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
289 If sharing multiple files between jobs, it is usually necessary
290 to have fio generate the exact names that you want. By default,
291 fio will name a file based on the default file format
292 specification of jobname.jobnumber.filenumber. With this
293 option, that can be customized. Fio will recognize and replace
294 the following keywords in this string:
297 The name of the worker thread or process.
300 The incremental number of the worker thread or
304 The incremental number of the file for that worker
307 To have dependent jobs share a set of files, this option can
308 be set to have fio generate filenames that are shared between
309 the two. For instance, if testfiles.$filenum is specified,
310 file number 4 for any job will be named testfiles.4. The
311 default of $jobname.$jobnum.$filenum will be used if
312 no other format specifier is given.
314 opendir=str Tell fio to recursively add any file it can find in this
315 directory and down the file system tree.
317 lockfile=str Fio defaults to not locking any files before it does
318 IO to them. If a file or file descriptor is shared, fio
319 can serialize IO to that file to make the end result
320 consistent. This is usual for emulating real workloads that
321 share files. The lock modes are:
323 none No locking. The default.
324 exclusive Only one thread/process may do IO,
325 excluding all others.
326 readwrite Read-write locking on the file. Many
327 readers may access the file at the
328 same time, but writes get exclusive
332 rw=str Type of io pattern. Accepted values are:
334 read Sequential reads
335 write Sequential writes
336 randwrite Random writes
337 randread Random reads
338 rw,readwrite Sequential mixed reads and writes
339 randrw Random mixed reads and writes
341 For the mixed io types, the default is to split them 50/50.
342 For certain types of io the result may still be skewed a bit,
343 since the speed may be different. It is possible to specify
344 a number of IO's to do before getting a new offset, this is
345 one by appending a ':<nr>' to the end of the string given.
346 For a random read, it would look like 'rw=randread:8' for
347 passing in an offset modifier with a value of 8. If the
348 suffix is used with a sequential IO pattern, then the value
349 specified will be added to the generated offset for each IO.
350 For instance, using rw=write:4k will skip 4k for every
351 write. It turns sequential IO into sequential IO with holes.
352 See the 'rw_sequencer' option.
354 rw_sequencer=str If an offset modifier is given by appending a number to
355 the rw=<str> line, then this option controls how that
356 number modifies the IO offset being generated. Accepted
359 sequential Generate sequential offset
360 identical Generate the same offset
362 'sequential' is only useful for random IO, where fio would
363 normally generate a new random offset for every IO. If you
364 append eg 8 to randread, you would get a new random offset for
365 every 8 IO's. The result would be a seek for only every 8
366 IO's, instead of for every IO. Use rw=randread:8 to specify
367 that. As sequential IO is already sequential, setting
368 'sequential' for that would not result in any differences.
369 'identical' behaves in a similar fashion, except it sends
370 the same offset 8 number of times before generating a new
373 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
374 Storage manufacturers like to use 10^3 or 1000 as a base
375 ten unit instead, for obvious reasons. Allow values are
376 1024 or 1000, with 1024 being the default.
378 unified_rw_reporting=bool Fio normally reports statistics on a per
379 data direction basis, meaning that read, write, and trim are
380 accounted and reported separately. If this option is set,
381 the fio will sum the results and report them as "mixed"
384 randrepeat=bool For random IO workloads, seed the generator in a predictable
385 way so that results are repeatable across repetitions.
387 randseed=int Seed the random number generators based on this seed value, to
388 be able to control what sequence of output is being generated.
389 If not set, the random sequence depends on the randrepeat
392 use_os_rand=bool Fio can either use the random generator supplied by the OS
393 to generator random offsets, or it can use it's own internal
394 generator (based on Tausworthe). Default is to use the
395 internal generator, which is often of better quality and
398 fallocate=str Whether pre-allocation is performed when laying down files.
401 none Do not pre-allocate space
402 posix Pre-allocate via posix_fallocate()
403 keep Pre-allocate via fallocate() with
404 FALLOC_FL_KEEP_SIZE set
405 0 Backward-compatible alias for 'none'
406 1 Backward-compatible alias for 'posix'
408 May not be available on all supported platforms. 'keep' is only
409 available on Linux.If using ZFS on Solaris this must be set to
410 'none' because ZFS doesn't support it. Default: 'posix'.
412 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
413 on what IO patterns it is likely to issue. Sometimes you
414 want to test specific IO patterns without telling the
415 kernel about it, in which case you can disable this option.
416 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
417 IO and POSIX_FADV_RANDOM for random IO.
419 size=int The total size of file io for this job. Fio will run until
420 this many bytes has been transferred, unless runtime is
421 limited by other options (such as 'runtime', for instance).
422 Unless specific nrfiles and filesize options are given,
423 fio will divide this size between the available files
424 specified by the job. If not set, fio will use the full
425 size of the given files or devices. If the the files
426 do not exist, size must be given. It is also possible to
427 give size as a percentage between 1 and 100. If size=20%
428 is given, fio will use 20% of the full size of the given
431 filesize=int Individual file sizes. May be a range, in which case fio
432 will select sizes for files at random within the given range
433 and limited to 'size' in total (if that is given). If not
434 given, each created file is the same size.
437 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
438 space left on device) as the terminating condition. Only makes
439 sense with sequential write. For a read workload, the mount
440 point will be filled first then IO started on the result. This
441 option doesn't make sense if operating on a raw device node,
442 since the size of that is already known by the file system.
443 Additionally, writing beyond end-of-device will not return
447 bs=int The block size used for the io units. Defaults to 4k. Values
448 can be given for both read and writes. If a single int is
449 given, it will apply to both. If a second int is specified
450 after a comma, it will apply to writes only. In other words,
451 the format is either bs=read_and_write or bs=read,write,trim.
452 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
453 writes, and 8k for trims. You can terminate the list with
454 a trailing comma. bs=4k,8k, would use the default value for
455 trims.. If you only wish to set the write size, you
456 can do so by passing an empty read size - bs=,8k will set
457 8k for writes and leave the read default value.
460 ba=int At what boundary to align random IO offsets. Defaults to
461 the same as 'blocksize' the minimum blocksize given.
462 Minimum alignment is typically 512b for using direct IO,
463 though it usually depends on the hardware block size. This
464 option is mutually exclusive with using a random map for
465 files, so it will turn off that option.
467 blocksize_range=irange
468 bsrange=irange Instead of giving a single block size, specify a range
469 and fio will mix the issued io block sizes. The issued
470 io unit will always be a multiple of the minimum value
471 given (also see bs_unaligned). Applies to both reads and
472 writes, however a second range can be given after a comma.
475 bssplit=str Sometimes you want even finer grained control of the
476 block sizes issued, not just an even split between them.
477 This option allows you to weight various block sizes,
478 so that you are able to define a specific amount of
479 block sizes issued. The format for this option is:
481 bssplit=blocksize/percentage:blocksize/percentage
483 for as many block sizes as needed. So if you want to define
484 a workload that has 50% 64k blocks, 10% 4k blocks, and
485 40% 32k blocks, you would write:
487 bssplit=4k/10:64k/50:32k/40
489 Ordering does not matter. If the percentage is left blank,
490 fio will fill in the remaining values evenly. So a bssplit
491 option like this one:
493 bssplit=4k/50:1k/:32k/
495 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
496 always add up to 100, if bssplit is given a range that adds
497 up to more, it will error out.
499 bssplit also supports giving separate splits to reads and
500 writes. The format is identical to what bs= accepts. You
501 have to separate the read and write parts with a comma. So
502 if you want a workload that has 50% 2k reads and 50% 4k reads,
503 while having 90% 4k writes and 10% 8k writes, you would
506 bssplit=2k/50:4k/50,4k/90,8k/10
509 bs_unaligned If this option is given, any byte size value within bsrange
510 may be used as a block range. This typically wont work with
511 direct IO, as that normally requires sector alignment.
513 bs_is_seq_rand If this option is set, fio will use the normal read,write
514 blocksize settings as sequential,random instead. Any random
515 read or write will use the WRITE blocksize settings, and any
516 sequential read or write will use the READ blocksize setting.
518 zero_buffers If this option is given, fio will init the IO buffers to
519 all zeroes. The default is to fill them with random data.
521 refill_buffers If this option is given, fio will refill the IO buffers
522 on every submit. The default is to only fill it at init
523 time and reuse that data. Only makes sense if zero_buffers
524 isn't specified, naturally. If data verification is enabled,
525 refill_buffers is also automatically enabled.
527 scramble_buffers=bool If refill_buffers is too costly and the target is
528 using data deduplication, then setting this option will
529 slightly modify the IO buffer contents to defeat normal
530 de-dupe attempts. This is not enough to defeat more clever
531 block compression attempts, but it will stop naive dedupe of
532 blocks. Default: true.
534 buffer_compress_percentage=int If this is set, then fio will attempt to
535 provide IO buffer content (on WRITEs) that compress to
536 the specified level. Fio does this by providing a mix of
537 random data and zeroes. Note that this is per block size
538 unit, for file/disk wide compression level that matches
539 this setting, you'll also want to set refill_buffers.
541 buffer_compress_chunk=int See buffer_compress_percentage. This
542 setting allows fio to manage how big the ranges of random
543 data and zeroed data is. Without this set, fio will
544 provide buffer_compress_percentage of blocksize random
545 data, followed by the remaining zeroed. With this set
546 to some chunk size smaller than the block size, fio can
547 alternate random and zeroed data throughout the IO
550 buffer_pattern=str If set, fio will fill the io buffers with this pattern.
551 If not set, the contents of io buffers is defined by the other
552 options related to buffer contents. The setting can be any
553 pattern of bytes, and can be prefixed with 0x for hex values.
555 nrfiles=int Number of files to use for this job. Defaults to 1.
557 openfiles=int Number of files to keep open at the same time. Defaults to
558 the same as nrfiles, can be set smaller to limit the number
561 file_service_type=str Defines how fio decides which file from a job to
562 service next. The following types are defined:
564 random Just choose a file at random.
566 roundrobin Round robin over open files. This
569 sequential Finish one file before moving on to
570 the next. Multiple files can still be
571 open depending on 'openfiles'.
573 The string can have a number appended, indicating how
574 often to switch to a new file. So if option random:4 is
575 given, fio will switch to a new random file after 4 ios
578 ioengine=str Defines how the job issues io to the file. The following
581 sync Basic read(2) or write(2) io. lseek(2) is
582 used to position the io location.
584 psync Basic pread(2) or pwrite(2) io.
586 vsync Basic readv(2) or writev(2) IO.
588 psyncv Basic preadv(2) or pwritev(2) IO.
590 libaio Linux native asynchronous io. Note that Linux
591 may only support queued behaviour with
592 non-buffered IO (set direct=1 or buffered=0).
593 This engine defines engine specific options.
595 posixaio glibc posix asynchronous io.
597 solarisaio Solaris native asynchronous io.
599 windowsaio Windows native asynchronous io.
601 mmap File is memory mapped and data copied
602 to/from using memcpy(3).
604 splice splice(2) is used to transfer the data and
605 vmsplice(2) to transfer data from user
608 syslet-rw Use the syslet system calls to make
609 regular read/write async.
611 sg SCSI generic sg v3 io. May either be
612 synchronous using the SG_IO ioctl, or if
613 the target is an sg character device
614 we use read(2) and write(2) for asynchronous
617 null Doesn't transfer any data, just pretends
618 to. This is mainly used to exercise fio
619 itself and for debugging/testing purposes.
621 net Transfer over the network to given host:port.
622 Depending on the protocol used, the hostname,
623 port, listen and filename options are used to
624 specify what sort of connection to make, while
625 the protocol option determines which protocol
627 This engine defines engine specific options.
629 netsplice Like net, but uses splice/vmsplice to
630 map data and send/receive.
631 This engine defines engine specific options.
633 cpuio Doesn't transfer any data, but burns CPU
634 cycles according to the cpuload= and
635 cpucycle= options. Setting cpuload=85
636 will cause that job to do nothing but burn
637 85% of the CPU. In case of SMP machines,
638 use numjobs=<no_of_cpu> to get desired CPU
639 usage, as the cpuload only loads a single
640 CPU at the desired rate.
642 guasi The GUASI IO engine is the Generic Userspace
643 Asyncronous Syscall Interface approach
646 http://www.xmailserver.org/guasi-lib.html
648 for more info on GUASI.
650 rdma The RDMA I/O engine supports both RDMA
651 memory semantics (RDMA_WRITE/RDMA_READ) and
652 channel semantics (Send/Recv) for the
653 InfiniBand, RoCE and iWARP protocols.
655 falloc IO engine that does regular fallocate to
656 simulate data transfer as fio ioengine.
657 DDIR_READ does fallocate(,mode = keep_size,)
658 DDIR_WRITE does fallocate(,mode = 0)
659 DDIR_TRIM does fallocate(,mode = punch_hole)
661 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
662 ioctls to simulate defragment activity in
663 request to DDIR_WRITE event
665 external Prefix to specify loading an external
666 IO engine object file. Append the engine
667 filename, eg ioengine=external:/tmp/foo.o
668 to load ioengine foo.o in /tmp.
670 iodepth=int This defines how many io units to keep in flight against
671 the file. The default is 1 for each file defined in this
672 job, can be overridden with a larger value for higher
673 concurrency. Note that increasing iodepth beyond 1 will not
674 affect synchronous ioengines (except for small degress when
675 verify_async is in use). Even async engines may impose OS
676 restrictions causing the desired depth not to be achieved.
677 This may happen on Linux when using libaio and not setting
678 direct=1, since buffered IO is not async on that OS. Keep an
679 eye on the IO depth distribution in the fio output to verify
680 that the achieved depth is as expected. Default: 1.
682 iodepth_batch_submit=int
683 iodepth_batch=int This defines how many pieces of IO to submit at once.
684 It defaults to 1 which means that we submit each IO
685 as soon as it is available, but can be raised to submit
686 bigger batches of IO at the time.
688 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
689 at once. It defaults to 1 which means that we'll ask
690 for a minimum of 1 IO in the retrieval process from
691 the kernel. The IO retrieval will go on until we
692 hit the limit set by iodepth_low. If this variable is
693 set to 0, then fio will always check for completed
694 events before queuing more IO. This helps reduce
695 IO latency, at the cost of more retrieval system calls.
697 iodepth_low=int The low water mark indicating when to start filling
698 the queue again. Defaults to the same as iodepth, meaning
699 that fio will attempt to keep the queue full at all times.
700 If iodepth is set to eg 16 and iodepth_low is set to 4, then
701 after fio has filled the queue of 16 requests, it will let
702 the depth drain down to 4 before starting to fill it again.
704 direct=bool If value is true, use non-buffered io. This is usually
705 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
706 On Windows the synchronous ioengines don't support direct io.
708 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
709 writes are guaranteed to be stable once acknowledged by
710 the operating system. Only Linux supports O_ATOMIC right
713 buffered=bool If value is true, use buffered io. This is the opposite
714 of the 'direct' option. Defaults to true.
716 offset=int Start io at the given offset in the file. The data before
717 the given offset will not be touched. This effectively
718 caps the file size at real_size - offset.
720 offset_increment=int If this is provided, then the real offset becomes
721 the offset + offset_increment * thread_number, where the
722 thread number is a counter that starts at 0 and is incremented
723 for each job. This option is useful if there are several jobs
724 which are intended to operate on a file in parallel in disjoint
725 segments, with even spacing between the starting points.
727 number_ios=int Fio will normally perform IOs until it has exhausted the size
728 of the region set by size=, or if it exhaust the allocated
729 time (or hits an error condition). With this setting, the
730 range/size can be set independently of the number of IOs to
731 perform. When fio reaches this number, it will exit normally
734 fsync=int If writing to a file, issue a sync of the dirty data
735 for every number of blocks given. For example, if you give
736 32 as a parameter, fio will sync the file for every 32
737 writes issued. If fio is using non-buffered io, we may
738 not sync the file. The exception is the sg io engine, which
739 synchronizes the disk cache anyway.
741 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
743 In FreeBSD and Windows there is no fdatasync(), this falls back to
746 sync_file_range=str:val Use sync_file_range() for every 'val' number of
747 write operations. Fio will track range of writes that
748 have happened since the last sync_file_range() call. 'str'
749 can currently be one or more of:
751 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
752 write SYNC_FILE_RANGE_WRITE
753 wait_after SYNC_FILE_RANGE_WAIT_AFTER
755 So if you do sync_file_range=wait_before,write:8, fio would
756 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
757 every 8 writes. Also see the sync_file_range(2) man page.
758 This option is Linux specific.
760 overwrite=bool If true, writes to a file will always overwrite existing
761 data. If the file doesn't already exist, it will be
762 created before the write phase begins. If the file exists
763 and is large enough for the specified write phase, nothing
766 end_fsync=bool If true, fsync file contents when a write stage has completed.
768 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
769 This differs from end_fsync in that it will happen on every
770 file close, not just at the end of the job.
772 rwmixread=int How large a percentage of the mix should be reads.
774 rwmixwrite=int How large a percentage of the mix should be writes. If both
775 rwmixread and rwmixwrite is given and the values do not add
776 up to 100%, the latter of the two will be used to override
777 the first. This may interfere with a given rate setting,
778 if fio is asked to limit reads or writes to a certain rate.
779 If that is the case, then the distribution may be skewed.
781 random_distribution=str:float By default, fio will use a completely uniform
782 random distribution when asked to perform random IO. Sometimes
783 it is useful to skew the distribution in specific ways,
784 ensuring that some parts of the data is more hot than others.
785 fio includes the following distribution models:
787 random Uniform random distribution
788 zipf Zipf distribution
789 pareto Pareto distribution
791 When using a zipf or pareto distribution, an input value
792 is also needed to define the access pattern. For zipf, this
793 is the zipf theta. For pareto, it's the pareto power. Fio
794 includes a test program, genzipf, that can be used visualize
795 what the given input values will yield in terms of hit rates.
796 If you wanted to use zipf with a theta of 1.2, you would use
797 random_distribution=zipf:1.2 as the option. If a non-uniform
798 model is used, fio will disable use of the random map.
800 percentage_random=int For a random workload, set how big a percentage should
801 be random. This defaults to 100%, in which case the workload
802 is fully random. It can be set from anywhere from 0 to 100.
803 Setting it to 0 would make the workload fully sequential. Any
804 setting in between will result in a random mix of sequential
805 and random IO, at the given percentages. It is possible to
806 set different values for reads, writes, and trim. To do so,
807 simply use a comma separated list. See blocksize.
809 norandommap Normally fio will cover every block of the file when doing
810 random IO. If this option is given, fio will just get a
811 new random offset without looking at past io history. This
812 means that some blocks may not be read or written, and that
813 some blocks may be read/written more than once. This option
814 is mutually exclusive with verify= if and only if multiple
815 blocksizes (via bsrange=) are used, since fio only tracks
816 complete rewrites of blocks.
818 softrandommap=bool See norandommap. If fio runs with the random block map
819 enabled and it fails to allocate the map, if this option is
820 set it will continue without a random block map. As coverage
821 will not be as complete as with random maps, this option is
824 random_generator=str Fio supports the following engines for generating
825 IO offsets for random IO:
827 tausworthe Strong 2^88 cycle random number generator
828 lfsr Linear feedback shift register generator
830 Tausworthe is a strong random number generator, but it
831 requires tracking on the side if we want to ensure that
832 blocks are only read or written once. LFSR guarantees
833 that we never generate the same offset twice, and it's
834 also less computationally expensive. It's not a true
835 random generator, however, though for IO purposes it's
836 typically good enough. LFSR only works with single
837 block sizes, not with workloads that use multiple block
838 sizes. If used with such a workload, fio may read or write
839 some blocks multiple times.
841 nice=int Run the job with the given nice value. See man nice(2).
843 prio=int Set the io priority value of this job. Linux limits us to
844 a positive value between 0 and 7, with 0 being the highest.
847 prioclass=int Set the io priority class. See man ionice(1).
849 thinktime=int Stall the job x microseconds after an io has completed before
850 issuing the next. May be used to simulate processing being
851 done by an application. See thinktime_blocks and
855 Only valid if thinktime is set - pretend to spend CPU time
856 doing something with the data received, before falling back
857 to sleeping for the rest of the period specified by
861 Only valid if thinktime is set - control how many blocks
862 to issue, before waiting 'thinktime' usecs. If not set,
863 defaults to 1 which will make fio wait 'thinktime' usecs
864 after every block. This effectively makes any queue depth
865 setting redundant, since no more than 1 IO will be queued
866 before we have to complete it and do our thinktime. In
867 other words, this setting effectively caps the queue depth
868 if the latter is larger.
870 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
871 the normal suffix rules apply. You can use rate=500k to limit
872 reads and writes to 500k each, or you can specify read and
873 writes separately. Using rate=1m,500k would limit reads to
874 1MB/sec and writes to 500KB/sec. Capping only reads or
875 writes can be done with rate=,500k or rate=500k,. The former
876 will only limit writes (to 500KB/sec), the latter will only
879 ratemin=int Tell fio to do whatever it can to maintain at least this
880 bandwidth. Failing to meet this requirement, will cause
881 the job to exit. The same format as rate is used for
882 read vs write separation.
884 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
885 as rate, just specified independently of bandwidth. If the
886 job is given a block size range instead of a fixed value,
887 the smallest block size is used as the metric. The same format
888 as rate is used for read vs write separation.
890 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
891 the job to exit. The same format as rate is used for read vs
894 latency_target=int If set, fio will attempt to find the max performance
895 point that the given workload will run at while maintaining a
896 latency below this target. The values is given in microseconds.
897 See latency_window and latency_percentile
899 latency_window=int Used with latency_target to specify the sample window
900 that the job is run at varying queue depths to test the
901 performance. The value is given in microseconds.
903 latency_percentile=float The percentage of IOs that must fall within the
904 criteria specified by latency_target and latency_window. If not
905 set, this defaults to 100.0, meaning that all IOs must be equal
906 or below to the value set by latency_target.
908 max_latency=int If set, fio will exit the job if it exceeds this maximum
909 latency. It will exit with an ETIME error.
911 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
914 cpumask=int Set the CPU affinity of this job. The parameter given is a
915 bitmask of allowed CPU's the job may run on. So if you want
916 the allowed CPUs to be 1 and 5, you would pass the decimal
917 value of (1 << 1 | 1 << 5), or 34. See man
918 sched_setaffinity(2). This may not work on all supported
919 operating systems or kernel versions. This option doesn't
920 work well for a higher CPU count than what you can store in
921 an integer mask, so it can only control cpus 1-32. For
922 boxes with larger CPU counts, use cpus_allowed.
924 cpus_allowed=str Controls the same options as cpumask, but it allows a text
925 setting of the permitted CPUs instead. So to use CPUs 1 and
926 5, you would specify cpus_allowed=1,5. This options also
927 allows a range of CPUs. Say you wanted a binding to CPUs
928 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
930 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
931 arguments allow comma delimited list of cpu numbers,
932 A-B ranges, or 'all'. Note, to enable numa options support,
933 fio must be built on a system with libnuma-dev(el) installed.
935 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
936 nodes. Format of the argements:
938 `mode' is one of the following memory policy:
939 default, prefer, bind, interleave, local
940 For `default' and `local' memory policy, no node is
941 needed to be specified.
942 For `prefer', only one node is allowed.
943 For `bind' and `interleave', it allow comma delimited
944 list of numbers, A-B ranges, or 'all'.
946 startdelay=time Start this job the specified number of seconds after fio
947 has started. Only useful if the job file contains several
948 jobs, and you want to delay starting some jobs to a certain
951 runtime=time Tell fio to terminate processing after the specified number
952 of seconds. It can be quite hard to determine for how long
953 a specified job will run, so this parameter is handy to
954 cap the total runtime to a given time.
956 time_based If set, fio will run for the duration of the runtime
957 specified even if the file(s) are completely read or
958 written. It will simply loop over the same workload
959 as many times as the runtime allows.
961 ramp_time=time If set, fio will run the specified workload for this amount
962 of time before logging any performance numbers. Useful for
963 letting performance settle before logging results, thus
964 minimizing the runtime required for stable results. Note
965 that the ramp_time is considered lead in time for a job,
966 thus it will increase the total runtime if a special timeout
967 or runtime is specified.
969 invalidate=bool Invalidate the buffer/page cache parts for this file prior
970 to starting io. Defaults to true.
972 sync=bool Use sync io for buffered writes. For the majority of the
973 io engines, this means using O_SYNC.
976 mem=str Fio can use various types of memory as the io unit buffer.
977 The allowed values are:
979 malloc Use memory from malloc(3) as the buffers.
981 shm Use shared memory as the buffers. Allocated
984 shmhuge Same as shm, but use huge pages as backing.
986 mmap Use mmap to allocate buffers. May either be
987 anonymous memory, or can be file backed if
988 a filename is given after the option. The
989 format is mem=mmap:/path/to/file.
991 mmaphuge Use a memory mapped huge file as the buffer
992 backing. Append filename after mmaphuge, ala
993 mem=mmaphuge:/hugetlbfs/file
995 The area allocated is a function of the maximum allowed
996 bs size for the job, multiplied by the io depth given. Note
997 that for shmhuge and mmaphuge to work, the system must have
998 free huge pages allocated. This can normally be checked
999 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1000 Linux system. Fio assumes a huge page is 4MB in size. So
1001 to calculate the number of huge pages you need for a given
1002 job file, add up the io depth of all jobs (normally one unless
1003 iodepth= is used) and multiply by the maximum bs set. Then
1004 divide that number by the huge page size. You can see the
1005 size of the huge pages in /proc/meminfo. If no huge pages
1006 are allocated by having a non-zero number in nr_hugepages,
1007 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1009 mmaphuge also needs to have hugetlbfs mounted and the file
1010 location should point there. So if it's mounted in /huge,
1011 you would use mem=mmaphuge:/huge/somefile.
1013 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1014 Note that the given alignment is applied to the first IO unit
1015 buffer, if using iodepth the alignment of the following buffers
1016 are given by the bs used. In other words, if using a bs that is
1017 a multiple of the page sized in the system, all buffers will
1018 be aligned to this value. If using a bs that is not page
1019 aligned, the alignment of subsequent IO memory buffers is the
1020 sum of the iomem_align and bs used.
1023 Defines the size of a huge page. Must at least be equal
1024 to the system setting, see /proc/meminfo. Defaults to 4MB.
1025 Should probably always be a multiple of megabytes, so using
1026 hugepage-size=Xm is the preferred way to set this to avoid
1027 setting a non-pow-2 bad value.
1029 exitall When one job finishes, terminate the rest. The default is
1030 to wait for each job to finish, sometimes that is not the
1033 bwavgtime=int Average the calculated bandwidth over the given time. Value
1034 is specified in milliseconds.
1036 iopsavgtime=int Average the calculated IOPS over the given time. Value
1037 is specified in milliseconds.
1039 create_serialize=bool If true, serialize the file creating for the jobs.
1040 This may be handy to avoid interleaving of data
1041 files, which may greatly depend on the filesystem
1042 used and even the number of processors in the system.
1044 create_fsync=bool fsync the data file after creation. This is the
1047 create_on_open=bool Don't pre-setup the files for IO, just create open()
1048 when it's time to do IO to that file.
1050 create_only=bool If true, fio will only run the setup phase of the job.
1051 If files need to be laid out or updated on disk, only
1052 that will be done. The actual job contents are not
1055 pre_read=bool If this is given, files will be pre-read into memory before
1056 starting the given IO operation. This will also clear
1057 the 'invalidate' flag, since it is pointless to pre-read
1058 and then drop the cache. This will only work for IO engines
1059 that are seekable, since they allow you to read the same data
1060 multiple times. Thus it will not work on eg network or splice
1063 unlink=bool Unlink the job files when done. Not the default, as repeated
1064 runs of that job would then waste time recreating the file
1065 set again and again.
1067 loops=int Run the specified number of iterations of this job. Used
1068 to repeat the same workload a given number of times. Defaults
1071 verify_only Do not perform specified workload---only verify data still
1072 matches previous invocation of this workload. This option
1073 allows one to check data multiple times at a later date
1074 without overwriting it. This option makes sense only for
1075 workloads that write data, and does not support workloads
1076 with the time_based option set.
1078 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1079 verify is set. Defaults to 1.
1081 verify=str If writing to a file, fio can verify the file contents
1082 after each iteration of the job. The allowed values are:
1084 md5 Use an md5 sum of the data area and store
1085 it in the header of each block.
1087 crc64 Use an experimental crc64 sum of the data
1088 area and store it in the header of each
1091 crc32c Use a crc32c sum of the data area and store
1092 it in the header of each block.
1094 crc32c-intel Use hardware assisted crc32c calcuation
1095 provided on SSE4.2 enabled processors. Falls
1096 back to regular software crc32c, if not
1097 supported by the system.
1099 crc32 Use a crc32 sum of the data area and store
1100 it in the header of each block.
1102 crc16 Use a crc16 sum of the data area and store
1103 it in the header of each block.
1105 crc7 Use a crc7 sum of the data area and store
1106 it in the header of each block.
1108 sha512 Use sha512 as the checksum function.
1110 sha256 Use sha256 as the checksum function.
1112 sha1 Use optimized sha1 as the checksum function.
1114 meta Write extra information about each io
1115 (timestamp, block number etc.). The block
1116 number is verified. The io sequence number is
1117 verified for workloads that write data.
1118 See also verify_pattern.
1120 null Only pretend to verify. Useful for testing
1121 internals with ioengine=null, not for much
1124 This option can be used for repeated burn-in tests of a
1125 system to make sure that the written data is also
1126 correctly read back. If the data direction given is
1127 a read or random read, fio will assume that it should
1128 verify a previously written file. If the data direction
1129 includes any form of write, the verify will be of the
1132 verifysort=bool If set, fio will sort written verify blocks when it deems
1133 it faster to read them back in a sorted manner. This is
1134 often the case when overwriting an existing file, since
1135 the blocks are already laid out in the file system. You
1136 can ignore this option unless doing huge amounts of really
1137 fast IO where the red-black tree sorting CPU time becomes
1140 verify_offset=int Swap the verification header with data somewhere else
1141 in the block before writing. Its swapped back before
1144 verify_interval=int Write the verification header at a finer granularity
1145 than the blocksize. It will be written for chunks the
1146 size of header_interval. blocksize should divide this
1149 verify_pattern=str If set, fio will fill the io buffers with this
1150 pattern. Fio defaults to filling with totally random
1151 bytes, but sometimes it's interesting to fill with a known
1152 pattern for io verification purposes. Depending on the
1153 width of the pattern, fio will fill 1/2/3/4 bytes of the
1154 buffer at the time(it can be either a decimal or a hex number).
1155 The verify_pattern if larger than a 32-bit quantity has to
1156 be a hex number that starts with either "0x" or "0X". Use
1159 verify_fatal=bool Normally fio will keep checking the entire contents
1160 before quitting on a block verification failure. If this
1161 option is set, fio will exit the job on the first observed
1164 verify_dump=bool If set, dump the contents of both the original data
1165 block and the data block we read off disk to files. This
1166 allows later analysis to inspect just what kind of data
1167 corruption occurred. Off by default.
1169 verify_async=int Fio will normally verify IO inline from the submitting
1170 thread. This option takes an integer describing how many
1171 async offload threads to create for IO verification instead,
1172 causing fio to offload the duty of verifying IO contents
1173 to one or more separate threads. If using this offload
1174 option, even sync IO engines can benefit from using an
1175 iodepth setting higher than 1, as it allows them to have
1176 IO in flight while verifies are running.
1178 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1179 async IO verification threads. See cpus_allowed for the
1182 verify_backlog=int Fio will normally verify the written contents of a
1183 job that utilizes verify once that job has completed. In
1184 other words, everything is written then everything is read
1185 back and verified. You may want to verify continually
1186 instead for a variety of reasons. Fio stores the meta data
1187 associated with an IO block in memory, so for large
1188 verify workloads, quite a bit of memory would be used up
1189 holding this meta data. If this option is enabled, fio
1190 will write only N blocks before verifying these blocks.
1192 verify_backlog_batch=int Control how many blocks fio will verify
1193 if verify_backlog is set. If not set, will default to
1194 the value of verify_backlog (meaning the entire queue
1195 is read back and verified). If verify_backlog_batch is
1196 less than verify_backlog then not all blocks will be verified,
1197 if verify_backlog_batch is larger than verify_backlog, some
1198 blocks will be verified more than once.
1201 wait_for_previous Wait for preceding jobs in the job file to exit, before
1202 starting this one. Can be used to insert serialization
1203 points in the job file. A stone wall also implies starting
1204 a new reporting group.
1206 new_group Start a new reporting group. See: group_reporting.
1208 numjobs=int Create the specified number of clones of this job. May be
1209 used to setup a larger number of threads/processes doing
1210 the same thing. Each thread is reported separately; to see
1211 statistics for all clones as a whole, use group_reporting in
1212 conjunction with new_group.
1214 group_reporting It may sometimes be interesting to display statistics for
1215 groups of jobs as a whole instead of for each individual job.
1216 This is especially true if 'numjobs' is used; looking at
1217 individual thread/process output quickly becomes unwieldy.
1218 To see the final report per-group instead of per-job, use
1219 'group_reporting'. Jobs in a file will be part of the same
1220 reporting group, unless if separated by a stonewall, or by
1223 thread fio defaults to forking jobs, however if this option is
1224 given, fio will use pthread_create(3) to create threads
1227 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1229 zoneskip=int Skip the specified number of bytes when zonesize data has
1230 been read. The two zone options can be used to only do
1231 io on zones of a file.
1233 write_iolog=str Write the issued io patterns to the specified file. See
1234 read_iolog. Specify a separate file for each job, otherwise
1235 the iologs will be interspersed and the file may be corrupt.
1237 read_iolog=str Open an iolog with the specified file name and replay the
1238 io patterns it contains. This can be used to store a
1239 workload and replay it sometime later. The iolog given
1240 may also be a blktrace binary file, which allows fio
1241 to replay a workload captured by blktrace. See blktrace
1242 for how to capture such logging data. For blktrace replay,
1243 the file needs to be turned into a blkparse binary data
1244 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1246 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1247 is to attempt to respect the time stamps within the log and
1248 replay them with the appropriate delay between IOPS. By
1249 setting this variable fio will not respect the timestamps and
1250 attempt to replay them as fast as possible while still
1251 respecting ordering. The result is the same I/O pattern to a
1252 given device, but different timings.
1254 replay_redirect=str While replaying I/O patterns using read_iolog the
1255 default behavior is to replay the IOPS onto the major/minor
1256 device that each IOP was recorded from. This is sometimes
1257 undesirable because on a different machine those major/minor
1258 numbers can map to a different device. Changing hardware on
1259 the same system can also result in a different major/minor
1260 mapping. Replay_redirect causes all IOPS to be replayed onto
1261 the single specified device regardless of the device it was
1262 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1263 IO in the blktrace to be replayed onto /dev/sdc. This means
1264 multiple devices will be replayed onto a single, if the trace
1265 contains multiple devices. If you want multiple devices to be
1266 replayed concurrently to multiple redirected devices you must
1267 blkparse your trace into separate traces and replay them with
1268 independent fio invocations. Unfortuantely this also breaks
1269 the strict time ordering between multiple device accesses.
1271 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1272 file. Can be used to store data of the bandwidth of the
1273 jobs in their lifetime. The included fio_generate_plots
1274 script uses gnuplot to turn these text files into nice
1275 graphs. See write_lat_log for behaviour of given
1276 filename. For this option, the suffix is _bw.log.
1278 write_lat_log=str Same as write_bw_log, except that this option stores io
1279 submission, completion, and total latencies instead. If no
1280 filename is given with this option, the default filename of
1281 "jobname_type.log" is used. Even if the filename is given,
1282 fio will still append the type of log. So if one specifies
1286 The actual log names will be foo_slat.log, foo_clat.log,
1287 and foo_lat.log. This helps fio_generate_plot fine the logs
1290 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1291 given with this option, the default filename of
1292 "jobname_type.log" is used. Even if the filename is given,
1293 fio will still append the type of log.
1295 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1296 or bw log for every IO that completes. When writing to the
1297 disk log, that can quickly grow to a very large size. Setting
1298 this option makes fio average the each log entry over the
1299 specified period of time, reducing the resolution of the log.
1302 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1303 potentially be used instead of removing memory or booting
1304 with less memory to simulate a smaller amount of memory.
1305 The amount specified is per worker.
1307 exec_prerun=str Before running this job, issue the command specified
1308 through system(3). Output is redirected in a file called
1311 exec_postrun=str After the job completes, issue the command specified
1312 though system(3). Output is redirected in a file called
1313 jobname.postrun.txt.
1315 ioscheduler=str Attempt to switch the device hosting the file to the specified
1316 io scheduler before running.
1318 disk_util=bool Generate disk utilization statistics, if the platform
1319 supports it. Defaults to on.
1321 disable_lat=bool Disable measurements of total latency numbers. Useful
1322 only for cutting back the number of calls to gettimeofday,
1323 as that does impact performance at really high IOPS rates.
1324 Note that to really get rid of a large amount of these
1325 calls, this option must be used with disable_slat and
1328 disable_clat=bool Disable measurements of completion latency numbers. See
1331 disable_slat=bool Disable measurements of submission latency numbers. See
1334 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1337 clat_percentiles=bool Enable the reporting of percentiles of
1338 completion latencies.
1340 percentile_list=float_list Overwrite the default list of percentiles
1341 for completion latencies. Each number is a floating
1342 number in the range (0,100], and the maximum length of
1343 the list is 20. Use ':' to separate the numbers, and
1344 list the numbers in ascending order. For example,
1345 --percentile_list=99.5:99.9 will cause fio to report
1346 the values of completion latency below which 99.5% and
1347 99.9% of the observed latencies fell, respectively.
1349 clocksource=str Use the given clocksource as the base of timing. The
1350 supported options are:
1352 gettimeofday gettimeofday(2)
1354 clock_gettime clock_gettime(2)
1356 cpu Internal CPU clock source
1358 cpu is the preferred clocksource if it is reliable, as it
1359 is very fast (and fio is heavy on time calls). Fio will
1360 automatically use this clocksource if it's supported and
1361 considered reliable on the system it is running on, unless
1362 another clocksource is specifically set. For x86/x86-64 CPUs,
1363 this means supporting TSC Invariant.
1365 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1366 (disable_clat, disable_slat, disable_bw) plus reduce
1367 precision of the timeout somewhat to really shrink
1368 the gettimeofday() call count. With this option enabled,
1369 we only do about 0.4% of the gtod() calls we would have
1370 done if all time keeping was enabled.
1372 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1373 execution to just getting the current time. Fio (and
1374 databases, for instance) are very intensive on gettimeofday()
1375 calls. With this option, you can set one CPU aside for
1376 doing nothing but logging current time to a shared memory
1377 location. Then the other threads/processes that run IO
1378 workloads need only copy that segment, instead of entering
1379 the kernel with a gettimeofday() call. The CPU set aside
1380 for doing these time calls will be excluded from other
1381 uses. Fio will manually clear it from the CPU mask of other
1384 continue_on_error=str Normally fio will exit the job on the first observed
1385 failure. If this option is set, fio will continue the job when
1386 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1387 is exceeded or the I/O size specified is completed. If this
1388 option is used, there are two more stats that are appended,
1389 the total error count and the first error. The error field
1390 given in the stats is the first error that was hit during the
1393 The allowed values are:
1395 none Exit on any IO or verify errors.
1397 read Continue on read errors, exit on all others.
1399 write Continue on write errors, exit on all others.
1401 io Continue on any IO error, exit on all others.
1403 verify Continue on verify errors, exit on all others.
1405 all Continue on all errors.
1407 0 Backward-compatible alias for 'none'.
1409 1 Backward-compatible alias for 'all'.
1411 ignore_error=str Sometimes you want to ignore some errors during test
1412 in that case you can specify error list for each error type.
1413 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1414 errors for given error type is separated with ':'. Error
1415 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1417 ignore_error=EAGAIN,ENOSPC:122
1418 This option will ignore EAGAIN from READ, and ENOSPC and
1419 122(EDQUOT) from WRITE.
1421 error_dump=bool If set dump every error even if it is non fatal, true
1422 by default. If disabled only fatal error will be dumped
1424 cgroup=str Add job to this control group. If it doesn't exist, it will
1425 be created. The system must have a mounted cgroup blkio
1426 mount point for this to work. If your system doesn't have it
1427 mounted, you can do so with:
1429 # mount -t cgroup -o blkio none /cgroup
1431 cgroup_weight=int Set the weight of the cgroup to this value. See
1432 the documentation that comes with the kernel, allowed values
1433 are in the range of 100..1000.
1435 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1436 the job completion. To override this behavior and to leave
1437 cgroups around after the job completion, set cgroup_nodelete=1.
1438 This can be useful if one wants to inspect various cgroup
1439 files after job completion. Default: false
1441 uid=int Instead of running as the invoking user, set the user ID to
1442 this value before the thread/process does any work.
1444 gid=int Set group ID, see uid.
1446 flow_id=int The ID of the flow. If not specified, it defaults to being a
1447 global flow. See flow.
1449 flow=int Weight in token-based flow control. If this value is used, then
1450 there is a 'flow counter' which is used to regulate the
1451 proportion of activity between two or more jobs. fio attempts
1452 to keep this flow counter near zero. The 'flow' parameter
1453 stands for how much should be added or subtracted to the flow
1454 counter on each iteration of the main I/O loop. That is, if
1455 one job has flow=8 and another job has flow=-1, then there
1456 will be a roughly 1:8 ratio in how much one runs vs the other.
1458 flow_watermark=int The maximum value that the absolute value of the flow
1459 counter is allowed to reach before the job must wait for a
1460 lower value of the counter.
1462 flow_sleep=int The period of time, in microseconds, to wait after the flow
1463 watermark has been exceeded before retrying operations
1465 In addition, there are some parameters which are only valid when a specific
1466 ioengine is in use. These are used identically to normal parameters, with the
1467 caveat that when used on the command line, they must come after the ioengine
1468 that defines them is selected.
1470 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1471 the io_getevents system call to reap newly returned events.
1472 With this flag turned on, the AIO ring will be read directly
1473 from user-space to reap events. The reaping mode is only
1474 enabled when polling for a minimum of 0 events (eg when
1475 iodepth_batch_complete=0).
1477 [cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1479 [cpu] cpuchunks=int Split the load into cycles of the given time. In
1482 [netsplice] hostname=str
1483 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1484 If the job is a TCP listener or UDP reader, the hostname is not
1485 used and must be omitted unless it is a valid UDP multicast
1488 [netsplice] port=int
1489 [net] port=int The TCP or UDP port to bind to or connect to.
1491 [netsplice] interface=str
1492 [net] interface=str The IP address of the network interface used to send or
1493 receive UDP multicast
1496 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1499 [netsplice] nodelay=bool
1500 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1502 [netsplice] protocol=str
1503 [netsplice] proto=str
1505 [net] proto=str The network protocol to use. Accepted values are:
1507 tcp Transmission control protocol
1508 tcpv6 Transmission control protocol V6
1509 udp User datagram protocol
1510 udpv6 User datagram protocol V6
1511 unix UNIX domain socket
1513 When the protocol is TCP or UDP, the port must also be given,
1514 as well as the hostname if the job is a TCP listener or UDP
1515 reader. For unix sockets, the normal filename option should be
1516 used and the port is invalid.
1518 [net] listen For TCP network connections, tell fio to listen for incoming
1519 connections rather than initiating an outgoing connection. The
1520 hostname must be omitted if this option is used.
1521 [net] pingpong Normaly a network writer will just continue writing data, and
1522 a network reader will just consume packages. If pingpong=1
1523 is set, a writer will send its normal payload to the reader,
1524 then wait for the reader to send the same payload back. This
1525 allows fio to measure network latencies. The submission
1526 and completion latencies then measure local time spent
1527 sending or receiving, and the completion latency measures
1528 how long it took for the other end to receive and send back.
1529 For UDP multicast traffic pingpong=1 should only be set for a
1530 single reader when multiple readers are listening to the same
1533 [e4defrag] donorname=str
1534 File will be used as a block donor(swap extents between files)
1535 [e4defrag] inplace=int
1536 Configure donor file blocks allocation strategy
1537 0(default): Preallocate donor's file on init
1538 1 : allocate space immidietly inside defragment event,
1539 and free right after event
1543 6.0 Interpreting the output
1544 ---------------------------
1546 fio spits out a lot of output. While running, fio will display the
1547 status of the jobs created. An example of that would be:
1549 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1551 The characters inside the square brackets denote the current status of
1552 each thread. The possible values (in typical life cycle order) are:
1556 P Thread setup, but not started.
1558 I Thread initialized, waiting or generating necessary data.
1559 p Thread running pre-reading file(s).
1560 R Running, doing sequential reads.
1561 r Running, doing random reads.
1562 W Running, doing sequential writes.
1563 w Running, doing random writes.
1564 M Running, doing mixed sequential reads/writes.
1565 m Running, doing mixed random reads/writes.
1566 F Running, currently waiting for fsync()
1567 V Running, doing verification of written data.
1568 E Thread exited, not reaped by main thread yet.
1570 X Thread reaped, exited with an error.
1571 K Thread reaped, exited due to signal.
1573 The other values are fairly self explanatory - number of threads
1574 currently running and doing io, rate of io since last check (read speed
1575 listed first, then write speed), and the estimated completion percentage
1576 and time for the running group. It's impossible to estimate runtime of
1577 the following groups (if any). Note that the string is displayed in order,
1578 so it's possible to tell which of the jobs are currently doing what. The
1579 first character is the first job defined in the job file, and so forth.
1581 When fio is done (or interrupted by ctrl-c), it will show the data for
1582 each thread, group of threads, and disks in that order. For each data
1583 direction, the output looks like:
1585 Client1 (g=0): err= 0:
1586 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1587 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1588 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1589 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1590 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1591 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1592 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1593 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1594 issued r/w: total=0/32768, short=0/0
1595 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1596 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1598 The client number is printed, along with the group id and error of that
1599 thread. Below is the io statistics, here for writes. In the order listed,
1602 io= Number of megabytes io performed
1603 bw= Average bandwidth rate
1604 iops= Average IOs performed per second
1605 runt= The runtime of that thread
1606 slat= Submission latency (avg being the average, stdev being the
1607 standard deviation). This is the time it took to submit
1608 the io. For sync io, the slat is really the completion
1609 latency, since queue/complete is one operation there. This
1610 value can be in milliseconds or microseconds, fio will choose
1611 the most appropriate base and print that. In the example
1612 above, milliseconds is the best scale. Note: in --minimal mode
1613 latencies are always expressed in microseconds.
1614 clat= Completion latency. Same names as slat, this denotes the
1615 time from submission to completion of the io pieces. For
1616 sync io, clat will usually be equal (or very close) to 0,
1617 as the time from submit to complete is basically just
1618 CPU time (io has already been done, see slat explanation).
1619 bw= Bandwidth. Same names as the xlat stats, but also includes
1620 an approximate percentage of total aggregate bandwidth
1621 this thread received in this group. This last value is
1622 only really useful if the threads in this group are on the
1623 same disk, since they are then competing for disk access.
1624 cpu= CPU usage. User and system time, along with the number
1625 of context switches this thread went through, usage of
1626 system and user time, and finally the number of major
1627 and minor page faults.
1628 IO depths= The distribution of io depths over the job life time. The
1629 numbers are divided into powers of 2, so for example the
1630 16= entries includes depths up to that value but higher
1631 than the previous entry. In other words, it covers the
1632 range from 16 to 31.
1633 IO submit= How many pieces of IO were submitting in a single submit
1634 call. Each entry denotes that amount and below, until
1635 the previous entry - eg, 8=100% mean that we submitted
1636 anywhere in between 5-8 ios per submit call.
1637 IO complete= Like the above submit number, but for completions instead.
1638 IO issued= The number of read/write requests issued, and how many
1640 IO latencies= The distribution of IO completion latencies. This is the
1641 time from when IO leaves fio and when it gets completed.
1642 The numbers follow the same pattern as the IO depths,
1643 meaning that 2=1.6% means that 1.6% of the IO completed
1644 within 2 msecs, 20=12.8% means that 12.8% of the IO
1645 took more than 10 msecs, but less than (or equal to) 20 msecs.
1647 After each client has been listed, the group statistics are printed. They
1648 will look like this:
1650 Run status group 0 (all jobs):
1651 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1652 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1654 For each data direction, it prints:
1656 io= Number of megabytes io performed.
1657 aggrb= Aggregate bandwidth of threads in this group.
1658 minb= The minimum average bandwidth a thread saw.
1659 maxb= The maximum average bandwidth a thread saw.
1660 mint= The smallest runtime of the threads in that group.
1661 maxt= The longest runtime of the threads in that group.
1663 And finally, the disk statistics are printed. They will look like this:
1665 Disk stats (read/write):
1666 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1668 Each value is printed for both reads and writes, with reads first. The
1671 ios= Number of ios performed by all groups.
1672 merge= Number of merges io the io scheduler.
1673 ticks= Number of ticks we kept the disk busy.
1674 io_queue= Total time spent in the disk queue.
1675 util= The disk utilization. A value of 100% means we kept the disk
1676 busy constantly, 50% would be a disk idling half of the time.
1678 It is also possible to get fio to dump the current output while it is
1679 running, without terminating the job. To do that, send fio the USR1 signal.
1680 You can also get regularly timed dumps by using the --status-interval
1681 parameter, or by creating a file in /tmp named fio-dump-status. If fio
1682 sees this file, it will unlink it and dump the current output status.
1688 For scripted usage where you typically want to generate tables or graphs
1689 of the results, fio can output the results in a semicolon separated format.
1690 The format is one long line of values, such as:
1692 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%
1693 A description of this job goes here.
1695 The job description (if provided) follows on a second line.
1697 To enable terse output, use the --minimal command line option. The first
1698 value is the version of the terse output format. If the output has to
1699 be changed for some reason, this number will be incremented by 1 to
1700 signify that change.
1702 Split up, the format is as follows:
1704 terse version, fio version, jobname, groupid, error
1706 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1707 Submission latency: min, max, mean, deviation (usec)
1708 Completion latency: min, max, mean, deviation (usec)
1709 Completion latency percentiles: 20 fields (see below)
1710 Total latency: min, max, mean, deviation (usec)
1711 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1713 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1714 Submission latency: min, max, mean, deviation (usec)
1715 Completion latency: min, max, mean, deviation (usec)
1716 Completion latency percentiles: 20 fields (see below)
1717 Total latency: min, max, mean, deviation (usec)
1718 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1719 CPU usage: user, system, context switches, major faults, minor faults
1720 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1721 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1722 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1723 Disk utilization: Disk name, Read ios, write ios,
1724 Read merges, write merges,
1725 Read ticks, write ticks,
1726 Time spent in queue, disk utilization percentage
1727 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
1729 Additional Info (dependent on description being set): Text description
1731 Completion latency percentiles can be a grouping of up to 20 sets, so
1732 for the terse output fio writes all of them. Each field will look like this:
1736 which is the Xth percentile, and the usec latency associated with it.
1738 For disk utilization, all disks used by fio are shown. So for each disk
1739 there will be a disk utilization section.
1742 8.0 Trace file format
1743 ---------------------
1744 There are two trace file format that you can encounter. The older (v1) format
1745 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1746 below in case that you get an old trace and want to understand it.
1748 In any case the trace is a simple text file with a single action per line.
1751 8.1 Trace file format v1
1752 ------------------------
1753 Each line represents a single io action in the following format:
1757 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1759 This format is not supported in Fio versions => 1.20-rc3.
1762 8.2 Trace file format v2
1763 ------------------------
1764 The second version of the trace file format was added in Fio version 1.17.
1765 It allows to access more then one file per trace and has a bigger set of
1766 possible file actions.
1768 The first line of the trace file has to be:
1772 Following this can be lines in two different formats, which are described below.
1774 The file management format:
1778 The filename is given as an absolute path. The action can be one of these:
1780 add Add the given filename to the trace
1781 open Open the file with the given filename. The filename has to have
1782 been added with the add action before.
1783 close Close the file with the given filename. The file has to have been
1787 The file io action format:
1789 filename action offset length
1791 The filename is given as an absolute path, and has to have been added and opened
1792 before it can be used with this format. The offset and length are given in
1793 bytes. The action can be one of these:
1795 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1796 read Read 'length' bytes beginning from 'offset'
1797 write Write 'length' bytes beginning from 'offset'
1798 sync fsync() the file
1799 datasync fdatasync() the file
1800 trim trim the given file from the given 'offset' for 'length' bytes
1803 9.0 CPU idleness profiling
1804 --------------------------
1805 In some cases, we want to understand CPU overhead in a test. For example,
1806 we test patches for the specific goodness of whether they reduce CPU usage.
1807 fio implements a balloon approach to create a thread per CPU that runs at
1808 idle priority, meaning that it only runs when nobody else needs the cpu.
1809 By measuring the amount of work completed by the thread, idleness of each
1810 CPU can be derived accordingly.
1812 An unit work is defined as touching a full page of unsigned characters. Mean
1813 and standard deviation of time to complete an unit work is reported in "unit
1814 work" section. Options can be chosen to report detailed percpu idleness or
1815 overall system idleness by aggregating percpu stats.