8 5. Detailed list of parameters
12 9. CPU idleness profiling
14 1.0 Overview and history
15 ------------------------
16 fio was originally written to save me the hassle of writing special test
17 case programs when I wanted to test a specific workload, either for
18 performance reasons or to find/reproduce a bug. The process of writing
19 such a test app can be tiresome, especially if you have to do it often.
20 Hence I needed a tool that would be able to simulate a given io workload
21 without resorting to writing a tailored test case again and again.
23 A test work load is difficult to define, though. There can be any number
24 of processes or threads involved, and they can each be using their own
25 way of generating io. You could have someone dirtying large amounts of
26 memory in an memory mapped file, or maybe several threads issuing
27 reads using asynchronous io. fio needed to be flexible enough to
28 simulate both of these cases, and many more.
32 The first step in getting fio to simulate a desired io workload, is
33 writing a job file describing that specific setup. A job file may contain
34 any number of threads and/or files - the typical contents of the job file
35 is a global section defining shared parameters, and one or more job
36 sections describing the jobs involved. When run, fio parses this file
37 and sets everything up as described. If we break down a job from top to
38 bottom, it contains the following basic parameters:
40 IO type Defines the io pattern issued to the file(s).
41 We may only be reading sequentially from this
42 file(s), or we may be writing randomly. Or even
43 mixing reads and writes, sequentially or randomly.
45 Block size In how large chunks are we issuing io? This may be
46 a single value, or it may describe a range of
49 IO size How much data are we going to be reading/writing.
51 IO engine How do we issue io? We could be memory mapping the
52 file, we could be using regular read/write, we
53 could be using splice, async io, syslet, or even
56 IO depth If the io engine is async, how large a queuing
57 depth do we want to maintain?
59 IO type Should we be doing buffered io, or direct/raw io?
61 Num files How many files are we spreading the workload over.
63 Num threads How many threads or processes should we spread
66 The above are the basic parameters defined for a workload, in addition
67 there's a multitude of parameters that modify other aspects of how this
73 See the README file for command line parameters, there are only a few
76 Running fio is normally the easiest part - you just give it the job file
77 (or job files) as parameters:
81 and it will start doing what the job_file tells it to do. You can give
82 more than one job file on the command line, fio will serialize the running
83 of those files. Internally that is the same as using the 'stonewall'
84 parameter described the the parameter section.
86 If the job file contains only one job, you may as well just give the
87 parameters on the command line. The command line parameters are identical
88 to the job parameters, with a few extra that control global parameters
89 (see README). For example, for the job file parameter iodepth=2, the
90 mirror command line option would be --iodepth 2 or --iodepth=2. You can
91 also use the command line for giving more than one job entry. For each
92 --name option that fio sees, it will start a new job with that name.
93 Command line entries following a --name entry will apply to that job,
94 until there are no more entries or a new --name entry is seen. This is
95 similar to the job file options, where each option applies to the current
96 job until a new [] job entry is seen.
98 fio does not need to run as root, except if the files or devices specified
99 in the job section requires that. Some other options may also be restricted,
100 such as memory locking, io scheduler switching, and decreasing the nice value.
105 As previously described, fio accepts one or more job files describing
106 what it is supposed to do. The job file format is the classic ini file,
107 where the names enclosed in [] brackets define the job name. You are free
108 to use any ascii name you want, except 'global' which has special meaning.
109 A global section sets defaults for the jobs described in that file. A job
110 may override a global section parameter, and a job file may even have
111 several global sections if so desired. A job is only affected by a global
112 section residing above it. If the first character in a line is a ';' or a
113 '#', the entire line is discarded as a comment.
115 So let's look at a really simple job file that defines two processes, each
116 randomly reading from a 128MB file.
118 ; -- start job file --
129 As you can see, the job file sections themselves are empty as all the
130 described parameters are shared. As no filename= option is given, fio
131 makes up a filename for each of the jobs as it sees fit. On the command
132 line, this job would look as follows:
134 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
137 Let's look at an example that has a number of processes writing randomly
140 ; -- start job file --
152 Here we have no global section, as we only have one job defined anyway.
153 We want to use async io here, with a depth of 4 for each file. We also
154 increased the buffer size used to 32KB and define numjobs to 4 to
155 fork 4 identical jobs. The result is 4 processes each randomly writing
156 to their own 64MB file. Instead of using the above job file, you could
157 have given the parameters on the command line. For this case, you would
160 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
162 4.1 Environment variables
163 -------------------------
165 fio also supports environment variable expansion in job files. Any
166 substring of the form "${VARNAME}" as part of an option value (in other
167 words, on the right of the `='), will be expanded to the value of the
168 environment variable called VARNAME. If no such environment variable
169 is defined, or VARNAME is the empty string, the empty string will be
172 As an example, let's look at a sample fio invocation and job file:
174 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
176 ; -- start job file --
183 This will expand to the following equivalent job file at runtime:
185 ; -- start job file --
192 fio ships with a few example job files, you can also look there for
195 4.2 Reserved keywords
196 ---------------------
198 Additionally, fio has a set of reserved keywords that will be replaced
199 internally with the appropriate value. Those keywords are:
201 $pagesize The architecture page size of the running system
202 $mb_memory Megabytes of total memory in the system
203 $ncpus Number of online available CPUs
205 These can be used on the command line or in the job file, and will be
206 automatically substituted with the current system values when the job
207 is run. Simple math is also supported on these keywords, so you can
208 perform actions like:
212 and get that properly expanded to 8 times the size of memory in the
216 5.0 Detailed list of parameters
217 -------------------------------
219 This section describes in details each parameter associated with a job.
220 Some parameters take an option of a given type, such as an integer or
221 a string. The following types are used:
223 str String. This is a sequence of alpha characters.
224 time Integer with possible time suffix. In seconds unless otherwise
225 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
227 int SI integer. A whole number value, which may contain a suffix
228 describing the base of the number. Accepted suffixes are k/m/g/t/p,
229 meaning kilo, mega, giga, tera, and peta. The suffix is not case
230 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
231 as 'k'). So if you want to specify 4096, you could either write
232 out '4096' or just give 4k. The suffixes signify base 2 values, so
233 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
234 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
235 case, then 1000 is used as the multiplier. This can be handy for
236 disks, since manufacturers generally use base 10 values when listing
237 the capacity of a drive. If the option accepts an upper and lower
238 range, use a colon ':' or minus '-' to separate such values. May also
239 include a prefix to indicate numbers base. If 0x is used, the number
240 is assumed to be hexadecimal. See irange.
241 bool Boolean. Usually parsed as an integer, however only defined for
242 true and false (1 and 0).
243 irange Integer range with suffix. Allows value range to be given, such
244 as 1024-4096. A colon may also be used as the separator, eg
245 1k:4k. If the option allows two sets of ranges, they can be
246 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
248 float_list A list of floating numbers, separated by a ':' character.
250 With the above in mind, here follows the complete list of fio job
253 name=str ASCII name of the job. This may be used to override the
254 name printed by fio for this job. Otherwise the job
255 name is used. On the command line this parameter has the
256 special purpose of also signaling the start of a new
259 description=str Text description of the job. Doesn't do anything except
260 dump this text description when this job is run. It's
263 directory=str Prefix filenames with this directory. Used to place files
264 in a different location than "./".
266 filename=str Fio normally makes up a filename based on the job name,
267 thread number, and file number. If you want to share
268 files between threads in a job or several jobs, specify
269 a filename for each of them to override the default. If
270 the ioengine used is 'net', the filename is the host, port,
271 and protocol to use in the format of =host,port,protocol.
272 See ioengine=net for more. If the ioengine is file based, you
273 can specify a number of files by separating the names with a
274 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
275 as the two working files, you would use
276 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
277 accessed as \\.\PhysicalDrive0 for the first device,
278 \\.\PhysicalDrive1 for the second etc. Note: Windows and
279 FreeBSD prevent write access to areas of the disk containing
280 in-use data (e.g. filesystems).
281 If the wanted filename does need to include a colon, then
282 escape that with a '\' character. For instance, if the filename
283 is "/dev/dsk/foo@3,0:c", then you would use
284 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
285 stdin or stdout. Which of the two depends on the read/write
288 opendir=str Tell fio to recursively add any file it can find in this
289 directory and down the file system tree.
291 lockfile=str Fio defaults to not locking any files before it does
292 IO to them. If a file or file descriptor is shared, fio
293 can serialize IO to that file to make the end result
294 consistent. This is usual for emulating real workloads that
295 share files. The lock modes are:
297 none No locking. The default.
298 exclusive Only one thread/process may do IO,
299 excluding all others.
300 readwrite Read-write locking on the file. Many
301 readers may access the file at the
302 same time, but writes get exclusive
306 rw=str Type of io pattern. Accepted values are:
308 read Sequential reads
309 write Sequential writes
310 randwrite Random writes
311 randread Random reads
312 rw,readwrite Sequential mixed reads and writes
313 randrw Random mixed reads and writes
315 For the mixed io types, the default is to split them 50/50.
316 For certain types of io the result may still be skewed a bit,
317 since the speed may be different. It is possible to specify
318 a number of IO's to do before getting a new offset, this is
319 one by appending a ':<nr>' to the end of the string given.
320 For a random read, it would look like 'rw=randread:8' for
321 passing in an offset modifier with a value of 8. If the
322 suffix is used with a sequential IO pattern, then the value
323 specified will be added to the generated offset for each IO.
324 For instance, using rw=write:4k will skip 4k for every
325 write. It turns sequential IO into sequential IO with holes.
326 See the 'rw_sequencer' option.
328 rw_sequencer=str If an offset modifier is given by appending a number to
329 the rw=<str> line, then this option controls how that
330 number modifies the IO offset being generated. Accepted
333 sequential Generate sequential offset
334 identical Generate the same offset
336 'sequential' is only useful for random IO, where fio would
337 normally generate a new random offset for every IO. If you
338 append eg 8 to randread, you would get a new random offset for
339 every 8 IO's. The result would be a seek for only every 8
340 IO's, instead of for every IO. Use rw=randread:8 to specify
341 that. As sequential IO is already sequential, setting
342 'sequential' for that would not result in any differences.
343 'identical' behaves in a similar fashion, except it sends
344 the same offset 8 number of times before generating a new
347 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
348 Storage manufacturers like to use 10^3 or 1000 as a base
349 ten unit instead, for obvious reasons. Allow values are
350 1024 or 1000, with 1024 being the default.
352 unified_rw_reporting=bool Fio normally reports statistics on a per
353 data direction basis, meaning that read, write, and trim are
354 accounted and reported separately. If this option is set,
355 the fio will sum the results and report them as "mixed"
358 randrepeat=bool For random IO workloads, seed the generator in a predictable
359 way so that results are repeatable across repetitions.
361 use_os_rand=bool Fio can either use the random generator supplied by the OS
362 to generator random offsets, or it can use it's own internal
363 generator (based on Tausworthe). Default is to use the
364 internal generator, which is often of better quality and
367 fallocate=str Whether pre-allocation is performed when laying down files.
370 none Do not pre-allocate space
371 posix Pre-allocate via posix_fallocate()
372 keep Pre-allocate via fallocate() with
373 FALLOC_FL_KEEP_SIZE set
374 0 Backward-compatible alias for 'none'
375 1 Backward-compatible alias for 'posix'
377 May not be available on all supported platforms. 'keep' is only
378 available on Linux.If using ZFS on Solaris this must be set to
379 'none' because ZFS doesn't support it. Default: 'posix'.
381 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
382 on what IO patterns it is likely to issue. Sometimes you
383 want to test specific IO patterns without telling the
384 kernel about it, in which case you can disable this option.
385 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
386 IO and POSIX_FADV_RANDOM for random IO.
388 size=int The total size of file io for this job. Fio will run until
389 this many bytes has been transferred, unless runtime is
390 limited by other options (such as 'runtime', for instance).
391 Unless specific nrfiles and filesize options are given,
392 fio will divide this size between the available files
393 specified by the job. If not set, fio will use the full
394 size of the given files or devices. If the the files
395 do not exist, size must be given. It is also possible to
396 give size as a percentage between 1 and 100. If size=20%
397 is given, fio will use 20% of the full size of the given
400 filesize=int Individual file sizes. May be a range, in which case fio
401 will select sizes for files at random within the given range
402 and limited to 'size' in total (if that is given). If not
403 given, each created file is the same size.
406 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
407 space left on device) as the terminating condition. Only makes
408 sense with sequential write. For a read workload, the mount
409 point will be filled first then IO started on the result. This
410 option doesn't make sense if operating on a raw device node,
411 since the size of that is already known by the file system.
412 Additionally, writing beyond end-of-device will not return
416 bs=int The block size used for the io units. Defaults to 4k. Values
417 can be given for both read and writes. If a single int is
418 given, it will apply to both. If a second int is specified
419 after a comma, it will apply to writes only. In other words,
420 the format is either bs=read_and_write or bs=read,write.
421 bs=4k,8k will thus use 4k blocks for reads, and 8k blocks
422 for writes. If you only wish to set the write size, you
423 can do so by passing an empty read size - bs=,8k will set
424 8k for writes and leave the read default value.
427 ba=int At what boundary to align random IO offsets. Defaults to
428 the same as 'blocksize' the minimum blocksize given.
429 Minimum alignment is typically 512b for using direct IO,
430 though it usually depends on the hardware block size. This
431 option is mutually exclusive with using a random map for
432 files, so it will turn off that option.
434 blocksize_range=irange
435 bsrange=irange Instead of giving a single block size, specify a range
436 and fio will mix the issued io block sizes. The issued
437 io unit will always be a multiple of the minimum value
438 given (also see bs_unaligned). Applies to both reads and
439 writes, however a second range can be given after a comma.
442 bssplit=str Sometimes you want even finer grained control of the
443 block sizes issued, not just an even split between them.
444 This option allows you to weight various block sizes,
445 so that you are able to define a specific amount of
446 block sizes issued. The format for this option is:
448 bssplit=blocksize/percentage:blocksize/percentage
450 for as many block sizes as needed. So if you want to define
451 a workload that has 50% 64k blocks, 10% 4k blocks, and
452 40% 32k blocks, you would write:
454 bssplit=4k/10:64k/50:32k/40
456 Ordering does not matter. If the percentage is left blank,
457 fio will fill in the remaining values evenly. So a bssplit
458 option like this one:
460 bssplit=4k/50:1k/:32k/
462 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
463 always add up to 100, if bssplit is given a range that adds
464 up to more, it will error out.
466 bssplit also supports giving separate splits to reads and
467 writes. The format is identical to what bs= accepts. You
468 have to separate the read and write parts with a comma. So
469 if you want a workload that has 50% 2k reads and 50% 4k reads,
470 while having 90% 4k writes and 10% 8k writes, you would
473 bssplit=2k/50:4k/50,4k/90,8k/10
476 bs_unaligned If this option is given, any byte size value within bsrange
477 may be used as a block range. This typically wont work with
478 direct IO, as that normally requires sector alignment.
480 zero_buffers If this option is given, fio will init the IO buffers to
481 all zeroes. The default is to fill them with random data.
483 refill_buffers If this option is given, fio will refill the IO buffers
484 on every submit. The default is to only fill it at init
485 time and reuse that data. Only makes sense if zero_buffers
486 isn't specified, naturally. If data verification is enabled,
487 refill_buffers is also automatically enabled.
489 scramble_buffers=bool If refill_buffers is too costly and the target is
490 using data deduplication, then setting this option will
491 slightly modify the IO buffer contents to defeat normal
492 de-dupe attempts. This is not enough to defeat more clever
493 block compression attempts, but it will stop naive dedupe of
494 blocks. Default: true.
496 buffer_compress_percentage=int If this is set, then fio will attempt to
497 provide IO buffer content (on WRITEs) that compress to
498 the specified level. Fio does this by providing a mix of
499 random data and zeroes. Note that this is per block size
500 unit, for file/disk wide compression level that matches
501 this setting, you'll also want to set refill_buffers.
503 buffer_compress_chunk=int See buffer_compress_percentage. This
504 setting allows fio to manage how big the ranges of random
505 data and zeroed data is. Without this set, fio will
506 provide buffer_compress_percentage of blocksize random
507 data, followed by the remaining zeroed. With this set
508 to some chunk size smaller than the block size, fio can
509 alternate random and zeroed data throughout the IO
512 nrfiles=int Number of files to use for this job. Defaults to 1.
514 openfiles=int Number of files to keep open at the same time. Defaults to
515 the same as nrfiles, can be set smaller to limit the number
518 file_service_type=str Defines how fio decides which file from a job to
519 service next. The following types are defined:
521 random Just choose a file at random.
523 roundrobin Round robin over open files. This
526 sequential Finish one file before moving on to
527 the next. Multiple files can still be
528 open depending on 'openfiles'.
530 The string can have a number appended, indicating how
531 often to switch to a new file. So if option random:4 is
532 given, fio will switch to a new random file after 4 ios
535 ioengine=str Defines how the job issues io to the file. The following
538 sync Basic read(2) or write(2) io. lseek(2) is
539 used to position the io location.
541 psync Basic pread(2) or pwrite(2) io.
543 vsync Basic readv(2) or writev(2) IO.
545 libaio Linux native asynchronous io. Note that Linux
546 may only support queued behaviour with
547 non-buffered IO (set direct=1 or buffered=0).
548 This engine defines engine specific options.
550 posixaio glibc posix asynchronous io.
552 solarisaio Solaris native asynchronous io.
554 windowsaio Windows native asynchronous io.
556 mmap File is memory mapped and data copied
557 to/from using memcpy(3).
559 splice splice(2) is used to transfer the data and
560 vmsplice(2) to transfer data from user
563 syslet-rw Use the syslet system calls to make
564 regular read/write async.
566 sg SCSI generic sg v3 io. May either be
567 synchronous using the SG_IO ioctl, or if
568 the target is an sg character device
569 we use read(2) and write(2) for asynchronous
572 null Doesn't transfer any data, just pretends
573 to. This is mainly used to exercise fio
574 itself and for debugging/testing purposes.
576 net Transfer over the network to given host:port.
577 Depending on the protocol used, the hostname,
578 port, listen and filename options are used to
579 specify what sort of connection to make, while
580 the protocol option determines which protocol
582 This engine defines engine specific options.
584 netsplice Like net, but uses splice/vmsplice to
585 map data and send/receive.
586 This engine defines engine specific options.
588 cpuio Doesn't transfer any data, but burns CPU
589 cycles according to the cpuload= and
590 cpucycle= options. Setting cpuload=85
591 will cause that job to do nothing but burn
592 85% of the CPU. In case of SMP machines,
593 use numjobs=<no_of_cpu> to get desired CPU
594 usage, as the cpuload only loads a single
595 CPU at the desired rate.
597 guasi The GUASI IO engine is the Generic Userspace
598 Asyncronous Syscall Interface approach
601 http://www.xmailserver.org/guasi-lib.html
603 for more info on GUASI.
605 rdma The RDMA I/O engine supports both RDMA
606 memory semantics (RDMA_WRITE/RDMA_READ) and
607 channel semantics (Send/Recv) for the
608 InfiniBand, RoCE and iWARP protocols.
610 falloc IO engine that does regular fallocate to
611 simulate data transfer as fio ioengine.
612 DDIR_READ does fallocate(,mode = keep_size,)
613 DDIR_WRITE does fallocate(,mode = 0)
614 DDIR_TRIM does fallocate(,mode = punch_hole)
616 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
617 ioctls to simulate defragment activity in
618 request to DDIR_WRITE event
620 external Prefix to specify loading an external
621 IO engine object file. Append the engine
622 filename, eg ioengine=external:/tmp/foo.o
623 to load ioengine foo.o in /tmp.
625 iodepth=int This defines how many io units to keep in flight against
626 the file. The default is 1 for each file defined in this
627 job, can be overridden with a larger value for higher
628 concurrency. Note that increasing iodepth beyond 1 will not
629 affect synchronous ioengines (except for small degress when
630 verify_async is in use). Even async engines may impose OS
631 restrictions causing the desired depth not to be achieved.
632 This may happen on Linux when using libaio and not setting
633 direct=1, since buffered IO is not async on that OS. Keep an
634 eye on the IO depth distribution in the fio output to verify
635 that the achieved depth is as expected. Default: 1.
637 iodepth_batch_submit=int
638 iodepth_batch=int This defines how many pieces of IO to submit at once.
639 It defaults to 1 which means that we submit each IO
640 as soon as it is available, but can be raised to submit
641 bigger batches of IO at the time.
643 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
644 at once. It defaults to 1 which means that we'll ask
645 for a minimum of 1 IO in the retrieval process from
646 the kernel. The IO retrieval will go on until we
647 hit the limit set by iodepth_low. If this variable is
648 set to 0, then fio will always check for completed
649 events before queuing more IO. This helps reduce
650 IO latency, at the cost of more retrieval system calls.
652 iodepth_low=int The low water mark indicating when to start filling
653 the queue again. Defaults to the same as iodepth, meaning
654 that fio will attempt to keep the queue full at all times.
655 If iodepth is set to eg 16 and iodepth_low is set to 4, then
656 after fio has filled the queue of 16 requests, it will let
657 the depth drain down to 4 before starting to fill it again.
659 direct=bool If value is true, use non-buffered io. This is usually
660 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
661 On Windows the synchronous ioengines don't support direct io.
663 buffered=bool If value is true, use buffered io. This is the opposite
664 of the 'direct' option. Defaults to true.
666 offset=int Start io at the given offset in the file. The data before
667 the given offset will not be touched. This effectively
668 caps the file size at real_size - offset.
670 offset_increment=int If this is provided, then the real offset becomes
671 the offset + offset_increment * thread_number, where the
672 thread number is a counter that starts at 0 and is incremented
673 for each job. This option is useful if there are several jobs
674 which are intended to operate on a file in parallel in disjoint
675 segments, with even spacing between the starting points.
677 fsync=int If writing to a file, issue a sync of the dirty data
678 for every number of blocks given. For example, if you give
679 32 as a parameter, fio will sync the file for every 32
680 writes issued. If fio is using non-buffered io, we may
681 not sync the file. The exception is the sg io engine, which
682 synchronizes the disk cache anyway.
684 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
686 In FreeBSD and Windows there is no fdatasync(), this falls back to
689 sync_file_range=str:val Use sync_file_range() for every 'val' number of
690 write operations. Fio will track range of writes that
691 have happened since the last sync_file_range() call. 'str'
692 can currently be one or more of:
694 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
695 write SYNC_FILE_RANGE_WRITE
696 wait_after SYNC_FILE_RANGE_WAIT_AFTER
698 So if you do sync_file_range=wait_before,write:8, fio would
699 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
700 every 8 writes. Also see the sync_file_range(2) man page.
701 This option is Linux specific.
703 overwrite=bool If true, writes to a file will always overwrite existing
704 data. If the file doesn't already exist, it will be
705 created before the write phase begins. If the file exists
706 and is large enough for the specified write phase, nothing
709 end_fsync=bool If true, fsync file contents when a write stage has completed.
711 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
712 This differs from end_fsync in that it will happen on every
713 file close, not just at the end of the job.
715 rwmixread=int How large a percentage of the mix should be reads.
717 rwmixwrite=int How large a percentage of the mix should be writes. If both
718 rwmixread and rwmixwrite is given and the values do not add
719 up to 100%, the latter of the two will be used to override
720 the first. This may interfere with a given rate setting,
721 if fio is asked to limit reads or writes to a certain rate.
722 If that is the case, then the distribution may be skewed.
724 random_distribution=str:float By default, fio will use a completely uniform
725 random distribution when asked to perform random IO. Sometimes
726 it is useful to skew the distribution in specific ways,
727 ensuring that some parts of the data is more hot than others.
728 fio includes the following distribution models:
730 random Uniform random distribution
731 zipf Zipf distribution
732 pareto Pareto distribution
734 When using a zipf or pareto distribution, an input value
735 is also needed to define the access pattern. For zipf, this
736 is the zipf theta. For pareto, it's the pareto power. Fio
737 includes a test program, genzipf, that can be used visualize
738 what the given input values will yield in terms of hit rates.
739 If you wanted to use zipf with a theta of 1.2, you would use
740 random_distribution=zipf:1.2 as the option. If a non-uniform
741 model is used, fio will disable use of the random map.
743 norandommap Normally fio will cover every block of the file when doing
744 random IO. If this option is given, fio will just get a
745 new random offset without looking at past io history. This
746 means that some blocks may not be read or written, and that
747 some blocks may be read/written more than once. This option
748 is mutually exclusive with verify= if and only if multiple
749 blocksizes (via bsrange=) are used, since fio only tracks
750 complete rewrites of blocks.
752 softrandommap=bool See norandommap. If fio runs with the random block map
753 enabled and it fails to allocate the map, if this option is
754 set it will continue without a random block map. As coverage
755 will not be as complete as with random maps, this option is
758 random_generator=str Fio supports the following engines for generating
759 IO offsets for random IO:
761 tausworthe Strong 2^88 cycle random number generator
762 lfsr Linear feedback shift register generator
764 Tausworthe is a strong random number generator, but it
765 requires tracking on the side if we want to ensure that
766 blocks are only read or written once. LFSR guarantees
767 that we never generate the same offset twice, and it's
768 also less computationally expensive. It's not a true
769 random generator, however, though for IO purposes it's
770 typically good enough. LFSR only works with single
771 block sizes, not with workloads that use multiple block
772 sizes. If used with such a workload, fio may read or write
773 some blocks multiple times.
775 nice=int Run the job with the given nice value. See man nice(2).
777 prio=int Set the io priority value of this job. Linux limits us to
778 a positive value between 0 and 7, with 0 being the highest.
781 prioclass=int Set the io priority class. See man ionice(1).
783 thinktime=int Stall the job x microseconds after an io has completed before
784 issuing the next. May be used to simulate processing being
785 done by an application. See thinktime_blocks and
789 Only valid if thinktime is set - pretend to spend CPU time
790 doing something with the data received, before falling back
791 to sleeping for the rest of the period specified by
795 Only valid if thinktime is set - control how many blocks
796 to issue, before waiting 'thinktime' usecs. If not set,
797 defaults to 1 which will make fio wait 'thinktime' usecs
800 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
801 the normal suffix rules apply. You can use rate=500k to limit
802 reads and writes to 500k each, or you can specify read and
803 writes separately. Using rate=1m,500k would limit reads to
804 1MB/sec and writes to 500KB/sec. Capping only reads or
805 writes can be done with rate=,500k or rate=500k,. The former
806 will only limit writes (to 500KB/sec), the latter will only
809 ratemin=int Tell fio to do whatever it can to maintain at least this
810 bandwidth. Failing to meet this requirement, will cause
811 the job to exit. The same format as rate is used for
812 read vs write separation.
814 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
815 as rate, just specified independently of bandwidth. If the
816 job is given a block size range instead of a fixed value,
817 the smallest block size is used as the metric. The same format
818 as rate is used for read vs write seperation.
820 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
821 the job to exit. The same format as rate is used for read vs
824 max_latency=int If set, fio will exit the job if it exceeds this maximum
825 latency. It will exit with an ETIME error.
827 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
830 cpumask=int Set the CPU affinity of this job. The parameter given is a
831 bitmask of allowed CPU's the job may run on. So if you want
832 the allowed CPUs to be 1 and 5, you would pass the decimal
833 value of (1 << 1 | 1 << 5), or 34. See man
834 sched_setaffinity(2). This may not work on all supported
835 operating systems or kernel versions. This option doesn't
836 work well for a higher CPU count than what you can store in
837 an integer mask, so it can only control cpus 1-32. For
838 boxes with larger CPU counts, use cpus_allowed.
840 cpus_allowed=str Controls the same options as cpumask, but it allows a text
841 setting of the permitted CPUs instead. So to use CPUs 1 and
842 5, you would specify cpus_allowed=1,5. This options also
843 allows a range of CPUs. Say you wanted a binding to CPUs
844 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
846 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
847 arguments allow comma delimited list of cpu numbers,
848 A-B ranges, or 'all'. Note, to enable numa options support,
849 fio must be built on a system with libnuma-dev(el) installed.
851 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
852 nodes. Format of the argements:
854 `mode' is one of the following memory policy:
855 default, prefer, bind, interleave, local
856 For `default' and `local' memory policy, no node is
857 needed to be specified.
858 For `prefer', only one node is allowed.
859 For `bind' and `interleave', it allow comma delimited
860 list of numbers, A-B ranges, or 'all'.
862 startdelay=time Start this job the specified number of seconds after fio
863 has started. Only useful if the job file contains several
864 jobs, and you want to delay starting some jobs to a certain
867 runtime=time Tell fio to terminate processing after the specified number
868 of seconds. It can be quite hard to determine for how long
869 a specified job will run, so this parameter is handy to
870 cap the total runtime to a given time.
872 time_based If set, fio will run for the duration of the runtime
873 specified even if the file(s) are completely read or
874 written. It will simply loop over the same workload
875 as many times as the runtime allows.
877 ramp_time=time If set, fio will run the specified workload for this amount
878 of time before logging any performance numbers. Useful for
879 letting performance settle before logging results, thus
880 minimizing the runtime required for stable results. Note
881 that the ramp_time is considered lead in time for a job,
882 thus it will increase the total runtime if a special timeout
883 or runtime is specified.
885 invalidate=bool Invalidate the buffer/page cache parts for this file prior
886 to starting io. Defaults to true.
888 sync=bool Use sync io for buffered writes. For the majority of the
889 io engines, this means using O_SYNC.
892 mem=str Fio can use various types of memory as the io unit buffer.
893 The allowed values are:
895 malloc Use memory from malloc(3) as the buffers.
897 shm Use shared memory as the buffers. Allocated
900 shmhuge Same as shm, but use huge pages as backing.
902 mmap Use mmap to allocate buffers. May either be
903 anonymous memory, or can be file backed if
904 a filename is given after the option. The
905 format is mem=mmap:/path/to/file.
907 mmaphuge Use a memory mapped huge file as the buffer
908 backing. Append filename after mmaphuge, ala
909 mem=mmaphuge:/hugetlbfs/file
911 The area allocated is a function of the maximum allowed
912 bs size for the job, multiplied by the io depth given. Note
913 that for shmhuge and mmaphuge to work, the system must have
914 free huge pages allocated. This can normally be checked
915 and set by reading/writing /proc/sys/vm/nr_hugepages on a
916 Linux system. Fio assumes a huge page is 4MB in size. So
917 to calculate the number of huge pages you need for a given
918 job file, add up the io depth of all jobs (normally one unless
919 iodepth= is used) and multiply by the maximum bs set. Then
920 divide that number by the huge page size. You can see the
921 size of the huge pages in /proc/meminfo. If no huge pages
922 are allocated by having a non-zero number in nr_hugepages,
923 using mmaphuge or shmhuge will fail. Also see hugepage-size.
925 mmaphuge also needs to have hugetlbfs mounted and the file
926 location should point there. So if it's mounted in /huge,
927 you would use mem=mmaphuge:/huge/somefile.
929 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
930 Note that the given alignment is applied to the first IO unit
931 buffer, if using iodepth the alignment of the following buffers
932 are given by the bs used. In other words, if using a bs that is
933 a multiple of the page sized in the system, all buffers will
934 be aligned to this value. If using a bs that is not page
935 aligned, the alignment of subsequent IO memory buffers is the
936 sum of the iomem_align and bs used.
939 Defines the size of a huge page. Must at least be equal
940 to the system setting, see /proc/meminfo. Defaults to 4MB.
941 Should probably always be a multiple of megabytes, so using
942 hugepage-size=Xm is the preferred way to set this to avoid
943 setting a non-pow-2 bad value.
945 exitall When one job finishes, terminate the rest. The default is
946 to wait for each job to finish, sometimes that is not the
949 bwavgtime=int Average the calculated bandwidth over the given time. Value
950 is specified in milliseconds.
952 iopsavgtime=int Average the calculated IOPS over the given time. Value
953 is specified in milliseconds.
955 create_serialize=bool If true, serialize the file creating for the jobs.
956 This may be handy to avoid interleaving of data
957 files, which may greatly depend on the filesystem
958 used and even the number of processors in the system.
960 create_fsync=bool fsync the data file after creation. This is the
963 create_on_open=bool Don't pre-setup the files for IO, just create open()
964 when it's time to do IO to that file.
966 create_only=bool If true, fio will only run the setup phase of the job.
967 If files need to be laid out or updated on disk, only
968 that will be done. The actual job contents are not
971 pre_read=bool If this is given, files will be pre-read into memory before
972 starting the given IO operation. This will also clear
973 the 'invalidate' flag, since it is pointless to pre-read
974 and then drop the cache. This will only work for IO engines
975 that are seekable, since they allow you to read the same data
976 multiple times. Thus it will not work on eg network or splice
979 unlink=bool Unlink the job files when done. Not the default, as repeated
980 runs of that job would then waste time recreating the file
983 loops=int Run the specified number of iterations of this job. Used
984 to repeat the same workload a given number of times. Defaults
987 do_verify=bool Run the verify phase after a write phase. Only makes sense if
988 verify is set. Defaults to 1.
990 verify=str If writing to a file, fio can verify the file contents
991 after each iteration of the job. The allowed values are:
993 md5 Use an md5 sum of the data area and store
994 it in the header of each block.
996 crc64 Use an experimental crc64 sum of the data
997 area and store it in the header of each
1000 crc32c Use a crc32c sum of the data area and store
1001 it in the header of each block.
1003 crc32c-intel Use hardware assisted crc32c calcuation
1004 provided on SSE4.2 enabled processors. Falls
1005 back to regular software crc32c, if not
1006 supported by the system.
1008 crc32 Use a crc32 sum of the data area and store
1009 it in the header of each block.
1011 crc16 Use a crc16 sum of the data area and store
1012 it in the header of each block.
1014 crc7 Use a crc7 sum of the data area and store
1015 it in the header of each block.
1017 sha512 Use sha512 as the checksum function.
1019 sha256 Use sha256 as the checksum function.
1021 sha1 Use optimized sha1 as the checksum function.
1023 meta Write extra information about each io
1024 (timestamp, block number etc.). The block
1025 number is verified. See also verify_pattern.
1027 null Only pretend to verify. Useful for testing
1028 internals with ioengine=null, not for much
1031 This option can be used for repeated burn-in tests of a
1032 system to make sure that the written data is also
1033 correctly read back. If the data direction given is
1034 a read or random read, fio will assume that it should
1035 verify a previously written file. If the data direction
1036 includes any form of write, the verify will be of the
1039 verifysort=bool If set, fio will sort written verify blocks when it deems
1040 it faster to read them back in a sorted manner. This is
1041 often the case when overwriting an existing file, since
1042 the blocks are already laid out in the file system. You
1043 can ignore this option unless doing huge amounts of really
1044 fast IO where the red-black tree sorting CPU time becomes
1047 verify_offset=int Swap the verification header with data somewhere else
1048 in the block before writing. Its swapped back before
1051 verify_interval=int Write the verification header at a finer granularity
1052 than the blocksize. It will be written for chunks the
1053 size of header_interval. blocksize should divide this
1056 verify_pattern=str If set, fio will fill the io buffers with this
1057 pattern. Fio defaults to filling with totally random
1058 bytes, but sometimes it's interesting to fill with a known
1059 pattern for io verification purposes. Depending on the
1060 width of the pattern, fio will fill 1/2/3/4 bytes of the
1061 buffer at the time(it can be either a decimal or a hex number).
1062 The verify_pattern if larger than a 32-bit quantity has to
1063 be a hex number that starts with either "0x" or "0X". Use
1066 verify_fatal=bool Normally fio will keep checking the entire contents
1067 before quitting on a block verification failure. If this
1068 option is set, fio will exit the job on the first observed
1071 verify_dump=bool If set, dump the contents of both the original data
1072 block and the data block we read off disk to files. This
1073 allows later analysis to inspect just what kind of data
1074 corruption occurred. Off by default.
1076 verify_async=int Fio will normally verify IO inline from the submitting
1077 thread. This option takes an integer describing how many
1078 async offload threads to create for IO verification instead,
1079 causing fio to offload the duty of verifying IO contents
1080 to one or more separate threads. If using this offload
1081 option, even sync IO engines can benefit from using an
1082 iodepth setting higher than 1, as it allows them to have
1083 IO in flight while verifies are running.
1085 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1086 async IO verification threads. See cpus_allowed for the
1089 verify_backlog=int Fio will normally verify the written contents of a
1090 job that utilizes verify once that job has completed. In
1091 other words, everything is written then everything is read
1092 back and verified. You may want to verify continually
1093 instead for a variety of reasons. Fio stores the meta data
1094 associated with an IO block in memory, so for large
1095 verify workloads, quite a bit of memory would be used up
1096 holding this meta data. If this option is enabled, fio
1097 will write only N blocks before verifying these blocks.
1099 will verify the previously written blocks before continuing
1102 verify_backlog_batch=int Control how many blocks fio will verify
1103 if verify_backlog is set. If not set, will default to
1104 the value of verify_backlog (meaning the entire queue
1105 is read back and verified). If verify_backlog_batch is
1106 less than verify_backlog then not all blocks will be verified,
1107 if verify_backlog_batch is larger than verify_backlog, some
1108 blocks will be verified more than once.
1111 wait_for_previous Wait for preceeding jobs in the job file to exit, before
1112 starting this one. Can be used to insert serialization
1113 points in the job file. A stone wall also implies starting
1114 a new reporting group.
1116 new_group Start a new reporting group. See: group_reporting.
1118 numjobs=int Create the specified number of clones of this job. May be
1119 used to setup a larger number of threads/processes doing
1120 the same thing. Each thread is reported separately; to see
1121 statistics for all clones as a whole, use group_reporting in
1122 conjunction with new_group.
1124 group_reporting It may sometimes be interesting to display statistics for
1125 groups of jobs as a whole instead of for each individual job.
1126 This is especially true if 'numjobs' is used; looking at
1127 individual thread/process output quickly becomes unwieldy.
1128 To see the final report per-group instead of per-job, use
1129 'group_reporting'. Jobs in a file will be part of the same
1130 reporting group, unless if separated by a stonewall, or by
1133 thread fio defaults to forking jobs, however if this option is
1134 given, fio will use pthread_create(3) to create threads
1137 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1139 zoneskip=int Skip the specified number of bytes when zonesize data has
1140 been read. The two zone options can be used to only do
1141 io on zones of a file.
1143 write_iolog=str Write the issued io patterns to the specified file. See
1144 read_iolog. Specify a separate file for each job, otherwise
1145 the iologs will be interspersed and the file may be corrupt.
1147 read_iolog=str Open an iolog with the specified file name and replay the
1148 io patterns it contains. This can be used to store a
1149 workload and replay it sometime later. The iolog given
1150 may also be a blktrace binary file, which allows fio
1151 to replay a workload captured by blktrace. See blktrace
1152 for how to capture such logging data. For blktrace replay,
1153 the file needs to be turned into a blkparse binary data
1154 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1156 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1157 is to attempt to respect the time stamps within the log and
1158 replay them with the appropriate delay between IOPS. By
1159 setting this variable fio will not respect the timestamps and
1160 attempt to replay them as fast as possible while still
1161 respecting ordering. The result is the same I/O pattern to a
1162 given device, but different timings.
1164 replay_redirect=str While replaying I/O patterns using read_iolog the
1165 default behavior is to replay the IOPS onto the major/minor
1166 device that each IOP was recorded from. This is sometimes
1167 undesireable because on a different machine those major/minor
1168 numbers can map to a different device. Changing hardware on
1169 the same system can also result in a different major/minor
1170 mapping. Replay_redirect causes all IOPS to be replayed onto
1171 the single specified device regardless of the device it was
1172 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1173 IO in the blktrace to be replayed onto /dev/sdc. This means
1174 multiple devices will be replayed onto a single, if the trace
1175 contains multiple devices. If you want multiple devices to be
1176 replayed concurrently to multiple redirected devices you must
1177 blkparse your trace into separate traces and replay them with
1178 independent fio invocations. Unfortuantely this also breaks
1179 the strict time ordering between multiple device accesses.
1181 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1182 file. Can be used to store data of the bandwidth of the
1183 jobs in their lifetime. The included fio_generate_plots
1184 script uses gnuplot to turn these text files into nice
1185 graphs. See write_lat_log for behaviour of given
1186 filename. For this option, the suffix is _bw.log.
1188 write_lat_log=str Same as write_bw_log, except that this option stores io
1189 submission, completion, and total latencies instead. If no
1190 filename is given with this option, the default filename of
1191 "jobname_type.log" is used. Even if the filename is given,
1192 fio will still append the type of log. So if one specifies
1196 The actual log names will be foo_slat.log, foo_slat.log,
1197 and foo_lat.log. This helps fio_generate_plot fine the logs
1200 write_bw_log=str If given, write an IOPS log of the jobs in this job
1201 file. See write_bw_log.
1203 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1204 given with this option, the default filename of
1205 "jobname_type.log" is used. Even if the filename is given,
1206 fio will still append the type of log.
1208 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1209 or bw log for every IO that completes. When writing to the
1210 disk log, that can quickly grow to a very large size. Setting
1211 this option makes fio average the each log entry over the
1212 specified period of time, reducing the resolution of the log.
1215 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1216 potentially be used instead of removing memory or booting
1217 with less memory to simulate a smaller amount of memory.
1219 exec_prerun=str Before running this job, issue the command specified
1222 exec_postrun=str After the job completes, issue the command specified
1225 ioscheduler=str Attempt to switch the device hosting the file to the specified
1226 io scheduler before running.
1228 cpuload=int If the job is a CPU cycle eater, attempt to use the specified
1229 percentage of CPU cycles.
1231 cpuchunks=int If the job is a CPU cycle eater, split the load into
1232 cycles of the given time. In microseconds.
1234 disk_util=bool Generate disk utilization statistics, if the platform
1235 supports it. Defaults to on.
1237 disable_lat=bool Disable measurements of total latency numbers. Useful
1238 only for cutting back the number of calls to gettimeofday,
1239 as that does impact performance at really high IOPS rates.
1240 Note that to really get rid of a large amount of these
1241 calls, this option must be used with disable_slat and
1244 disable_clat=bool Disable measurements of completion latency numbers. See
1247 disable_slat=bool Disable measurements of submission latency numbers. See
1250 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1253 clat_percentiles=bool Enable the reporting of percentiles of
1254 completion latencies.
1256 percentile_list=float_list Overwrite the default list of percentiles
1257 for completion latencies. Each number is a floating
1258 number in the range (0,100], and the maximum length of
1259 the list is 20. Use ':' to separate the numbers, and
1260 list the numbers in ascending order. For example,
1261 --percentile_list=99.5:99.9 will cause fio to report
1262 the values of completion latency below which 99.5% and
1263 99.9% of the observed latencies fell, respectively.
1265 clocksource=str Use the given clocksource as the base of timing. The
1266 supported options are:
1268 gettimeofday gettimeofday(2)
1270 clock_gettime clock_gettime(2)
1272 cpu Internal CPU clock source
1274 cpu is the preferred clocksource if it is reliable, as it
1275 is very fast (and fio is heavy on time calls). Fio will
1276 automatically use this clocksource if it's supported and
1277 considered reliable on the system it is running on, unless
1278 another clocksource is specifically set. For x86/x86-64 CPUs,
1279 this means supporting TSC Invariant.
1281 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1282 (disable_clat, disable_slat, disable_bw) plus reduce
1283 precision of the timeout somewhat to really shrink
1284 the gettimeofday() call count. With this option enabled,
1285 we only do about 0.4% of the gtod() calls we would have
1286 done if all time keeping was enabled.
1288 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1289 execution to just getting the current time. Fio (and
1290 databases, for instance) are very intensive on gettimeofday()
1291 calls. With this option, you can set one CPU aside for
1292 doing nothing but logging current time to a shared memory
1293 location. Then the other threads/processes that run IO
1294 workloads need only copy that segment, instead of entering
1295 the kernel with a gettimeofday() call. The CPU set aside
1296 for doing these time calls will be excluded from other
1297 uses. Fio will manually clear it from the CPU mask of other
1300 continue_on_error=str Normally fio will exit the job on the first observed
1301 failure. If this option is set, fio will continue the job when
1302 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1303 is exceeded or the I/O size specified is completed. If this
1304 option is used, there are two more stats that are appended,
1305 the total error count and the first error. The error field
1306 given in the stats is the first error that was hit during the
1309 The allowed values are:
1311 none Exit on any IO or verify errors.
1313 read Continue on read errors, exit on all others.
1315 write Continue on write errors, exit on all others.
1317 io Continue on any IO error, exit on all others.
1319 verify Continue on verify errors, exit on all others.
1321 all Continue on all errors.
1323 0 Backward-compatible alias for 'none'.
1325 1 Backward-compatible alias for 'all'.
1327 ignore_error=str Sometimes you want to ignore some errors during test
1328 in that case you can specify error list for each error type.
1329 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1330 errors for given error type is separated with ':'. Error
1331 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1333 ignore_error=EAGAIN,ENOSPC:122
1334 This option will ignore EAGAIN from READ, and ENOSPC and
1335 122(EDQUOT) from WRITE.
1337 error_dump=bool If set dump every error even if it is non fatal, true
1338 by default. If disabled only fatal error will be dumped
1340 cgroup=str Add job to this control group. If it doesn't exist, it will
1341 be created. The system must have a mounted cgroup blkio
1342 mount point for this to work. If your system doesn't have it
1343 mounted, you can do so with:
1345 # mount -t cgroup -o blkio none /cgroup
1347 cgroup_weight=int Set the weight of the cgroup to this value. See
1348 the documentation that comes with the kernel, allowed values
1349 are in the range of 100..1000.
1351 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1352 the job completion. To override this behavior and to leave
1353 cgroups around after the job completion, set cgroup_nodelete=1.
1354 This can be useful if one wants to inspect various cgroup
1355 files after job completion. Default: false
1357 uid=int Instead of running as the invoking user, set the user ID to
1358 this value before the thread/process does any work.
1360 gid=int Set group ID, see uid.
1362 flow_id=int The ID of the flow. If not specified, it defaults to being a
1363 global flow. See flow.
1365 flow=int Weight in token-based flow control. If this value is used, then
1366 there is a 'flow counter' which is used to regulate the
1367 proportion of activity between two or more jobs. fio attempts
1368 to keep this flow counter near zero. The 'flow' parameter
1369 stands for how much should be added or subtracted to the flow
1370 counter on each iteration of the main I/O loop. That is, if
1371 one job has flow=8 and another job has flow=-1, then there
1372 will be a roughly 1:8 ratio in how much one runs vs the other.
1374 flow_watermark=int The maximum value that the absolute value of the flow
1375 counter is allowed to reach before the job must wait for a
1376 lower value of the counter.
1378 flow_sleep=int The period of time, in microseconds, to wait after the flow
1379 watermark has been exceeded before retrying operations
1381 In addition, there are some parameters which are only valid when a specific
1382 ioengine is in use. These are used identically to normal parameters, with the
1383 caveat that when used on the command line, they must come after the ioengine
1384 that defines them is selected.
1386 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1387 the io_getevents system call to reap newly returned events.
1388 With this flag turned on, the AIO ring will be read directly
1389 from user-space to reap events. The reaping mode is only
1390 enabled when polling for a minimum of 0 events (eg when
1391 iodepth_batch_complete=0).
1393 [netsplice] hostname=str
1394 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1395 If the job is a TCP listener or UDP reader, the hostname is not
1396 used and must be omitted.
1398 [netsplice] port=int
1399 [net] port=int The TCP or UDP port to bind to or connect to.
1401 [netsplice] nodelay=bool
1402 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1404 [netsplice] protocol=str
1405 [netsplice] proto=str
1407 [net] proto=str The network protocol to use. Accepted values are:
1409 tcp Transmission control protocol
1410 udp User datagram protocol
1411 unix UNIX domain socket
1413 When the protocol is TCP or UDP, the port must also be given,
1414 as well as the hostname if the job is a TCP listener or UDP
1415 reader. For unix sockets, the normal filename option should be
1416 used and the port is invalid.
1418 [net] listen For TCP network connections, tell fio to listen for incoming
1419 connections rather than initiating an outgoing connection. The
1420 hostname must be omitted if this option is used.
1421 [net] pingpong Normal a network writer will just continue writing data, and
1422 a network reader will just consume packages. If pingpong=1
1423 is set, a writer will send its normal payload to the reader,
1424 then wait for the reader to send the same payload back. This
1425 allows fio to measure network latencies. The submission
1426 and completion latencies then measure local time spent
1427 sending or receiving, and the completion latency measures
1428 how long it took for the other end to receive and send back.
1430 [e4defrag] donorname=str
1431 File will be used as a block donor(swap extents between files)
1432 [e4defrag] inplace=int
1433 Configure donor file blocks allocation strategy
1434 0(default): Preallocate donor's file on init
1435 1 : allocate space immidietly inside defragment event,
1436 and free right after event
1440 6.0 Interpreting the output
1441 ---------------------------
1443 fio spits out a lot of output. While running, fio will display the
1444 status of the jobs created. An example of that would be:
1446 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1448 The characters inside the square brackets denote the current status of
1449 each thread. The possible values (in typical life cycle order) are:
1453 P Thread setup, but not started.
1455 I Thread initialized, waiting or generating necessary data.
1456 p Thread running pre-reading file(s).
1457 R Running, doing sequential reads.
1458 r Running, doing random reads.
1459 W Running, doing sequential writes.
1460 w Running, doing random writes.
1461 M Running, doing mixed sequential reads/writes.
1462 m Running, doing mixed random reads/writes.
1463 F Running, currently waiting for fsync()
1464 V Running, doing verification of written data.
1465 E Thread exited, not reaped by main thread yet.
1467 X Thread reaped, exited with an error.
1468 K Thread reaped, exited due to signal.
1470 The other values are fairly self explanatory - number of threads
1471 currently running and doing io, rate of io since last check (read speed
1472 listed first, then write speed), and the estimated completion percentage
1473 and time for the running group. It's impossible to estimate runtime of
1474 the following groups (if any). Note that the string is displayed in order,
1475 so it's possible to tell which of the jobs are currently doing what. The
1476 first character is the first job defined in the job file, and so forth.
1478 When fio is done (or interrupted by ctrl-c), it will show the data for
1479 each thread, group of threads, and disks in that order. For each data
1480 direction, the output looks like:
1482 Client1 (g=0): err= 0:
1483 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1484 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1485 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1486 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1487 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1488 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1489 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1490 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1491 issued r/w: total=0/32768, short=0/0
1492 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1493 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1495 The client number is printed, along with the group id and error of that
1496 thread. Below is the io statistics, here for writes. In the order listed,
1499 io= Number of megabytes io performed
1500 bw= Average bandwidth rate
1501 iops= Average IOs performed per second
1502 runt= The runtime of that thread
1503 slat= Submission latency (avg being the average, stdev being the
1504 standard deviation). This is the time it took to submit
1505 the io. For sync io, the slat is really the completion
1506 latency, since queue/complete is one operation there. This
1507 value can be in milliseconds or microseconds, fio will choose
1508 the most appropriate base and print that. In the example
1509 above, milliseconds is the best scale. Note: in --minimal mode
1510 latencies are always expressed in microseconds.
1511 clat= Completion latency. Same names as slat, this denotes the
1512 time from submission to completion of the io pieces. For
1513 sync io, clat will usually be equal (or very close) to 0,
1514 as the time from submit to complete is basically just
1515 CPU time (io has already been done, see slat explanation).
1516 bw= Bandwidth. Same names as the xlat stats, but also includes
1517 an approximate percentage of total aggregate bandwidth
1518 this thread received in this group. This last value is
1519 only really useful if the threads in this group are on the
1520 same disk, since they are then competing for disk access.
1521 cpu= CPU usage. User and system time, along with the number
1522 of context switches this thread went through, usage of
1523 system and user time, and finally the number of major
1524 and minor page faults.
1525 IO depths= The distribution of io depths over the job life time. The
1526 numbers are divided into powers of 2, so for example the
1527 16= entries includes depths up to that value but higher
1528 than the previous entry. In other words, it covers the
1529 range from 16 to 31.
1530 IO submit= How many pieces of IO were submitting in a single submit
1531 call. Each entry denotes that amount and below, until
1532 the previous entry - eg, 8=100% mean that we submitted
1533 anywhere in between 5-8 ios per submit call.
1534 IO complete= Like the above submit number, but for completions instead.
1535 IO issued= The number of read/write requests issued, and how many
1537 IO latencies= The distribution of IO completion latencies. This is the
1538 time from when IO leaves fio and when it gets completed.
1539 The numbers follow the same pattern as the IO depths,
1540 meaning that 2=1.6% means that 1.6% of the IO completed
1541 within 2 msecs, 20=12.8% means that 12.8% of the IO
1542 took more than 10 msecs, but less than (or equal to) 20 msecs.
1544 After each client has been listed, the group statistics are printed. They
1545 will look like this:
1547 Run status group 0 (all jobs):
1548 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1549 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1551 For each data direction, it prints:
1553 io= Number of megabytes io performed.
1554 aggrb= Aggregate bandwidth of threads in this group.
1555 minb= The minimum average bandwidth a thread saw.
1556 maxb= The maximum average bandwidth a thread saw.
1557 mint= The smallest runtime of the threads in that group.
1558 maxt= The longest runtime of the threads in that group.
1560 And finally, the disk statistics are printed. They will look like this:
1562 Disk stats (read/write):
1563 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1565 Each value is printed for both reads and writes, with reads first. The
1568 ios= Number of ios performed by all groups.
1569 merge= Number of merges io the io scheduler.
1570 ticks= Number of ticks we kept the disk busy.
1571 io_queue= Total time spent in the disk queue.
1572 util= The disk utilization. A value of 100% means we kept the disk
1573 busy constantly, 50% would be a disk idling half of the time.
1575 It is also possible to get fio to dump the current output while it is
1576 running, without terminating the job. To do that, send fio the USR1 signal.
1582 For scripted usage where you typically want to generate tables or graphs
1583 of the results, fio can output the results in a semicolon separated format.
1584 The format is one long line of values, such as:
1586 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%
1587 A description of this job goes here.
1589 The job description (if provided) follows on a second line.
1591 To enable terse output, use the --minimal command line option. The first
1592 value is the version of the terse output format. If the output has to
1593 be changed for some reason, this number will be incremented by 1 to
1594 signify that change.
1596 Split up, the format is as follows:
1598 terse version, fio version, jobname, groupid, error
1600 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1601 Submission latency: min, max, mean, deviation (usec)
1602 Completion latency: min, max, mean, deviation (usec)
1603 Completion latency percentiles: 20 fields (see below)
1604 Total latency: min, max, mean, deviation (usec)
1605 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1607 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1608 Submission latency: min, max, mean, deviation (usec)
1609 Completion latency: min, max, mean, deviation (usec)
1610 Completion latency percentiles: 20 fields (see below)
1611 Total latency: min, max, mean, deviation (usec)
1612 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1613 CPU usage: user, system, context switches, major faults, minor faults
1614 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1615 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1616 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1617 Disk utilization: Disk name, Read ios, write ios,
1618 Read merges, write merges,
1619 Read ticks, write ticks,
1620 Time spent in queue, disk utilization percentage
1621 Additional Info (dependant on continue_on_error, default off): total # errors, first error code
1623 Additional Info (dependant on description being set): Text description
1625 Completion latency percentiles can be a grouping of up to 20 sets, so
1626 for the terse output fio writes all of them. Each field will look like this:
1630 which is the Xth percentile, and the usec latency associated with it.
1632 For disk utilization, all disks used by fio are shown. So for each disk
1633 there will be a disk utilization section.
1636 8.0 Trace file format
1637 ---------------------
1638 There are two trace file format that you can encounter. The older (v1) format
1639 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1640 below in case that you get an old trace and want to understand it.
1642 In any case the trace is a simple text file with a single action per line.
1645 8.1 Trace file format v1
1646 ------------------------
1647 Each line represents a single io action in the following format:
1651 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1653 This format is not supported in Fio versions => 1.20-rc3.
1656 8.2 Trace file format v2
1657 ------------------------
1658 The second version of the trace file format was added in Fio version 1.17.
1659 It allows to access more then one file per trace and has a bigger set of
1660 possible file actions.
1662 The first line of the trace file has to be:
1666 Following this can be lines in two different formats, which are described below.
1668 The file management format:
1672 The filename is given as an absolute path. The action can be one of these:
1674 add Add the given filename to the trace
1675 open Open the file with the given filename. The filename has to have
1676 been added with the add action before.
1677 close Close the file with the given filename. The file has to have been
1681 The file io action format:
1683 filename action offset length
1685 The filename is given as an absolute path, and has to have been added and opened
1686 before it can be used with this format. The offset and length are given in
1687 bytes. The action can be one of these:
1689 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1690 read Read 'length' bytes beginning from 'offset'
1691 write Write 'length' bytes beginning from 'offset'
1692 sync fsync() the file
1693 datasync fdatasync() the file
1694 trim trim the given file from the given 'offset' for 'length' bytes
1697 9.0 CPU idleness profiling
1699 In some cases, we want to understand CPU overhead in a test. For example,
1700 we test patches for the specific goodness of whether they reduce CPU usage.
1701 fio implements a balloon approach to create a thread per CPU that runs at
1702 idle priority, meaning that it only runs when nobody else needs the cpu.
1703 By measuring the amount of work completed by the thread, idleness of each
1704 CPU can be derived accordingly.
1706 An unit work is defined as touching a full page of unsigned characters. Mean
1707 and standard deviation of time to complete an unit work is reported in "unit
1708 work" section. Options can be chosen to report detailed percpu idleness or
1709 overall system idleness by aggregating percpu stats.