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,
227 int SI integer. A whole number value, which may contain a suffix
228 describing the base of the number. Accepted suffixes are k/m/g/t/p,
229 meaning kilo, mega, giga, tera, and peta. The suffix is not case
230 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
231 as 'k'). So if you want to specify 4096, you could either write
232 out '4096' or just give 4k. The suffixes signify base 2 values, so
233 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
234 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
235 case, then 1000 is used as the multiplier. This can be handy for
236 disks, since manufacturers generally use base 10 values when listing
237 the capacity of a drive. If the option accepts an upper and lower
238 range, use a colon ':' or minus '-' to separate such values. May also
239 include a prefix to indicate numbers base. If 0x is used, the number
240 is assumed to be hexadecimal. See irange.
241 bool Boolean. Usually parsed as an integer, however only defined for
242 true and false (1 and 0).
243 irange Integer range with suffix. Allows value range to be given, such
244 as 1024-4096. A colon may also be used as the separator, eg
245 1k:4k. If the option allows two sets of ranges, they can be
246 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
248 float_list A list of floating numbers, separated by a ':' character.
250 With the above in mind, here follows the complete list of fio job
253 name=str ASCII name of the job. This may be used to override the
254 name printed by fio for this job. Otherwise the job
255 name is used. On the command line this parameter has the
256 special purpose of also signaling the start of a new
259 description=str Text description of the job. Doesn't do anything except
260 dump this text description when this job is run. It's
263 directory=str Prefix filenames with this directory. Used to place files
264 in a different location than "./".
266 filename=str Fio normally makes up a filename based on the job name,
267 thread number, and file number. If you want to share
268 files between threads in a job or several jobs, specify
269 a filename for each of them to override the default. If
270 the ioengine used is 'net', the filename is the host, port,
271 and protocol to use in the format of =host,port,protocol.
272 See ioengine=net for more. If the ioengine is file based, you
273 can specify a number of files by separating the names with a
274 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
275 as the two working files, you would use
276 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
277 accessed as \\.\PhysicalDrive0 for the first device,
278 \\.\PhysicalDrive1 for the second etc. Note: Windows and
279 FreeBSD prevent write access to areas of the disk containing
280 in-use data (e.g. filesystems).
281 If the wanted filename does need to include a colon, then
282 escape that with a '\' character. For instance, if the filename
283 is "/dev/dsk/foo@3,0:c", then you would use
284 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
285 stdin or stdout. Which of the two depends on the read/write
288 opendir=str Tell fio to recursively add any file it can find in this
289 directory and down the file system tree.
291 lockfile=str Fio defaults to not locking any files before it does
292 IO to them. If a file or file descriptor is shared, fio
293 can serialize IO to that file to make the end result
294 consistent. This is usual for emulating real workloads that
295 share files. The lock modes are:
297 none No locking. The default.
298 exclusive Only one thread/process may do IO,
299 excluding all others.
300 readwrite Read-write locking on the file. Many
301 readers may access the file at the
302 same time, but writes get exclusive
305 The option may be post-fixed with a lock batch number. If
306 set, then each thread/process may do that amount of IOs to
307 the file before giving up the lock. Since lock acquisition is
308 expensive, batching the lock/unlocks will speed up IO.
311 rw=str Type of io pattern. Accepted values are:
313 read Sequential reads
314 write Sequential writes
315 randwrite Random writes
316 randread Random reads
317 rw,readwrite Sequential mixed reads and writes
318 randrw Random mixed reads and writes
320 For the mixed io types, the default is to split them 50/50.
321 For certain types of io the result may still be skewed a bit,
322 since the speed may be different. It is possible to specify
323 a number of IO's to do before getting a new offset, this is
324 one by appending a ':<nr>' to the end of the string given.
325 For a random read, it would look like 'rw=randread:8' for
326 passing in an offset modifier with a value of 8. If the
327 suffix is used with a sequential IO pattern, then the value
328 specified will be added to the generated offset for each IO.
329 For instance, using rw=write:4k will skip 4k for every
330 write. It turns sequential IO into sequential IO with holes.
331 See the 'rw_sequencer' option.
333 rw_sequencer=str If an offset modifier is given by appending a number to
334 the rw=<str> line, then this option controls how that
335 number modifies the IO offset being generated. Accepted
338 sequential Generate sequential offset
339 identical Generate the same offset
341 'sequential' is only useful for random IO, where fio would
342 normally generate a new random offset for every IO. If you
343 append eg 8 to randread, you would get a new random offset for
344 every 8 IO's. The result would be a seek for only every 8
345 IO's, instead of for every IO. Use rw=randread:8 to specify
346 that. As sequential IO is already sequential, setting
347 'sequential' for that would not result in any differences.
348 'identical' behaves in a similar fashion, except it sends
349 the same offset 8 number of times before generating a new
352 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
353 Storage manufacturers like to use 10^3 or 1000 as a base
354 ten unit instead, for obvious reasons. Allow values are
355 1024 or 1000, with 1024 being the default.
357 unified_rw_reporting=bool Fio normally reports statistics on a per
358 data direction basis, meaning that read, write, and trim are
359 accounted and reported separately. If this option is set,
360 the fio will sum the results and report them as "mixed"
363 randrepeat=bool For random IO workloads, seed the generator in a predictable
364 way so that results are repeatable across repetitions.
366 use_os_rand=bool Fio can either use the random generator supplied by the OS
367 to generator random offsets, or it can use it's own internal
368 generator (based on Tausworthe). Default is to use the
369 internal generator, which is often of better quality and
372 fallocate=str Whether pre-allocation is performed when laying down files.
375 none Do not pre-allocate space
376 posix Pre-allocate via posix_fallocate()
377 keep Pre-allocate via fallocate() with
378 FALLOC_FL_KEEP_SIZE set
379 0 Backward-compatible alias for 'none'
380 1 Backward-compatible alias for 'posix'
382 May not be available on all supported platforms. 'keep' is only
383 available on Linux.If using ZFS on Solaris this must be set to
384 'none' because ZFS doesn't support it. Default: 'posix'.
386 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
387 on what IO patterns it is likely to issue. Sometimes you
388 want to test specific IO patterns without telling the
389 kernel about it, in which case you can disable this option.
390 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
391 IO and POSIX_FADV_RANDOM for random IO.
393 size=int The total size of file io for this job. Fio will run until
394 this many bytes has been transferred, unless runtime is
395 limited by other options (such as 'runtime', for instance).
396 Unless specific nrfiles and filesize options are given,
397 fio will divide this size between the available files
398 specified by the job. If not set, fio will use the full
399 size of the given files or devices. If the the files
400 do not exist, size must be given. It is also possible to
401 give size as a percentage between 1 and 100. If size=20%
402 is given, fio will use 20% of the full size of the given
405 filesize=int Individual file sizes. May be a range, in which case fio
406 will select sizes for files at random within the given range
407 and limited to 'size' in total (if that is given). If not
408 given, each created file is the same size.
411 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
412 space left on device) as the terminating condition. Only makes
413 sense with sequential write. For a read workload, the mount
414 point will be filled first then IO started on the result. This
415 option doesn't make sense if operating on a raw device node,
416 since the size of that is already known by the file system.
417 Additionally, writing beyond end-of-device will not return
421 bs=int The block size used for the io units. Defaults to 4k. Values
422 can be given for both read and writes. If a single int is
423 given, it will apply to both. If a second int is specified
424 after a comma, it will apply to writes only. In other words,
425 the format is either bs=read_and_write or bs=read,write.
426 bs=4k,8k will thus use 4k blocks for reads, and 8k blocks
427 for writes. If you only wish to set the write size, you
428 can do so by passing an empty read size - bs=,8k will set
429 8k for writes and leave the read default value.
432 ba=int At what boundary to align random IO offsets. Defaults to
433 the same as 'blocksize' the minimum blocksize given.
434 Minimum alignment is typically 512b for using direct IO,
435 though it usually depends on the hardware block size. This
436 option is mutually exclusive with using a random map for
437 files, so it will turn off that option.
439 blocksize_range=irange
440 bsrange=irange Instead of giving a single block size, specify a range
441 and fio will mix the issued io block sizes. The issued
442 io unit will always be a multiple of the minimum value
443 given (also see bs_unaligned). Applies to both reads and
444 writes, however a second range can be given after a comma.
447 bssplit=str Sometimes you want even finer grained control of the
448 block sizes issued, not just an even split between them.
449 This option allows you to weight various block sizes,
450 so that you are able to define a specific amount of
451 block sizes issued. The format for this option is:
453 bssplit=blocksize/percentage:blocksize/percentage
455 for as many block sizes as needed. So if you want to define
456 a workload that has 50% 64k blocks, 10% 4k blocks, and
457 40% 32k blocks, you would write:
459 bssplit=4k/10:64k/50:32k/40
461 Ordering does not matter. If the percentage is left blank,
462 fio will fill in the remaining values evenly. So a bssplit
463 option like this one:
465 bssplit=4k/50:1k/:32k/
467 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
468 always add up to 100, if bssplit is given a range that adds
469 up to more, it will error out.
471 bssplit also supports giving separate splits to reads and
472 writes. The format is identical to what bs= accepts. You
473 have to separate the read and write parts with a comma. So
474 if you want a workload that has 50% 2k reads and 50% 4k reads,
475 while having 90% 4k writes and 10% 8k writes, you would
478 bssplit=2k/50:4k/50,4k/90,8k/10
481 bs_unaligned If this option is given, any byte size value within bsrange
482 may be used as a block range. This typically wont work with
483 direct IO, as that normally requires sector alignment.
485 zero_buffers If this option is given, fio will init the IO buffers to
486 all zeroes. The default is to fill them with random data.
488 refill_buffers If this option is given, fio will refill the IO buffers
489 on every submit. The default is to only fill it at init
490 time and reuse that data. Only makes sense if zero_buffers
491 isn't specified, naturally. If data verification is enabled,
492 refill_buffers is also automatically enabled.
494 scramble_buffers=bool If refill_buffers is too costly and the target is
495 using data deduplication, then setting this option will
496 slightly modify the IO buffer contents to defeat normal
497 de-dupe attempts. This is not enough to defeat more clever
498 block compression attempts, but it will stop naive dedupe of
499 blocks. Default: true.
501 buffer_compress_percentage=int If this is set, then fio will attempt to
502 provide IO buffer content (on WRITEs) that compress to
503 the specified level. Fio does this by providing a mix of
504 random data and zeroes. Note that this is per block size
505 unit, for file/disk wide compression level that matches
506 this setting, you'll also want to set refill_buffers.
508 buffer_compress_chunk=int See buffer_compress_percentage. This
509 setting allows fio to manage how big the ranges of random
510 data and zeroed data is. Without this set, fio will
511 provide buffer_compress_percentage of blocksize random
512 data, followed by the remaining zeroed. With this set
513 to some chunk size smaller than the block size, fio can
514 alternate random and zeroed data throughout the IO
517 nrfiles=int Number of files to use for this job. Defaults to 1.
519 openfiles=int Number of files to keep open at the same time. Defaults to
520 the same as nrfiles, can be set smaller to limit the number
523 file_service_type=str Defines how fio decides which file from a job to
524 service next. The following types are defined:
526 random Just choose a file at random.
528 roundrobin Round robin over open files. This
531 sequential Finish one file before moving on to
532 the next. Multiple files can still be
533 open depending on 'openfiles'.
535 The string can have a number appended, indicating how
536 often to switch to a new file. So if option random:4 is
537 given, fio will switch to a new random file after 4 ios
540 ioengine=str Defines how the job issues io to the file. The following
543 sync Basic read(2) or write(2) io. lseek(2) is
544 used to position the io location.
546 psync Basic pread(2) or pwrite(2) io.
548 vsync Basic readv(2) or writev(2) IO.
550 libaio Linux native asynchronous io. Note that Linux
551 may only support queued behaviour with
552 non-buffered IO (set direct=1 or buffered=0).
553 This engine defines engine specific options.
555 posixaio glibc posix asynchronous io.
557 solarisaio Solaris native asynchronous io.
559 windowsaio Windows native asynchronous io.
561 mmap File is memory mapped and data copied
562 to/from using memcpy(3).
564 splice splice(2) is used to transfer the data and
565 vmsplice(2) to transfer data from user
568 syslet-rw Use the syslet system calls to make
569 regular read/write async.
571 sg SCSI generic sg v3 io. May either be
572 synchronous using the SG_IO ioctl, or if
573 the target is an sg character device
574 we use read(2) and write(2) for asynchronous
577 null Doesn't transfer any data, just pretends
578 to. This is mainly used to exercise fio
579 itself and for debugging/testing purposes.
581 net Transfer over the network to given host:port.
582 Depending on the protocol used, the hostname,
583 port, listen and filename options are used to
584 specify what sort of connection to make, while
585 the protocol option determines which protocol
587 This engine defines engine specific options.
589 netsplice Like net, but uses splice/vmsplice to
590 map data and send/receive.
591 This engine defines engine specific options.
593 cpuio Doesn't transfer any data, but burns CPU
594 cycles according to the cpuload= and
595 cpucycle= options. Setting cpuload=85
596 will cause that job to do nothing but burn
597 85% of the CPU. In case of SMP machines,
598 use numjobs=<no_of_cpu> to get desired CPU
599 usage, as the cpuload only loads a single
600 CPU at the desired rate.
602 guasi The GUASI IO engine is the Generic Userspace
603 Asyncronous Syscall Interface approach
606 http://www.xmailserver.org/guasi-lib.html
608 for more info on GUASI.
610 rdma The RDMA I/O engine supports both RDMA
611 memory semantics (RDMA_WRITE/RDMA_READ) and
612 channel semantics (Send/Recv) for the
613 InfiniBand, RoCE and iWARP protocols.
615 falloc IO engine that does regular fallocate to
616 simulate data transfer as fio ioengine.
617 DDIR_READ does fallocate(,mode = keep_size,)
618 DDIR_WRITE does fallocate(,mode = 0)
619 DDIR_TRIM does fallocate(,mode = punch_hole)
621 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
622 ioctls to simulate defragment activity in
623 request to DDIR_WRITE event
625 external Prefix to specify loading an external
626 IO engine object file. Append the engine
627 filename, eg ioengine=external:/tmp/foo.o
628 to load ioengine foo.o in /tmp.
630 iodepth=int This defines how many io units to keep in flight against
631 the file. The default is 1 for each file defined in this
632 job, can be overridden with a larger value for higher
633 concurrency. Note that increasing iodepth beyond 1 will not
634 affect synchronous ioengines (except for small degress when
635 verify_async is in use). Even async engines may impose OS
636 restrictions causing the desired depth not to be achieved.
637 This may happen on Linux when using libaio and not setting
638 direct=1, since buffered IO is not async on that OS. Keep an
639 eye on the IO depth distribution in the fio output to verify
640 that the achieved depth is as expected. Default: 1.
642 iodepth_batch_submit=int
643 iodepth_batch=int This defines how many pieces of IO to submit at once.
644 It defaults to 1 which means that we submit each IO
645 as soon as it is available, but can be raised to submit
646 bigger batches of IO at the time.
648 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
649 at once. It defaults to 1 which means that we'll ask
650 for a minimum of 1 IO in the retrieval process from
651 the kernel. The IO retrieval will go on until we
652 hit the limit set by iodepth_low. If this variable is
653 set to 0, then fio will always check for completed
654 events before queuing more IO. This helps reduce
655 IO latency, at the cost of more retrieval system calls.
657 iodepth_low=int The low water mark indicating when to start filling
658 the queue again. Defaults to the same as iodepth, meaning
659 that fio will attempt to keep the queue full at all times.
660 If iodepth is set to eg 16 and iodepth_low is set to 4, then
661 after fio has filled the queue of 16 requests, it will let
662 the depth drain down to 4 before starting to fill it again.
664 direct=bool If value is true, use non-buffered io. This is usually
665 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
666 On Windows the synchronous ioengines don't support direct io.
668 buffered=bool If value is true, use buffered io. This is the opposite
669 of the 'direct' option. Defaults to true.
671 offset=int Start io at the given offset in the file. The data before
672 the given offset will not be touched. This effectively
673 caps the file size at real_size - offset.
675 offset_increment=int If this is provided, then the real offset becomes
676 the offset + offset_increment * thread_number, where the
677 thread number is a counter that starts at 0 and is incremented
678 for each job. This option is useful if there are several jobs
679 which are intended to operate on a file in parallel in disjoint
680 segments, with even spacing between the starting points.
682 fsync=int If writing to a file, issue a sync of the dirty data
683 for every number of blocks given. For example, if you give
684 32 as a parameter, fio will sync the file for every 32
685 writes issued. If fio is using non-buffered io, we may
686 not sync the file. The exception is the sg io engine, which
687 synchronizes the disk cache anyway.
689 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
691 In FreeBSD and Windows there is no fdatasync(), this falls back to
694 sync_file_range=str:val Use sync_file_range() for every 'val' number of
695 write operations. Fio will track range of writes that
696 have happened since the last sync_file_range() call. 'str'
697 can currently be one or more of:
699 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
700 write SYNC_FILE_RANGE_WRITE
701 wait_after SYNC_FILE_RANGE_WAIT_AFTER
703 So if you do sync_file_range=wait_before,write:8, fio would
704 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
705 every 8 writes. Also see the sync_file_range(2) man page.
706 This option is Linux specific.
708 overwrite=bool If true, writes to a file will always overwrite existing
709 data. If the file doesn't already exist, it will be
710 created before the write phase begins. If the file exists
711 and is large enough for the specified write phase, nothing
714 end_fsync=bool If true, fsync file contents when a write stage has completed.
716 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
717 This differs from end_fsync in that it will happen on every
718 file close, not just at the end of the job.
720 rwmixread=int How large a percentage of the mix should be reads.
722 rwmixwrite=int How large a percentage of the mix should be writes. If both
723 rwmixread and rwmixwrite is given and the values do not add
724 up to 100%, the latter of the two will be used to override
725 the first. This may interfere with a given rate setting,
726 if fio is asked to limit reads or writes to a certain rate.
727 If that is the case, then the distribution may be skewed.
729 random_distribution=str:float By default, fio will use a completely uniform
730 random distribution when asked to perform random IO. Sometimes
731 it is useful to skew the distribution in specific ways,
732 ensuring that some parts of the data is more hot than others.
733 fio includes the following distribution models:
735 random Uniform random distribution
736 zipf Zipf distribution
737 pareto Pareto distribution
739 When using a zipf or pareto distribution, an input value
740 is also needed to define the access pattern. For zipf, this
741 is the zipf theta. For pareto, it's the pareto power. Fio
742 includes a test program, genzipf, that can be used visualize
743 what the given input values will yield in terms of hit rates.
744 If you wanted to use zipf with a theta of 1.2, you would use
745 random_distribution=zipf:1.2 as the option. If a non-uniform
746 model is used, fio will disable use of the random map.
748 norandommap Normally fio will cover every block of the file when doing
749 random IO. If this option is given, fio will just get a
750 new random offset without looking at past io history. This
751 means that some blocks may not be read or written, and that
752 some blocks may be read/written more than once. This option
753 is mutually exclusive with verify= if and only if multiple
754 blocksizes (via bsrange=) are used, since fio only tracks
755 complete rewrites of blocks.
757 softrandommap=bool See norandommap. If fio runs with the random block map
758 enabled and it fails to allocate the map, if this option is
759 set it will continue without a random block map. As coverage
760 will not be as complete as with random maps, this option is
763 random_generator=str Fio supports the following engines for generating
764 IO offsets for random IO:
766 tausworthe Strong 2^88 cycle random number generator
767 lfsr Linear feedback shift register generator
769 Tausworthe is a strong random number generator, but it
770 requires tracking on the side if we want to ensure that
771 blocks are only read or written once. LFSR guarantees
772 that we never generate the same offset twice, and it's
773 also less computationally expensive. It's not a true
774 random generator, however, though for IO purposes it's
775 typically good enough. LFSR only works with single
776 block sizes, not with workloads that use multiple block
777 sizes. If used with such a workload, fio may read or write
778 some blocks multiple times.
780 nice=int Run the job with the given nice value. See man nice(2).
782 prio=int Set the io priority value of this job. Linux limits us to
783 a positive value between 0 and 7, with 0 being the highest.
786 prioclass=int Set the io priority class. See man ionice(1).
788 thinktime=int Stall the job x microseconds after an io has completed before
789 issuing the next. May be used to simulate processing being
790 done by an application. See thinktime_blocks and
794 Only valid if thinktime is set - pretend to spend CPU time
795 doing something with the data received, before falling back
796 to sleeping for the rest of the period specified by
800 Only valid if thinktime is set - control how many blocks
801 to issue, before waiting 'thinktime' usecs. If not set,
802 defaults to 1 which will make fio wait 'thinktime' usecs
805 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
806 the normal suffix rules apply. You can use rate=500k to limit
807 reads and writes to 500k each, or you can specify read and
808 writes separately. Using rate=1m,500k would limit reads to
809 1MB/sec and writes to 500KB/sec. Capping only reads or
810 writes can be done with rate=,500k or rate=500k,. The former
811 will only limit writes (to 500KB/sec), the latter will only
814 ratemin=int Tell fio to do whatever it can to maintain at least this
815 bandwidth. Failing to meet this requirement, will cause
816 the job to exit. The same format as rate is used for
817 read vs write separation.
819 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
820 as rate, just specified independently of bandwidth. If the
821 job is given a block size range instead of a fixed value,
822 the smallest block size is used as the metric. The same format
823 as rate is used for read vs write seperation.
825 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
826 the job to exit. The same format as rate is used for read vs
829 max_latency=int If set, fio will exit the job if it exceeds this maximum
830 latency. It will exit with an ETIME error.
832 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
835 cpumask=int Set the CPU affinity of this job. The parameter given is a
836 bitmask of allowed CPU's the job may run on. So if you want
837 the allowed CPUs to be 1 and 5, you would pass the decimal
838 value of (1 << 1 | 1 << 5), or 34. See man
839 sched_setaffinity(2). This may not work on all supported
840 operating systems or kernel versions. This option doesn't
841 work well for a higher CPU count than what you can store in
842 an integer mask, so it can only control cpus 1-32. For
843 boxes with larger CPU counts, use cpus_allowed.
845 cpus_allowed=str Controls the same options as cpumask, but it allows a text
846 setting of the permitted CPUs instead. So to use CPUs 1 and
847 5, you would specify cpus_allowed=1,5. This options also
848 allows a range of CPUs. Say you wanted a binding to CPUs
849 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
851 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
852 arguments allow comma delimited list of cpu numbers,
853 A-B ranges, or 'all'. Note, to enable numa options support,
854 fio must be built on a system with libnuma-dev(el) installed.
856 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
857 nodes. Format of the argements:
859 `mode' is one of the following memory policy:
860 default, prefer, bind, interleave, local
861 For `default' and `local' memory policy, no node is
862 needed to be specified.
863 For `prefer', only one node is allowed.
864 For `bind' and `interleave', it allow comma delimited
865 list of numbers, A-B ranges, or 'all'.
867 startdelay=time Start this job the specified number of seconds after fio
868 has started. Only useful if the job file contains several
869 jobs, and you want to delay starting some jobs to a certain
872 runtime=time Tell fio to terminate processing after the specified number
873 of seconds. It can be quite hard to determine for how long
874 a specified job will run, so this parameter is handy to
875 cap the total runtime to a given time.
877 time_based If set, fio will run for the duration of the runtime
878 specified even if the file(s) are completely read or
879 written. It will simply loop over the same workload
880 as many times as the runtime allows.
882 ramp_time=time If set, fio will run the specified workload for this amount
883 of time before logging any performance numbers. Useful for
884 letting performance settle before logging results, thus
885 minimizing the runtime required for stable results. Note
886 that the ramp_time is considered lead in time for a job,
887 thus it will increase the total runtime if a special timeout
888 or runtime is specified.
890 invalidate=bool Invalidate the buffer/page cache parts for this file prior
891 to starting io. Defaults to true.
893 sync=bool Use sync io for buffered writes. For the majority of the
894 io engines, this means using O_SYNC.
897 mem=str Fio can use various types of memory as the io unit buffer.
898 The allowed values are:
900 malloc Use memory from malloc(3) as the buffers.
902 shm Use shared memory as the buffers. Allocated
905 shmhuge Same as shm, but use huge pages as backing.
907 mmap Use mmap to allocate buffers. May either be
908 anonymous memory, or can be file backed if
909 a filename is given after the option. The
910 format is mem=mmap:/path/to/file.
912 mmaphuge Use a memory mapped huge file as the buffer
913 backing. Append filename after mmaphuge, ala
914 mem=mmaphuge:/hugetlbfs/file
916 The area allocated is a function of the maximum allowed
917 bs size for the job, multiplied by the io depth given. Note
918 that for shmhuge and mmaphuge to work, the system must have
919 free huge pages allocated. This can normally be checked
920 and set by reading/writing /proc/sys/vm/nr_hugepages on a
921 Linux system. Fio assumes a huge page is 4MB in size. So
922 to calculate the number of huge pages you need for a given
923 job file, add up the io depth of all jobs (normally one unless
924 iodepth= is used) and multiply by the maximum bs set. Then
925 divide that number by the huge page size. You can see the
926 size of the huge pages in /proc/meminfo. If no huge pages
927 are allocated by having a non-zero number in nr_hugepages,
928 using mmaphuge or shmhuge will fail. Also see hugepage-size.
930 mmaphuge also needs to have hugetlbfs mounted and the file
931 location should point there. So if it's mounted in /huge,
932 you would use mem=mmaphuge:/huge/somefile.
934 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
935 Note that the given alignment is applied to the first IO unit
936 buffer, if using iodepth the alignment of the following buffers
937 are given by the bs used. In other words, if using a bs that is
938 a multiple of the page sized in the system, all buffers will
939 be aligned to this value. If using a bs that is not page
940 aligned, the alignment of subsequent IO memory buffers is the
941 sum of the iomem_align and bs used.
944 Defines the size of a huge page. Must at least be equal
945 to the system setting, see /proc/meminfo. Defaults to 4MB.
946 Should probably always be a multiple of megabytes, so using
947 hugepage-size=Xm is the preferred way to set this to avoid
948 setting a non-pow-2 bad value.
950 exitall When one job finishes, terminate the rest. The default is
951 to wait for each job to finish, sometimes that is not the
954 bwavgtime=int Average the calculated bandwidth over the given time. Value
955 is specified in milliseconds.
957 iopsavgtime=int Average the calculated IOPS over the given time. Value
958 is specified in milliseconds.
960 create_serialize=bool If true, serialize the file creating for the jobs.
961 This may be handy to avoid interleaving of data
962 files, which may greatly depend on the filesystem
963 used and even the number of processors in the system.
965 create_fsync=bool fsync the data file after creation. This is the
968 create_on_open=bool Don't pre-setup the files for IO, just create open()
969 when it's time to do IO to that file.
971 create_only=bool If true, fio will only run the setup phase of the job.
972 If files need to be laid out or updated on disk, only
973 that will be done. The actual job contents are not
976 pre_read=bool If this is given, files will be pre-read into memory before
977 starting the given IO operation. This will also clear
978 the 'invalidate' flag, since it is pointless to pre-read
979 and then drop the cache. This will only work for IO engines
980 that are seekable, since they allow you to read the same data
981 multiple times. Thus it will not work on eg network or splice
984 unlink=bool Unlink the job files when done. Not the default, as repeated
985 runs of that job would then waste time recreating the file
988 loops=int Run the specified number of iterations of this job. Used
989 to repeat the same workload a given number of times. Defaults
992 do_verify=bool Run the verify phase after a write phase. Only makes sense if
993 verify is set. Defaults to 1.
995 verify=str If writing to a file, fio can verify the file contents
996 after each iteration of the job. The allowed values are:
998 md5 Use an md5 sum of the data area and store
999 it in the header of each block.
1001 crc64 Use an experimental crc64 sum of the data
1002 area and store it in the header of each
1005 crc32c Use a crc32c sum of the data area and store
1006 it in the header of each block.
1008 crc32c-intel Use hardware assisted crc32c calcuation
1009 provided on SSE4.2 enabled processors. Falls
1010 back to regular software crc32c, if not
1011 supported by the system.
1013 crc32 Use a crc32 sum of the data area and store
1014 it in the header of each block.
1016 crc16 Use a crc16 sum of the data area and store
1017 it in the header of each block.
1019 crc7 Use a crc7 sum of the data area and store
1020 it in the header of each block.
1022 sha512 Use sha512 as the checksum function.
1024 sha256 Use sha256 as the checksum function.
1026 sha1 Use optimized sha1 as the checksum function.
1028 meta Write extra information about each io
1029 (timestamp, block number etc.). The block
1030 number is verified. See also verify_pattern.
1032 null Only pretend to verify. Useful for testing
1033 internals with ioengine=null, not for much
1036 This option can be used for repeated burn-in tests of a
1037 system to make sure that the written data is also
1038 correctly read back. If the data direction given is
1039 a read or random read, fio will assume that it should
1040 verify a previously written file. If the data direction
1041 includes any form of write, the verify will be of the
1044 verifysort=bool If set, fio will sort written verify blocks when it deems
1045 it faster to read them back in a sorted manner. This is
1046 often the case when overwriting an existing file, since
1047 the blocks are already laid out in the file system. You
1048 can ignore this option unless doing huge amounts of really
1049 fast IO where the red-black tree sorting CPU time becomes
1052 verify_offset=int Swap the verification header with data somewhere else
1053 in the block before writing. Its swapped back before
1056 verify_interval=int Write the verification header at a finer granularity
1057 than the blocksize. It will be written for chunks the
1058 size of header_interval. blocksize should divide this
1061 verify_pattern=str If set, fio will fill the io buffers with this
1062 pattern. Fio defaults to filling with totally random
1063 bytes, but sometimes it's interesting to fill with a known
1064 pattern for io verification purposes. Depending on the
1065 width of the pattern, fio will fill 1/2/3/4 bytes of the
1066 buffer at the time(it can be either a decimal or a hex number).
1067 The verify_pattern if larger than a 32-bit quantity has to
1068 be a hex number that starts with either "0x" or "0X". Use
1071 verify_fatal=bool Normally fio will keep checking the entire contents
1072 before quitting on a block verification failure. If this
1073 option is set, fio will exit the job on the first observed
1076 verify_dump=bool If set, dump the contents of both the original data
1077 block and the data block we read off disk to files. This
1078 allows later analysis to inspect just what kind of data
1079 corruption occurred. Off by default.
1081 verify_async=int Fio will normally verify IO inline from the submitting
1082 thread. This option takes an integer describing how many
1083 async offload threads to create for IO verification instead,
1084 causing fio to offload the duty of verifying IO contents
1085 to one or more separate threads. If using this offload
1086 option, even sync IO engines can benefit from using an
1087 iodepth setting higher than 1, as it allows them to have
1088 IO in flight while verifies are running.
1090 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1091 async IO verification threads. See cpus_allowed for the
1094 verify_backlog=int Fio will normally verify the written contents of a
1095 job that utilizes verify once that job has completed. In
1096 other words, everything is written then everything is read
1097 back and verified. You may want to verify continually
1098 instead for a variety of reasons. Fio stores the meta data
1099 associated with an IO block in memory, so for large
1100 verify workloads, quite a bit of memory would be used up
1101 holding this meta data. If this option is enabled, fio
1102 will write only N blocks before verifying these blocks.
1104 will verify the previously written blocks before continuing
1107 verify_backlog_batch=int Control how many blocks fio will verify
1108 if verify_backlog is set. If not set, will default to
1109 the value of verify_backlog (meaning the entire queue
1110 is read back and verified). If verify_backlog_batch is
1111 less than verify_backlog then not all blocks will be verified,
1112 if verify_backlog_batch is larger than verify_backlog, some
1113 blocks will be verified more than once.
1116 wait_for_previous Wait for preceeding jobs in the job file to exit, before
1117 starting this one. Can be used to insert serialization
1118 points in the job file. A stone wall also implies starting
1119 a new reporting group.
1121 new_group Start a new reporting group. See: group_reporting.
1123 numjobs=int Create the specified number of clones of this job. May be
1124 used to setup a larger number of threads/processes doing
1125 the same thing. Each thread is reported separately; to see
1126 statistics for all clones as a whole, use group_reporting in
1127 conjunction with new_group.
1129 group_reporting It may sometimes be interesting to display statistics for
1130 groups of jobs as a whole instead of for each individual job.
1131 This is especially true if 'numjobs' is used; looking at
1132 individual thread/process output quickly becomes unwieldy.
1133 To see the final report per-group instead of per-job, use
1134 'group_reporting'. Jobs in a file will be part of the same
1135 reporting group, unless if separated by a stonewall, or by
1138 thread fio defaults to forking jobs, however if this option is
1139 given, fio will use pthread_create(3) to create threads
1142 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1144 zoneskip=int Skip the specified number of bytes when zonesize data has
1145 been read. The two zone options can be used to only do
1146 io on zones of a file.
1148 write_iolog=str Write the issued io patterns to the specified file. See
1149 read_iolog. Specify a separate file for each job, otherwise
1150 the iologs will be interspersed and the file may be corrupt.
1152 read_iolog=str Open an iolog with the specified file name and replay the
1153 io patterns it contains. This can be used to store a
1154 workload and replay it sometime later. The iolog given
1155 may also be a blktrace binary file, which allows fio
1156 to replay a workload captured by blktrace. See blktrace
1157 for how to capture such logging data. For blktrace replay,
1158 the file needs to be turned into a blkparse binary data
1159 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1161 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1162 is to attempt to respect the time stamps within the log and
1163 replay them with the appropriate delay between IOPS. By
1164 setting this variable fio will not respect the timestamps and
1165 attempt to replay them as fast as possible while still
1166 respecting ordering. The result is the same I/O pattern to a
1167 given device, but different timings.
1169 replay_redirect=str While replaying I/O patterns using read_iolog the
1170 default behavior is to replay the IOPS onto the major/minor
1171 device that each IOP was recorded from. This is sometimes
1172 undesireable because on a different machine those major/minor
1173 numbers can map to a different device. Changing hardware on
1174 the same system can also result in a different major/minor
1175 mapping. Replay_redirect causes all IOPS to be replayed onto
1176 the single specified device regardless of the device it was
1177 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1178 IO in the blktrace to be replayed onto /dev/sdc. This means
1179 multiple devices will be replayed onto a single, if the trace
1180 contains multiple devices. If you want multiple devices to be
1181 replayed concurrently to multiple redirected devices you must
1182 blkparse your trace into separate traces and replay them with
1183 independent fio invocations. Unfortuantely this also breaks
1184 the strict time ordering between multiple device accesses.
1186 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1187 file. Can be used to store data of the bandwidth of the
1188 jobs in their lifetime. The included fio_generate_plots
1189 script uses gnuplot to turn these text files into nice
1190 graphs. See write_lat_log for behaviour of given
1191 filename. For this option, the suffix is _bw.log.
1193 write_lat_log=str Same as write_bw_log, except that this option stores io
1194 submission, completion, and total latencies instead. If no
1195 filename is given with this option, the default filename of
1196 "jobname_type.log" is used. Even if the filename is given,
1197 fio will still append the type of log. So if one specifies
1201 The actual log names will be foo_slat.log, foo_slat.log,
1202 and foo_lat.log. This helps fio_generate_plot fine the logs
1205 write_bw_log=str If given, write an IOPS log of the jobs in this job
1206 file. See write_bw_log.
1208 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1209 given with this option, the default filename of
1210 "jobname_type.log" is used. Even if the filename is given,
1211 fio will still append the type of log.
1213 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1214 or bw log for every IO that completes. When writing to the
1215 disk log, that can quickly grow to a very large size. Setting
1216 this option makes fio average the each log entry over the
1217 specified period of time, reducing the resolution of the log.
1220 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1221 potentially be used instead of removing memory or booting
1222 with less memory to simulate a smaller amount of memory.
1224 exec_prerun=str Before running this job, issue the command specified
1227 exec_postrun=str After the job completes, issue the command specified
1230 ioscheduler=str Attempt to switch the device hosting the file to the specified
1231 io scheduler before running.
1233 cpuload=int If the job is a CPU cycle eater, attempt to use the specified
1234 percentage of CPU cycles.
1236 cpuchunks=int If the job is a CPU cycle eater, split the load into
1237 cycles of the given time. In microseconds.
1239 disk_util=bool Generate disk utilization statistics, if the platform
1240 supports it. Defaults to on.
1242 disable_lat=bool Disable measurements of total latency numbers. Useful
1243 only for cutting back the number of calls to gettimeofday,
1244 as that does impact performance at really high IOPS rates.
1245 Note that to really get rid of a large amount of these
1246 calls, this option must be used with disable_slat and
1249 disable_clat=bool Disable measurements of completion latency numbers. See
1252 disable_slat=bool Disable measurements of submission latency numbers. See
1255 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1258 clat_percentiles=bool Enable the reporting of percentiles of
1259 completion latencies.
1261 percentile_list=float_list Overwrite the default list of percentiles
1262 for completion latencies. Each number is a floating
1263 number in the range (0,100], and the maximum length of
1264 the list is 20. Use ':' to separate the numbers, and
1265 list the numbers in ascending order. For example,
1266 --percentile_list=99.5:99.9 will cause fio to report
1267 the values of completion latency below which 99.5% and
1268 99.9% of the observed latencies fell, respectively.
1270 clocksource=str Use the given clocksource as the base of timing. The
1271 supported options are:
1273 gettimeofday gettimeofday(2)
1275 clock_gettime clock_gettime(2)
1277 cpu Internal CPU clock source
1279 cpu is the preferred clocksource if it is reliable, as it
1280 is very fast (and fio is heavy on time calls). Fio will
1281 automatically use this clocksource if it's supported and
1282 considered reliable on the system it is running on, unless
1283 another clocksource is specifically set. For x86/x86-64 CPUs,
1284 this means supporting TSC Invariant.
1286 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1287 (disable_clat, disable_slat, disable_bw) plus reduce
1288 precision of the timeout somewhat to really shrink
1289 the gettimeofday() call count. With this option enabled,
1290 we only do about 0.4% of the gtod() calls we would have
1291 done if all time keeping was enabled.
1293 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1294 execution to just getting the current time. Fio (and
1295 databases, for instance) are very intensive on gettimeofday()
1296 calls. With this option, you can set one CPU aside for
1297 doing nothing but logging current time to a shared memory
1298 location. Then the other threads/processes that run IO
1299 workloads need only copy that segment, instead of entering
1300 the kernel with a gettimeofday() call. The CPU set aside
1301 for doing these time calls will be excluded from other
1302 uses. Fio will manually clear it from the CPU mask of other
1305 continue_on_error=str Normally fio will exit the job on the first observed
1306 failure. If this option is set, fio will continue the job when
1307 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1308 is exceeded or the I/O size specified is completed. If this
1309 option is used, there are two more stats that are appended,
1310 the total error count and the first error. The error field
1311 given in the stats is the first error that was hit during the
1314 The allowed values are:
1316 none Exit on any IO or verify errors.
1318 read Continue on read errors, exit on all others.
1320 write Continue on write errors, exit on all others.
1322 io Continue on any IO error, exit on all others.
1324 verify Continue on verify errors, exit on all others.
1326 all Continue on all errors.
1328 0 Backward-compatible alias for 'none'.
1330 1 Backward-compatible alias for 'all'.
1332 ignore_error=str Sometimes you want to ignore some errors during test
1333 in that case you can specify error list for each error type.
1334 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1335 errors for given error type is separated with ':'. Error
1336 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1338 ignore_error=EAGAIN,ENOSPC:122
1339 This option will ignore EAGAIN from READ, and ENOSPC and
1340 122(EDQUOT) from WRITE.
1342 error_dump=bool If set dump every error even if it is non fatal, true
1343 by default. If disabled only fatal error will be dumped
1345 cgroup=str Add job to this control group. If it doesn't exist, it will
1346 be created. The system must have a mounted cgroup blkio
1347 mount point for this to work. If your system doesn't have it
1348 mounted, you can do so with:
1350 # mount -t cgroup -o blkio none /cgroup
1352 cgroup_weight=int Set the weight of the cgroup to this value. See
1353 the documentation that comes with the kernel, allowed values
1354 are in the range of 100..1000.
1356 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1357 the job completion. To override this behavior and to leave
1358 cgroups around after the job completion, set cgroup_nodelete=1.
1359 This can be useful if one wants to inspect various cgroup
1360 files after job completion. Default: false
1362 uid=int Instead of running as the invoking user, set the user ID to
1363 this value before the thread/process does any work.
1365 gid=int Set group ID, see uid.
1367 flow_id=int The ID of the flow. If not specified, it defaults to being a
1368 global flow. See flow.
1370 flow=int Weight in token-based flow control. If this value is used, then
1371 there is a 'flow counter' which is used to regulate the
1372 proportion of activity between two or more jobs. fio attempts
1373 to keep this flow counter near zero. The 'flow' parameter
1374 stands for how much should be added or subtracted to the flow
1375 counter on each iteration of the main I/O loop. That is, if
1376 one job has flow=8 and another job has flow=-1, then there
1377 will be a roughly 1:8 ratio in how much one runs vs the other.
1379 flow_watermark=int The maximum value that the absolute value of the flow
1380 counter is allowed to reach before the job must wait for a
1381 lower value of the counter.
1383 flow_sleep=int The period of time, in microseconds, to wait after the flow
1384 watermark has been exceeded before retrying operations
1386 In addition, there are some parameters which are only valid when a specific
1387 ioengine is in use. These are used identically to normal parameters, with the
1388 caveat that when used on the command line, they must come after the ioengine
1389 that defines them is selected.
1391 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1392 the io_getevents system call to reap newly returned events.
1393 With this flag turned on, the AIO ring will be read directly
1394 from user-space to reap events. The reaping mode is only
1395 enabled when polling for a minimum of 0 events (eg when
1396 iodepth_batch_complete=0).
1398 [netsplice] hostname=str
1399 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1400 If the job is a TCP listener or UDP reader, the hostname is not
1401 used and must be omitted.
1403 [netsplice] port=int
1404 [net] port=int The TCP or UDP port to bind to or connect to.
1406 [netsplice] nodelay=bool
1407 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1409 [netsplice] protocol=str
1410 [netsplice] proto=str
1412 [net] proto=str The network protocol to use. Accepted values are:
1414 tcp Transmission control protocol
1415 udp User datagram protocol
1416 unix UNIX domain socket
1418 When the protocol is TCP or UDP, the port must also be given,
1419 as well as the hostname if the job is a TCP listener or UDP
1420 reader. For unix sockets, the normal filename option should be
1421 used and the port is invalid.
1423 [net] listen For TCP network connections, tell fio to listen for incoming
1424 connections rather than initiating an outgoing connection. The
1425 hostname must be omitted if this option is used.
1426 [net] pingpong Normal a network writer will just continue writing data, and
1427 a network reader will just consume packages. If pingpong=1
1428 is set, a writer will send its normal payload to the reader,
1429 then wait for the reader to send the same payload back. This
1430 allows fio to measure network latencies. The submission
1431 and completion latencies then measure local time spent
1432 sending or receiving, and the completion latency measures
1433 how long it took for the other end to receive and send back.
1435 [e4defrag] donorname=str
1436 File will be used as a block donor(swap extents between files)
1437 [e4defrag] inplace=int
1438 Configure donor file blocks allocation strategy
1439 0(default): Preallocate donor's file on init
1440 1 : allocate space immidietly inside defragment event,
1441 and free right after event
1445 6.0 Interpreting the output
1446 ---------------------------
1448 fio spits out a lot of output. While running, fio will display the
1449 status of the jobs created. An example of that would be:
1451 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1453 The characters inside the square brackets denote the current status of
1454 each thread. The possible values (in typical life cycle order) are:
1458 P Thread setup, but not started.
1460 I Thread initialized, waiting or generating necessary data.
1461 p Thread running pre-reading file(s).
1462 R Running, doing sequential reads.
1463 r Running, doing random reads.
1464 W Running, doing sequential writes.
1465 w Running, doing random writes.
1466 M Running, doing mixed sequential reads/writes.
1467 m Running, doing mixed random reads/writes.
1468 F Running, currently waiting for fsync()
1469 V Running, doing verification of written data.
1470 E Thread exited, not reaped by main thread yet.
1472 X Thread reaped, exited with an error.
1473 K Thread reaped, exited due to signal.
1475 The other values are fairly self explanatory - number of threads
1476 currently running and doing io, rate of io since last check (read speed
1477 listed first, then write speed), and the estimated completion percentage
1478 and time for the running group. It's impossible to estimate runtime of
1479 the following groups (if any). Note that the string is displayed in order,
1480 so it's possible to tell which of the jobs are currently doing what. The
1481 first character is the first job defined in the job file, and so forth.
1483 When fio is done (or interrupted by ctrl-c), it will show the data for
1484 each thread, group of threads, and disks in that order. For each data
1485 direction, the output looks like:
1487 Client1 (g=0): err= 0:
1488 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1489 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1490 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1491 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1492 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1493 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1494 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1495 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1496 issued r/w: total=0/32768, short=0/0
1497 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1498 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1500 The client number is printed, along with the group id and error of that
1501 thread. Below is the io statistics, here for writes. In the order listed,
1504 io= Number of megabytes io performed
1505 bw= Average bandwidth rate
1506 iops= Average IOs performed per second
1507 runt= The runtime of that thread
1508 slat= Submission latency (avg being the average, stdev being the
1509 standard deviation). This is the time it took to submit
1510 the io. For sync io, the slat is really the completion
1511 latency, since queue/complete is one operation there. This
1512 value can be in milliseconds or microseconds, fio will choose
1513 the most appropriate base and print that. In the example
1514 above, milliseconds is the best scale. Note: in --minimal mode
1515 latencies are always expressed in microseconds.
1516 clat= Completion latency. Same names as slat, this denotes the
1517 time from submission to completion of the io pieces. For
1518 sync io, clat will usually be equal (or very close) to 0,
1519 as the time from submit to complete is basically just
1520 CPU time (io has already been done, see slat explanation).
1521 bw= Bandwidth. Same names as the xlat stats, but also includes
1522 an approximate percentage of total aggregate bandwidth
1523 this thread received in this group. This last value is
1524 only really useful if the threads in this group are on the
1525 same disk, since they are then competing for disk access.
1526 cpu= CPU usage. User and system time, along with the number
1527 of context switches this thread went through, usage of
1528 system and user time, and finally the number of major
1529 and minor page faults.
1530 IO depths= The distribution of io depths over the job life time. The
1531 numbers are divided into powers of 2, so for example the
1532 16= entries includes depths up to that value but higher
1533 than the previous entry. In other words, it covers the
1534 range from 16 to 31.
1535 IO submit= How many pieces of IO were submitting in a single submit
1536 call. Each entry denotes that amount and below, until
1537 the previous entry - eg, 8=100% mean that we submitted
1538 anywhere in between 5-8 ios per submit call.
1539 IO complete= Like the above submit number, but for completions instead.
1540 IO issued= The number of read/write requests issued, and how many
1542 IO latencies= The distribution of IO completion latencies. This is the
1543 time from when IO leaves fio and when it gets completed.
1544 The numbers follow the same pattern as the IO depths,
1545 meaning that 2=1.6% means that 1.6% of the IO completed
1546 within 2 msecs, 20=12.8% means that 12.8% of the IO
1547 took more than 10 msecs, but less than (or equal to) 20 msecs.
1549 After each client has been listed, the group statistics are printed. They
1550 will look like this:
1552 Run status group 0 (all jobs):
1553 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1554 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1556 For each data direction, it prints:
1558 io= Number of megabytes io performed.
1559 aggrb= Aggregate bandwidth of threads in this group.
1560 minb= The minimum average bandwidth a thread saw.
1561 maxb= The maximum average bandwidth a thread saw.
1562 mint= The smallest runtime of the threads in that group.
1563 maxt= The longest runtime of the threads in that group.
1565 And finally, the disk statistics are printed. They will look like this:
1567 Disk stats (read/write):
1568 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1570 Each value is printed for both reads and writes, with reads first. The
1573 ios= Number of ios performed by all groups.
1574 merge= Number of merges io the io scheduler.
1575 ticks= Number of ticks we kept the disk busy.
1576 io_queue= Total time spent in the disk queue.
1577 util= The disk utilization. A value of 100% means we kept the disk
1578 busy constantly, 50% would be a disk idling half of the time.
1580 It is also possible to get fio to dump the current output while it is
1581 running, without terminating the job. To do that, send fio the USR1 signal.
1587 For scripted usage where you typically want to generate tables or graphs
1588 of the results, fio can output the results in a semicolon separated format.
1589 The format is one long line of values, such as:
1591 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%
1592 A description of this job goes here.
1594 The job description (if provided) follows on a second line.
1596 To enable terse output, use the --minimal command line option. The first
1597 value is the version of the terse output format. If the output has to
1598 be changed for some reason, this number will be incremented by 1 to
1599 signify that change.
1601 Split up, the format is as follows:
1603 terse version, fio version, jobname, groupid, error
1605 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1606 Submission latency: min, max, mean, deviation (usec)
1607 Completion latency: min, max, mean, deviation (usec)
1608 Completion latency percentiles: 20 fields (see below)
1609 Total latency: min, max, mean, deviation (usec)
1610 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1612 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1613 Submission latency: min, max, mean, deviation (usec)
1614 Completion latency: min, max, mean, deviation (usec)
1615 Completion latency percentiles: 20 fields (see below)
1616 Total latency: min, max, mean, deviation (usec)
1617 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1618 CPU usage: user, system, context switches, major faults, minor faults
1619 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1620 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1621 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1622 Disk utilization: Disk name, Read ios, write ios,
1623 Read merges, write merges,
1624 Read ticks, write ticks,
1625 Time spent in queue, disk utilization percentage
1626 Additional Info (dependant on continue_on_error, default off): total # errors, first error code
1628 Additional Info (dependant on description being set): Text description
1630 Completion latency percentiles can be a grouping of up to 20 sets, so
1631 for the terse output fio writes all of them. Each field will look like this:
1635 which is the Xth percentile, and the usec latency associated with it.
1637 For disk utilization, all disks used by fio are shown. So for each disk
1638 there will be a disk utilization section.
1641 8.0 Trace file format
1642 ---------------------
1643 There are two trace file format that you can encounter. The older (v1) format
1644 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1645 below in case that you get an old trace and want to understand it.
1647 In any case the trace is a simple text file with a single action per line.
1650 8.1 Trace file format v1
1651 ------------------------
1652 Each line represents a single io action in the following format:
1656 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1658 This format is not supported in Fio versions => 1.20-rc3.
1661 8.2 Trace file format v2
1662 ------------------------
1663 The second version of the trace file format was added in Fio version 1.17.
1664 It allows to access more then one file per trace and has a bigger set of
1665 possible file actions.
1667 The first line of the trace file has to be:
1671 Following this can be lines in two different formats, which are described below.
1673 The file management format:
1677 The filename is given as an absolute path. The action can be one of these:
1679 add Add the given filename to the trace
1680 open Open the file with the given filename. The filename has to have
1681 been added with the add action before.
1682 close Close the file with the given filename. The file has to have been
1686 The file io action format:
1688 filename action offset length
1690 The filename is given as an absolute path, and has to have been added and opened
1691 before it can be used with this format. The offset and length are given in
1692 bytes. The action can be one of these:
1694 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1695 read Read 'length' bytes beginning from 'offset'
1696 write Write 'length' bytes beginning from 'offset'
1697 sync fsync() the file
1698 datasync fdatasync() the file
1699 trim trim the given file from the given 'offset' for 'length' bytes
1702 9.0 CPU idleness profiling
1704 In some cases, we want to understand CPU overhead in a test. For example,
1705 we test patches for the specific goodness of whether they reduce CPU usage.
1706 fio implements a balloon approach to create a thread per CPU that runs at
1707 idle priority, meaning that it only runs when nobody else needs the cpu.
1708 By measuring the amount of work completed by the thread, idleness of each
1709 CPU can be derived accordingly.
1711 An unit work is defined as touching a full page of unsigned characters. Mean
1712 and standard deviation of time to complete an unit work is reported in "unit
1713 work" section. Options can be chosen to report detailed percpu idleness or
1714 overall system idleness by aggregating percpu stats.