP multicast traffic pingpong=1 should only be set for a
1550 single reader when multiple readers are listening to the same
1553 [e4defrag] donorname=str
1554 File will be used as a block donor(swap extents between files)
1555 [e4defrag] inplace=int
1556 Configure donor file blocks allocation strategy
1557 0(default): Preallocate donor's file on init
1558 1 : allocate space immidietly inside defragment event,
1559 and free right after event
1563 6.0 Interpreting the output
1564 ---------------------------
1566 fio spits out a lot of output. While running, fio will display the
1567 status of the jobs created. An example of that would be:
1569 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1571 The characters inside the square brackets denote the current status of
1572 each thread. The possible values (in typical life cycle order) are:
1576 P Thread setup, but not started.
1578 I Thread initialized, waiting or generating necessary data.
1579 p Thread running pre-reading file(s).
1580 R Running, doing sequential reads.
1581 r Running, doing random reads.
1582 W Running, doing sequential writes.
1583 w Running, doing random writes.
1584 M Running, doing mixed sequential reads/writes.
1585 m Running, doing mixed random reads/writes.
1586 F Running, currently waiting for fsync()
1587 V Running, doing verification of written data.
1588 E Thread exited, not reaped by main thread yet.
1590 X Thread reaped, exited with an error.
1591 K Thread reaped, exited due to signal.
1593 The other values are fairly self explanatory - number of threads
1594 currently running and doing io, rate of io since last check (read speed
1595 listed first, then write speed), and the estimated completion percentage
1596 and time for the running group. It's impossible to estimate runtime of
1597 the following groups (if any). Note that the string is displayed in order,
1598 so it's possible to tell which of the jobs are currently doing what. The
1599 first character is the first job defined in the job file, and so forth.
1601 When fio is done (or interrupted by ctrl-c), it will show the data for
1602 each thread, group of threads, and disks in that order. For each data
1603 direction, the output looks like:
1605 Client1 (g=0): err= 0:
1606 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1607 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1608 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1609 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1610 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1611 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1612 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1613 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1614 issued r/w: total=0/32768, short=0/0
1615 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1616 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1618 The client number is printed, along with the group id and error of that
1619 thread. Below is the io statistics, here for writes. In the order listed,
1622 io= Number of megabytes io performed
1623 bw= Average bandwidth rate
1624 iops= Average IOs performed per second
1625 runt= The runtime of that thread
1626 slat= Submission latency (avg being the average, stdev being the
1627 standard deviation). This is the time it took to submit
1628 the io. For sync io, the slat is really the completion
1629 latency, since queue/complete is one operation there. This
1630 value can be in milliseconds or microseconds, fio will choose
1631 the most appropriate base and print that. In the example
1632 above, milliseconds is the best scale. Note: in --minimal mode
1633 latencies are always expressed in microseconds.
1634 clat= Completion latency. Same names as slat, this denotes the
1635 time from submission to completion of the io pieces. For
1636 sync io, clat will usually be equal (or very close) to 0,
1637 as the time from submit to complete is basically just
1638 CPU time (io has already been done, see slat explanation).
1639 bw= Bandwidth. Same names as the xlat stats, but also includes
1640 an approximate percentage of total aggregate bandwidth
1641 this thread received in this group. This last value is
1642 only really useful if the threads in this group are on the
1643 same disk, since they are then competing for disk access.
1644 cpu= CPU usage. User and system time, along with the number
1645 of context switches this thread went through, usage of
1646 system and user time, and finally the number of major
1647 and minor page faults.
1648 IO depths= The distribution of io depths over the job life time. The
1649 numbers are divided into powers of 2, so for example the
1650 16= entries includes depths up to that value but higher
1651 than the previous entry. In other words, it covers the
1652 range from 16 to 31.
1653 IO submit= How many pieces of IO were submitting in a single submit
1654 call. Each entry denotes that amount and below, until
1655 the previous entry - eg, 8=100% mean that we submitted
1656 anywhere in between 5-8 ios per submit call.
1657 IO complete= Like the above submit number, but for completions instead.
1658 IO issued= The number of read/write requests issued, and how many
1660 IO latencies= The distribution of IO completion latencies. This is the
1661 time from when IO leaves fio and when it gets completed.
1662 The numbers follow the same pattern as the IO depths,
1663 meaning that 2=1.6% means that 1.6% of the IO completed
1664 within 2 msecs, 20=12.8% means that 12.8% of the IO
1665 took more than 10 msecs, but less than (or equal to) 20 msecs.
1667 After each client has been listed, the group statistics are printed. They
1668 will look like this:
1670 Run status group 0 (all jobs):
1671 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1672 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1674 For each data direction, it prints:
1676 io= Number of megabytes io performed.
1677 aggrb= Aggregate bandwidth of threads in this group.
1678 minb= The minimum average bandwidth a thread saw.
1679 maxb= The maximum average bandwidth a thread saw.
1680 mint= The smallest runtime of the threads in that group.
1681 maxt= The longest runtime of the threads in that group.
1683 And finally, the disk statistics are printed. They will look like this:
1685 Disk stats (read/write):
1686 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1688 Each value is printed for both reads and writes, with reads first. The
1691 ios= Number of ios performed by all groups.
1692 merge= Number of merges io the io scheduler.
1693 ticks= Number of ticks we kept the disk busy.
1694 io_queue= Total time spent in the disk queue.
1695 util= The disk utilization. A value of 100% means we kept the disk
1696 busy constantly, 50% would be a disk idling half of the time.
1698 It is also possible to get fio to dump the current output while it is
1699 running, without terminating the job. To do that, send fio the USR1 signal.
1700 You can also get regularly timed dumps by using the --status-interval
1701 parameter, or by creating a file in /tmp named fio-dump-status. If fio
1702 sees this file, it will unlink it and dump the current output status.
1708 For scripted usage where you typically want to generate tables or graphs
1709 of the results, fio can output the results in a semicolon separated format.
1710 The format is one long line of values, such as:
1712 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%
1713 A description of this job goes here.
1715 The job description (if provided) follows on a second line.
1717 To enable terse output, use the --minimal command line option. The first
1718 value is the version of the terse output format. If the output has to
1719 be changed for some reason, this number will be incremented by 1 to
1720 signify that change.
1722 Split up, the format is as follows:
1724 terse version, fio version, jobname, groupid, error
1726 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1727 Submission latency: min, max, mean, deviation (usec)
1728 Completion latency: min, max, mean, deviation (usec)
1729 Completion latency percentiles: 20 fields (see below)
1730 Total latency: min, max, mean, deviation (usec)
1731 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1733 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1734 Submission latency: min, max, mean, deviation (usec)
1735 Completion latency: min, max, mean, deviation (usec)
1736 Completion latency percentiles: 20 fields (see below)
1737 Total latency: min, max, mean, deviation (usec)
1738 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1739 CPU usage: user, system, context switches, major faults, minor faults
1740 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1741 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1742 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1743 Disk utilization: Disk name, Read ios, write ios,
1744 Read merges, write merges,
1745 Read ticks, write ticks,
1746 Time spent in queue, disk utilization percentage
1747 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
1749 Additional Info (dependent on description being set): Text description
1751 Completion latency percentiles can be a grouping of up to 20 sets, so
1752 for the terse output fio writes all of them. Each field will look like this:
1756 which is the Xth percentile, and the usec latency associated with it.
1758 For disk utilization, all disks used by fio are shown. So for each disk
1759 there will be a disk utilization section.
1762 8.0 Trace file format
1763 ---------------------
1764 There are two trace file format that you can encounter. The older (v1) format
1765 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1766 below in case that you get an old trace and want to understand it.
1768 In any case the trace is a simple text file with a single action per line.
1771 8.1 Trace file format v1
1772 ------------------------
1773 Each line represents a single io action in the following format:
1777 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1779 This format is not supported in Fio versions => 1.20-rc3.
1782 8.2 Trace file format v2
1783 ------------------------
1784 The second version of the trace file format was added in Fio version 1.17.
1785 It allows to access more then one file per trace and has a bigger set of
1786 possible file actions.
1788 The first line of the trace file has to be:
1792 Following this can be lines in two different formats, which are described below.
1794 The file management format:
1798 The filename is given as an absolute path. The action can be one of these:
1800 add Add the given filename to the trace
1801 open Open the file with the given filename. The filename has to have
1802 been added with the add action before.
1803 close Close the file with the given filename. The file has to have been
1807 The file io action format:
1809 filename action offset length
1811 The filename is given as an absolute path, and has to have been added and opened
1812 before it can be used with this format. The offset and length are given in
1813 bytes. The action can be one of these:
1815 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1816 read Read 'length' bytes beginning from 'offset'
1817 write Write 'length' bytes beginning from 'offset'
1818 sync fsync() the file
1819 datasync fdatasync() the file
1820 trim trim the given file from the given 'offset' for 'length' bytes
1823 9.0 CPU idleness profiling
1824 --------------------------
1825 In some cases, we want to understand CPU overhead in a test. For example,
1826 we test patches for the specific goodness of whether they reduce CPU usage.
1827 fio implements a balloon approach to create a thread per CPU that runs at
1828 idle priority, meaning that it only runs when nobody else needs the cpu.
1829 By measuring the amount of work completed by the thread, idleness of each
1830 CPU can be derived accordingly.
1832 An unit work is defined as touching a full page of unsigned characters. Mean
1833 and standard deviation of time to complete an unit work is reported in "unit
1834 work" section. Options can be chosen to report detailed percpu idleness or
1835 overall system idleness by aggregating percpu stats.