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 "./".
267 filename=str Fio normally makes up a filename based on the job name,
268 thread number, and file number. If you want to share
269 files between threads in a job or several jobs, specify
270 a filename for each of them to override the default. If
271 the ioengine used is 'net', the filename is the host, port,
272 and protocol to use in the format of =host,port,protocol.
273 See ioengine=net for more. If the ioengine is file based, you
274 can specify a number of files by separating the names with a
275 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
276 as the two working files, you would use
277 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
278 accessed as \\.\PhysicalDrive0 for the first device,
279 \\.\PhysicalDrive1 for the second etc. Note: Windows and
280 FreeBSD prevent write access to areas of the disk containing
281 in-use data (e.g. filesystems).
282 If the wanted filename does need to include a colon, then
283 escape that with a '\' character. For instance, if the filename
284 is "/dev/dsk/foo@3,0:c", then you would use
285 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
286 stdin or stdout. Which of the two depends on the read/write
290 If sharing multiple files between jobs, it is usually necessary
291 to have fio generate the exact names that you want. By default,
292 fio will name a file based on the default file format
293 specification of jobname.jobnumber.filenumber. With this
294 option, that can be customized. Fio will recognize and replace
295 the following keywords in this string:
298 The name of the worker thread or process.
301 The incremental number of the worker thread or
305 The incremental number of the file for that worker
308 To have dependent jobs share a set of files, this option can
309 be set to have fio generate filenames that are shared between
310 the two. For instance, if testfiles.$filenum is specified,
311 file number 4 for any job will be named testfiles.4. The
312 default of $jobname.$jobnum.$filenum will be used if
313 no other format specifier is given.
315 opendir=str Tell fio to recursively add any file it can find in this
316 directory and down the file system tree.
318 lockfile=str Fio defaults to not locking any files before it does
319 IO to them. If a file or file descriptor is shared, fio
320 can serialize IO to that file to make the end result
321 consistent. This is usual for emulating real workloads that
322 share files. The lock modes are:
324 none No locking. The default.
325 exclusive Only one thread/process may do IO,
326 excluding all others.
327 readwrite Read-write locking on the file. Many
328 readers may access the file at the
329 same time, but writes get exclusive
333 rw=str Type of io pattern. Accepted values are:
335 read Sequential reads
336 write Sequential writes
337 randwrite Random writes
338 randread Random reads
339 rw,readwrite Sequential mixed reads and writes
340 randrw Random mixed reads and writes
342 For the mixed io types, the default is to split them 50/50.
343 For certain types of io the result may still be skewed a bit,
344 since the speed may be different. It is possible to specify
345 a number of IO's to do before getting a new offset, this is
346 one by appending a ':<nr>' to the end of the string given.
347 For a random read, it would look like 'rw=randread:8' for
348 passing in an offset modifier with a value of 8. If the
349 suffix is used with a sequential IO pattern, then the value
350 specified will be added to the generated offset for each IO.
351 For instance, using rw=write:4k will skip 4k for every
352 write. It turns sequential IO into sequential IO with holes.
353 See the 'rw_sequencer' option.
355 rw_sequencer=str If an offset modifier is given by appending a number to
356 the rw=<str> line, then this option controls how that
357 number modifies the IO offset being generated. Accepted
360 sequential Generate sequential offset
361 identical Generate the same offset
363 'sequential' is only useful for random IO, where fio would
364 normally generate a new random offset for every IO. If you
365 append eg 8 to randread, you would get a new random offset for
366 every 8 IO's. The result would be a seek for only every 8
367 IO's, instead of for every IO. Use rw=randread:8 to specify
368 that. As sequential IO is already sequential, setting
369 'sequential' for that would not result in any differences.
370 'identical' behaves in a similar fashion, except it sends
371 the same offset 8 number of times before generating a new
374 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
375 Storage manufacturers like to use 10^3 or 1000 as a base
376 ten unit instead, for obvious reasons. Allow values are
377 1024 or 1000, with 1024 being the default.
379 unified_rw_reporting=bool Fio normally reports statistics on a per
380 data direction basis, meaning that read, write, and trim are
381 accounted and reported separately. If this option is set,
382 the fio will sum the results and report them as "mixed"
385 randrepeat=bool For random IO workloads, seed the generator in a predictable
386 way so that results are repeatable across repetitions.
388 randseed=int Seed the random number generators based on this seed value, to
389 be able to control what sequence of output is being generated.
390 If not set, the random sequence depends on the randrepeat
393 use_os_rand=bool Fio can either use the random generator supplied by the OS
394 to generator random offsets, or it can use it's own internal
395 generator (based on Tausworthe). Default is to use the
396 internal generator, which is often of better quality and
399 fallocate=str Whether pre-allocation is performed when laying down files.
402 none Do not pre-allocate space
403 posix Pre-allocate via posix_fallocate()
404 keep Pre-allocate via fallocate() with
405 FALLOC_FL_KEEP_SIZE set
406 0 Backward-compatible alias for 'none'
407 1 Backward-compatible alias for 'posix'
409 May not be available on all supported platforms. 'keep' is only
410 available on Linux.If using ZFS on Solaris this must be set to
411 'none' because ZFS doesn't support it. Default: 'posix'.
413 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
414 on what IO patterns it is likely to issue. Sometimes you
415 want to test specific IO patterns without telling the
416 kernel about it, in which case you can disable this option.
417 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
418 IO and POSIX_FADV_RANDOM for random IO.
420 size=int The total size of file io for this job. Fio will run until
421 this many bytes has been transferred, unless runtime is
422 limited by other options (such as 'runtime', for instance).
423 Unless specific nrfiles and filesize options are given,
424 fio will divide this size between the available files
425 specified by the job. If not set, fio will use the full
426 size of the given files or devices. If the the files
427 do not exist, size must be given. It is also possible to
428 give size as a percentage between 1 and 100. If size=20%
429 is given, fio will use 20% of the full size of the given
432 filesize=int Individual file sizes. May be a range, in which case fio
433 will select sizes for files at random within the given range
434 and limited to 'size' in total (if that is given). If not
435 given, each created file is the same size.
438 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
439 space left on device) as the terminating condition. Only makes
440 sense with sequential write. For a read workload, the mount
441 point will be filled first then IO started on the result. This
442 option doesn't make sense if operating on a raw device node,
443 since the size of that is already known by the file system.
444 Additionally, writing beyond end-of-device will not return
448 bs=int The block size used for the io units. Defaults to 4k. Values
449 can be given for both read and writes. If a single int is
450 given, it will apply to both. If a second int is specified
451 after a comma, it will apply to writes only. In other words,
452 the format is either bs=read_and_write or bs=read,write,trim.
453 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
454 writes, and 8k for trims. You can terminate the list with
455 a trailing comma. bs=4k,8k, would use the default value for
456 trims.. If you only wish to set the write size, you
457 can do so by passing an empty read size - bs=,8k will set
458 8k for writes and leave the read default value.
461 ba=int At what boundary to align random IO offsets. Defaults to
462 the same as 'blocksize' the minimum blocksize given.
463 Minimum alignment is typically 512b for using direct IO,
464 though it usually depends on the hardware block size. This
465 option is mutually exclusive with using a random map for
466 files, so it will turn off that option.
468 blocksize_range=irange
469 bsrange=irange Instead of giving a single block size, specify a range
470 and fio will mix the issued io block sizes. The issued
471 io unit will always be a multiple of the minimum value
472 given (also see bs_unaligned). Applies to both reads and
473 writes, however a second range can be given after a comma.
476 bssplit=str Sometimes you want even finer grained control of the
477 block sizes issued, not just an even split between them.
478 This option allows you to weight various block sizes,
479 so that you are able to define a specific amount of
480 block sizes issued. The format for this option is:
482 bssplit=blocksize/percentage:blocksize/percentage
484 for as many block sizes as needed. So if you want to define
485 a workload that has 50% 64k blocks, 10% 4k blocks, and
486 40% 32k blocks, you would write:
488 bssplit=4k/10:64k/50:32k/40
490 Ordering does not matter. If the percentage is left blank,
491 fio will fill in the remaining values evenly. So a bssplit
492 option like this one:
494 bssplit=4k/50:1k/:32k/
496 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
497 always add up to 100, if bssplit is given a range that adds
498 up to more, it will error out.
500 bssplit also supports giving separate splits to reads and
501 writes. The format is identical to what bs= accepts. You
502 have to separate the read and write parts with a comma. So
503 if you want a workload that has 50% 2k reads and 50% 4k reads,
504 while having 90% 4k writes and 10% 8k writes, you would
507 bssplit=2k/50:4k/50,4k/90,8k/10
510 bs_unaligned If this option is given, any byte size value within bsrange
511 may be used as a block range. This typically wont work with
512 direct IO, as that normally requires sector alignment.
514 bs_is_seq_rand If this option is set, fio will use the normal read,write
515 blocksize settings as sequential,random instead. Any random
516 read or write will use the WRITE blocksize settings, and any
517 sequential read or write will use the READ blocksize setting.
519 zero_buffers If this option is given, fio will init the IO buffers to
520 all zeroes. The default is to fill them with random data.
522 refill_buffers If this option is given, fio will refill the IO buffers
523 on every submit. The default is to only fill it at init
524 time and reuse that data. Only makes sense if zero_buffers
525 isn't specified, naturally. If data verification is enabled,
526 refill_buffers is also automatically enabled.
528 scramble_buffers=bool If refill_buffers is too costly and the target is
529 using data deduplication, then setting this option will
530 slightly modify the IO buffer contents to defeat normal
531 de-dupe attempts. This is not enough to defeat more clever
532 block compression attempts, but it will stop naive dedupe of
533 blocks. Default: true.
535 buffer_compress_percentage=int If this is set, then fio will attempt to
536 provide IO buffer content (on WRITEs) that compress to
537 the specified level. Fio does this by providing a mix of
538 random data and zeroes. Note that this is per block size
539 unit, for file/disk wide compression level that matches
540 this setting, you'll also want to set refill_buffers.
542 buffer_compress_chunk=int See buffer_compress_percentage. This
543 setting allows fio to manage how big the ranges of random
544 data and zeroed data is. Without this set, fio will
545 provide buffer_compress_percentage of blocksize random
546 data, followed by the remaining zeroed. With this set
547 to some chunk size smaller than the block size, fio can
548 alternate random and zeroed data throughout the IO
551 buffer_pattern=str If set, fio will fill the io buffers with this pattern.
552 If not set, the contents of io buffers is defined by the other
553 options related to buffer contents. The setting can be any
554 pattern of bytes, and can be prefixed with 0x for hex values.
556 nrfiles=int Number of files to use for this job. Defaults to 1.
558 openfiles=int Number of files to keep open at the same time. Defaults to
559 the same as nrfiles, can be set smaller to limit the number
562 file_service_type=str Defines how fio decides which file from a job to
563 service next. The following types are defined:
565 random Just choose a file at random.
567 roundrobin Round robin over open files. This
570 sequential Finish one file before moving on to
571 the next. Multiple files can still be
572 open depending on 'openfiles'.
574 The string can have a number appended, indicating how
575 often to switch to a new file. So if option random:4 is
576 given, fio will switch to a new random file after 4 ios
579 ioengine=str Defines how the job issues io to the file. The following
582 sync Basic read(2) or write(2) io. lseek(2) is
583 used to position the io location.
585 psync Basic pread(2) or pwrite(2) io.
587 vsync Basic readv(2) or writev(2) IO.
589 psyncv Basic preadv(2) or pwritev(2) IO.
591 libaio Linux native asynchronous io. Note that Linux
592 may only support queued behaviour with
593 non-buffered IO (set direct=1 or buffered=0).
594 This engine defines engine specific options.
596 posixaio glibc posix asynchronous io.
598 solarisaio Solaris native asynchronous io.
600 windowsaio Windows native asynchronous io.
602 mmap File is memory mapped and data copied
603 to/from using memcpy(3).
605 splice splice(2) is used to transfer the data and
606 vmsplice(2) to transfer data from user
609 syslet-rw Use the syslet system calls to make
610 regular read/write async.
612 sg SCSI generic sg v3 io. May either be
613 synchronous using the SG_IO ioctl, or if
614 the target is an sg character device
615 we use read(2) and write(2) for asynchronous
618 null Doesn't transfer any data, just pretends
619 to. This is mainly used to exercise fio
620 itself and for debugging/testing purposes.
622 net Transfer over the network to given host:port.
623 Depending on the protocol used, the hostname,
624 port, listen and filename options are used to
625 specify what sort of connection to make, while
626 the protocol option determines which protocol
628 This engine defines engine specific options.
630 netsplice Like net, but uses splice/vmsplice to
631 map data and send/receive.
632 This engine defines engine specific options.
634 cpuio Doesn't transfer any data, but burns CPU
635 cycles according to the cpuload= and
636 cpucycle= options. Setting cpuload=85
637 will cause that job to do nothing but burn
638 85% of the CPU. In case of SMP machines,
639 use numjobs=<no_of_cpu> to get desired CPU
640 usage, as the cpuload only loads a single
641 CPU at the desired rate.
643 guasi The GUASI IO engine is the Generic Userspace
644 Asyncronous Syscall Interface approach
647 http://www.xmailserver.org/guasi-lib.html
649 for more info on GUASI.
651 rdma The RDMA I/O engine supports both RDMA
652 memory semantics (RDMA_WRITE/RDMA_READ) and
653 channel semantics (Send/Recv) for the
654 InfiniBand, RoCE and iWARP protocols.
656 falloc IO engine that does regular fallocate to
657 simulate data transfer as fio ioengine.
658 DDIR_READ does fallocate(,mode = keep_size,)
659 DDIR_WRITE does fallocate(,mode = 0)
660 DDIR_TRIM does fallocate(,mode = punch_hole)
662 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
663 ioctls to simulate defragment activity in
664 request to DDIR_WRITE event
666 external Prefix to specify loading an external
667 IO engine object file. Append the engine
668 filename, eg ioengine=external:/tmp/foo.o
669 to load ioengine foo.o in /tmp.
671 iodepth=int This defines how many io units to keep in flight against
672 the file. The default is 1 for each file defined in this
673 job, can be overridden with a larger value for higher
674 concurrency. Note that increasing iodepth beyond 1 will not
675 affect synchronous ioengines (except for small degress when
676 verify_async is in use). Even async engines may impose OS
677 restrictions causing the desired depth not to be achieved.
678 This may happen on Linux when using libaio and not setting
679 direct=1, since buffered IO is not async on that OS. Keep an
680 eye on the IO depth distribution in the fio output to verify
681 that the achieved depth is as expected. Default: 1.
683 iodepth_batch_submit=int
684 iodepth_batch=int This defines how many pieces of IO to submit at once.
685 It defaults to 1 which means that we submit each IO
686 as soon as it is available, but can be raised to submit
687 bigger batches of IO at the time.
689 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
690 at once. It defaults to 1 which means that we'll ask
691 for a minimum of 1 IO in the retrieval process from
692 the kernel. The IO retrieval will go on until we
693 hit the limit set by iodepth_low. If this variable is
694 set to 0, then fio will always check for completed
695 events before queuing more IO. This helps reduce
696 IO latency, at the cost of more retrieval system calls.
698 iodepth_low=int The low water mark indicating when to start filling
699 the queue again. Defaults to the same as iodepth, meaning
700 that fio will attempt to keep the queue full at all times.
701 If iodepth is set to eg 16 and iodepth_low is set to 4, then
702 after fio has filled the queue of 16 requests, it will let
703 the depth drain down to 4 before starting to fill it again.
705 direct=bool If value is true, use non-buffered io. This is usually
706 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
707 On Windows the synchronous ioengines don't support direct io.
709 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
710 writes are guaranteed to be stable once acknowledged by
711 the operating system. Only Linux supports O_ATOMIC right
714 buffered=bool If value is true, use buffered io. This is the opposite
715 of the 'direct' option. Defaults to true.
717 offset=int Start io at the given offset in the file. The data before
718 the given offset will not be touched. This effectively
719 caps the file size at real_size - offset.
721 offset_increment=int If this is provided, then the real offset becomes
722 the offset + offset_increment * thread_number, where the
723 thread number is a counter that starts at 0 and is incremented
724 for each job. This option is useful if there are several jobs
725 which are intended to operate on a file in parallel in disjoint
726 segments, with even spacing between the starting points.
728 number_ios=int Fio will normally perform IOs until it has exhausted the size
729 of the region set by size=, or if it exhaust the allocated
730 time (or hits an error condition). With this setting, the
731 range/size can be set independently of the number of IOs to
732 perform. When fio reaches this number, it will exit normally
735 fsync=int If writing to a file, issue a sync of the dirty data
736 for every number of blocks given. For example, if you give
737 32 as a parameter, fio will sync the file for every 32
738 writes issued. If fio is using non-buffered io, we may
739 not sync the file. The exception is the sg io engine, which
740 synchronizes the disk cache anyway.
742 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
744 In FreeBSD and Windows there is no fdatasync(), this falls back to
747 sync_file_range=str:val Use sync_file_range() for every 'val' number of
748 write operations. Fio will track range of writes that
749 have happened since the last sync_file_range() call. 'str'
750 can currently be one or more of:
752 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
753 write SYNC_FILE_RANGE_WRITE
754 wait_after SYNC_FILE_RANGE_WAIT_AFTER
756 So if you do sync_file_range=wait_before,write:8, fio would
757 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
758 every 8 writes. Also see the sync_file_range(2) man page.
759 This option is Linux specific.
761 overwrite=bool If true, writes to a file will always overwrite existing
762 data. If the file doesn't already exist, it will be
763 created before the write phase begins. If the file exists
764 and is large enough for the specified write phase, nothing
767 end_fsync=bool If true, fsync file contents when a write stage has completed.
769 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
770 This differs from end_fsync in that it will happen on every
771 file close, not just at the end of the job.
773 rwmixread=int How large a percentage of the mix should be reads.
775 rwmixwrite=int How large a percentage of the mix should be writes. If both
776 rwmixread and rwmixwrite is given and the values do not add
777 up to 100%, the latter of the two will be used to override
778 the first. This may interfere with a given rate setting,
779 if fio is asked to limit reads or writes to a certain rate.
780 If that is the case, then the distribution may be skewed.
782 random_distribution=str:float By default, fio will use a completely uniform
783 random distribution when asked to perform random IO. Sometimes
784 it is useful to skew the distribution in specific ways,
785 ensuring that some parts of the data is more hot than others.
786 fio includes the following distribution models:
788 random Uniform random distribution
789 zipf Zipf distribution
790 pareto Pareto distribution
792 When using a zipf or pareto distribution, an input value
793 is also needed to define the access pattern. For zipf, this
794 is the zipf theta. For pareto, it's the pareto power. Fio
795 includes a test program, genzipf, that can be used visualize
796 what the given input values will yield in terms of hit rates.
797 If you wanted to use zipf with a theta of 1.2, you would use
798 random_distribution=zipf:1.2 as the option. If a non-uniform
799 model is used, fio will disable use of the random map.
801 percentage_random=int For a random workload, set how big a percentage should
802 be random. This defaults to 100%, in which case the workload
803 is fully random. It can be set from anywhere from 0 to 100.
804 Setting it to 0 would make the workload fully sequential. Any
805 setting in between will result in a random mix of sequential
806 and random IO, at the given percentages. It is possible to
807 set different values for reads, writes, and trim. To do so,
808 simply use a comma separated list. See blocksize.
810 norandommap Normally fio will cover every block of the file when doing
811 random IO. If this option is given, fio will just get a
812 new random offset without looking at past io history. This
813 means that some blocks may not be read or written, and that
814 some blocks may be read/written more than once. This option
815 is mutually exclusive with verify= if and only if multiple
816 blocksizes (via bsrange=) are used, since fio only tracks
817 complete rewrites of blocks.
819 softrandommap=bool See norandommap. If fio runs with the random block map
820 enabled and it fails to allocate the map, if this option is
821 set it will continue without a random block map. As coverage
822 will not be as complete as with random maps, this option is
825 random_generator=str Fio supports the following engines for generating
826 IO offsets for random IO:
828 tausworthe Strong 2^88 cycle random number generator
829 lfsr Linear feedback shift register generator
831 Tausworthe is a strong random number generator, but it
832 requires tracking on the side if we want to ensure that
833 blocks are only read or written once. LFSR guarantees
834 that we never generate the same offset twice, and it's
835 also less computationally expensive. It's not a true
836 random generator, however, though for IO purposes it's
837 typically good enough. LFSR only works with single
838 block sizes, not with workloads that use multiple block
839 sizes. If used with such a workload, fio may read or write
840 some blocks multiple times.
842 nice=int Run the job with the given nice value. See man nice(2).
844 prio=int Set the io priority value of this job. Linux limits us to
845 a positive value between 0 and 7, with 0 being the highest.
848 prioclass=int Set the io priority class. See man ionice(1).
850 thinktime=int Stall the job x microseconds after an io has completed before
851 issuing the next. May be used to simulate processing being
852 done by an application. See thinktime_blocks and
856 Only valid if thinktime is set - pretend to spend CPU time
857 doing something with the data received, before falling back
858 to sleeping for the rest of the period specified by
862 Only valid if thinktime is set - control how many blocks
863 to issue, before waiting 'thinktime' usecs. If not set,
864 defaults to 1 which will make fio wait 'thinktime' usecs
865 after every block. This effectively makes any queue depth
866 setting redundant, since no more than 1 IO will be queued
867 before we have to complete it and do our thinktime. In
868 other words, this setting effectively caps the queue depth
869 if the latter is larger.
871 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
872 the normal suffix rules apply. You can use rate=500k to limit
873 reads and writes to 500k each, or you can specify read and
874 writes separately. Using rate=1m,500k would limit reads to
875 1MB/sec and writes to 500KB/sec. Capping only reads or
876 writes can be done with rate=,500k or rate=500k,. The former
877 will only limit writes (to 500KB/sec), the latter will only
880 ratemin=int Tell fio to do whatever it can to maintain at least this
881 bandwidth. Failing to meet this requirement, will cause
882 the job to exit. The same format as rate is used for
883 read vs write separation.
885 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
886 as rate, just specified independently of bandwidth. If the
887 job is given a block size range instead of a fixed value,
888 the smallest block size is used as the metric. The same format
889 as rate is used for read vs write separation.
891 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
892 the job to exit. The same format as rate is used for read vs
895 latency_target=int If set, fio will attempt to find the max performance
896 point that the given workload will run at while maintaining a
897 latency below this target. The values is given in microseconds.
898 See latency_window and latency_percentile
900 latency_window=int Used with latency_target to specify the sample window
901 that the job is run at varying queue depths to test the
902 performance. The value is given in microseconds.
904 latency_percentile=float The percentage of IOs that must fall within the
905 criteria specified by latency_target and latency_window. If not
906 set, this defaults to 100.0, meaning that all IOs must be equal
907 or below to the value set by latency_target.
909 max_latency=int If set, fio will exit the job if it exceeds this maximum
910 latency. It will exit with an ETIME error.
912 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
915 cpumask=int Set the CPU affinity of this job. The parameter given is a
916 bitmask of allowed CPU's the job may run on. So if you want
917 the allowed CPUs to be 1 and 5, you would pass the decimal
918 value of (1 << 1 | 1 << 5), or 34. See man
919 sched_setaffinity(2). This may not work on all supported
920 operating systems or kernel versions. This option doesn't
921 work well for a higher CPU count than what you can store in
922 an integer mask, so it can only control cpus 1-32. For
923 boxes with larger CPU counts, use cpus_allowed.
925 cpus_allowed=str Controls the same options as cpumask, but it allows a text
926 setting of the permitted CPUs instead. So to use CPUs 1 and
927 5, you would specify cpus_allowed=1,5. This options also
928 allows a range of CPUs. Say you wanted a binding to CPUs
929 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
931 cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
932 specified by cpus_allowed or cpumask. Two policies are
935 shared All jobs will share the CPU set specified.
936 split Each job will get a unique CPU from the CPU set.
938 'shared' is the default behaviour, if the option isn't
939 specified. If split is specified, then fio will error out if
940 there are more jobs defined than CPUs given in the set.
942 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
943 arguments allow comma delimited list of cpu numbers,
944 A-B ranges, or 'all'. Note, to enable numa options support,
945 fio must be built on a system with libnuma-dev(el) installed.
947 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
948 nodes. Format of the argements:
950 `mode' is one of the following memory policy:
951 default, prefer, bind, interleave, local
952 For `default' and `local' memory policy, no node is
953 needed to be specified.
954 For `prefer', only one node is allowed.
955 For `bind' and `interleave', it allow comma delimited
956 list of numbers, A-B ranges, or 'all'.
958 startdelay=time Start this job the specified number of seconds after fio
959 has started. Only useful if the job file contains several
960 jobs, and you want to delay starting some jobs to a certain
963 runtime=time Tell fio to terminate processing after the specified number
964 of seconds. It can be quite hard to determine for how long
965 a specified job will run, so this parameter is handy to
966 cap the total runtime to a given time.
968 time_based If set, fio will run for the duration of the runtime
969 specified even if the file(s) are completely read or
970 written. It will simply loop over the same workload
971 as many times as the runtime allows.
973 ramp_time=time If set, fio will run the specified workload for this amount
974 of time before logging any performance numbers. Useful for
975 letting performance settle before logging results, thus
976 minimizing the runtime required for stable results. Note
977 that the ramp_time is considered lead in time for a job,
978 thus it will increase the total runtime if a special timeout
979 or runtime is specified.
981 invalidate=bool Invalidate the buffer/page cache parts for this file prior
982 to starting io. Defaults to true.
984 sync=bool Use sync io for buffered writes. For the majority of the
985 io engines, this means using O_SYNC.
988 mem=str Fio can use various types of memory as the io unit buffer.
989 The allowed values are:
991 malloc Use memory from malloc(3) as the buffers.
993 shm Use shared memory as the buffers. Allocated
996 shmhuge Same as shm, but use huge pages as backing.
998 mmap Use mmap to allocate buffers. May either be
999 anonymous memory, or can be file backed if
1000 a filename is given after the option. The
1001 format is mem=mmap:/path/to/file.
1003 mmaphuge Use a memory mapped huge file as the buffer
1004 backing. Append filename after mmaphuge, ala
1005 mem=mmaphuge:/hugetlbfs/file
1007 The area allocated is a function of the maximum allowed
1008 bs size for the job, multiplied by the io depth given. Note
1009 that for shmhuge and mmaphuge to work, the system must have
1010 free huge pages allocated. This can normally be checked
1011 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1012 Linux system. Fio assumes a huge page is 4MB in size. So
1013 to calculate the number of huge pages you need for a given
1014 job file, add up the io depth of all jobs (normally one unless
1015 iodepth= is used) and multiply by the maximum bs set. Then
1016 divide that number by the huge page size. You can see the
1017 size of the huge pages in /proc/meminfo. If no huge pages
1018 are allocated by having a non-zero number in nr_hugepages,
1019 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1021 mmaphuge also needs to have hugetlbfs mounted and the file
1022 location should point there. So if it's mounted in /huge,
1023 you would use mem=mmaphuge:/huge/somefile.
1025 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1026 Note that the given alignment is applied to the first IO unit
1027 buffer, if using iodepth the alignment of the following buffers
1028 are given by the bs used. In other words, if using a bs that is
1029 a multiple of the page sized in the system, all buffers will
1030 be aligned to this value. If using a bs that is not page
1031 aligned, the alignment of subsequent IO memory buffers is the
1032 sum of the iomem_align and bs used.
1035 Defines the size of a huge page. Must at least be equal
1036 to the system setting, see /proc/meminfo. Defaults to 4MB.
1037 Should probably always be a multiple of megabytes, so using
1038 hugepage-size=Xm is the preferred way to set this to avoid
1039 setting a non-pow-2 bad value.
1041 exitall When one job finishes, terminate the rest. The default is
1042 to wait for each job to finish, sometimes that is not the
1045 bwavgtime=int Average the calculated bandwidth over the given time. Value
1046 is specified in milliseconds.
1048 iopsavgtime=int Average the calculated IOPS over the given time. Value
1049 is specified in milliseconds.
1051 create_serialize=bool If true, serialize the file creating for the jobs.
1052 This may be handy to avoid interleaving of data
1053 files, which may greatly depend on the filesystem
1054 used and even the number of processors in the system.
1056 create_fsync=bool fsync the data file after creation. This is the
1059 create_on_open=bool Don't pre-setup the files for IO, just create open()
1060 when it's time to do IO to that file.
1062 create_only=bool If true, fio will only run the setup phase of the job.
1063 If files need to be laid out or updated on disk, only
1064 that will be done. The actual job contents are not
1067 pre_read=bool If this is given, files will be pre-read into memory before
1068 starting the given IO operation. This will also clear
1069 the 'invalidate' flag, since it is pointless to pre-read
1070 and then drop the cache. This will only work for IO engines
1071 that are seekable, since they allow you to read the same data
1072 multiple times. Thus it will not work on eg network or splice
1075 unlink=bool Unlink the job files when done. Not the default, as repeated
1076 runs of that job would then waste time recreating the file
1077 set again and again.
1079 loops=int Run the specified number of iterations of this job. Used
1080 to repeat the same workload a given number of times. Defaults
1083 verify_only Do not perform specified workload---only verify data still
1084 matches previous invocation of this workload. This option
1085 allows one to check data multiple times at a later date
1086 without overwriting it. This option makes sense only for
1087 workloads that write data, and does not support workloads
1088 with the time_based option set.
1090 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1091 verify is set. Defaults to 1.
1093 verify=str If writing to a file, fio can verify the file contents
1094 after each iteration of the job. The allowed values are:
1096 md5 Use an md5 sum of the data area and store
1097 it in the header of each block.
1099 crc64 Use an experimental crc64 sum of the data
1100 area and store it in the header of each
1103 crc32c Use a crc32c sum of the data area and store
1104 it in the header of each block.
1106 crc32c-intel Use hardware assisted crc32c calcuation
1107 provided on SSE4.2 enabled processors. Falls
1108 back to regular software crc32c, if not
1109 supported by the system.
1111 crc32 Use a crc32 sum of the data area and store
1112 it in the header of each block.
1114 crc16 Use a crc16 sum of the data area and store
1115 it in the header of each block.
1117 crc7 Use a crc7 sum of the data area and store
1118 it in the header of each block.
1120 xxhash Use xxhash as the checksum function. Generally
1121 the fastest software checksum that fio
1124 sha512 Use sha512 as the checksum function.
1126 sha256 Use sha256 as the checksum function.
1128 sha1 Use optimized sha1 as the checksum function.
1130 meta Write extra information about each io
1131 (timestamp, block number etc.). The block
1132 number is verified. The io sequence number is
1133 verified for workloads that write data.
1134 See also verify_pattern.
1136 null Only pretend to verify. Useful for testing
1137 internals with ioengine=null, not for much
1140 This option can be used for repeated burn-in tests of a
1141 system to make sure that the written data is also
1142 correctly read back. If the data direction given is
1143 a read or random read, fio will assume that it should
1144 verify a previously written file. If the data direction
1145 includes any form of write, the verify will be of the
1148 verifysort=bool If set, fio will sort written verify blocks when it deems
1149 it faster to read them back in a sorted manner. This is
1150 often the case when overwriting an existing file, since
1151 the blocks are already laid out in the file system. You
1152 can ignore this option unless doing huge amounts of really
1153 fast IO where the red-black tree sorting CPU time becomes
1156 verify_offset=int Swap the verification header with data somewhere else
1157 in the block before writing. Its swapped back before
1160 verify_interval=int Write the verification header at a finer granularity
1161 than the blocksize. It will be written for chunks the
1162 size of header_interval. blocksize should divide this
1165 verify_pattern=str If set, fio will fill the io buffers with this
1166 pattern. Fio defaults to filling with totally random
1167 bytes, but sometimes it's interesting to fill with a known
1168 pattern for io verification purposes. Depending on the
1169 width of the pattern, fio will fill 1/2/3/4 bytes of the
1170 buffer at the time(it can be either a decimal or a hex number).
1171 The verify_pattern if larger than a 32-bit quantity has to
1172 be a hex number that starts with either "0x" or "0X". Use
1175 verify_fatal=bool Normally fio will keep checking the entire contents
1176 before quitting on a block verification failure. If this
1177 option is set, fio will exit the job on the first observed
1180 verify_dump=bool If set, dump the contents of both the original data
1181 block and the data block we read off disk to files. This
1182 allows later analysis to inspect just what kind of data
1183 corruption occurred. Off by default.
1185 verify_async=int Fio will normally verify IO inline from the submitting
1186 thread. This option takes an integer describing how many
1187 async offload threads to create for IO verification instead,
1188 causing fio to offload the duty of verifying IO contents
1189 to one or more separate threads. If using this offload
1190 option, even sync IO engines can benefit from using an
1191 iodepth setting higher than 1, as it allows them to have
1192 IO in flight while verifies are running.
1194 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1195 async IO verification threads. See cpus_allowed for the
1198 verify_backlog=int Fio will normally verify the written contents of a
1199 job that utilizes verify once that job has completed. In
1200 other words, everything is written then everything is read
1201 back and verified. You may want to verify continually
1202 instead for a variety of reasons. Fio stores the meta data
1203 associated with an IO block in memory, so for large
1204 verify workloads, quite a bit of memory would be used up
1205 holding this meta data. If this option is enabled, fio
1206 will write only N blocks before verifying these blocks.
1208 verify_backlog_batch=int Control how many blocks fio will verify
1209 if verify_backlog is set. If not set, will default to
1210 the value of verify_backlog (meaning the entire queue
1211 is read back and verified). If verify_backlog_batch is
1212 less than verify_backlog then not all blocks will be verified,
1213 if verify_backlog_batch is larger than verify_backlog, some
1214 blocks will be verified more than once.
1217 wait_for_previous Wait for preceding jobs in the job file to exit, before
1218 starting this one. Can be used to insert serialization
1219 points in the job file. A stone wall also implies starting
1220 a new reporting group.
1222 new_group Start a new reporting group. See: group_reporting.
1224 numjobs=int Create the specified number of clones of this job. May be
1225 used to setup a larger number of threads/processes doing
1226 the same thing. Each thread is reported separately; to see
1227 statistics for all clones as a whole, use group_reporting in
1228 conjunction with new_group.
1230 group_reporting It may sometimes be interesting to display statistics for
1231 groups of jobs as a whole instead of for each individual job.
1232 This is especially true if 'numjobs' is used; looking at
1233 individual thread/process output quickly becomes unwieldy.
1234 To see the final report per-group instead of per-job, use
1235 'group_reporting'. Jobs in a file will be part of the same
1236 reporting group, unless if separated by a stonewall, or by
1239 thread fio defaults to forking jobs, however if this option is
1240 given, fio will use pthread_create(3) to create threads
1243 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1245 zoneskip=int Skip the specified number of bytes when zonesize data has
1246 been read. The two zone options can be used to only do
1247 io on zones of a file.
1249 write_iolog=str Write the issued io patterns to the specified file. See
1250 read_iolog. Specify a separate file for each job, otherwise
1251 the iologs will be interspersed and the file may be corrupt.
1253 read_iolog=str Open an iolog with the specified file name and replay the
1254 io patterns it contains. This can be used to store a
1255 workload and replay it sometime later. The iolog given
1256 may also be a blktrace binary file, which allows fio
1257 to replay a workload captured by blktrace. See blktrace
1258 for how to capture such logging data. For blktrace replay,
1259 the file needs to be turned into a blkparse binary data
1260 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1262 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1263 is to attempt to respect the time stamps within the log and
1264 replay them with the appropriate delay between IOPS. By
1265 setting this variable fio will not respect the timestamps and
1266 attempt to replay them as fast as possible while still
1267 respecting ordering. The result is the same I/O pattern to a
1268 given device, but different timings.
1270 replay_redirect=str While replaying I/O patterns using read_iolog the
1271 default behavior is to replay the IOPS onto the major/minor
1272 device that each IOP was recorded from. This is sometimes
1273 undesirable because on a different machine those major/minor
1274 numbers can map to a different device. Changing hardware on
1275 the same system can also result in a different major/minor
1276 mapping. Replay_redirect causes all IOPS to be replayed onto
1277 the single specified device regardless of the device it was
1278 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1279 IO in the blktrace to be replayed onto /dev/sdc. This means
1280 multiple devices will be replayed onto a single, if the trace
1281 contains multiple devices. If you want multiple devices to be
1282 replayed concurrently to multiple redirected devices you must
1283 blkparse your trace into separate traces and replay them with
1284 independent fio invocations. Unfortuantely this also breaks
1285 the strict time ordering between multiple device accesses.
1287 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1288 file. Can be used to store data of the bandwidth of the
1289 jobs in their lifetime. The included fio_generate_plots
1290 script uses gnuplot to turn these text files into nice
1291 graphs. See write_lat_log for behaviour of given
1292 filename. For this option, the suffix is _bw.log.
1294 write_lat_log=str Same as write_bw_log, except that this option stores io
1295 submission, completion, and total latencies instead. If no
1296 filename is given with this option, the default filename of
1297 "jobname_type.log" is used. Even if the filename is given,
1298 fio will still append the type of log. So if one specifies
1302 The actual log names will be foo_slat.log, foo_clat.log,
1303 and foo_lat.log. This helps fio_generate_plot fine the logs
1306 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1307 given with this option, the default filename of
1308 "jobname_type.log" is used. Even if the filename is given,
1309 fio will still append the type of log.
1311 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1312 or bw log for every IO that completes. When writing to the
1313 disk log, that can quickly grow to a very large size. Setting
1314 this option makes fio average the each log entry over the
1315 specified period of time, reducing the resolution of the log.
1318 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1319 potentially be used instead of removing memory or booting
1320 with less memory to simulate a smaller amount of memory.
1321 The amount specified is per worker.
1323 exec_prerun=str Before running this job, issue the command specified
1324 through system(3). Output is redirected in a file called
1327 exec_postrun=str After the job completes, issue the command specified
1328 though system(3). Output is redirected in a file called
1329 jobname.postrun.txt.
1331 ioscheduler=str Attempt to switch the device hosting the file to the specified
1332 io scheduler before running.
1334 disk_util=bool Generate disk utilization statistics, if the platform
1335 supports it. Defaults to on.
1337 disable_lat=bool Disable measurements of total latency numbers. Useful
1338 only for cutting back the number of calls to gettimeofday,
1339 as that does impact performance at really high IOPS rates.
1340 Note that to really get rid of a large amount of these
1341 calls, this option must be used with disable_slat and
1344 disable_clat=bool Disable measurements of completion latency numbers. See
1347 disable_slat=bool Disable measurements of submission latency numbers. See
1350 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1353 clat_percentiles=bool Enable the reporting of percentiles of
1354 completion latencies.
1356 percentile_list=float_list Overwrite the default list of percentiles
1357 for completion latencies. Each number is a floating
1358 number in the range (0,100], and the maximum length of
1359 the list is 20. Use ':' to separate the numbers, and
1360 list the numbers in ascending order. For example,
1361 --percentile_list=99.5:99.9 will cause fio to report
1362 the values of completion latency below which 99.5% and
1363 99.9% of the observed latencies fell, respectively.
1365 clocksource=str Use the given clocksource as the base of timing. The
1366 supported options are:
1368 gettimeofday gettimeofday(2)
1370 clock_gettime clock_gettime(2)
1372 cpu Internal CPU clock source
1374 cpu is the preferred clocksource if it is reliable, as it
1375 is very fast (and fio is heavy on time calls). Fio will
1376 automatically use this clocksource if it's supported and
1377 considered reliable on the system it is running on, unless
1378 another clocksource is specifically set. For x86/x86-64 CPUs,
1379 this means supporting TSC Invariant.
1381 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1382 (disable_clat, disable_slat, disable_bw) plus reduce
1383 precision of the timeout somewhat to really shrink
1384 the gettimeofday() call count. With this option enabled,
1385 we only do about 0.4% of the gtod() calls we would have
1386 done if all time keeping was enabled.
1388 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1389 execution to just getting the current time. Fio (and
1390 databases, for instance) are very intensive on gettimeofday()
1391 calls. With this option, you can set one CPU aside for
1392 doing nothing but logging current time to a shared memory
1393 location. Then the other threads/processes that run IO
1394 workloads need only copy that segment, instead of entering
1395 the kernel with a gettimeofday() call. The CPU set aside
1396 for doing these time calls will be excluded from other
1397 uses. Fio will manually clear it from the CPU mask of other
1400 continue_on_error=str Normally fio will exit the job on the first observed
1401 failure. If this option is set, fio will continue the job when
1402 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1403 is exceeded or the I/O size specified is completed. If this
1404 option is used, there are two more stats that are appended,
1405 the total error count and the first error. The error field
1406 given in the stats is the first error that was hit during the
1409 The allowed values are:
1411 none Exit on any IO or verify errors.
1413 read Continue on read errors, exit on all others.
1415 write Continue on write errors, exit on all others.
1417 io Continue on any IO error, exit on all others.
1419 verify Continue on verify errors, exit on all others.
1421 all Continue on all errors.
1423 0 Backward-compatible alias for 'none'.
1425 1 Backward-compatible alias for 'all'.
1427 ignore_error=str Sometimes you want to ignore some errors during test
1428 in that case you can specify error list for each error type.
1429 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1430 errors for given error type is separated with ':'. Error
1431 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1433 ignore_error=EAGAIN,ENOSPC:122
1434 This option will ignore EAGAIN from READ, and ENOSPC and
1435 122(EDQUOT) from WRITE.
1437 error_dump=bool If set dump every error even if it is non fatal, true
1438 by default. If disabled only fatal error will be dumped
1440 cgroup=str Add job to this control group. If it doesn't exist, it will
1441 be created. The system must have a mounted cgroup blkio
1442 mount point for this to work. If your system doesn't have it
1443 mounted, you can do so with:
1445 # mount -t cgroup -o blkio none /cgroup
1447 cgroup_weight=int Set the weight of the cgroup to this value. See
1448 the documentation that comes with the kernel, allowed values
1449 are in the range of 100..1000.
1451 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1452 the job completion. To override this behavior and to leave
1453 cgroups around after the job completion, set cgroup_nodelete=1.
1454 This can be useful if one wants to inspect various cgroup
1455 files after job completion. Default: false
1457 uid=int Instead of running as the invoking user, set the user ID to
1458 this value before the thread/process does any work.
1460 gid=int Set group ID, see uid.
1462 flow_id=int The ID of the flow. If not specified, it defaults to being a
1463 global flow. See flow.
1465 flow=int Weight in token-based flow control. If this value is used, then
1466 there is a 'flow counter' which is used to regulate the
1467 proportion of activity between two or more jobs. fio attempts
1468 to keep this flow counter near zero. The 'flow' parameter
1469 stands for how much should be added or subtracted to the flow
1470 counter on each iteration of the main I/O loop. That is, if
1471 one job has flow=8 and another job has flow=-1, then there
1472 will be a roughly 1:8 ratio in how much one runs vs the other.
1474 flow_watermark=int The maximum value that the absolute value of the flow
1475 counter is allowed to reach before the job must wait for a
1476 lower value of the counter.
1478 flow_sleep=int The period of time, in microseconds, to wait after the flow
1479 watermark has been exceeded before retrying operations
1481 In addition, there are some parameters which are only valid when a specific
1482 ioengine is in use. These are used identically to normal parameters, with the
1483 caveat that when used on the command line, they must come after the ioengine
1484 that defines them is selected.
1486 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1487 the io_getevents system call to reap newly returned events.
1488 With this flag turned on, the AIO ring will be read directly
1489 from user-space to reap events. The reaping mode is only
1490 enabled when polling for a minimum of 0 events (eg when
1491 iodepth_batch_complete=0).
1493 [cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1495 [cpu] cpuchunks=int Split the load into cycles of the given time. In
1498 [netsplice] hostname=str
1499 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1500 If the job is a TCP listener or UDP reader, the hostname is not
1501 used and must be omitted unless it is a valid UDP multicast
1504 [netsplice] port=int
1505 [net] port=int The TCP or UDP port to bind to or connect to.
1507 [netsplice] interface=str
1508 [net] interface=str The IP address of the network interface used to send or
1509 receive UDP multicast
1512 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1515 [netsplice] nodelay=bool
1516 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1518 [netsplice] protocol=str
1519 [netsplice] proto=str
1521 [net] proto=str The network protocol to use. Accepted values are:
1523 tcp Transmission control protocol
1524 tcpv6 Transmission control protocol V6
1525 udp User datagram protocol
1526 udpv6 User datagram protocol V6
1527 unix UNIX domain socket
1529 When the protocol is TCP or UDP, the port must also be given,
1530 as well as the hostname if the job is a TCP listener or UDP
1531 reader. For unix sockets, the normal filename option should be
1532 used and the port is invalid.
1534 [net] listen For TCP network connections, tell fio to listen for incoming
1535 connections rather than initiating an outgoing connection. The
1536 hostname must be omitted if this option is used.
1537 [net] pingpong Normaly a network writer will just continue writing data, and
1538 a network reader will just consume packages. If pingpong=1
1539 is set, a writer will send its normal payload to the reader,
1540 then wait for the reader to send the same payload back. This
1541 allows fio to measure network latencies. The submission
1542 and completion latencies then measure local time spent
1543 sending or receiving, and the completion latency measures
1544 how long it took for the other end to receive and send back.
1545 For UDP multicast traffic pingpong=1 should only be set for a
1546 single reader when multiple readers are listening to the same
1549 [e4defrag] donorname=str
1550 File will be used as a block donor(swap extents between files)
1551 [e4defrag] inplace=int
1552 Configure donor file blocks allocation strategy
1553 0(default): Preallocate donor's file on init
1554 1 : allocate space immidietly inside defragment event,
1555 and free right after event
1559 6.0 Interpreting the output
1560 ---------------------------
1562 fio spits out a lot of output. While running, fio will display the
1563 status of the jobs created. An example of that would be:
1565 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1567 The characters inside the square brackets denote the current status of
1568 each thread. The possible values (in typical life cycle order) are:
1572 P Thread setup, but not started.
1574 I Thread initialized, waiting or generating necessary data.
1575 p Thread running pre-reading file(s).
1576 R Running, doing sequential reads.
1577 r Running, doing random reads.
1578 W Running, doing sequential writes.
1579 w Running, doing random writes.
1580 M Running, doing mixed sequential reads/writes.
1581 m Running, doing mixed random reads/writes.
1582 F Running, currently waiting for fsync()
1583 V Running, doing verification of written data.
1584 E Thread exited, not reaped by main thread yet.
1586 X Thread reaped, exited with an error.
1587 K Thread reaped, exited due to signal.
1589 The other values are fairly self explanatory - number of threads
1590 currently running and doing io, rate of io since last check (read speed
1591 listed first, then write speed), and the estimated completion percentage
1592 and time for the running group. It's impossible to estimate runtime of
1593 the following groups (if any). Note that the string is displayed in order,
1594 so it's possible to tell which of the jobs are currently doing what. The
1595 first character is the first job defined in the job file, and so forth.
1597 When fio is done (or interrupted by ctrl-c), it will show the data for
1598 each thread, group of threads, and disks in that order. For each data
1599 direction, the output looks like:
1601 Client1 (g=0): err= 0:
1602 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1603 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1604 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1605 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1606 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1607 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1608 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1609 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1610 issued r/w: total=0/32768, short=0/0
1611 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1612 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1614 The client number is printed, along with the group id and error of that
1615 thread. Below is the io statistics, here for writes. In the order listed,
1618 io= Number of megabytes io performed
1619 bw= Average bandwidth rate
1620 iops= Average IOs performed per second
1621 runt= The runtime of that thread
1622 slat= Submission latency (avg being the average, stdev being the
1623 standard deviation). This is the time it took to submit
1624 the io. For sync io, the slat is really the completion
1625 latency, since queue/complete is one operation there. This
1626 value can be in milliseconds or microseconds, fio will choose
1627 the most appropriate base and print that. In the example
1628 above, milliseconds is the best scale. Note: in --minimal mode
1629 latencies are always expressed in microseconds.
1630 clat= Completion latency. Same names as slat, this denotes the
1631 time from submission to completion of the io pieces. For
1632 sync io, clat will usually be equal (or very close) to 0,
1633 as the time from submit to complete is basically just
1634 CPU time (io has already been done, see slat explanation).
1635 bw= Bandwidth. Same names as the xlat stats, but also includes
1636 an approximate percentage of total aggregate bandwidth
1637 this thread received in this group. This last value is
1638 only really useful if the threads in this group are on the
1639 same disk, since they are then competing for disk access.
1640 cpu= CPU usage. User and system time, along with the number
1641 of context switches this thread went through, usage of
1642 system and user time, and finally the number of major
1643 and minor page faults.
1644 IO depths= The distribution of io depths over the job life time. The
1645 numbers are divided into powers of 2, so for example the
1646 16= entries includes depths up to that value but higher
1647 than the previous entry. In other words, it covers the
1648 range from 16 to 31.
1649 IO submit= How many pieces of IO were submitting in a single submit
1650 call. Each entry denotes that amount and below, until
1651 the previous entry - eg, 8=100% mean that we submitted
1652 anywhere in between 5-8 ios per submit call.
1653 IO complete= Like the above submit number, but for completions instead.
1654 IO issued= The number of read/write requests issued, and how many
1656 IO latencies= The distribution of IO completion latencies. This is the
1657 time from when IO leaves fio and when it gets completed.
1658 The numbers follow the same pattern as the IO depths,
1659 meaning that 2=1.6% means that 1.6% of the IO completed
1660 within 2 msecs, 20=12.8% means that 12.8% of the IO
1661 took more than 10 msecs, but less than (or equal to) 20 msecs.
1663 After each client has been listed, the group statistics are printed. They
1664 will look like this:
1666 Run status group 0 (all jobs):
1667 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1668 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1670 For each data direction, it prints:
1672 io= Number of megabytes io performed.
1673 aggrb= Aggregate bandwidth of threads in this group.
1674 minb= The minimum average bandwidth a thread saw.
1675 maxb= The maximum average bandwidth a thread saw.
1676 mint= The smallest runtime of the threads in that group.
1677 maxt= The longest runtime of the threads in that group.
1679 And finally, the disk statistics are printed. They will look like this:
1681 Disk stats (read/write):
1682 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1684 Each value is printed for both reads and writes, with reads first. The
1687 ios= Number of ios performed by all groups.
1688 merge= Number of merges io the io scheduler.
1689 ticks= Number of ticks we kept the disk busy.
1690 io_queue= Total time spent in the disk queue.
1691 util= The disk utilization. A value of 100% means we kept the disk
1692 busy constantly, 50% would be a disk idling half of the time.
1694 It is also possible to get fio to dump the current output while it is
1695 running, without terminating the job. To do that, send fio the USR1 signal.
1696 You can also get regularly timed dumps by using the --status-interval
1697 parameter, or by creating a file in /tmp named fio-dump-status. If fio
1698 sees this file, it will unlink it and dump the current output status.
1704 For scripted usage where you typically want to generate tables or graphs
1705 of the results, fio can output the results in a semicolon separated format.
1706 The format is one long line of values, such as:
1708 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%
1709 A description of this job goes here.
1711 The job description (if provided) follows on a second line.
1713 To enable terse output, use the --minimal command line option. The first
1714 value is the version of the terse output format. If the output has to
1715 be changed for some reason, this number will be incremented by 1 to
1716 signify that change.
1718 Split up, the format is as follows:
1720 terse version, fio version, jobname, groupid, error
1722 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1723 Submission latency: min, max, mean, deviation (usec)
1724 Completion latency: min, max, mean, deviation (usec)
1725 Completion latency percentiles: 20 fields (see below)
1726 Total latency: min, max, mean, deviation (usec)
1727 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1729 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1730 Submission latency: min, max, mean, deviation (usec)
1731 Completion latency: min, max, mean, deviation (usec)
1732 Completion latency percentiles: 20 fields (see below)
1733 Total latency: min, max, mean, deviation (usec)
1734 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1735 CPU usage: user, system, context switches, major faults, minor faults
1736 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1737 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1738 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1739 Disk utilization: Disk name, Read ios, write ios,
1740 Read merges, write merges,
1741 Read ticks, write ticks,
1742 Time spent in queue, disk utilization percentage
1743 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
1745 Additional Info (dependent on description being set): Text description
1747 Completion latency percentiles can be a grouping of up to 20 sets, so
1748 for the terse output fio writes all of them. Each field will look like this:
1752 which is the Xth percentile, and the usec latency associated with it.
1754 For disk utilization, all disks used by fio are shown. So for each disk
1755 there will be a disk utilization section.
1758 8.0 Trace file format
1759 ---------------------
1760 There are two trace file format that you can encounter. The older (v1) format
1761 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1762 below in case that you get an old trace and want to understand it.
1764 In any case the trace is a simple text file with a single action per line.
1767 8.1 Trace file format v1
1768 ------------------------
1769 Each line represents a single io action in the following format:
1773 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1775 This format is not supported in Fio versions => 1.20-rc3.
1778 8.2 Trace file format v2
1779 ------------------------
1780 The second version of the trace file format was added in Fio version 1.17.
1781 It allows to access more then one file per trace and has a bigger set of
1782 possible file actions.
1784 The first line of the trace file has to be:
1788 Following this can be lines in two different formats, which are described below.
1790 The file management format:
1794 The filename is given as an absolute path. The action can be one of these:
1796 add Add the given filename to the trace
1797 open Open the file with the given filename. The filename has to have
1798 been added with the add action before.
1799 close Close the file with the given filename. The file has to have been
1803 The file io action format:
1805 filename action offset length
1807 The filename is given as an absolute path, and has to have been added and opened
1808 before it can be used with this format. The offset and length are given in
1809 bytes. The action can be one of these:
1811 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1812 read Read 'length' bytes beginning from 'offset'
1813 write Write 'length' bytes beginning from 'offset'
1814 sync fsync() the file
1815 datasync fdatasync() the file
1816 trim trim the given file from the given 'offset' for 'length' bytes
1819 9.0 CPU idleness profiling
1820 --------------------------
1821 In some cases, we want to understand CPU overhead in a test. For example,
1822 we test patches for the specific goodness of whether they reduce CPU usage.
1823 fio implements a balloon approach to create a thread per CPU that runs at
1824 idle priority, meaning that it only runs when nobody else needs the cpu.
1825 By measuring the amount of work completed by the thread, idleness of each
1826 CPU can be derived accordingly.
1828 An unit work is defined as touching a full page of unsigned characters. Mean
1829 and standard deviation of time to complete an unit work is reported in "unit
1830 work" section. Options can be chosen to report detailed percpu idleness or
1831 overall system idleness by aggregating percpu stats.