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 When fio is utilized as a basis of any reasonably large test suite, it might be
163 desirable to share a set of standardized settings across multiple job files.
164 Instead of copy/pasting such settings, any section may pull in an external
165 .fio file with 'include filename' directive, as in the following example:
167 ; -- start job file including.fio --
171 include glob-include.fio
178 include test-include.fio
179 ; -- end job file including.fio --
181 ; -- start job file glob-include.fio --
184 ; -- end job file glob-include.fio --
186 ; -- start job file test-include.fio --
189 ; -- end job file test-include.fio --
191 Settings pulled into a section apply to that section only (except global
192 section). Include directives may be nested in that any included file may
193 contain further include directive(s).
196 4.1 Environment variables
197 -------------------------
199 fio also supports environment variable expansion in job files. Any
200 substring of the form "${VARNAME}" as part of an option value (in other
201 words, on the right of the `='), will be expanded to the value of the
202 environment variable called VARNAME. If no such environment variable
203 is defined, or VARNAME is the empty string, the empty string will be
206 As an example, let's look at a sample fio invocation and job file:
208 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
210 ; -- start job file --
217 This will expand to the following equivalent job file at runtime:
219 ; -- start job file --
226 fio ships with a few example job files, you can also look there for
229 4.2 Reserved keywords
230 ---------------------
232 Additionally, fio has a set of reserved keywords that will be replaced
233 internally with the appropriate value. Those keywords are:
235 $pagesize The architecture page size of the running system
236 $mb_memory Megabytes of total memory in the system
237 $ncpus Number of online available CPUs
239 These can be used on the command line or in the job file, and will be
240 automatically substituted with the current system values when the job
241 is run. Simple math is also supported on these keywords, so you can
242 perform actions like:
246 and get that properly expanded to 8 times the size of memory in the
250 5.0 Detailed list of parameters
251 -------------------------------
253 This section describes in details each parameter associated with a job.
254 Some parameters take an option of a given type, such as an integer or
255 a string. The following types are used:
257 str String. This is a sequence of alpha characters.
258 time Integer with possible time suffix. In seconds unless otherwise
259 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
260 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
261 and 'us' (or 'usec') for microseconds.
262 int SI integer. A whole number value, which may contain a suffix
263 describing the base of the number. Accepted suffixes are k/m/g/t/p,
264 meaning kilo, mega, giga, tera, and peta. The suffix is not case
265 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
266 as 'k'). So if you want to specify 4096, you could either write
267 out '4096' or just give 4k. The suffixes signify base 2 values, so
268 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
269 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
270 case, then 1000 is used as the multiplier. This can be handy for
271 disks, since manufacturers generally use base 10 values when listing
272 the capacity of a drive. If the option accepts an upper and lower
273 range, use a colon ':' or minus '-' to separate such values. May also
274 include a prefix to indicate numbers base. If 0x is used, the number
275 is assumed to be hexadecimal. See irange.
276 bool Boolean. Usually parsed as an integer, however only defined for
277 true and false (1 and 0).
278 irange Integer range with suffix. Allows value range to be given, such
279 as 1024-4096. A colon may also be used as the separator, eg
280 1k:4k. If the option allows two sets of ranges, they can be
281 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
283 float_list A list of floating numbers, separated by a ':' character.
285 With the above in mind, here follows the complete list of fio job
288 name=str ASCII name of the job. This may be used to override the
289 name printed by fio for this job. Otherwise the job
290 name is used. On the command line this parameter has the
291 special purpose of also signaling the start of a new
294 description=str Text description of the job. Doesn't do anything except
295 dump this text description when this job is run. It's
298 directory=str Prefix filenames with this directory. Used to place files
299 in a different location than "./". See the 'filename' option
300 for escaping certain characters.
302 filename=str Fio normally makes up a filename based on the job name,
303 thread number, and file number. If you want to share
304 files between threads in a job or several jobs, specify
305 a filename for each of them to override the default. If
306 the ioengine used is 'net', the filename is the host, port,
307 and protocol to use in the format of =host,port,protocol.
308 See ioengine=net for more. If the ioengine is file based, you
309 can specify a number of files by separating the names with a
310 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
311 as the two working files, you would use
312 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
313 accessed as \\.\PhysicalDrive0 for the first device,
314 \\.\PhysicalDrive1 for the second etc. Note: Windows and
315 FreeBSD prevent write access to areas of the disk containing
316 in-use data (e.g. filesystems).
317 If the wanted filename does need to include a colon, then
318 escape that with a '\' character. For instance, if the filename
319 is "/dev/dsk/foo@3,0:c", then you would use
320 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
321 stdin or stdout. Which of the two depends on the read/write
325 If sharing multiple files between jobs, it is usually necessary
326 to have fio generate the exact names that you want. By default,
327 fio will name a file based on the default file format
328 specification of jobname.jobnumber.filenumber. With this
329 option, that can be customized. Fio will recognize and replace
330 the following keywords in this string:
333 The name of the worker thread or process.
336 The incremental number of the worker thread or
340 The incremental number of the file for that worker
343 To have dependent jobs share a set of files, this option can
344 be set to have fio generate filenames that are shared between
345 the two. For instance, if testfiles.$filenum is specified,
346 file number 4 for any job will be named testfiles.4. The
347 default of $jobname.$jobnum.$filenum will be used if
348 no other format specifier is given.
350 opendir=str Tell fio to recursively add any file it can find in this
351 directory and down the file system tree.
353 lockfile=str Fio defaults to not locking any files before it does
354 IO to them. If a file or file descriptor is shared, fio
355 can serialize IO to that file to make the end result
356 consistent. This is usual for emulating real workloads that
357 share files. The lock modes are:
359 none No locking. The default.
360 exclusive Only one thread/process may do IO,
361 excluding all others.
362 readwrite Read-write locking on the file. Many
363 readers may access the file at the
364 same time, but writes get exclusive
368 rw=str Type of io pattern. Accepted values are:
370 read Sequential reads
371 write Sequential writes
372 randwrite Random writes
373 randread Random reads
374 rw,readwrite Sequential mixed reads and writes
375 randrw Random mixed reads and writes
377 For the mixed io types, the default is to split them 50/50.
378 For certain types of io the result may still be skewed a bit,
379 since the speed may be different. It is possible to specify
380 a number of IO's to do before getting a new offset, this is
381 one by appending a ':<nr>' to the end of the string given.
382 For a random read, it would look like 'rw=randread:8' for
383 passing in an offset modifier with a value of 8. If the
384 suffix is used with a sequential IO pattern, then the value
385 specified will be added to the generated offset for each IO.
386 For instance, using rw=write:4k will skip 4k for every
387 write. It turns sequential IO into sequential IO with holes.
388 See the 'rw_sequencer' option.
390 rw_sequencer=str If an offset modifier is given by appending a number to
391 the rw=<str> line, then this option controls how that
392 number modifies the IO offset being generated. Accepted
395 sequential Generate sequential offset
396 identical Generate the same offset
398 'sequential' is only useful for random IO, where fio would
399 normally generate a new random offset for every IO. If you
400 append eg 8 to randread, you would get a new random offset for
401 every 8 IO's. The result would be a seek for only every 8
402 IO's, instead of for every IO. Use rw=randread:8 to specify
403 that. As sequential IO is already sequential, setting
404 'sequential' for that would not result in any differences.
405 'identical' behaves in a similar fashion, except it sends
406 the same offset 8 number of times before generating a new
409 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
410 Storage manufacturers like to use 10^3 or 1000 as a base
411 ten unit instead, for obvious reasons. Allow values are
412 1024 or 1000, with 1024 being the default.
414 unified_rw_reporting=bool Fio normally reports statistics on a per
415 data direction basis, meaning that read, write, and trim are
416 accounted and reported separately. If this option is set,
417 the fio will sum the results and report them as "mixed"
420 randrepeat=bool For random IO workloads, seed the generator in a predictable
421 way so that results are repeatable across repetitions.
423 randseed=int Seed the random number generators based on this seed value, to
424 be able to control what sequence of output is being generated.
425 If not set, the random sequence depends on the randrepeat
428 use_os_rand=bool Fio can either use the random generator supplied by the OS
429 to generator random offsets, or it can use it's own internal
430 generator (based on Tausworthe). Default is to use the
431 internal generator, which is often of better quality and
434 fallocate=str Whether pre-allocation is performed when laying down files.
437 none Do not pre-allocate space
438 posix Pre-allocate via posix_fallocate()
439 keep Pre-allocate via fallocate() with
440 FALLOC_FL_KEEP_SIZE set
441 0 Backward-compatible alias for 'none'
442 1 Backward-compatible alias for 'posix'
444 May not be available on all supported platforms. 'keep' is only
445 available on Linux.If using ZFS on Solaris this must be set to
446 'none' because ZFS doesn't support it. Default: 'posix'.
448 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
449 on what IO patterns it is likely to issue. Sometimes you
450 want to test specific IO patterns without telling the
451 kernel about it, in which case you can disable this option.
452 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
453 IO and POSIX_FADV_RANDOM for random IO.
455 size=int The total size of file io for this job. Fio will run until
456 this many bytes has been transferred, unless runtime is
457 limited by other options (such as 'runtime', for instance).
458 Unless specific nrfiles and filesize options are given,
459 fio will divide this size between the available files
460 specified by the job. If not set, fio will use the full
461 size of the given files or devices. If the files do not
462 exist, size must be given. It is also possible to give
463 size as a percentage between 1 and 100. If size=20% is
464 given, fio will use 20% of the full size of the given
467 io_limit=int Normally fio operates within the region set by 'size', which
468 means that the 'size' option sets both the region and size of
469 IO to be performed. Sometimes that is not what you want. With
470 this option, it is possible to define just the amount of IO
471 that fio should do. For instance, if 'size' is set to 20G and
472 'io_limit' is set to 5G, fio will perform IO within the first
473 20G but exit when 5G have been done.
475 filesize=int Individual file sizes. May be a range, in which case fio
476 will select sizes for files at random within the given range
477 and limited to 'size' in total (if that is given). If not
478 given, each created file is the same size.
480 file_append=bool Perform IO after the end of the file. Normally fio will
481 operate within the size of a file. If this option is set, then
482 fio will append to the file instead. This has identical
483 behavior to setting offset to the size of a file. This option
484 is ignored on non-regular files.
487 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
488 space left on device) as the terminating condition. Only makes
489 sense with sequential write. For a read workload, the mount
490 point will be filled first then IO started on the result. This
491 option doesn't make sense if operating on a raw device node,
492 since the size of that is already known by the file system.
493 Additionally, writing beyond end-of-device will not return
497 bs=int The block size used for the io units. Defaults to 4k. Values
498 can be given for both read and writes. If a single int is
499 given, it will apply to both. If a second int is specified
500 after a comma, it will apply to writes only. In other words,
501 the format is either bs=read_and_write or bs=read,write,trim.
502 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
503 writes, and 8k for trims. You can terminate the list with
504 a trailing comma. bs=4k,8k, would use the default value for
505 trims.. If you only wish to set the write size, you
506 can do so by passing an empty read size - bs=,8k will set
507 8k for writes and leave the read default value.
510 ba=int At what boundary to align random IO offsets. Defaults to
511 the same as 'blocksize' the minimum blocksize given.
512 Minimum alignment is typically 512b for using direct IO,
513 though it usually depends on the hardware block size. This
514 option is mutually exclusive with using a random map for
515 files, so it will turn off that option.
517 blocksize_range=irange
518 bsrange=irange Instead of giving a single block size, specify a range
519 and fio will mix the issued io block sizes. The issued
520 io unit will always be a multiple of the minimum value
521 given (also see bs_unaligned). Applies to both reads and
522 writes, however a second range can be given after a comma.
525 bssplit=str Sometimes you want even finer grained control of the
526 block sizes issued, not just an even split between them.
527 This option allows you to weight various block sizes,
528 so that you are able to define a specific amount of
529 block sizes issued. The format for this option is:
531 bssplit=blocksize/percentage:blocksize/percentage
533 for as many block sizes as needed. So if you want to define
534 a workload that has 50% 64k blocks, 10% 4k blocks, and
535 40% 32k blocks, you would write:
537 bssplit=4k/10:64k/50:32k/40
539 Ordering does not matter. If the percentage is left blank,
540 fio will fill in the remaining values evenly. So a bssplit
541 option like this one:
543 bssplit=4k/50:1k/:32k/
545 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
546 always add up to 100, if bssplit is given a range that adds
547 up to more, it will error out.
549 bssplit also supports giving separate splits to reads and
550 writes. The format is identical to what bs= accepts. You
551 have to separate the read and write parts with a comma. So
552 if you want a workload that has 50% 2k reads and 50% 4k reads,
553 while having 90% 4k writes and 10% 8k writes, you would
556 bssplit=2k/50:4k/50,4k/90,8k/10
559 bs_unaligned If this option is given, any byte size value within bsrange
560 may be used as a block range. This typically wont work with
561 direct IO, as that normally requires sector alignment.
563 bs_is_seq_rand If this option is set, fio will use the normal read,write
564 blocksize settings as sequential,random instead. Any random
565 read or write will use the WRITE blocksize settings, and any
566 sequential read or write will use the READ blocksize setting.
568 zero_buffers If this option is given, fio will init the IO buffers to
569 all zeroes. The default is to fill them with random data.
570 The resulting IO buffers will not be completely zeroed,
571 unless scramble_buffers is also turned off.
573 refill_buffers If this option is given, fio will refill the IO buffers
574 on every submit. The default is to only fill it at init
575 time and reuse that data. Only makes sense if zero_buffers
576 isn't specified, naturally. If data verification is enabled,
577 refill_buffers is also automatically enabled.
579 scramble_buffers=bool If refill_buffers is too costly and the target is
580 using data deduplication, then setting this option will
581 slightly modify the IO buffer contents to defeat normal
582 de-dupe attempts. This is not enough to defeat more clever
583 block compression attempts, but it will stop naive dedupe of
584 blocks. Default: true.
586 buffer_compress_percentage=int If this is set, then fio will attempt to
587 provide IO buffer content (on WRITEs) that compress to
588 the specified level. Fio does this by providing a mix of
589 random data and zeroes. Note that this is per block size
590 unit, for file/disk wide compression level that matches
591 this setting, you'll also want to set refill_buffers.
593 buffer_compress_chunk=int See buffer_compress_percentage. This
594 setting allows fio to manage how big the ranges of random
595 data and zeroed data is. Without this set, fio will
596 provide buffer_compress_percentage of blocksize random
597 data, followed by the remaining zeroed. With this set
598 to some chunk size smaller than the block size, fio can
599 alternate random and zeroed data throughout the IO
602 buffer_pattern=str If set, fio will fill the io buffers with this
603 pattern. If not set, the contents of io buffers is defined by
604 the other options related to buffer contents. The setting can
605 be any pattern of bytes, and can be prefixed with 0x for hex
606 values. It may also be a string, where the string must then
609 dedupe_percentage=int If set, fio will generate this percentage of
610 identical buffers when writing. These buffers will be
611 naturally dedupable. The contents of the buffers depend on
612 what other buffer compression settings have been set. It's
613 possible to have the individual buffers either fully
614 compressible, or not at all. This option only controls the
615 distribution of unique buffers.
617 nrfiles=int Number of files to use for this job. Defaults to 1.
619 openfiles=int Number of files to keep open at the same time. Defaults to
620 the same as nrfiles, can be set smaller to limit the number
623 file_service_type=str Defines how fio decides which file from a job to
624 service next. The following types are defined:
626 random Just choose a file at random.
628 roundrobin Round robin over open files. This
631 sequential Finish one file before moving on to
632 the next. Multiple files can still be
633 open depending on 'openfiles'.
635 The string can have a number appended, indicating how
636 often to switch to a new file. So if option random:4 is
637 given, fio will switch to a new random file after 4 ios
640 ioengine=str Defines how the job issues io to the file. The following
643 sync Basic read(2) or write(2) io. lseek(2) is
644 used to position the io location.
646 psync Basic pread(2) or pwrite(2) io.
648 vsync Basic readv(2) or writev(2) IO.
650 psyncv Basic preadv(2) or pwritev(2) IO.
652 libaio Linux native asynchronous io. Note that Linux
653 may only support queued behaviour with
654 non-buffered IO (set direct=1 or buffered=0).
655 This engine defines engine specific options.
657 posixaio glibc posix asynchronous io.
659 solarisaio Solaris native asynchronous io.
661 windowsaio Windows native asynchronous io.
663 mmap File is memory mapped and data copied
664 to/from using memcpy(3).
666 splice splice(2) is used to transfer the data and
667 vmsplice(2) to transfer data from user
670 syslet-rw Use the syslet system calls to make
671 regular read/write async.
673 sg SCSI generic sg v3 io. May either be
674 synchronous using the SG_IO ioctl, or if
675 the target is an sg character device
676 we use read(2) and write(2) for asynchronous
679 null Doesn't transfer any data, just pretends
680 to. This is mainly used to exercise fio
681 itself and for debugging/testing purposes.
683 net Transfer over the network to given host:port.
684 Depending on the protocol used, the hostname,
685 port, listen and filename options are used to
686 specify what sort of connection to make, while
687 the protocol option determines which protocol
689 This engine defines engine specific options.
691 netsplice Like net, but uses splice/vmsplice to
692 map data and send/receive.
693 This engine defines engine specific options.
695 cpuio Doesn't transfer any data, but burns CPU
696 cycles according to the cpuload= and
697 cpucycle= options. Setting cpuload=85
698 will cause that job to do nothing but burn
699 85% of the CPU. In case of SMP machines,
700 use numjobs=<no_of_cpu> to get desired CPU
701 usage, as the cpuload only loads a single
702 CPU at the desired rate.
704 guasi The GUASI IO engine is the Generic Userspace
705 Asyncronous Syscall Interface approach
708 http://www.xmailserver.org/guasi-lib.html
710 for more info on GUASI.
712 rdma The RDMA I/O engine supports both RDMA
713 memory semantics (RDMA_WRITE/RDMA_READ) and
714 channel semantics (Send/Recv) for the
715 InfiniBand, RoCE and iWARP protocols.
717 falloc IO engine that does regular fallocate to
718 simulate data transfer as fio ioengine.
719 DDIR_READ does fallocate(,mode = keep_size,)
720 DDIR_WRITE does fallocate(,mode = 0)
721 DDIR_TRIM does fallocate(,mode = punch_hole)
723 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
724 ioctls to simulate defragment activity in
725 request to DDIR_WRITE event
727 rbd IO engine supporting direct access to Ceph
728 Rados Block Devices (RBD) via librbd without
729 the need to use the kernel rbd driver. This
730 ioengine defines engine specific options.
732 gfapi Using Glusterfs libgfapi sync interface to
733 direct access to Glusterfs volumes without
736 gfapi_async Using Glusterfs libgfapi async interface
737 to direct access to Glusterfs volumes without
738 having to go through FUSE. This ioengine
739 defines engine specific options.
741 libhdfs Read and write through Hadoop (HDFS).
742 The 'filename' option is used to specify host,
743 port of the hdfs name-node to connect. This
744 engine interprets offsets a little
745 differently. In HDFS, files once created
746 cannot be modified. So random writes are not
747 possible. To imitate this, libhdfs engine
748 expects bunch of small files to be created
749 over HDFS, and engine will randomly pick a
750 file out of those files based on the offset
751 generated by fio backend. (see the example
752 job file to create such files, use rw=write
753 option). Please note, you might want to set
754 necessary environment variables to work with
755 hdfs/libhdfs properly.
757 external Prefix to specify loading an external
758 IO engine object file. Append the engine
759 filename, eg ioengine=external:/tmp/foo.o
760 to load ioengine foo.o in /tmp.
762 iodepth=int This defines how many io units to keep in flight against
763 the file. The default is 1 for each file defined in this
764 job, can be overridden with a larger value for higher
765 concurrency. Note that increasing iodepth beyond 1 will not
766 affect synchronous ioengines (except for small degress when
767 verify_async is in use). Even async engines may impose OS
768 restrictions causing the desired depth not to be achieved.
769 This may happen on Linux when using libaio and not setting
770 direct=1, since buffered IO is not async on that OS. Keep an
771 eye on the IO depth distribution in the fio output to verify
772 that the achieved depth is as expected. Default: 1.
774 iodepth_batch_submit=int
775 iodepth_batch=int This defines how many pieces of IO to submit at once.
776 It defaults to 1 which means that we submit each IO
777 as soon as it is available, but can be raised to submit
778 bigger batches of IO at the time.
780 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
781 at once. It defaults to 1 which means that we'll ask
782 for a minimum of 1 IO in the retrieval process from
783 the kernel. The IO retrieval will go on until we
784 hit the limit set by iodepth_low. If this variable is
785 set to 0, then fio will always check for completed
786 events before queuing more IO. This helps reduce
787 IO latency, at the cost of more retrieval system calls.
789 iodepth_low=int The low water mark indicating when to start filling
790 the queue again. Defaults to the same as iodepth, meaning
791 that fio will attempt to keep the queue full at all times.
792 If iodepth is set to eg 16 and iodepth_low is set to 4, then
793 after fio has filled the queue of 16 requests, it will let
794 the depth drain down to 4 before starting to fill it again.
796 direct=bool If value is true, use non-buffered io. This is usually
797 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
798 On Windows the synchronous ioengines don't support direct io.
800 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
801 writes are guaranteed to be stable once acknowledged by
802 the operating system. Only Linux supports O_ATOMIC right
805 buffered=bool If value is true, use buffered io. This is the opposite
806 of the 'direct' option. Defaults to true.
808 offset=int Start io at the given offset in the file. The data before
809 the given offset will not be touched. This effectively
810 caps the file size at real_size - offset.
812 offset_increment=int If this is provided, then the real offset becomes
813 offset + offset_increment * thread_number, where the thread
814 number is a counter that starts at 0 and is incremented for
815 each sub-job (i.e. when numjobs option is specified). This
816 option is useful if there are several jobs which are intended
817 to operate on a file in parallel disjoint segments, with
818 even spacing between the starting points.
820 number_ios=int Fio will normally perform IOs until it has exhausted the size
821 of the region set by size=, or if it exhaust the allocated
822 time (or hits an error condition). With this setting, the
823 range/size can be set independently of the number of IOs to
824 perform. When fio reaches this number, it will exit normally
827 fsync=int If writing to a file, issue a sync of the dirty data
828 for every number of blocks given. For example, if you give
829 32 as a parameter, fio will sync the file for every 32
830 writes issued. If fio is using non-buffered io, we may
831 not sync the file. The exception is the sg io engine, which
832 synchronizes the disk cache anyway.
834 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
836 In FreeBSD and Windows there is no fdatasync(), this falls back to
839 sync_file_range=str:val Use sync_file_range() for every 'val' number of
840 write operations. Fio will track range of writes that
841 have happened since the last sync_file_range() call. 'str'
842 can currently be one or more of:
844 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
845 write SYNC_FILE_RANGE_WRITE
846 wait_after SYNC_FILE_RANGE_WAIT_AFTER
848 So if you do sync_file_range=wait_before,write:8, fio would
849 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
850 every 8 writes. Also see the sync_file_range(2) man page.
851 This option is Linux specific.
853 overwrite=bool If true, writes to a file will always overwrite existing
854 data. If the file doesn't already exist, it will be
855 created before the write phase begins. If the file exists
856 and is large enough for the specified write phase, nothing
859 end_fsync=bool If true, fsync file contents when a write stage has completed.
861 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
862 This differs from end_fsync in that it will happen on every
863 file close, not just at the end of the job.
865 rwmixread=int How large a percentage of the mix should be reads.
867 rwmixwrite=int How large a percentage of the mix should be writes. If both
868 rwmixread and rwmixwrite is given and the values do not add
869 up to 100%, the latter of the two will be used to override
870 the first. This may interfere with a given rate setting,
871 if fio is asked to limit reads or writes to a certain rate.
872 If that is the case, then the distribution may be skewed.
874 random_distribution=str:float By default, fio will use a completely uniform
875 random distribution when asked to perform random IO. Sometimes
876 it is useful to skew the distribution in specific ways,
877 ensuring that some parts of the data is more hot than others.
878 fio includes the following distribution models:
880 random Uniform random distribution
881 zipf Zipf distribution
882 pareto Pareto distribution
884 When using a zipf or pareto distribution, an input value
885 is also needed to define the access pattern. For zipf, this
886 is the zipf theta. For pareto, it's the pareto power. Fio
887 includes a test program, genzipf, that can be used visualize
888 what the given input values will yield in terms of hit rates.
889 If you wanted to use zipf with a theta of 1.2, you would use
890 random_distribution=zipf:1.2 as the option. If a non-uniform
891 model is used, fio will disable use of the random map.
893 percentage_random=int For a random workload, set how big a percentage should
894 be random. This defaults to 100%, in which case the workload
895 is fully random. It can be set from anywhere from 0 to 100.
896 Setting it to 0 would make the workload fully sequential. Any
897 setting in between will result in a random mix of sequential
898 and random IO, at the given percentages. It is possible to
899 set different values for reads, writes, and trim. To do so,
900 simply use a comma separated list. See blocksize.
902 norandommap Normally fio will cover every block of the file when doing
903 random IO. If this option is given, fio will just get a
904 new random offset without looking at past io history. This
905 means that some blocks may not be read or written, and that
906 some blocks may be read/written more than once. This option
907 is mutually exclusive with verify= if and only if multiple
908 blocksizes (via bsrange=) are used, since fio only tracks
909 complete rewrites of blocks.
911 softrandommap=bool See norandommap. If fio runs with the random block map
912 enabled and it fails to allocate the map, if this option is
913 set it will continue without a random block map. As coverage
914 will not be as complete as with random maps, this option is
917 random_generator=str Fio supports the following engines for generating
918 IO offsets for random IO:
920 tausworthe Strong 2^88 cycle random number generator
921 lfsr Linear feedback shift register generator
923 Tausworthe is a strong random number generator, but it
924 requires tracking on the side if we want to ensure that
925 blocks are only read or written once. LFSR guarantees
926 that we never generate the same offset twice, and it's
927 also less computationally expensive. It's not a true
928 random generator, however, though for IO purposes it's
929 typically good enough. LFSR only works with single
930 block sizes, not with workloads that use multiple block
931 sizes. If used with such a workload, fio may read or write
932 some blocks multiple times.
934 nice=int Run the job with the given nice value. See man nice(2).
936 prio=int Set the io priority value of this job. Linux limits us to
937 a positive value between 0 and 7, with 0 being the highest.
940 prioclass=int Set the io priority class. See man ionice(1).
942 thinktime=int Stall the job x microseconds after an io has completed before
943 issuing the next. May be used to simulate processing being
944 done by an application. See thinktime_blocks and
948 Only valid if thinktime is set - pretend to spend CPU time
949 doing something with the data received, before falling back
950 to sleeping for the rest of the period specified by
954 Only valid if thinktime is set - control how many blocks
955 to issue, before waiting 'thinktime' usecs. If not set,
956 defaults to 1 which will make fio wait 'thinktime' usecs
957 after every block. This effectively makes any queue depth
958 setting redundant, since no more than 1 IO will be queued
959 before we have to complete it and do our thinktime. In
960 other words, this setting effectively caps the queue depth
961 if the latter is larger.
963 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
964 the normal suffix rules apply. You can use rate=500k to limit
965 reads and writes to 500k each, or you can specify read and
966 writes separately. Using rate=1m,500k would limit reads to
967 1MB/sec and writes to 500KB/sec. Capping only reads or
968 writes can be done with rate=,500k or rate=500k,. The former
969 will only limit writes (to 500KB/sec), the latter will only
972 ratemin=int Tell fio to do whatever it can to maintain at least this
973 bandwidth. Failing to meet this requirement, will cause
974 the job to exit. The same format as rate is used for
975 read vs write separation.
977 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
978 as rate, just specified independently of bandwidth. If the
979 job is given a block size range instead of a fixed value,
980 the smallest block size is used as the metric. The same format
981 as rate is used for read vs write separation.
983 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
984 the job to exit. The same format as rate is used for read vs
987 latency_target=int If set, fio will attempt to find the max performance
988 point that the given workload will run at while maintaining a
989 latency below this target. The values is given in microseconds.
990 See latency_window and latency_percentile
992 latency_window=int Used with latency_target to specify the sample window
993 that the job is run at varying queue depths to test the
994 performance. The value is given in microseconds.
996 latency_percentile=float The percentage of IOs that must fall within the
997 criteria specified by latency_target and latency_window. If not
998 set, this defaults to 100.0, meaning that all IOs must be equal
999 or below to the value set by latency_target.
1001 max_latency=int If set, fio will exit the job if it exceeds this maximum
1002 latency. It will exit with an ETIME error.
1004 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
1007 cpumask=int Set the CPU affinity of this job. The parameter given is a
1008 bitmask of allowed CPU's the job may run on. So if you want
1009 the allowed CPUs to be 1 and 5, you would pass the decimal
1010 value of (1 << 1 | 1 << 5), or 34. See man
1011 sched_setaffinity(2). This may not work on all supported
1012 operating systems or kernel versions. This option doesn't
1013 work well for a higher CPU count than what you can store in
1014 an integer mask, so it can only control cpus 1-32. For
1015 boxes with larger CPU counts, use cpus_allowed.
1017 cpus_allowed=str Controls the same options as cpumask, but it allows a text
1018 setting of the permitted CPUs instead. So to use CPUs 1 and
1019 5, you would specify cpus_allowed=1,5. This options also
1020 allows a range of CPUs. Say you wanted a binding to CPUs
1021 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
1023 cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
1024 specified by cpus_allowed or cpumask. Two policies are
1027 shared All jobs will share the CPU set specified.
1028 split Each job will get a unique CPU from the CPU set.
1030 'shared' is the default behaviour, if the option isn't
1031 specified. If split is specified, then fio will will assign
1032 one cpu per job. If not enough CPUs are given for the jobs
1033 listed, then fio will roundrobin the CPUs in the set.
1035 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
1036 arguments allow comma delimited list of cpu numbers,
1037 A-B ranges, or 'all'. Note, to enable numa options support,
1038 fio must be built on a system with libnuma-dev(el) installed.
1040 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
1041 nodes. Format of the argements:
1043 `mode' is one of the following memory policy:
1044 default, prefer, bind, interleave, local
1045 For `default' and `local' memory policy, no node is
1046 needed to be specified.
1047 For `prefer', only one node is allowed.
1048 For `bind' and `interleave', it allow comma delimited
1049 list of numbers, A-B ranges, or 'all'.
1051 startdelay=time Start this job the specified number of seconds after fio
1052 has started. Only useful if the job file contains several
1053 jobs, and you want to delay starting some jobs to a certain
1056 runtime=time Tell fio to terminate processing after the specified number
1057 of seconds. It can be quite hard to determine for how long
1058 a specified job will run, so this parameter is handy to
1059 cap the total runtime to a given time.
1061 time_based If set, fio will run for the duration of the runtime
1062 specified even if the file(s) are completely read or
1063 written. It will simply loop over the same workload
1064 as many times as the runtime allows.
1066 ramp_time=time If set, fio will run the specified workload for this amount
1067 of time before logging any performance numbers. Useful for
1068 letting performance settle before logging results, thus
1069 minimizing the runtime required for stable results. Note
1070 that the ramp_time is considered lead in time for a job,
1071 thus it will increase the total runtime if a special timeout
1072 or runtime is specified.
1074 invalidate=bool Invalidate the buffer/page cache parts for this file prior
1075 to starting io. Defaults to true.
1077 sync=bool Use sync io for buffered writes. For the majority of the
1078 io engines, this means using O_SYNC.
1081 mem=str Fio can use various types of memory as the io unit buffer.
1082 The allowed values are:
1084 malloc Use memory from malloc(3) as the buffers.
1086 shm Use shared memory as the buffers. Allocated
1089 shmhuge Same as shm, but use huge pages as backing.
1091 mmap Use mmap to allocate buffers. May either be
1092 anonymous memory, or can be file backed if
1093 a filename is given after the option. The
1094 format is mem=mmap:/path/to/file.
1096 mmaphuge Use a memory mapped huge file as the buffer
1097 backing. Append filename after mmaphuge, ala
1098 mem=mmaphuge:/hugetlbfs/file
1100 The area allocated is a function of the maximum allowed
1101 bs size for the job, multiplied by the io depth given. Note
1102 that for shmhuge and mmaphuge to work, the system must have
1103 free huge pages allocated. This can normally be checked
1104 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1105 Linux system. Fio assumes a huge page is 4MB in size. So
1106 to calculate the number of huge pages you need for a given
1107 job file, add up the io depth of all jobs (normally one unless
1108 iodepth= is used) and multiply by the maximum bs set. Then
1109 divide that number by the huge page size. You can see the
1110 size of the huge pages in /proc/meminfo. If no huge pages
1111 are allocated by having a non-zero number in nr_hugepages,
1112 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1114 mmaphuge also needs to have hugetlbfs mounted and the file
1115 location should point there. So if it's mounted in /huge,
1116 you would use mem=mmaphuge:/huge/somefile.
1118 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1119 Note that the given alignment is applied to the first IO unit
1120 buffer, if using iodepth the alignment of the following buffers
1121 are given by the bs used. In other words, if using a bs that is
1122 a multiple of the page sized in the system, all buffers will
1123 be aligned to this value. If using a bs that is not page
1124 aligned, the alignment of subsequent IO memory buffers is the
1125 sum of the iomem_align and bs used.
1128 Defines the size of a huge page. Must at least be equal
1129 to the system setting, see /proc/meminfo. Defaults to 4MB.
1130 Should probably always be a multiple of megabytes, so using
1131 hugepage-size=Xm is the preferred way to set this to avoid
1132 setting a non-pow-2 bad value.
1134 exitall When one job finishes, terminate the rest. The default is
1135 to wait for each job to finish, sometimes that is not the
1138 bwavgtime=int Average the calculated bandwidth over the given time. Value
1139 is specified in milliseconds.
1141 iopsavgtime=int Average the calculated IOPS over the given time. Value
1142 is specified in milliseconds.
1144 create_serialize=bool If true, serialize the file creating for the jobs.
1145 This may be handy to avoid interleaving of data
1146 files, which may greatly depend on the filesystem
1147 used and even the number of processors in the system.
1149 create_fsync=bool fsync the data file after creation. This is the
1152 create_on_open=bool Don't pre-setup the files for IO, just create open()
1153 when it's time to do IO to that file.
1155 create_only=bool If true, fio will only run the setup phase of the job.
1156 If files need to be laid out or updated on disk, only
1157 that will be done. The actual job contents are not
1160 pre_read=bool If this is given, files will be pre-read into memory before
1161 starting the given IO operation. This will also clear
1162 the 'invalidate' flag, since it is pointless to pre-read
1163 and then drop the cache. This will only work for IO engines
1164 that are seekable, since they allow you to read the same data
1165 multiple times. Thus it will not work on eg network or splice
1168 unlink=bool Unlink the job files when done. Not the default, as repeated
1169 runs of that job would then waste time recreating the file
1170 set again and again.
1172 loops=int Run the specified number of iterations of this job. Used
1173 to repeat the same workload a given number of times. Defaults
1176 verify_only Do not perform specified workload---only verify data still
1177 matches previous invocation of this workload. This option
1178 allows one to check data multiple times at a later date
1179 without overwriting it. This option makes sense only for
1180 workloads that write data, and does not support workloads
1181 with the time_based option set.
1183 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1184 verify is set. Defaults to 1.
1186 verify=str If writing to a file, fio can verify the file contents
1187 after each iteration of the job. The allowed values are:
1189 md5 Use an md5 sum of the data area and store
1190 it in the header of each block.
1192 crc64 Use an experimental crc64 sum of the data
1193 area and store it in the header of each
1196 crc32c Use a crc32c sum of the data area and store
1197 it in the header of each block.
1199 crc32c-intel Use hardware assisted crc32c calcuation
1200 provided on SSE4.2 enabled processors. Falls
1201 back to regular software crc32c, if not
1202 supported by the system.
1204 crc32 Use a crc32 sum of the data area and store
1205 it in the header of each block.
1207 crc16 Use a crc16 sum of the data area and store
1208 it in the header of each block.
1210 crc7 Use a crc7 sum of the data area and store
1211 it in the header of each block.
1213 xxhash Use xxhash as the checksum function. Generally
1214 the fastest software checksum that fio
1217 sha512 Use sha512 as the checksum function.
1219 sha256 Use sha256 as the checksum function.
1221 sha1 Use optimized sha1 as the checksum function.
1223 meta Write extra information about each io
1224 (timestamp, block number etc.). The block
1225 number is verified. The io sequence number is
1226 verified for workloads that write data.
1227 See also verify_pattern.
1229 null Only pretend to verify. Useful for testing
1230 internals with ioengine=null, not for much
1233 This option can be used for repeated burn-in tests of a
1234 system to make sure that the written data is also
1235 correctly read back. If the data direction given is
1236 a read or random read, fio will assume that it should
1237 verify a previously written file. If the data direction
1238 includes any form of write, the verify will be of the
1241 verifysort=bool If set, fio will sort written verify blocks when it deems
1242 it faster to read them back in a sorted manner. This is
1243 often the case when overwriting an existing file, since
1244 the blocks are already laid out in the file system. You
1245 can ignore this option unless doing huge amounts of really
1246 fast IO where the red-black tree sorting CPU time becomes
1249 verify_offset=int Swap the verification header with data somewhere else
1250 in the block before writing. Its swapped back before
1253 verify_interval=int Write the verification header at a finer granularity
1254 than the blocksize. It will be written for chunks the
1255 size of header_interval. blocksize should divide this
1258 verify_pattern=str If set, fio will fill the io buffers with this
1259 pattern. Fio defaults to filling with totally random
1260 bytes, but sometimes it's interesting to fill with a known
1261 pattern for io verification purposes. Depending on the
1262 width of the pattern, fio will fill 1/2/3/4 bytes of the
1263 buffer at the time(it can be either a decimal or a hex number).
1264 The verify_pattern if larger than a 32-bit quantity has to
1265 be a hex number that starts with either "0x" or "0X". Use
1268 verify_fatal=bool Normally fio will keep checking the entire contents
1269 before quitting on a block verification failure. If this
1270 option is set, fio will exit the job on the first observed
1273 verify_dump=bool If set, dump the contents of both the original data
1274 block and the data block we read off disk to files. This
1275 allows later analysis to inspect just what kind of data
1276 corruption occurred. Off by default.
1278 verify_async=int Fio will normally verify IO inline from the submitting
1279 thread. This option takes an integer describing how many
1280 async offload threads to create for IO verification instead,
1281 causing fio to offload the duty of verifying IO contents
1282 to one or more separate threads. If using this offload
1283 option, even sync IO engines can benefit from using an
1284 iodepth setting higher than 1, as it allows them to have
1285 IO in flight while verifies are running.
1287 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1288 async IO verification threads. See cpus_allowed for the
1291 verify_backlog=int Fio will normally verify the written contents of a
1292 job that utilizes verify once that job has completed. In
1293 other words, everything is written then everything is read
1294 back and verified. You may want to verify continually
1295 instead for a variety of reasons. Fio stores the meta data
1296 associated with an IO block in memory, so for large
1297 verify workloads, quite a bit of memory would be used up
1298 holding this meta data. If this option is enabled, fio
1299 will write only N blocks before verifying these blocks.
1301 verify_backlog_batch=int Control how many blocks fio will verify
1302 if verify_backlog is set. If not set, will default to
1303 the value of verify_backlog (meaning the entire queue
1304 is read back and verified). If verify_backlog_batch is
1305 less than verify_backlog then not all blocks will be verified,
1306 if verify_backlog_batch is larger than verify_backlog, some
1307 blocks will be verified more than once.
1310 wait_for_previous Wait for preceding jobs in the job file to exit, before
1311 starting this one. Can be used to insert serialization
1312 points in the job file. A stone wall also implies starting
1313 a new reporting group.
1315 new_group Start a new reporting group. See: group_reporting.
1317 numjobs=int Create the specified number of clones of this job. May be
1318 used to setup a larger number of threads/processes doing
1319 the same thing. Each thread is reported separately; to see
1320 statistics for all clones as a whole, use group_reporting in
1321 conjunction with new_group.
1323 group_reporting It may sometimes be interesting to display statistics for
1324 groups of jobs as a whole instead of for each individual job.
1325 This is especially true if 'numjobs' is used; looking at
1326 individual thread/process output quickly becomes unwieldy.
1327 To see the final report per-group instead of per-job, use
1328 'group_reporting'. Jobs in a file will be part of the same
1329 reporting group, unless if separated by a stonewall, or by
1332 thread fio defaults to forking jobs, however if this option is
1333 given, fio will use pthread_create(3) to create threads
1336 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1338 zoneskip=int Skip the specified number of bytes when zonesize data has
1339 been read. The two zone options can be used to only do
1340 io on zones of a file.
1342 write_iolog=str Write the issued io patterns to the specified file. See
1343 read_iolog. Specify a separate file for each job, otherwise
1344 the iologs will be interspersed and the file may be corrupt.
1346 read_iolog=str Open an iolog with the specified file name and replay the
1347 io patterns it contains. This can be used to store a
1348 workload and replay it sometime later. The iolog given
1349 may also be a blktrace binary file, which allows fio
1350 to replay a workload captured by blktrace. See blktrace
1351 for how to capture such logging data. For blktrace replay,
1352 the file needs to be turned into a blkparse binary data
1353 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1355 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1356 is to attempt to respect the time stamps within the log and
1357 replay them with the appropriate delay between IOPS. By
1358 setting this variable fio will not respect the timestamps and
1359 attempt to replay them as fast as possible while still
1360 respecting ordering. The result is the same I/O pattern to a
1361 given device, but different timings.
1363 replay_redirect=str While replaying I/O patterns using read_iolog the
1364 default behavior is to replay the IOPS onto the major/minor
1365 device that each IOP was recorded from. This is sometimes
1366 undesirable because on a different machine those major/minor
1367 numbers can map to a different device. Changing hardware on
1368 the same system can also result in a different major/minor
1369 mapping. Replay_redirect causes all IOPS to be replayed onto
1370 the single specified device regardless of the device it was
1371 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1372 IO in the blktrace to be replayed onto /dev/sdc. This means
1373 multiple devices will be replayed onto a single, if the trace
1374 contains multiple devices. If you want multiple devices to be
1375 replayed concurrently to multiple redirected devices you must
1376 blkparse your trace into separate traces and replay them with
1377 independent fio invocations. Unfortuantely this also breaks
1378 the strict time ordering between multiple device accesses.
1380 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1381 file. Can be used to store data of the bandwidth of the
1382 jobs in their lifetime. The included fio_generate_plots
1383 script uses gnuplot to turn these text files into nice
1384 graphs. See write_lat_log for behaviour of given
1385 filename. For this option, the suffix is _bw.x.log, where
1386 x is the index of the job (1..N, where N is the number of
1389 write_lat_log=str Same as write_bw_log, except that this option stores io
1390 submission, completion, and total latencies instead. If no
1391 filename is given with this option, the default filename of
1392 "jobname_type.log" is used. Even if the filename is given,
1393 fio will still append the type of log. So if one specifies
1397 The actual log names will be foo_slat.x.log, foo_clat.x.log,
1398 and foo_lat.x.log, where x is the index of the job (1..N,
1399 where N is the number of jobs). This helps fio_generate_plot
1400 fine the logs automatically.
1402 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1403 given with this option, the default filename of
1404 "jobname_type.x.log" is used,where x is the index of the job
1405 (1..N, where N is the number of jobs). Even if the filename
1406 is given, fio will still append the type of log.
1408 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1409 or bw log for every IO that completes. When writing to the
1410 disk log, that can quickly grow to a very large size. Setting
1411 this option makes fio average the each log entry over the
1412 specified period of time, reducing the resolution of the log.
1415 log_offset=int If this is set, the iolog options will include the byte
1416 offset for the IO entry as well as the other data values.
1418 log_compression=int If this is set, fio will compress the IO logs as
1419 it goes, to keep the memory footprint lower. When a log
1420 reaches the specified size, that chunk is removed and
1421 compressed in the background. Given that IO logs are
1422 fairly highly compressible, this yields a nice memory
1423 savings for longer runs. The downside is that the
1424 compression will consume some background CPU cycles, so
1425 it may impact the run. This, however, is also true if
1426 the logging ends up consuming most of the system memory.
1427 So pick your poison. The IO logs are saved normally at the
1428 end of a run, by decompressing the chunks and storing them
1429 in the specified log file. This feature depends on the
1430 availability of zlib.
1432 log_store_compressed=bool If set, and log_compression is also set,
1433 fio will store the log files in a compressed format. They
1434 can be decompressed with fio, using the --inflate-log
1435 command line parameter. The files will be stored with a
1438 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1439 potentially be used instead of removing memory or booting
1440 with less memory to simulate a smaller amount of memory.
1441 The amount specified is per worker.
1443 exec_prerun=str Before running this job, issue the command specified
1444 through system(3). Output is redirected in a file called
1447 exec_postrun=str After the job completes, issue the command specified
1448 though system(3). Output is redirected in a file called
1449 jobname.postrun.txt.
1451 ioscheduler=str Attempt to switch the device hosting the file to the specified
1452 io scheduler before running.
1454 disk_util=bool Generate disk utilization statistics, if the platform
1455 supports it. Defaults to on.
1457 disable_lat=bool Disable measurements of total latency numbers. Useful
1458 only for cutting back the number of calls to gettimeofday,
1459 as that does impact performance at really high IOPS rates.
1460 Note that to really get rid of a large amount of these
1461 calls, this option must be used with disable_slat and
1464 disable_clat=bool Disable measurements of completion latency numbers. See
1467 disable_slat=bool Disable measurements of submission latency numbers. See
1470 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1473 clat_percentiles=bool Enable the reporting of percentiles of
1474 completion latencies.
1476 percentile_list=float_list Overwrite the default list of percentiles
1477 for completion latencies. Each number is a floating
1478 number in the range (0,100], and the maximum length of
1479 the list is 20. Use ':' to separate the numbers, and
1480 list the numbers in ascending order. For example,
1481 --percentile_list=99.5:99.9 will cause fio to report
1482 the values of completion latency below which 99.5% and
1483 99.9% of the observed latencies fell, respectively.
1485 clocksource=str Use the given clocksource as the base of timing. The
1486 supported options are:
1488 gettimeofday gettimeofday(2)
1490 clock_gettime clock_gettime(2)
1492 cpu Internal CPU clock source
1494 cpu is the preferred clocksource if it is reliable, as it
1495 is very fast (and fio is heavy on time calls). Fio will
1496 automatically use this clocksource if it's supported and
1497 considered reliable on the system it is running on, unless
1498 another clocksource is specifically set. For x86/x86-64 CPUs,
1499 this means supporting TSC Invariant.
1501 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1502 (disable_clat, disable_slat, disable_bw) plus reduce
1503 precision of the timeout somewhat to really shrink
1504 the gettimeofday() call count. With this option enabled,
1505 we only do about 0.4% of the gtod() calls we would have
1506 done if all time keeping was enabled.
1508 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1509 execution to just getting the current time. Fio (and
1510 databases, for instance) are very intensive on gettimeofday()
1511 calls. With this option, you can set one CPU aside for
1512 doing nothing but logging current time to a shared memory
1513 location. Then the other threads/processes that run IO
1514 workloads need only copy that segment, instead of entering
1515 the kernel with a gettimeofday() call. The CPU set aside
1516 for doing these time calls will be excluded from other
1517 uses. Fio will manually clear it from the CPU mask of other
1520 continue_on_error=str Normally fio will exit the job on the first observed
1521 failure. If this option is set, fio will continue the job when
1522 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1523 is exceeded or the I/O size specified is completed. If this
1524 option is used, there are two more stats that are appended,
1525 the total error count and the first error. The error field
1526 given in the stats is the first error that was hit during the
1529 The allowed values are:
1531 none Exit on any IO or verify errors.
1533 read Continue on read errors, exit on all others.
1535 write Continue on write errors, exit on all others.
1537 io Continue on any IO error, exit on all others.
1539 verify Continue on verify errors, exit on all others.
1541 all Continue on all errors.
1543 0 Backward-compatible alias for 'none'.
1545 1 Backward-compatible alias for 'all'.
1547 ignore_error=str Sometimes you want to ignore some errors during test
1548 in that case you can specify error list for each error type.
1549 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1550 errors for given error type is separated with ':'. Error
1551 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1553 ignore_error=EAGAIN,ENOSPC:122
1554 This option will ignore EAGAIN from READ, and ENOSPC and
1555 122(EDQUOT) from WRITE.
1557 error_dump=bool If set dump every error even if it is non fatal, true
1558 by default. If disabled only fatal error will be dumped
1560 cgroup=str Add job to this control group. If it doesn't exist, it will
1561 be created. The system must have a mounted cgroup blkio
1562 mount point for this to work. If your system doesn't have it
1563 mounted, you can do so with:
1565 # mount -t cgroup -o blkio none /cgroup
1567 cgroup_weight=int Set the weight of the cgroup to this value. See
1568 the documentation that comes with the kernel, allowed values
1569 are in the range of 100..1000.
1571 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1572 the job completion. To override this behavior and to leave
1573 cgroups around after the job completion, set cgroup_nodelete=1.
1574 This can be useful if one wants to inspect various cgroup
1575 files after job completion. Default: false
1577 uid=int Instead of running as the invoking user, set the user ID to
1578 this value before the thread/process does any work.
1580 gid=int Set group ID, see uid.
1582 flow_id=int The ID of the flow. If not specified, it defaults to being a
1583 global flow. See flow.
1585 flow=int Weight in token-based flow control. If this value is used, then
1586 there is a 'flow counter' which is used to regulate the
1587 proportion of activity between two or more jobs. fio attempts
1588 to keep this flow counter near zero. The 'flow' parameter
1589 stands for how much should be added or subtracted to the flow
1590 counter on each iteration of the main I/O loop. That is, if
1591 one job has flow=8 and another job has flow=-1, then there
1592 will be a roughly 1:8 ratio in how much one runs vs the other.
1594 flow_watermark=int The maximum value that the absolute value of the flow
1595 counter is allowed to reach before the job must wait for a
1596 lower value of the counter.
1598 flow_sleep=int The period of time, in microseconds, to wait after the flow
1599 watermark has been exceeded before retrying operations
1601 In addition, there are some parameters which are only valid when a specific
1602 ioengine is in use. These are used identically to normal parameters, with the
1603 caveat that when used on the command line, they must come after the ioengine
1604 that defines them is selected.
1606 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1607 the io_getevents system call to reap newly returned events.
1608 With this flag turned on, the AIO ring will be read directly
1609 from user-space to reap events. The reaping mode is only
1610 enabled when polling for a minimum of 0 events (eg when
1611 iodepth_batch_complete=0).
1613 [cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1615 [cpu] cpuchunks=int Split the load into cycles of the given time. In
1618 [cpu] exit_on_io_done=bool Detect when IO threads are done, then exit.
1620 [netsplice] hostname=str
1621 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1622 If the job is a TCP listener or UDP reader, the hostname is not
1623 used and must be omitted unless it is a valid UDP multicast
1626 [netsplice] port=int
1627 [net] port=int The TCP or UDP port to bind to or connect to.
1629 [netsplice] interface=str
1630 [net] interface=str The IP address of the network interface used to send or
1631 receive UDP multicast
1634 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1637 [netsplice] nodelay=bool
1638 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1640 [netsplice] protocol=str
1641 [netsplice] proto=str
1643 [net] proto=str The network protocol to use. Accepted values are:
1645 tcp Transmission control protocol
1646 tcpv6 Transmission control protocol V6
1647 udp User datagram protocol
1648 udpv6 User datagram protocol V6
1649 unix UNIX domain socket
1651 When the protocol is TCP or UDP, the port must also be given,
1652 as well as the hostname if the job is a TCP listener or UDP
1653 reader. For unix sockets, the normal filename option should be
1654 used and the port is invalid.
1656 [net] listen For TCP network connections, tell fio to listen for incoming
1657 connections rather than initiating an outgoing connection. The
1658 hostname must be omitted if this option is used.
1659 [net] pingpong Normaly a network writer will just continue writing data, and
1660 a network reader will just consume packages. If pingpong=1
1661 is set, a writer will send its normal payload to the reader,
1662 then wait for the reader to send the same payload back. This
1663 allows fio to measure network latencies. The submission
1664 and completion latencies then measure local time spent
1665 sending or receiving, and the completion latency measures
1666 how long it took for the other end to receive and send back.
1667 For UDP multicast traffic pingpong=1 should only be set for a
1668 single reader when multiple readers are listening to the same
1671 [e4defrag] donorname=str
1672 File will be used as a block donor(swap extents between files)
1673 [e4defrag] inplace=int
1674 Configure donor file blocks allocation strategy
1675 0(default): Preallocate donor's file on init
1676 1 : allocate space immidietly inside defragment event,
1677 and free right after event
1681 6.0 Interpreting the output
1682 ---------------------------
1684 fio spits out a lot of output. While running, fio will display the
1685 status of the jobs created. An example of that would be:
1687 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1689 The characters inside the square brackets denote the current status of
1690 each thread. The possible values (in typical life cycle order) are:
1694 P Thread setup, but not started.
1696 I Thread initialized, waiting or generating necessary data.
1697 p Thread running pre-reading file(s).
1698 R Running, doing sequential reads.
1699 r Running, doing random reads.
1700 W Running, doing sequential writes.
1701 w Running, doing random writes.
1702 M Running, doing mixed sequential reads/writes.
1703 m Running, doing mixed random reads/writes.
1704 F Running, currently waiting for fsync()
1705 f Running, finishing up (writing IO logs, etc)
1706 V Running, doing verification of written data.
1707 E Thread exited, not reaped by main thread yet.
1709 X Thread reaped, exited with an error.
1710 K Thread reaped, exited due to signal.
1712 Fio will condense the thread string as not to take up more space on the
1713 command line as is needed. For instance, if you have 10 readers and 10
1714 writers running, the output would look like this:
1716 Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s]
1718 Fio will still maintain the ordering, though. So the above means that jobs
1719 1..10 are readers, and 11..20 are writers.
1721 The other values are fairly self explanatory - number of threads
1722 currently running and doing io, rate of io since last check (read speed
1723 listed first, then write speed), and the estimated completion percentage
1724 and time for the running group. It's impossible to estimate runtime of
1725 the following groups (if any). Note that the string is displayed in order,
1726 so it's possible to tell which of the jobs are currently doing what. The
1727 first character is the first job defined in the job file, and so forth.
1729 When fio is done (or interrupted by ctrl-c), it will show the data for
1730 each thread, group of threads, and disks in that order. For each data
1731 direction, the output looks like:
1733 Client1 (g=0): err= 0:
1734 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1735 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1736 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1737 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1738 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1739 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1740 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1741 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1742 issued r/w: total=0/32768, short=0/0
1743 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1744 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1746 The client number is printed, along with the group id and error of that
1747 thread. Below is the io statistics, here for writes. In the order listed,
1750 io= Number of megabytes io performed
1751 bw= Average bandwidth rate
1752 iops= Average IOs performed per second
1753 runt= The runtime of that thread
1754 slat= Submission latency (avg being the average, stdev being the
1755 standard deviation). This is the time it took to submit
1756 the io. For sync io, the slat is really the completion
1757 latency, since queue/complete is one operation there. This
1758 value can be in milliseconds or microseconds, fio will choose
1759 the most appropriate base and print that. In the example
1760 above, milliseconds is the best scale. Note: in --minimal mode
1761 latencies are always expressed in microseconds.
1762 clat= Completion latency. Same names as slat, this denotes the
1763 time from submission to completion of the io pieces. For
1764 sync io, clat will usually be equal (or very close) to 0,
1765 as the time from submit to complete is basically just
1766 CPU time (io has already been done, see slat explanation).
1767 bw= Bandwidth. Same names as the xlat stats, but also includes
1768 an approximate percentage of total aggregate bandwidth
1769 this thread received in this group. This last value is
1770 only really useful if the threads in this group are on the
1771 same disk, since they are then competing for disk access.
1772 cpu= CPU usage. User and system time, along with the number
1773 of context switches this thread went through, usage of
1774 system and user time, and finally the number of major
1775 and minor page faults.
1776 IO depths= The distribution of io depths over the job life time. The
1777 numbers are divided into powers of 2, so for example the
1778 16= entries includes depths up to that value but higher
1779 than the previous entry. In other words, it covers the
1780 range from 16 to 31.
1781 IO submit= How many pieces of IO were submitting in a single submit
1782 call. Each entry denotes that amount and below, until
1783 the previous entry - eg, 8=100% mean that we submitted
1784 anywhere in between 5-8 ios per submit call.
1785 IO complete= Like the above submit number, but for completions instead.
1786 IO issued= The number of read/write requests issued, and how many
1788 IO latencies= The distribution of IO completion latencies. This is the
1789 time from when IO leaves fio and when it gets completed.
1790 The numbers follow the same pattern as the IO depths,
1791 meaning that 2=1.6% means that 1.6% of the IO completed
1792 within 2 msecs, 20=12.8% means that 12.8% of the IO
1793 took more than 10 msecs, but less than (or equal to) 20 msecs.
1795 After each client has been listed, the group statistics are printed. They
1796 will look like this:
1798 Run status group 0 (all jobs):
1799 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1800 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1802 For each data direction, it prints:
1804 io= Number of megabytes io performed.
1805 aggrb= Aggregate bandwidth of threads in this group.
1806 minb= The minimum average bandwidth a thread saw.
1807 maxb= The maximum average bandwidth a thread saw.
1808 mint= The smallest runtime of the threads in that group.
1809 maxt= The longest runtime of the threads in that group.
1811 And finally, the disk statistics are printed. They will look like this:
1813 Disk stats (read/write):
1814 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1816 Each value is printed for both reads and writes, with reads first. The
1819 ios= Number of ios performed by all groups.
1820 merge= Number of merges io the io scheduler.
1821 ticks= Number of ticks we kept the disk busy.
1822 io_queue= Total time spent in the disk queue.
1823 util= The disk utilization. A value of 100% means we kept the disk
1824 busy constantly, 50% would be a disk idling half of the time.
1826 It is also possible to get fio to dump the current output while it is
1827 running, without terminating the job. To do that, send fio the USR1 signal.
1828 You can also get regularly timed dumps by using the --status-interval
1829 parameter, or by creating a file in /tmp named fio-dump-status. If fio
1830 sees this file, it will unlink it and dump the current output status.
1836 For scripted usage where you typically want to generate tables or graphs
1837 of the results, fio can output the results in a semicolon separated format.
1838 The format is one long line of values, such as:
1840 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%
1841 A description of this job goes here.
1843 The job description (if provided) follows on a second line.
1845 To enable terse output, use the --minimal command line option. The first
1846 value is the version of the terse output format. If the output has to
1847 be changed for some reason, this number will be incremented by 1 to
1848 signify that change.
1850 Split up, the format is as follows:
1852 terse version, fio version, jobname, groupid, error
1854 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1855 Submission latency: min, max, mean, deviation (usec)
1856 Completion latency: min, max, mean, deviation (usec)
1857 Completion latency percentiles: 20 fields (see below)
1858 Total latency: min, max, mean, deviation (usec)
1859 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1861 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1862 Submission latency: min, max, mean, deviation (usec)
1863 Completion latency: min, max, mean, deviation (usec)
1864 Completion latency percentiles: 20 fields (see below)
1865 Total latency: min, max, mean, deviation (usec)
1866 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1867 CPU usage: user, system, context switches, major faults, minor faults
1868 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1869 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1870 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1871 Disk utilization: Disk name, Read ios, write ios,
1872 Read merges, write merges,
1873 Read ticks, write ticks,
1874 Time spent in queue, disk utilization percentage
1875 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
1877 Additional Info (dependent on description being set): Text description
1879 Completion latency percentiles can be a grouping of up to 20 sets, so
1880 for the terse output fio writes all of them. Each field will look like this:
1884 which is the Xth percentile, and the usec latency associated with it.
1886 For disk utilization, all disks used by fio are shown. So for each disk
1887 there will be a disk utilization section.
1890 8.0 Trace file format
1891 ---------------------
1892 There are two trace file format that you can encounter. The older (v1) format
1893 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1894 below in case that you get an old trace and want to understand it.
1896 In any case the trace is a simple text file with a single action per line.
1899 8.1 Trace file format v1
1900 ------------------------
1901 Each line represents a single io action in the following format:
1905 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1907 This format is not supported in Fio versions => 1.20-rc3.
1910 8.2 Trace file format v2
1911 ------------------------
1912 The second version of the trace file format was added in Fio version 1.17.
1913 It allows to access more then one file per trace and has a bigger set of
1914 possible file actions.
1916 The first line of the trace file has to be:
1920 Following this can be lines in two different formats, which are described below.
1922 The file management format:
1926 The filename is given as an absolute path. The action can be one of these:
1928 add Add the given filename to the trace
1929 open Open the file with the given filename. The filename has to have
1930 been added with the add action before.
1931 close Close the file with the given filename. The file has to have been
1935 The file io action format:
1937 filename action offset length
1939 The filename is given as an absolute path, and has to have been added and opened
1940 before it can be used with this format. The offset and length are given in
1941 bytes. The action can be one of these:
1943 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1944 read Read 'length' bytes beginning from 'offset'
1945 write Write 'length' bytes beginning from 'offset'
1946 sync fsync() the file
1947 datasync fdatasync() the file
1948 trim trim the given file from the given 'offset' for 'length' bytes
1951 9.0 CPU idleness profiling
1952 --------------------------
1953 In some cases, we want to understand CPU overhead in a test. For example,
1954 we test patches for the specific goodness of whether they reduce CPU usage.
1955 fio implements a balloon approach to create a thread per CPU that runs at
1956 idle priority, meaning that it only runs when nobody else needs the cpu.
1957 By measuring the amount of work completed by the thread, idleness of each
1958 CPU can be derived accordingly.
1960 An unit work is defined as touching a full page of unsigned characters. Mean
1961 and standard deviation of time to complete an unit work is reported in "unit
1962 work" section. Options can be chosen to report detailed percpu idleness or
1963 overall system idleness by aggregating percpu stats.