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 in the parameter section.
86 If the job file contains only one job, you may as well just give the
87 parameters on the command line. The command line parameters are identical
88 to the job parameters, with a few extra that control global parameters
89 (see README). For example, for the job file parameter iodepth=2, the
90 mirror command line option would be --iodepth 2 or --iodepth=2. You can
91 also use the command line for giving more than one job entry. For each
92 --name option that fio sees, it will start a new job with that name.
93 Command line entries following a --name entry will apply to that job,
94 until there are no more entries or a new --name entry is seen. This is
95 similar to the job file options, where each option applies to the current
96 job until a new [] job entry is seen.
98 fio does not need to run as root, except if the files or devices specified
99 in the job section requires that. Some other options may also be restricted,
100 such as memory locking, io scheduler switching, and decreasing the nice value.
105 As previously described, fio accepts one or more job files describing
106 what it is supposed to do. The job file format is the classic ini file,
107 where the names enclosed in [] brackets define the job name. You are free
108 to use any ascii name you want, except 'global' which has special meaning.
109 A global section sets defaults for the jobs described in that file. A job
110 may override a global section parameter, and a job file may even have
111 several global sections if so desired. A job is only affected by a global
112 section residing above it. If the first character in a line is a ';' or a
113 '#', the entire line is discarded as a comment.
115 So let's look at a really simple job file that defines two processes, each
116 randomly reading from a 128MB file.
118 ; -- start job file --
129 As you can see, the job file sections themselves are empty as all the
130 described parameters are shared. As no filename= option is given, fio
131 makes up a filename for each of the jobs as it sees fit. On the command
132 line, this job would look as follows:
134 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
137 Let's look at an example that has a number of processes writing randomly
140 ; -- start job file --
152 Here we have no global section, as we only have one job defined anyway.
153 We want to use async io here, with a depth of 4 for each file. We also
154 increased the buffer size used to 32KB and define numjobs to 4 to
155 fork 4 identical jobs. The result is 4 processes each randomly writing
156 to their own 64MB file. Instead of using the above job file, you could
157 have given the parameters on the command line. For this case, you would
160 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
162 4.1 Environment variables
163 -------------------------
165 fio also supports environment variable expansion in job files. Any
166 substring of the form "${VARNAME}" as part of an option value (in other
167 words, on the right of the `='), will be expanded to the value of the
168 environment variable called VARNAME. If no such environment variable
169 is defined, or VARNAME is the empty string, the empty string will be
172 As an example, let's look at a sample fio invocation and job file:
174 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
176 ; -- start job file --
183 This will expand to the following equivalent job file at runtime:
185 ; -- start job file --
192 fio ships with a few example job files, you can also look there for
195 4.2 Reserved keywords
196 ---------------------
198 Additionally, fio has a set of reserved keywords that will be replaced
199 internally with the appropriate value. Those keywords are:
201 $pagesize The architecture page size of the running system
202 $mb_memory Megabytes of total memory in the system
203 $ncpus Number of online available CPUs
205 These can be used on the command line or in the job file, and will be
206 automatically substituted with the current system values when the job
207 is run. Simple math is also supported on these keywords, so you can
208 perform actions like:
212 and get that properly expanded to 8 times the size of memory in the
216 5.0 Detailed list of parameters
217 -------------------------------
219 This section describes in details each parameter associated with a job.
220 Some parameters take an option of a given type, such as an integer or
221 a string. The following types are used:
223 str String. This is a sequence of alpha characters.
224 time Integer with possible time suffix. In seconds unless otherwise
225 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
226 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
227 and 'us' (or 'usec') for microseconds.
228 int SI integer. A whole number value, which may contain a suffix
229 describing the base of the number. Accepted suffixes are k/m/g/t/p,
230 meaning kilo, mega, giga, tera, and peta. The suffix is not case
231 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
232 as 'k'). So if you want to specify 4096, you could either write
233 out '4096' or just give 4k. The suffixes signify base 2 values, so
234 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
235 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
236 case, then 1000 is used as the multiplier. This can be handy for
237 disks, since manufacturers generally use base 10 values when listing
238 the capacity of a drive. If the option accepts an upper and lower
239 range, use a colon ':' or minus '-' to separate such values. May also
240 include a prefix to indicate numbers base. If 0x is used, the number
241 is assumed to be hexadecimal. See irange.
242 bool Boolean. Usually parsed as an integer, however only defined for
243 true and false (1 and 0).
244 irange Integer range with suffix. Allows value range to be given, such
245 as 1024-4096. A colon may also be used as the separator, eg
246 1k:4k. If the option allows two sets of ranges, they can be
247 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
249 float_list A list of floating numbers, separated by a ':' character.
251 With the above in mind, here follows the complete list of fio job
254 name=str ASCII name of the job. This may be used to override the
255 name printed by fio for this job. Otherwise the job
256 name is used. On the command line this parameter has the
257 special purpose of also signaling the start of a new
260 description=str Text description of the job. Doesn't do anything except
261 dump this text description when this job is run. It's
264 directory=str Prefix filenames with this directory. Used to place files
265 in a different location than "./". See the 'filename' option
266 for escaping certain characters.
268 filename=str Fio normally makes up a filename based on the job name,
269 thread number, and file number. If you want to share
270 files between threads in a job or several jobs, specify
271 a filename for each of them to override the default. If
272 the ioengine used is 'net', the filename is the host, port,
273 and protocol to use in the format of =host,port,protocol.
274 See ioengine=net for more. If the ioengine is file based, you
275 can specify a number of files by separating the names with a
276 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
277 as the two working files, you would use
278 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
279 accessed as \\.\PhysicalDrive0 for the first device,
280 \\.\PhysicalDrive1 for the second etc. Note: Windows and
281 FreeBSD prevent write access to areas of the disk containing
282 in-use data (e.g. filesystems).
283 If the wanted filename does need to include a colon, then
284 escape that with a '\' character. For instance, if the filename
285 is "/dev/dsk/foo@3,0:c", then you would use
286 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
287 stdin or stdout. Which of the two depends on the read/write
291 If sharing multiple files between jobs, it is usually necessary
292 to have fio generate the exact names that you want. By default,
293 fio will name a file based on the default file format
294 specification of jobname.jobnumber.filenumber. With this
295 option, that can be customized. Fio will recognize and replace
296 the following keywords in this string:
299 The name of the worker thread or process.
302 The incremental number of the worker thread or
306 The incremental number of the file for that worker
309 To have dependent jobs share a set of files, this option can
310 be set to have fio generate filenames that are shared between
311 the two. For instance, if testfiles.$filenum is specified,
312 file number 4 for any job will be named testfiles.4. The
313 default of $jobname.$jobnum.$filenum will be used if
314 no other format specifier is given.
316 opendir=str Tell fio to recursively add any file it can find in this
317 directory and down the file system tree.
319 lockfile=str Fio defaults to not locking any files before it does
320 IO to them. If a file or file descriptor is shared, fio
321 can serialize IO to that file to make the end result
322 consistent. This is usual for emulating real workloads that
323 share files. The lock modes are:
325 none No locking. The default.
326 exclusive Only one thread/process may do IO,
327 excluding all others.
328 readwrite Read-write locking on the file. Many
329 readers may access the file at the
330 same time, but writes get exclusive
334 rw=str Type of io pattern. Accepted values are:
336 read Sequential reads
337 write Sequential writes
338 randwrite Random writes
339 randread Random reads
340 rw,readwrite Sequential mixed reads and writes
341 randrw Random mixed reads and writes
343 For the mixed io types, the default is to split them 50/50.
344 For certain types of io the result may still be skewed a bit,
345 since the speed may be different. It is possible to specify
346 a number of IO's to do before getting a new offset, this is
347 one by appending a ':<nr>' to the end of the string given.
348 For a random read, it would look like 'rw=randread:8' for
349 passing in an offset modifier with a value of 8. If the
350 suffix is used with a sequential IO pattern, then the value
351 specified will be added to the generated offset for each IO.
352 For instance, using rw=write:4k will skip 4k for every
353 write. It turns sequential IO into sequential IO with holes.
354 See the 'rw_sequencer' option.
356 rw_sequencer=str If an offset modifier is given by appending a number to
357 the rw=<str> line, then this option controls how that
358 number modifies the IO offset being generated. Accepted
361 sequential Generate sequential offset
362 identical Generate the same offset
364 'sequential' is only useful for random IO, where fio would
365 normally generate a new random offset for every IO. If you
366 append eg 8 to randread, you would get a new random offset for
367 every 8 IO's. The result would be a seek for only every 8
368 IO's, instead of for every IO. Use rw=randread:8 to specify
369 that. As sequential IO is already sequential, setting
370 'sequential' for that would not result in any differences.
371 'identical' behaves in a similar fashion, except it sends
372 the same offset 8 number of times before generating a new
375 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
376 Storage manufacturers like to use 10^3 or 1000 as a base
377 ten unit instead, for obvious reasons. Allow values are
378 1024 or 1000, with 1024 being the default.
380 unified_rw_reporting=bool Fio normally reports statistics on a per
381 data direction basis, meaning that read, write, and trim are
382 accounted and reported separately. If this option is set,
383 the fio will sum the results and report them as "mixed"
386 randrepeat=bool For random IO workloads, seed the generator in a predictable
387 way so that results are repeatable across repetitions.
389 randseed=int Seed the random number generators based on this seed value, to
390 be able to control what sequence of output is being generated.
391 If not set, the random sequence depends on the randrepeat
394 use_os_rand=bool Fio can either use the random generator supplied by the OS
395 to generator random offsets, or it can use it's own internal
396 generator (based on Tausworthe). Default is to use the
397 internal generator, which is often of better quality and
400 fallocate=str Whether pre-allocation is performed when laying down files.
403 none Do not pre-allocate space
404 posix Pre-allocate via posix_fallocate()
405 keep Pre-allocate via fallocate() with
406 FALLOC_FL_KEEP_SIZE set
407 0 Backward-compatible alias for 'none'
408 1 Backward-compatible alias for 'posix'
410 May not be available on all supported platforms. 'keep' is only
411 available on Linux.If using ZFS on Solaris this must be set to
412 'none' because ZFS doesn't support it. Default: 'posix'.
414 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
415 on what IO patterns it is likely to issue. Sometimes you
416 want to test specific IO patterns without telling the
417 kernel about it, in which case you can disable this option.
418 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
419 IO and POSIX_FADV_RANDOM for random IO.
421 size=int The total size of file io for this job. Fio will run until
422 this many bytes has been transferred, unless runtime is
423 limited by other options (such as 'runtime', for instance).
424 Unless specific nrfiles and filesize options are given,
425 fio will divide this size between the available files
426 specified by the job. If not set, fio will use the full
427 size of the given files or devices. If the files do not
428 exist, size must be given. It is also possible to give
429 size as a percentage between 1 and 100. If size=20% is
430 given, fio will use 20% of the full size of the given
433 io_limit=int Normally fio operates within the region set by 'size', which
434 means that the 'size' option sets both the region and size of
435 IO to be performed. Sometimes that is not what you want. With
436 this option, it is possible to define just the amount of IO
437 that fio should do. For instance, if 'size' is set to 20G and
438 'io_limit' is set to 5G, fio will perform IO within the first
439 20G but exit when 5G have been done.
441 filesize=int Individual file sizes. May be a range, in which case fio
442 will select sizes for files at random within the given range
443 and limited to 'size' in total (if that is given). If not
444 given, each created file is the same size.
446 file_append=bool Perform IO after the end of the file. Normally fio will
447 operate within the size of a file. If this option is set, then
448 fio will append to the file instead. This has identical
449 behavior to setting offset to the size of a file. This option
450 is ignored on non-regular files.
453 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
454 space left on device) as the terminating condition. Only makes
455 sense with sequential write. For a read workload, the mount
456 point will be filled first then IO started on the result. This
457 option doesn't make sense if operating on a raw device node,
458 since the size of that is already known by the file system.
459 Additionally, writing beyond end-of-device will not return
463 bs=int The block size used for the io units. Defaults to 4k. Values
464 can be given for both read and writes. If a single int is
465 given, it will apply to both. If a second int is specified
466 after a comma, it will apply to writes only. In other words,
467 the format is either bs=read_and_write or bs=read,write,trim.
468 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
469 writes, and 8k for trims. You can terminate the list with
470 a trailing comma. bs=4k,8k, would use the default value for
471 trims.. If you only wish to set the write size, you
472 can do so by passing an empty read size - bs=,8k will set
473 8k for writes and leave the read default value.
476 ba=int At what boundary to align random IO offsets. Defaults to
477 the same as 'blocksize' the minimum blocksize given.
478 Minimum alignment is typically 512b for using direct IO,
479 though it usually depends on the hardware block size. This
480 option is mutually exclusive with using a random map for
481 files, so it will turn off that option.
483 blocksize_range=irange
484 bsrange=irange Instead of giving a single block size, specify a range
485 and fio will mix the issued io block sizes. The issued
486 io unit will always be a multiple of the minimum value
487 given (also see bs_unaligned). Applies to both reads and
488 writes, however a second range can be given after a comma.
491 bssplit=str Sometimes you want even finer grained control of the
492 block sizes issued, not just an even split between them.
493 This option allows you to weight various block sizes,
494 so that you are able to define a specific amount of
495 block sizes issued. The format for this option is:
497 bssplit=blocksize/percentage:blocksize/percentage
499 for as many block sizes as needed. So if you want to define
500 a workload that has 50% 64k blocks, 10% 4k blocks, and
501 40% 32k blocks, you would write:
503 bssplit=4k/10:64k/50:32k/40
505 Ordering does not matter. If the percentage is left blank,
506 fio will fill in the remaining values evenly. So a bssplit
507 option like this one:
509 bssplit=4k/50:1k/:32k/
511 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
512 always add up to 100, if bssplit is given a range that adds
513 up to more, it will error out.
515 bssplit also supports giving separate splits to reads and
516 writes. The format is identical to what bs= accepts. You
517 have to separate the read and write parts with a comma. So
518 if you want a workload that has 50% 2k reads and 50% 4k reads,
519 while having 90% 4k writes and 10% 8k writes, you would
522 bssplit=2k/50:4k/50,4k/90,8k/10
525 bs_unaligned If this option is given, any byte size value within bsrange
526 may be used as a block range. This typically wont work with
527 direct IO, as that normally requires sector alignment.
529 bs_is_seq_rand If this option is set, fio will use the normal read,write
530 blocksize settings as sequential,random instead. Any random
531 read or write will use the WRITE blocksize settings, and any
532 sequential read or write will use the READ blocksize setting.
534 zero_buffers If this option is given, fio will init the IO buffers to
535 all zeroes. The default is to fill them with random data.
536 The resulting IO buffers will not be completely zeroed,
537 unless scramble_buffers is also turned off.
539 refill_buffers If this option is given, fio will refill the IO buffers
540 on every submit. The default is to only fill it at init
541 time and reuse that data. Only makes sense if zero_buffers
542 isn't specified, naturally. If data verification is enabled,
543 refill_buffers is also automatically enabled.
545 scramble_buffers=bool If refill_buffers is too costly and the target is
546 using data deduplication, then setting this option will
547 slightly modify the IO buffer contents to defeat normal
548 de-dupe attempts. This is not enough to defeat more clever
549 block compression attempts, but it will stop naive dedupe of
550 blocks. Default: true.
552 buffer_compress_percentage=int If this is set, then fio will attempt to
553 provide IO buffer content (on WRITEs) that compress to
554 the specified level. Fio does this by providing a mix of
555 random data and zeroes. Note that this is per block size
556 unit, for file/disk wide compression level that matches
557 this setting, you'll also want to set refill_buffers.
559 buffer_compress_chunk=int See buffer_compress_percentage. This
560 setting allows fio to manage how big the ranges of random
561 data and zeroed data is. Without this set, fio will
562 provide buffer_compress_percentage of blocksize random
563 data, followed by the remaining zeroed. With this set
564 to some chunk size smaller than the block size, fio can
565 alternate random and zeroed data throughout the IO
568 buffer_pattern=str If set, fio will fill the io buffers with this pattern.
569 If not set, the contents of io buffers is defined by the other
570 options related to buffer contents. The setting can be any
571 pattern of bytes, and can be prefixed with 0x for hex values.
573 nrfiles=int Number of files to use for this job. Defaults to 1.
575 openfiles=int Number of files to keep open at the same time. Defaults to
576 the same as nrfiles, can be set smaller to limit the number
579 file_service_type=str Defines how fio decides which file from a job to
580 service next. The following types are defined:
582 random Just choose a file at random.
584 roundrobin Round robin over open files. This
587 sequential Finish one file before moving on to
588 the next. Multiple files can still be
589 open depending on 'openfiles'.
591 The string can have a number appended, indicating how
592 often to switch to a new file. So if option random:4 is
593 given, fio will switch to a new random file after 4 ios
596 ioengine=str Defines how the job issues io to the file. The following
599 sync Basic read(2) or write(2) io. lseek(2) is
600 used to position the io location.
602 psync Basic pread(2) or pwrite(2) io.
604 vsync Basic readv(2) or writev(2) IO.
606 psyncv Basic preadv(2) or pwritev(2) IO.
608 libaio Linux native asynchronous io. Note that Linux
609 may only support queued behaviour with
610 non-buffered IO (set direct=1 or buffered=0).
611 This engine defines engine specific options.
613 posixaio glibc posix asynchronous io.
615 solarisaio Solaris native asynchronous io.
617 windowsaio Windows native asynchronous io.
619 mmap File is memory mapped and data copied
620 to/from using memcpy(3).
622 splice splice(2) is used to transfer the data and
623 vmsplice(2) to transfer data from user
626 syslet-rw Use the syslet system calls to make
627 regular read/write async.
629 sg SCSI generic sg v3 io. May either be
630 synchronous using the SG_IO ioctl, or if
631 the target is an sg character device
632 we use read(2) and write(2) for asynchronous
635 null Doesn't transfer any data, just pretends
636 to. This is mainly used to exercise fio
637 itself and for debugging/testing purposes.
639 net Transfer over the network to given host:port.
640 Depending on the protocol used, the hostname,
641 port, listen and filename options are used to
642 specify what sort of connection to make, while
643 the protocol option determines which protocol
645 This engine defines engine specific options.
647 netsplice Like net, but uses splice/vmsplice to
648 map data and send/receive.
649 This engine defines engine specific options.
651 cpuio Doesn't transfer any data, but burns CPU
652 cycles according to the cpuload= and
653 cpucycle= options. Setting cpuload=85
654 will cause that job to do nothing but burn
655 85% of the CPU. In case of SMP machines,
656 use numjobs=<no_of_cpu> to get desired CPU
657 usage, as the cpuload only loads a single
658 CPU at the desired rate.
660 guasi The GUASI IO engine is the Generic Userspace
661 Asyncronous Syscall Interface approach
664 http://www.xmailserver.org/guasi-lib.html
666 for more info on GUASI.
668 rdma The RDMA I/O engine supports both RDMA
669 memory semantics (RDMA_WRITE/RDMA_READ) and
670 channel semantics (Send/Recv) for the
671 InfiniBand, RoCE and iWARP protocols.
673 falloc IO engine that does regular fallocate to
674 simulate data transfer as fio ioengine.
675 DDIR_READ does fallocate(,mode = keep_size,)
676 DDIR_WRITE does fallocate(,mode = 0)
677 DDIR_TRIM does fallocate(,mode = punch_hole)
679 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
680 ioctls to simulate defragment activity in
681 request to DDIR_WRITE event
683 external Prefix to specify loading an external
684 IO engine object file. Append the engine
685 filename, eg ioengine=external:/tmp/foo.o
686 to load ioengine foo.o in /tmp.
688 iodepth=int This defines how many io units to keep in flight against
689 the file. The default is 1 for each file defined in this
690 job, can be overridden with a larger value for higher
691 concurrency. Note that increasing iodepth beyond 1 will not
692 affect synchronous ioengines (except for small degress when
693 verify_async is in use). Even async engines may impose OS
694 restrictions causing the desired depth not to be achieved.
695 This may happen on Linux when using libaio and not setting
696 direct=1, since buffered IO is not async on that OS. Keep an
697 eye on the IO depth distribution in the fio output to verify
698 that the achieved depth is as expected. Default: 1.
700 iodepth_batch_submit=int
701 iodepth_batch=int This defines how many pieces of IO to submit at once.
702 It defaults to 1 which means that we submit each IO
703 as soon as it is available, but can be raised to submit
704 bigger batches of IO at the time.
706 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
707 at once. It defaults to 1 which means that we'll ask
708 for a minimum of 1 IO in the retrieval process from
709 the kernel. The IO retrieval will go on until we
710 hit the limit set by iodepth_low. If this variable is
711 set to 0, then fio will always check for completed
712 events before queuing more IO. This helps reduce
713 IO latency, at the cost of more retrieval system calls.
715 iodepth_low=int The low water mark indicating when to start filling
716 the queue again. Defaults to the same as iodepth, meaning
717 that fio will attempt to keep the queue full at all times.
718 If iodepth is set to eg 16 and iodepth_low is set to 4, then
719 after fio has filled the queue of 16 requests, it will let
720 the depth drain down to 4 before starting to fill it again.
722 direct=bool If value is true, use non-buffered io. This is usually
723 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
724 On Windows the synchronous ioengines don't support direct io.
726 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
727 writes are guaranteed to be stable once acknowledged by
728 the operating system. Only Linux supports O_ATOMIC right
731 buffered=bool If value is true, use buffered io. This is the opposite
732 of the 'direct' option. Defaults to true.
734 offset=int Start io at the given offset in the file. The data before
735 the given offset will not be touched. This effectively
736 caps the file size at real_size - offset.
738 offset_increment=int If this is provided, then the real offset becomes
739 the offset + offset_increment * thread_number, where the
740 thread number is a counter that starts at 0 and is incremented
741 for each job. This option is useful if there are several jobs
742 which are intended to operate on a file in parallel in disjoint
743 segments, with even spacing between the starting points.
745 number_ios=int Fio will normally perform IOs until it has exhausted the size
746 of the region set by size=, or if it exhaust the allocated
747 time (or hits an error condition). With this setting, the
748 range/size can be set independently of the number of IOs to
749 perform. When fio reaches this number, it will exit normally
752 fsync=int If writing to a file, issue a sync of the dirty data
753 for every number of blocks given. For example, if you give
754 32 as a parameter, fio will sync the file for every 32
755 writes issued. If fio is using non-buffered io, we may
756 not sync the file. The exception is the sg io engine, which
757 synchronizes the disk cache anyway.
759 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
761 In FreeBSD and Windows there is no fdatasync(), this falls back to
764 sync_file_range=str:val Use sync_file_range() for every 'val' number of
765 write operations. Fio will track range of writes that
766 have happened since the last sync_file_range() call. 'str'
767 can currently be one or more of:
769 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
770 write SYNC_FILE_RANGE_WRITE
771 wait_after SYNC_FILE_RANGE_WAIT_AFTER
773 So if you do sync_file_range=wait_before,write:8, fio would
774 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
775 every 8 writes. Also see the sync_file_range(2) man page.
776 This option is Linux specific.
778 overwrite=bool If true, writes to a file will always overwrite existing
779 data. If the file doesn't already exist, it will be
780 created before the write phase begins. If the file exists
781 and is large enough for the specified write phase, nothing
784 end_fsync=bool If true, fsync file contents when a write stage has completed.
786 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
787 This differs from end_fsync in that it will happen on every
788 file close, not just at the end of the job.
790 rwmixread=int How large a percentage of the mix should be reads.
792 rwmixwrite=int How large a percentage of the mix should be writes. If both
793 rwmixread and rwmixwrite is given and the values do not add
794 up to 100%, the latter of the two will be used to override
795 the first. This may interfere with a given rate setting,
796 if fio is asked to limit reads or writes to a certain rate.
797 If that is the case, then the distribution may be skewed.
799 random_distribution=str:float By default, fio will use a completely uniform
800 random distribution when asked to perform random IO. Sometimes
801 it is useful to skew the distribution in specific ways,
802 ensuring that some parts of the data is more hot than others.
803 fio includes the following distribution models:
805 random Uniform random distribution
806 zipf Zipf distribution
807 pareto Pareto distribution
809 When using a zipf or pareto distribution, an input value
810 is also needed to define the access pattern. For zipf, this
811 is the zipf theta. For pareto, it's the pareto power. Fio
812 includes a test program, genzipf, that can be used visualize
813 what the given input values will yield in terms of hit rates.
814 If you wanted to use zipf with a theta of 1.2, you would use
815 random_distribution=zipf:1.2 as the option. If a non-uniform
816 model is used, fio will disable use of the random map.
818 percentage_random=int For a random workload, set how big a percentage should
819 be random. This defaults to 100%, in which case the workload
820 is fully random. It can be set from anywhere from 0 to 100.
821 Setting it to 0 would make the workload fully sequential. Any
822 setting in between will result in a random mix of sequential
823 and random IO, at the given percentages. It is possible to
824 set different values for reads, writes, and trim. To do so,
825 simply use a comma separated list. See blocksize.
827 norandommap Normally fio will cover every block of the file when doing
828 random IO. If this option is given, fio will just get a
829 new random offset without looking at past io history. This
830 means that some blocks may not be read or written, and that
831 some blocks may be read/written more than once. This option
832 is mutually exclusive with verify= if and only if multiple
833 blocksizes (via bsrange=) are used, since fio only tracks
834 complete rewrites of blocks.
836 softrandommap=bool See norandommap. If fio runs with the random block map
837 enabled and it fails to allocate the map, if this option is
838 set it will continue without a random block map. As coverage
839 will not be as complete as with random maps, this option is
842 random_generator=str Fio supports the following engines for generating
843 IO offsets for random IO:
845 tausworthe Strong 2^88 cycle random number generator
846 lfsr Linear feedback shift register generator
848 Tausworthe is a strong random number generator, but it
849 requires tracking on the side if we want to ensure that
850 blocks are only read or written once. LFSR guarantees
851 that we never generate the same offset twice, and it's
852 also less computationally expensive. It's not a true
853 random generator, however, though for IO purposes it's
854 typically good enough. LFSR only works with single
855 block sizes, not with workloads that use multiple block
856 sizes. If used with such a workload, fio may read or write
857 some blocks multiple times.
859 nice=int Run the job with the given nice value. See man nice(2).
861 prio=int Set the io priority value of this job. Linux limits us to
862 a positive value between 0 and 7, with 0 being the highest.
865 prioclass=int Set the io priority class. See man ionice(1).
867 thinktime=int Stall the job x microseconds after an io has completed before
868 issuing the next. May be used to simulate processing being
869 done by an application. See thinktime_blocks and
873 Only valid if thinktime is set - pretend to spend CPU time
874 doing something with the data received, before falling back
875 to sleeping for the rest of the period specified by
879 Only valid if thinktime is set - control how many blocks
880 to issue, before waiting 'thinktime' usecs. If not set,
881 defaults to 1 which will make fio wait 'thinktime' usecs
882 after every block. This effectively makes any queue depth
883 setting redundant, since no more than 1 IO will be queued
884 before we have to complete it and do our thinktime. In
885 other words, this setting effectively caps the queue depth
886 if the latter is larger.
888 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
889 the normal suffix rules apply. You can use rate=500k to limit
890 reads and writes to 500k each, or you can specify read and
891 writes separately. Using rate=1m,500k would limit reads to
892 1MB/sec and writes to 500KB/sec. Capping only reads or
893 writes can be done with rate=,500k or rate=500k,. The former
894 will only limit writes (to 500KB/sec), the latter will only
897 ratemin=int Tell fio to do whatever it can to maintain at least this
898 bandwidth. Failing to meet this requirement, will cause
899 the job to exit. The same format as rate is used for
900 read vs write separation.
902 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
903 as rate, just specified independently of bandwidth. If the
904 job is given a block size range instead of a fixed value,
905 the smallest block size is used as the metric. The same format
906 as rate is used for read vs write separation.
908 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
909 the job to exit. The same format as rate is used for read vs
912 latency_target=int If set, fio will attempt to find the max performance
913 point that the given workload will run at while maintaining a
914 latency below this target. The values is given in microseconds.
915 See latency_window and latency_percentile
917 latency_window=int Used with latency_target to specify the sample window
918 that the job is run at varying queue depths to test the
919 performance. The value is given in microseconds.
921 latency_percentile=float The percentage of IOs that must fall within the
922 criteria specified by latency_target and latency_window. If not
923 set, this defaults to 100.0, meaning that all IOs must be equal
924 or below to the value set by latency_target.
926 max_latency=int If set, fio will exit the job if it exceeds this maximum
927 latency. It will exit with an ETIME error.
929 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
932 cpumask=int Set the CPU affinity of this job. The parameter given is a
933 bitmask of allowed CPU's the job may run on. So if you want
934 the allowed CPUs to be 1 and 5, you would pass the decimal
935 value of (1 << 1 | 1 << 5), or 34. See man
936 sched_setaffinity(2). This may not work on all supported
937 operating systems or kernel versions. This option doesn't
938 work well for a higher CPU count than what you can store in
939 an integer mask, so it can only control cpus 1-32. For
940 boxes with larger CPU counts, use cpus_allowed.
942 cpus_allowed=str Controls the same options as cpumask, but it allows a text
943 setting of the permitted CPUs instead. So to use CPUs 1 and
944 5, you would specify cpus_allowed=1,5. This options also
945 allows a range of CPUs. Say you wanted a binding to CPUs
946 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
948 cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
949 specified by cpus_allowed or cpumask. Two policies are
952 shared All jobs will share the CPU set specified.
953 split Each job will get a unique CPU from the CPU set.
955 'shared' is the default behaviour, if the option isn't
956 specified. If split is specified, then fio will will assign
957 one cpu per job. If not enough CPUs are given for the jobs
958 listed, then fio will roundrobin the CPUs in the set.
960 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
961 arguments allow comma delimited list of cpu numbers,
962 A-B ranges, or 'all'. Note, to enable numa options support,
963 fio must be built on a system with libnuma-dev(el) installed.
965 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
966 nodes. Format of the argements:
968 `mode' is one of the following memory policy:
969 default, prefer, bind, interleave, local
970 For `default' and `local' memory policy, no node is
971 needed to be specified.
972 For `prefer', only one node is allowed.
973 For `bind' and `interleave', it allow comma delimited
974 list of numbers, A-B ranges, or 'all'.
976 startdelay=time Start this job the specified number of seconds after fio
977 has started. Only useful if the job file contains several
978 jobs, and you want to delay starting some jobs to a certain
981 runtime=time Tell fio to terminate processing after the specified number
982 of seconds. It can be quite hard to determine for how long
983 a specified job will run, so this parameter is handy to
984 cap the total runtime to a given time.
986 time_based If set, fio will run for the duration of the runtime
987 specified even if the file(s) are completely read or
988 written. It will simply loop over the same workload
989 as many times as the runtime allows.
991 ramp_time=time If set, fio will run the specified workload for this amount
992 of time before logging any performance numbers. Useful for
993 letting performance settle before logging results, thus
994 minimizing the runtime required for stable results. Note
995 that the ramp_time is considered lead in time for a job,
996 thus it will increase the total runtime if a special timeout
997 or runtime is specified.
999 invalidate=bool Invalidate the buffer/page cache parts for this file prior
1000 to starting io. Defaults to true.
1002 sync=bool Use sync io for buffered writes. For the majority of the
1003 io engines, this means using O_SYNC.
1006 mem=str Fio can use various types of memory as the io unit buffer.
1007 The allowed values are:
1009 malloc Use memory from malloc(3) as the buffers.
1011 shm Use shared memory as the buffers. Allocated
1014 shmhuge Same as shm, but use huge pages as backing.
1016 mmap Use mmap to allocate buffers. May either be
1017 anonymous memory, or can be file backed if
1018 a filename is given after the option. The
1019 format is mem=mmap:/path/to/file.
1021 mmaphuge Use a memory mapped huge file as the buffer
1022 backing. Append filename after mmaphuge, ala
1023 mem=mmaphuge:/hugetlbfs/file
1025 The area allocated is a function of the maximum allowed
1026 bs size for the job, multiplied by the io depth given. Note
1027 that for shmhuge and mmaphuge to work, the system must have
1028 free huge pages allocated. This can normally be checked
1029 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1030 Linux system. Fio assumes a huge page is 4MB in size. So
1031 to calculate the number of huge pages you need for a given
1032 job file, add up the io depth of all jobs (normally one unless
1033 iodepth= is used) and multiply by the maximum bs set. Then
1034 divide that number by the huge page size. You can see the
1035 size of the huge pages in /proc/meminfo. If no huge pages
1036 are allocated by having a non-zero number in nr_hugepages,
1037 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1039 mmaphuge also needs to have hugetlbfs mounted and the file
1040 location should point there. So if it's mounted in /huge,
1041 you would use mem=mmaphuge:/huge/somefile.
1043 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1044 Note that the given alignment is applied to the first IO unit
1045 buffer, if using iodepth the alignment of the following buffers
1046 are given by the bs used. In other words, if using a bs that is
1047 a multiple of the page sized in the system, all buffers will
1048 be aligned to this value. If using a bs that is not page
1049 aligned, the alignment of subsequent IO memory buffers is the
1050 sum of the iomem_align and bs used.
1053 Defines the size of a huge page. Must at least be equal
1054 to the system setting, see /proc/meminfo. Defaults to 4MB.
1055 Should probably always be a multiple of megabytes, so using
1056 hugepage-size=Xm is the preferred way to set this to avoid
1057 setting a non-pow-2 bad value.
1059 exitall When one job finishes, terminate the rest. The default is
1060 to wait for each job to finish, sometimes that is not the
1063 bwavgtime=int Average the calculated bandwidth over the given time. Value
1064 is specified in milliseconds.
1066 iopsavgtime=int Average the calculated IOPS over the given time. Value
1067 is specified in milliseconds.
1069 create_serialize=bool If true, serialize the file creating for the jobs.
1070 This may be handy to avoid interleaving of data
1071 files, which may greatly depend on the filesystem
1072 used and even the number of processors in the system.
1074 create_fsync=bool fsync the data file after creation. This is the
1077 create_on_open=bool Don't pre-setup the files for IO, just create open()
1078 when it's time to do IO to that file.
1080 create_only=bool If true, fio will only run the setup phase of the job.
1081 If files need to be laid out or updated on disk, only
1082 that will be done. The actual job contents are not
1085 pre_read=bool If this is given, files will be pre-read into memory before
1086 starting the given IO operation. This will also clear
1087 the 'invalidate' flag, since it is pointless to pre-read
1088 and then drop the cache. This will only work for IO engines
1089 that are seekable, since they allow you to read the same data
1090 multiple times. Thus it will not work on eg network or splice
1093 unlink=bool Unlink the job files when done. Not the default, as repeated
1094 runs of that job would then waste time recreating the file
1095 set again and again.
1097 loops=int Run the specified number of iterations of this job. Used
1098 to repeat the same workload a given number of times. Defaults
1101 verify_only Do not perform specified workload---only verify data still
1102 matches previous invocation of this workload. This option
1103 allows one to check data multiple times at a later date
1104 without overwriting it. This option makes sense only for
1105 workloads that write data, and does not support workloads
1106 with the time_based option set.
1108 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1109 verify is set. Defaults to 1.
1111 verify=str If writing to a file, fio can verify the file contents
1112 after each iteration of the job. The allowed values are:
1114 md5 Use an md5 sum of the data area and store
1115 it in the header of each block.
1117 crc64 Use an experimental crc64 sum of the data
1118 area and store it in the header of each
1121 crc32c Use a crc32c sum of the data area and store
1122 it in the header of each block.
1124 crc32c-intel Use hardware assisted crc32c calcuation
1125 provided on SSE4.2 enabled processors. Falls
1126 back to regular software crc32c, if not
1127 supported by the system.
1129 crc32 Use a crc32 sum of the data area and store
1130 it in the header of each block.
1132 crc16 Use a crc16 sum of the data area and store
1133 it in the header of each block.
1135 crc7 Use a crc7 sum of the data area and store
1136 it in the header of each block.
1138 xxhash Use xxhash as the checksum function. Generally
1139 the fastest software checksum that fio
1142 sha512 Use sha512 as the checksum function.
1144 sha256 Use sha256 as the checksum function.
1146 sha1 Use optimized sha1 as the checksum function.
1148 meta Write extra information about each io
1149 (timestamp, block number etc.). The block
1150 number is verified. The io sequence number is
1151 verified for workloads that write data.
1152 See also verify_pattern.
1154 null Only pretend to verify. Useful for testing
1155 internals with ioengine=null, not for much
1158 This option can be used for repeated burn-in tests of a
1159 system to make sure that the written data is also
1160 correctly read back. If the data direction given is
1161 a read or random read, fio will assume that it should
1162 verify a previously written file. If the data direction
1163 includes any form of write, the verify will be of the
1166 verifysort=bool If set, fio will sort written verify blocks when it deems
1167 it faster to read them back in a sorted manner. This is
1168 often the case when overwriting an existing file, since
1169 the blocks are already laid out in the file system. You
1170 can ignore this option unless doing huge amounts of really
1171 fast IO where the red-black tree sorting CPU time becomes
1174 verify_offset=int Swap the verification header with data somewhere else
1175 in the block before writing. Its swapped back before
1178 verify_interval=int Write the verification header at a finer granularity
1179 than the blocksize. It will be written for chunks the
1180 size of header_interval. blocksize should divide this
1183 verify_pattern=str If set, fio will fill the io buffers with this
1184 pattern. Fio defaults to filling with totally random
1185 bytes, but sometimes it's interesting to fill with a known
1186 pattern for io verification purposes. Depending on the
1187 width of the pattern, fio will fill 1/2/3/4 bytes of the
1188 buffer at the time(it can be either a decimal or a hex number).
1189 The verify_pattern if larger than a 32-bit quantity has to
1190 be a hex number that starts with either "0x" or "0X". Use
1193 verify_fatal=bool Normally fio will keep checking the entire contents
1194 before quitting on a block verification failure. If this
1195 option is set, fio will exit the job on the first observed
1198 verify_dump=bool If set, dump the contents of both the original data
1199 block and the data block we read off disk to files. This
1200 allows later analysis to inspect just what kind of data
1201 corruption occurred. Off by default.
1203 verify_async=int Fio will normally verify IO inline from the submitting
1204 thread. This option takes an integer describing how many
1205 async offload threads to create for IO verification instead,
1206 causing fio to offload the duty of verifying IO contents
1207 to one or more separate threads. If using this offload
1208 option, even sync IO engines can benefit from using an
1209 iodepth setting higher than 1, as it allows them to have
1210 IO in flight while verifies are running.
1212 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1213 async IO verification threads. See cpus_allowed for the
1216 verify_backlog=int Fio will normally verify the written contents of a
1217 job that utilizes verify once that job has completed. In
1218 other words, everything is written then everything is read
1219 back and verified. You may want to verify continually
1220 instead for a variety of reasons. Fio stores the meta data
1221 associated with an IO block in memory, so for large
1222 verify workloads, quite a bit of memory would be used up
1223 holding this meta data. If this option is enabled, fio
1224 will write only N blocks before verifying these blocks.
1226 verify_backlog_batch=int Control how many blocks fio will verify
1227 if verify_backlog is set. If not set, will default to
1228 the value of verify_backlog (meaning the entire queue
1229 is read back and verified). If verify_backlog_batch is
1230 less than verify_backlog then not all blocks will be verified,
1231 if verify_backlog_batch is larger than verify_backlog, some
1232 blocks will be verified more than once.
1235 wait_for_previous Wait for preceding jobs in the job file to exit, before
1236 starting this one. Can be used to insert serialization
1237 points in the job file. A stone wall also implies starting
1238 a new reporting group.
1240 new_group Start a new reporting group. See: group_reporting.
1242 numjobs=int Create the specified number of clones of this job. May be
1243 used to setup a larger number of threads/processes doing
1244 the same thing. Each thread is reported separately; to see
1245 statistics for all clones as a whole, use group_reporting in
1246 conjunction with new_group.
1248 group_reporting It may sometimes be interesting to display statistics for
1249 groups of jobs as a whole instead of for each individual job.
1250 This is especially true if 'numjobs' is used; looking at
1251 individual thread/process output quickly becomes unwieldy.
1252 To see the final report per-group instead of per-job, use
1253 'group_reporting'. Jobs in a file will be part of the same
1254 reporting group, unless if separated by a stonewall, or by
1257 thread fio defaults to forking jobs, however if this option is
1258 given, fio will use pthread_create(3) to create threads
1261 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1263 zoneskip=int Skip the specified number of bytes when zonesize data has
1264 been read. The two zone options can be used to only do
1265 io on zones of a file.
1267 write_iolog=str Write the issued io patterns to the specified file. See
1268 read_iolog. Specify a separate file for each job, otherwise
1269 the iologs will be interspersed and the file may be corrupt.
1271 read_iolog=str Open an iolog with the specified file name and replay the
1272 io patterns it contains. This can be used to store a
1273 workload and replay it sometime later. The iolog given
1274 may also be a blktrace binary file, which allows fio
1275 to replay a workload captured by blktrace. See blktrace
1276 for how to capture such logging data. For blktrace replay,
1277 the file needs to be turned into a blkparse binary data
1278 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1280 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1281 is to attempt to respect the time stamps within the log and
1282 replay them with the appropriate delay between IOPS. By
1283 setting this variable fio will not respect the timestamps and
1284 attempt to replay them as fast as possible while still
1285 respecting ordering. The result is the same I/O pattern to a
1286 given device, but different timings.
1288 replay_redirect=str While replaying I/O patterns using read_iolog the
1289 default behavior is to replay the IOPS onto the major/minor
1290 device that each IOP was recorded from. This is sometimes
1291 undesirable because on a different machine those major/minor
1292 numbers can map to a different device. Changing hardware on
1293 the same system can also result in a different major/minor
1294 mapping. Replay_redirect causes all IOPS to be replayed onto
1295 the single specified device regardless of the device it was
1296 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1297 IO in the blktrace to be replayed onto /dev/sdc. This means
1298 multiple devices will be replayed onto a single, if the trace
1299 contains multiple devices. If you want multiple devices to be
1300 replayed concurrently to multiple redirected devices you must
1301 blkparse your trace into separate traces and replay them with
1302 independent fio invocations. Unfortuantely this also breaks
1303 the strict time ordering between multiple device accesses.
1305 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1306 file. Can be used to store data of the bandwidth of the
1307 jobs in their lifetime. The included fio_generate_plots
1308 script uses gnuplot to turn these text files into nice
1309 graphs. See write_lat_log for behaviour of given
1310 filename. For this option, the suffix is _bw.log.
1312 write_lat_log=str Same as write_bw_log, except that this option stores io
1313 submission, completion, and total latencies instead. If no
1314 filename is given with this option, the default filename of
1315 "jobname_type.log" is used. Even if the filename is given,
1316 fio will still append the type of log. So if one specifies
1320 The actual log names will be foo_slat.log, foo_clat.log,
1321 and foo_lat.log. This helps fio_generate_plot fine the logs
1324 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1325 given with this option, the default filename of
1326 "jobname_type.log" is used. Even if the filename is given,
1327 fio will still append the type of log.
1329 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1330 or bw log for every IO that completes. When writing to the
1331 disk log, that can quickly grow to a very large size. Setting
1332 this option makes fio average the each log entry over the
1333 specified period of time, reducing the resolution of the log.
1336 log_offset=int If this is set, the iolog options will include the byte
1337 offset for the IO entry as well as the other data values.
1339 log_compression=int If this is set, fio will compress the IO logs as
1340 it goes, to keep the memory footprint lower. When a log
1341 reaches the specified size, that chunk is removed and
1342 compressed in the background. Given that IO logs are
1343 fairly highly compressible, this yields a nice memory
1344 savings for longer runs. The downside is that the
1345 compression will consume some background CPU cycles, so
1346 it may impact the run. This, however, is also true if
1347 the logging ends up consuming most of the system memory.
1348 So pick your poison. The IO logs are saved normally at the
1349 end of a run, by decompressing the chunks and storing them
1350 in the specified log file. This feature depends on the
1351 availability of zlib.
1353 log_store_compressed=bool If set, and log_compression is also set,
1354 fio will store the log files in a compressed format. They
1355 can be decompressed with fio, using the --inflate-log
1356 command line parameter. The files will be stored with a
1359 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1360 potentially be used instead of removing memory or booting
1361 with less memory to simulate a smaller amount of memory.
1362 The amount specified is per worker.
1364 exec_prerun=str Before running this job, issue the command specified
1365 through system(3). Output is redirected in a file called
1368 exec_postrun=str After the job completes, issue the command specified
1369 though system(3). Output is redirected in a file called
1370 jobname.postrun.txt.
1372 ioscheduler=str Attempt to switch the device hosting the file to the specified
1373 io scheduler before running.
1375 disk_util=bool Generate disk utilization statistics, if the platform
1376 supports it. Defaults to on.
1378 disable_lat=bool Disable measurements of total latency numbers. Useful
1379 only for cutting back the number of calls to gettimeofday,
1380 as that does impact performance at really high IOPS rates.
1381 Note that to really get rid of a large amount of these
1382 calls, this option must be used with disable_slat and
1385 disable_clat=bool Disable measurements of completion latency numbers. See
1388 disable_slat=bool Disable measurements of submission latency numbers. See
1391 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1394 clat_percentiles=bool Enable the reporting of percentiles of
1395 completion latencies.
1397 percentile_list=float_list Overwrite the default list of percentiles
1398 for completion latencies. Each number is a floating
1399 number in the range (0,100], and the maximum length of
1400 the list is 20. Use ':' to separate the numbers, and
1401 list the numbers in ascending order. For example,
1402 --percentile_list=99.5:99.9 will cause fio to report
1403 the values of completion latency below which 99.5% and
1404 99.9% of the observed latencies fell, respectively.
1406 clocksource=str Use the given clocksource as the base of timing. The
1407 supported options are:
1409 gettimeofday gettimeofday(2)
1411 clock_gettime clock_gettime(2)
1413 cpu Internal CPU clock source
1415 cpu is the preferred clocksource if it is reliable, as it
1416 is very fast (and fio is heavy on time calls). Fio will
1417 automatically use this clocksource if it's supported and
1418 considered reliable on the system it is running on, unless
1419 another clocksource is specifically set. For x86/x86-64 CPUs,
1420 this means supporting TSC Invariant.
1422 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1423 (disable_clat, disable_slat, disable_bw) plus reduce
1424 precision of the timeout somewhat to really shrink
1425 the gettimeofday() call count. With this option enabled,
1426 we only do about 0.4% of the gtod() calls we would have
1427 done if all time keeping was enabled.
1429 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1430 execution to just getting the current time. Fio (and
1431 databases, for instance) are very intensive on gettimeofday()
1432 calls. With this option, you can set one CPU aside for
1433 doing nothing but logging current time to a shared memory
1434 location. Then the other threads/processes that run IO
1435 workloads need only copy that segment, instead of entering
1436 the kernel with a gettimeofday() call. The CPU set aside
1437 for doing these time calls will be excluded from other
1438 uses. Fio will manually clear it from the CPU mask of other
1441 continue_on_error=str Normally fio will exit the job on the first observed
1442 failure. If this option is set, fio will continue the job when
1443 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1444 is exceeded or the I/O size specified is completed. If this
1445 option is used, there are two more stats that are appended,
1446 the total error count and the first error. The error field
1447 given in the stats is the first error that was hit during the
1450 The allowed values are:
1452 none Exit on any IO or verify errors.
1454 read Continue on read errors, exit on all others.
1456 write Continue on write errors, exit on all others.
1458 io Continue on any IO error, exit on all others.
1460 verify Continue on verify errors, exit on all others.
1462 all Continue on all errors.
1464 0 Backward-compatible alias for 'none'.
1466 1 Backward-compatible alias for 'all'.
1468 ignore_error=str Sometimes you want to ignore some errors during test
1469 in that case you can specify error list for each error type.
1470 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1471 errors for given error type is separated with ':'. Error
1472 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1474 ignore_error=EAGAIN,ENOSPC:122
1475 This option will ignore EAGAIN from READ, and ENOSPC and
1476 122(EDQUOT) from WRITE.
1478 error_dump=bool If set dump every error even if it is non fatal, true
1479 by default. If disabled only fatal error will be dumped
1481 cgroup=str Add job to this control group. If it doesn't exist, it will
1482 be created. The system must have a mounted cgroup blkio
1483 mount point for this to work. If your system doesn't have it
1484 mounted, you can do so with:
1486 # mount -t cgroup -o blkio none /cgroup
1488 cgroup_weight=int Set the weight of the cgroup to this value. See
1489 the documentation that comes with the kernel, allowed values
1490 are in the range of 100..1000.
1492 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1493 the job completion. To override this behavior and to leave
1494 cgroups around after the job completion, set cgroup_nodelete=1.
1495 This can be useful if one wants to inspect various cgroup
1496 files after job completion. Default: false
1498 uid=int Instead of running as the invoking user, set the user ID to
1499 this value before the thread/process does any work.
1501 gid=int Set group ID, see uid.
1503 flow_id=int The ID of the flow. If not specified, it defaults to being a
1504 global flow. See flow.
1506 flow=int Weight in token-based flow control. If this value is used, then
1507 there is a 'flow counter' which is used to regulate the
1508 proportion of activity between two or more jobs. fio attempts
1509 to keep this flow counter near zero. The 'flow' parameter
1510 stands for how much should be added or subtracted to the flow
1511 counter on each iteration of the main I/O loop. That is, if
1512 one job has flow=8 and another job has flow=-1, then there
1513 will be a roughly 1:8 ratio in how much one runs vs the other.
1515 flow_watermark=int The maximum value that the absolute value of the flow
1516 counter is allowed to reach before the job must wait for a
1517 lower value of the counter.
1519 flow_sleep=int The period of time, in microseconds, to wait after the flow
1520 watermark has been exceeded before retrying operations
1522 In addition, there are some parameters which are only valid when a specific
1523 ioengine is in use. These are used identically to normal parameters, with the
1524 caveat that when used on the command line, they must come after the ioengine
1525 that defines them is selected.
1527 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1528 the io_getevents system call to reap newly returned events.
1529 With this flag turned on, the AIO ring will be read directly
1530 from user-space to reap events. The reaping mode is only
1531 enabled when polling for a minimum of 0 events (eg when
1532 iodepth_batch_complete=0).
1534 [cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1536 [cpu] cpuchunks=int Split the load into cycles of the given time. In
1539 [cpu] exit_on_io_done=bool Detect when IO threads are done, then exit.
1541 [netsplice] hostname=str
1542 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1543 If the job is a TCP listener or UDP reader, the hostname is not
1544 used and must be omitted unless it is a valid UDP multicast
1547 [netsplice] port=int
1548 [net] port=int The TCP or UDP port to bind to or connect to.
1550 [netsplice] interface=str
1551 [net] interface=str The IP address of the network interface used to send or
1552 receive UDP multicast
1555 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1558 [netsplice] nodelay=bool
1559 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1561 [netsplice] protocol=str
1562 [netsplice] proto=str
1564 [net] proto=str The network protocol to use. Accepted values are:
1566 tcp Transmission control protocol
1567 tcpv6 Transmission control protocol V6
1568 udp User datagram protocol
1569 udpv6 User datagram protocol V6
1570 unix UNIX domain socket
1572 When the protocol is TCP or UDP, the port must also be given,
1573 as well as the hostname if the job is a TCP listener or UDP
1574 reader. For unix sockets, the normal filename option should be
1575 used and the port is invalid.
1577 [net] listen For TCP network connections, tell fio to listen for incoming
1578 connections rather than initiating an outgoing connection. The
1579 hostname must be omitted if this option is used.
1580 [net] pingpong Normaly a network writer will just continue writing data, and
1581 a network reader will just consume packages. If pingpong=1
1582 is set, a writer will send its normal payload to the reader,
1583 then wait for the reader to send the same payload back. This
1584 allows fio to measure network latencies. The submission
1585 and completion latencies then measure local time spent
1586 sending or receiving, and the completion latency measures
1587 how long it took for the other end to receive and send back.
1588 For UDP multicast traffic pingpong=1 should only be set for a
1589 single reader when multiple readers are listening to the same
1592 [e4defrag] donorname=str
1593 File will be used as a block donor(swap extents between files)
1594 [e4defrag] inplace=int
1595 Configure donor file blocks allocation strategy
1596 0(default): Preallocate donor's file on init
1597 1 : allocate space immidietly inside defragment event,
1598 and free right after event
1602 6.0 Interpreting the output
1603 ---------------------------
1605 fio spits out a lot of output. While running, fio will display the
1606 status of the jobs created. An example of that would be:
1608 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1610 The characters inside the square brackets denote the current status of
1611 each thread. The possible values (in typical life cycle order) are:
1615 P Thread setup, but not started.
1617 I Thread initialized, waiting or generating necessary data.
1618 p Thread running pre-reading file(s).
1619 R Running, doing sequential reads.
1620 r Running, doing random reads.
1621 W Running, doing sequential writes.
1622 w Running, doing random writes.
1623 M Running, doing mixed sequential reads/writes.
1624 m Running, doing mixed random reads/writes.
1625 F Running, currently waiting for fsync()
1626 f Running, finishing up (writing IO logs, etc)
1627 V Running, doing verification of written data.
1628 E Thread exited, not reaped by main thread yet.
1630 X Thread reaped, exited with an error.
1631 K Thread reaped, exited due to signal.
1633 Fio will condense the thread string as not to take up more space on the
1634 command line as is needed. For instance, if you have 10 readers and 10
1635 writers running, the output would look like this:
1637 Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s]
1639 Fio will still maintain the ordering, though. So the above means that jobs
1640 1..10 are readers, and 11..20 are writers.
1642 The other values are fairly self explanatory - number of threads
1643 currently running and doing io, rate of io since last check (read speed
1644 listed first, then write speed), and the estimated completion percentage
1645 and time for the running group. It's impossible to estimate runtime of
1646 the following groups (if any). Note that the string is displayed in order,
1647 so it's possible to tell which of the jobs are currently doing what. The
1648 first character is the first job defined in the job file, and so forth.
1650 When fio is done (or interrupted by ctrl-c), it will show the data for
1651 each thread, group of threads, and disks in that order. For each data
1652 direction, the output looks like:
1654 Client1 (g=0): err= 0:
1655 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1656 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1657 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1658 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1659 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1660 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1661 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1662 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1663 issued r/w: total=0/32768, short=0/0
1664 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1665 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1667 The client number is printed, along with the group id and error of that
1668 thread. Below is the io statistics, here for writes. In the order listed,
1671 io= Number of megabytes io performed
1672 bw= Average bandwidth rate
1673 iops= Average IOs performed per second
1674 runt= The runtime of that thread
1675 slat= Submission latency (avg being the average, stdev being the
1676 standard deviation). This is the time it took to submit
1677 the io. For sync io, the slat is really the completion
1678 latency, since queue/complete is one operation there. This
1679 value can be in milliseconds or microseconds, fio will choose
1680 the most appropriate base and print that. In the example
1681 above, milliseconds is the best scale. Note: in --minimal mode
1682 latencies are always expressed in microseconds.
1683 clat= Completion latency. Same names as slat, this denotes the
1684 time from submission to completion of the io pieces. For
1685 sync io, clat will usually be equal (or very close) to 0,
1686 as the time from submit to complete is basically just
1687 CPU time (io has already been done, see slat explanation).
1688 bw= Bandwidth. Same names as the xlat stats, but also includes
1689 an approximate percentage of total aggregate bandwidth
1690 this thread received in this group. This last value is
1691 only really useful if the threads in this group are on the
1692 same disk, since they are then competing for disk access.
1693 cpu= CPU usage. User and system time, along with the number
1694 of context switches this thread went through, usage of
1695 system and user time, and finally the number of major
1696 and minor page faults.
1697 IO depths= The distribution of io depths over the job life time. The
1698 numbers are divided into powers of 2, so for example the
1699 16= entries includes depths up to that value but higher
1700 than the previous entry. In other words, it covers the
1701 range from 16 to 31.
1702 IO submit= How many pieces of IO were submitting in a single submit
1703 call. Each entry denotes that amount and below, until
1704 the previous entry - eg, 8=100% mean that we submitted
1705 anywhere in between 5-8 ios per submit call.
1706 IO complete= Like the above submit number, but for completions instead.
1707 IO issued= The number of read/write requests issued, and how many
1709 IO latencies= The distribution of IO completion latencies. This is the
1710 time from when IO leaves fio and when it gets completed.
1711 The numbers follow the same pattern as the IO depths,
1712 meaning that 2=1.6% means that 1.6% of the IO completed
1713 within 2 msecs, 20=12.8% means that 12.8% of the IO
1714 took more than 10 msecs, but less than (or equal to) 20 msecs.
1716 After each client has been listed, the group statistics are printed. They
1717 will look like this:
1719 Run status group 0 (all jobs):
1720 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1721 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1723 For each data direction, it prints:
1725 io= Number of megabytes io performed.
1726 aggrb= Aggregate bandwidth of threads in this group.
1727 minb= The minimum average bandwidth a thread saw.
1728 maxb= The maximum average bandwidth a thread saw.
1729 mint= The smallest runtime of the threads in that group.
1730 maxt= The longest runtime of the threads in that group.
1732 And finally, the disk statistics are printed. They will look like this:
1734 Disk stats (read/write):
1735 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1737 Each value is printed for both reads and writes, with reads first. The
1740 ios= Number of ios performed by all groups.
1741 merge= Number of merges io the io scheduler.
1742 ticks= Number of ticks we kept the disk busy.
1743 io_queue= Total time spent in the disk queue.
1744 util= The disk utilization. A value of 100% means we kept the disk
1745 busy constantly, 50% would be a disk idling half of the time.
1747 It is also possible to get fio to dump the current output while it is
1748 running, without terminating the job. To do that, send fio the USR1 signal.
1749 You can also get regularly timed dumps by using the --status-interval
1750 parameter, or by creating a file in /tmp named fio-dump-status. If fio
1751 sees this file, it will unlink it and dump the current output status.
1757 For scripted usage where you typically want to generate tables or graphs
1758 of the results, fio can output the results in a semicolon separated format.
1759 The format is one long line of values, such as:
1761 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%
1762 A description of this job goes here.
1764 The job description (if provided) follows on a second line.
1766 To enable terse output, use the --minimal command line option. The first
1767 value is the version of the terse output format. If the output has to
1768 be changed for some reason, this number will be incremented by 1 to
1769 signify that change.
1771 Split up, the format is as follows:
1773 terse version, fio version, jobname, groupid, error
1775 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1776 Submission latency: min, max, mean, deviation (usec)
1777 Completion latency: min, max, mean, deviation (usec)
1778 Completion latency percentiles: 20 fields (see below)
1779 Total latency: min, max, mean, deviation (usec)
1780 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1782 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1783 Submission latency: min, max, mean, deviation (usec)
1784 Completion latency: min, max, mean, deviation (usec)
1785 Completion latency percentiles: 20 fields (see below)
1786 Total latency: min, max, mean, deviation (usec)
1787 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1788 CPU usage: user, system, context switches, major faults, minor faults
1789 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1790 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1791 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1792 Disk utilization: Disk name, Read ios, write ios,
1793 Read merges, write merges,
1794 Read ticks, write ticks,
1795 Time spent in queue, disk utilization percentage
1796 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
1798 Additional Info (dependent on description being set): Text description
1800 Completion latency percentiles can be a grouping of up to 20 sets, so
1801 for the terse output fio writes all of them. Each field will look like this:
1805 which is the Xth percentile, and the usec latency associated with it.
1807 For disk utilization, all disks used by fio are shown. So for each disk
1808 there will be a disk utilization section.
1811 8.0 Trace file format
1812 ---------------------
1813 There are two trace file format that you can encounter. The older (v1) format
1814 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1815 below in case that you get an old trace and want to understand it.
1817 In any case the trace is a simple text file with a single action per line.
1820 8.1 Trace file format v1
1821 ------------------------
1822 Each line represents a single io action in the following format:
1826 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1828 This format is not supported in Fio versions => 1.20-rc3.
1831 8.2 Trace file format v2
1832 ------------------------
1833 The second version of the trace file format was added in Fio version 1.17.
1834 It allows to access more then one file per trace and has a bigger set of
1835 possible file actions.
1837 The first line of the trace file has to be:
1841 Following this can be lines in two different formats, which are described below.
1843 The file management format:
1847 The filename is given as an absolute path. The action can be one of these:
1849 add Add the given filename to the trace
1850 open Open the file with the given filename. The filename has to have
1851 been added with the add action before.
1852 close Close the file with the given filename. The file has to have been
1856 The file io action format:
1858 filename action offset length
1860 The filename is given as an absolute path, and has to have been added and opened
1861 before it can be used with this format. The offset and length are given in
1862 bytes. The action can be one of these:
1864 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1865 read Read 'length' bytes beginning from 'offset'
1866 write Write 'length' bytes beginning from 'offset'
1867 sync fsync() the file
1868 datasync fdatasync() the file
1869 trim trim the given file from the given 'offset' for 'length' bytes
1872 9.0 CPU idleness profiling
1873 --------------------------
1874 In some cases, we want to understand CPU overhead in a test. For example,
1875 we test patches for the specific goodness of whether they reduce CPU usage.
1876 fio implements a balloon approach to create a thread per CPU that runs at
1877 idle priority, meaning that it only runs when nobody else needs the cpu.
1878 By measuring the amount of work completed by the thread, idleness of each
1879 CPU can be derived accordingly.
1881 An unit work is defined as touching a full page of unsigned characters. Mean
1882 and standard deviation of time to complete an unit work is reported in "unit
1883 work" section. Options can be chosen to report detailed percpu idleness or
1884 overall system idleness by aggregating percpu stats.