8 5. Detailed list of parameters
12 9. CPU idleness profiling
14 1.0 Overview and history
15 ------------------------
16 fio was originally written to save me the hassle of writing special test
17 case programs when I wanted to test a specific workload, either for
18 performance reasons or to find/reproduce a bug. The process of writing
19 such a test app can be tiresome, especially if you have to do it often.
20 Hence I needed a tool that would be able to simulate a given io workload
21 without resorting to writing a tailored test case again and again.
23 A test work load is difficult to define, though. There can be any number
24 of processes or threads involved, and they can each be using their own
25 way of generating io. You could have someone dirtying large amounts of
26 memory in an memory mapped file, or maybe several threads issuing
27 reads using asynchronous io. fio needed to be flexible enough to
28 simulate both of these cases, and many more.
32 The first step in getting fio to simulate a desired io workload, is
33 writing a job file describing that specific setup. A job file may contain
34 any number of threads and/or files - the typical contents of the job file
35 is a global section defining shared parameters, and one or more job
36 sections describing the jobs involved. When run, fio parses this file
37 and sets everything up as described. If we break down a job from top to
38 bottom, it contains the following basic parameters:
40 IO type Defines the io pattern issued to the file(s).
41 We may only be reading sequentially from this
42 file(s), or we may be writing randomly. Or even
43 mixing reads and writes, sequentially or randomly.
45 Block size In how large chunks are we issuing io? This may be
46 a single value, or it may describe a range of
49 IO size How much data are we going to be reading/writing.
51 IO engine How do we issue io? We could be memory mapping the
52 file, we could be using regular read/write, we
53 could be using splice, async io, syslet, or even
56 IO depth If the io engine is async, how large a queuing
57 depth do we want to maintain?
59 IO type Should we be doing buffered io, or direct/raw io?
61 Num files How many files are we spreading the workload over.
63 Num threads How many threads or processes should we spread
66 The above are the basic parameters defined for a workload, in addition
67 there's a multitude of parameters that modify other aspects of how this
73 See the README file for command line parameters, there are only a few
76 Running fio is normally the easiest part - you just give it the job file
77 (or job files) as parameters:
81 and it will start doing what the job_file tells it to do. You can give
82 more than one job file on the command line, fio will serialize the running
83 of those files. Internally that is the same as using the 'stonewall'
84 parameter described the the parameter section.
86 If the job file contains only one job, you may as well just give the
87 parameters on the command line. The command line parameters are identical
88 to the job parameters, with a few extra that control global parameters
89 (see README). For example, for the job file parameter iodepth=2, the
90 mirror command line option would be --iodepth 2 or --iodepth=2. You can
91 also use the command line for giving more than one job entry. For each
92 --name option that fio sees, it will start a new job with that name.
93 Command line entries following a --name entry will apply to that job,
94 until there are no more entries or a new --name entry is seen. This is
95 similar to the job file options, where each option applies to the current
96 job until a new [] job entry is seen.
98 fio does not need to run as root, except if the files or devices specified
99 in the job section requires that. Some other options may also be restricted,
100 such as memory locking, io scheduler switching, and decreasing the nice value.
105 As previously described, fio accepts one or more job files describing
106 what it is supposed to do. The job file format is the classic ini file,
107 where the names enclosed in [] brackets define the job name. You are free
108 to use any ascii name you want, except 'global' which has special meaning.
109 A global section sets defaults for the jobs described in that file. A job
110 may override a global section parameter, and a job file may even have
111 several global sections if so desired. A job is only affected by a global
112 section residing above it. If the first character in a line is a ';' or a
113 '#', the entire line is discarded as a comment.
115 So let's look at a really simple job file that defines two processes, each
116 randomly reading from a 128MB file.
118 ; -- start job file --
129 As you can see, the job file sections themselves are empty as all the
130 described parameters are shared. As no filename= option is given, fio
131 makes up a filename for each of the jobs as it sees fit. On the command
132 line, this job would look as follows:
134 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
137 Let's look at an example that has a number of processes writing randomly
140 ; -- start job file --
152 Here we have no global section, as we only have one job defined anyway.
153 We want to use async io here, with a depth of 4 for each file. We also
154 increased the buffer size used to 32KB and define numjobs to 4 to
155 fork 4 identical jobs. The result is 4 processes each randomly writing
156 to their own 64MB file. Instead of using the above job file, you could
157 have given the parameters on the command line. For this case, you would
160 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
162 4.1 Environment variables
163 -------------------------
165 fio also supports environment variable expansion in job files. Any
166 substring of the form "${VARNAME}" as part of an option value (in other
167 words, on the right of the `='), will be expanded to the value of the
168 environment variable called VARNAME. If no such environment variable
169 is defined, or VARNAME is the empty string, the empty string will be
172 As an example, let's look at a sample fio invocation and job file:
174 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
176 ; -- start job file --
183 This will expand to the following equivalent job file at runtime:
185 ; -- start job file --
192 fio ships with a few example job files, you can also look there for
195 4.2 Reserved keywords
196 ---------------------
198 Additionally, fio has a set of reserved keywords that will be replaced
199 internally with the appropriate value. Those keywords are:
201 $pagesize The architecture page size of the running system
202 $mb_memory Megabytes of total memory in the system
203 $ncpus Number of online available CPUs
205 These can be used on the command line or in the job file, and will be
206 automatically substituted with the current system values when the job
207 is run. Simple math is also supported on these keywords, so you can
208 perform actions like:
212 and get that properly expanded to 8 times the size of memory in the
216 5.0 Detailed list of parameters
217 -------------------------------
219 This section describes in details each parameter associated with a job.
220 Some parameters take an option of a given type, such as an integer or
221 a string. The following types are used:
223 str String. This is a sequence of alpha characters.
224 time Integer with possible time suffix. In seconds unless otherwise
225 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
227 int SI integer. A whole number value, which may contain a suffix
228 describing the base of the number. Accepted suffixes are k/m/g/t/p,
229 meaning kilo, mega, giga, tera, and peta. The suffix is not case
230 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
231 as 'k'). So if you want to specify 4096, you could either write
232 out '4096' or just give 4k. The suffixes signify base 2 values, so
233 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
234 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
235 case, then 1000 is used as the multiplier. This can be handy for
236 disks, since manufacturers generally use base 10 values when listing
237 the capacity of a drive. If the option accepts an upper and lower
238 range, use a colon ':' or minus '-' to separate such values. May also
239 include a prefix to indicate numbers base. If 0x is used, the number
240 is assumed to be hexadecimal. See irange.
241 bool Boolean. Usually parsed as an integer, however only defined for
242 true and false (1 and 0).
243 irange Integer range with suffix. Allows value range to be given, such
244 as 1024-4096. A colon may also be used as the separator, eg
245 1k:4k. If the option allows two sets of ranges, they can be
246 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
248 float_list A list of floating numbers, separated by a ':' character.
250 With the above in mind, here follows the complete list of fio job
253 name=str ASCII name of the job. This may be used to override the
254 name printed by fio for this job. Otherwise the job
255 name is used. On the command line this parameter has the
256 special purpose of also signaling the start of a new
259 description=str Text description of the job. Doesn't do anything except
260 dump this text description when this job is run. It's
263 directory=str Prefix filenames with this directory. Used to place files
264 in a different location than "./".
266 filename=str Fio normally makes up a filename based on the job name,
267 thread number, and file number. If you want to share
268 files between threads in a job or several jobs, specify
269 a filename for each of them to override the default. If
270 the ioengine used is 'net', the filename is the host, port,
271 and protocol to use in the format of =host,port,protocol.
272 See ioengine=net for more. If the ioengine is file based, you
273 can specify a number of files by separating the names with a
274 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
275 as the two working files, you would use
276 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
277 accessed as \\.\PhysicalDrive0 for the first device,
278 \\.\PhysicalDrive1 for the second etc. Note: Windows and
279 FreeBSD prevent write access to areas of the disk containing
280 in-use data (e.g. filesystems).
281 If the wanted filename does need to include a colon, then
282 escape that with a '\' character. For instance, if the filename
283 is "/dev/dsk/foo@3,0:c", then you would use
284 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
285 stdin or stdout. Which of the two depends on the read/write
289 If sharing multiple files between jobs, it is usually necessary
290 to have fio generate the exact names that you want. By default,
291 fio will name a file based on the default file format
292 specification of jobname.jobnumber.filenumber. With this
293 option, that can be customized. Fio will recognize and replace
294 the following keywords in this string:
297 The name of the worker thread or process.
300 The incremental number of the worker thread or
304 The incremental number of the file for that worker
307 To have dependent jobs share a set of files, this option can
308 be set to have fio generate filenames that are shared between
309 the two. For instance, if testfiles.$filenum is specified,
310 file number 4 for any job will be named testfiles.4. The
311 default of $jobname.$jobnum.$filenum will be used if
312 no other format specifier is given.
314 opendir=str Tell fio to recursively add any file it can find in this
315 directory and down the file system tree.
317 lockfile=str Fio defaults to not locking any files before it does
318 IO to them. If a file or file descriptor is shared, fio
319 can serialize IO to that file to make the end result
320 consistent. This is usual for emulating real workloads that
321 share files. The lock modes are:
323 none No locking. The default.
324 exclusive Only one thread/process may do IO,
325 excluding all others.
326 readwrite Read-write locking on the file. Many
327 readers may access the file at the
328 same time, but writes get exclusive
332 rw=str Type of io pattern. Accepted values are:
334 read Sequential reads
335 write Sequential writes
336 randwrite Random writes
337 randread Random reads
338 rw,readwrite Sequential mixed reads and writes
339 randrw Random mixed reads and writes
341 For the mixed io types, the default is to split them 50/50.
342 For certain types of io the result may still be skewed a bit,
343 since the speed may be different. It is possible to specify
344 a number of IO's to do before getting a new offset, this is
345 one by appending a ':<nr>' to the end of the string given.
346 For a random read, it would look like 'rw=randread:8' for
347 passing in an offset modifier with a value of 8. If the
348 suffix is used with a sequential IO pattern, then the value
349 specified will be added to the generated offset for each IO.
350 For instance, using rw=write:4k will skip 4k for every
351 write. It turns sequential IO into sequential IO with holes.
352 See the 'rw_sequencer' option.
354 rw_sequencer=str If an offset modifier is given by appending a number to
355 the rw=<str> line, then this option controls how that
356 number modifies the IO offset being generated. Accepted
359 sequential Generate sequential offset
360 identical Generate the same offset
362 'sequential' is only useful for random IO, where fio would
363 normally generate a new random offset for every IO. If you
364 append eg 8 to randread, you would get a new random offset for
365 every 8 IO's. The result would be a seek for only every 8
366 IO's, instead of for every IO. Use rw=randread:8 to specify
367 that. As sequential IO is already sequential, setting
368 'sequential' for that would not result in any differences.
369 'identical' behaves in a similar fashion, except it sends
370 the same offset 8 number of times before generating a new
373 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
374 Storage manufacturers like to use 10^3 or 1000 as a base
375 ten unit instead, for obvious reasons. Allow values are
376 1024 or 1000, with 1024 being the default.
378 unified_rw_reporting=bool Fio normally reports statistics on a per
379 data direction basis, meaning that read, write, and trim are
380 accounted and reported separately. If this option is set,
381 the fio will sum the results and report them as "mixed"
384 randrepeat=bool For random IO workloads, seed the generator in a predictable
385 way so that results are repeatable across repetitions.
387 use_os_rand=bool Fio can either use the random generator supplied by the OS
388 to generator random offsets, or it can use it's own internal
389 generator (based on Tausworthe). Default is to use the
390 internal generator, which is often of better quality and
393 fallocate=str Whether pre-allocation is performed when laying down files.
396 none Do not pre-allocate space
397 posix Pre-allocate via posix_fallocate()
398 keep Pre-allocate via fallocate() with
399 FALLOC_FL_KEEP_SIZE set
400 0 Backward-compatible alias for 'none'
401 1 Backward-compatible alias for 'posix'
403 May not be available on all supported platforms. 'keep' is only
404 available on Linux.If using ZFS on Solaris this must be set to
405 'none' because ZFS doesn't support it. Default: 'posix'.
407 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
408 on what IO patterns it is likely to issue. Sometimes you
409 want to test specific IO patterns without telling the
410 kernel about it, in which case you can disable this option.
411 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
412 IO and POSIX_FADV_RANDOM for random IO.
414 size=int The total size of file io for this job. Fio will run until
415 this many bytes has been transferred, unless runtime is
416 limited by other options (such as 'runtime', for instance).
417 Unless specific nrfiles and filesize options are given,
418 fio will divide this size between the available files
419 specified by the job. If not set, fio will use the full
420 size of the given files or devices. If the the files
421 do not exist, size must be given. It is also possible to
422 give size as a percentage between 1 and 100. If size=20%
423 is given, fio will use 20% of the full size of the given
426 filesize=int Individual file sizes. May be a range, in which case fio
427 will select sizes for files at random within the given range
428 and limited to 'size' in total (if that is given). If not
429 given, each created file is the same size.
432 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
433 space left on device) as the terminating condition. Only makes
434 sense with sequential write. For a read workload, the mount
435 point will be filled first then IO started on the result. This
436 option doesn't make sense if operating on a raw device node,
437 since the size of that is already known by the file system.
438 Additionally, writing beyond end-of-device will not return
442 bs=int The block size used for the io units. Defaults to 4k. Values
443 can be given for both read and writes. If a single int is
444 given, it will apply to both. If a second int is specified
445 after a comma, it will apply to writes only. In other words,
446 the format is either bs=read_and_write or bs=read,write.
447 bs=4k,8k will thus use 4k blocks for reads, and 8k blocks
448 for writes. If you only wish to set the write size, you
449 can do so by passing an empty read size - bs=,8k will set
450 8k for writes and leave the read default value.
453 ba=int At what boundary to align random IO offsets. Defaults to
454 the same as 'blocksize' the minimum blocksize given.
455 Minimum alignment is typically 512b for using direct IO,
456 though it usually depends on the hardware block size. This
457 option is mutually exclusive with using a random map for
458 files, so it will turn off that option.
460 blocksize_range=irange
461 bsrange=irange Instead of giving a single block size, specify a range
462 and fio will mix the issued io block sizes. The issued
463 io unit will always be a multiple of the minimum value
464 given (also see bs_unaligned). Applies to both reads and
465 writes, however a second range can be given after a comma.
468 bssplit=str Sometimes you want even finer grained control of the
469 block sizes issued, not just an even split between them.
470 This option allows you to weight various block sizes,
471 so that you are able to define a specific amount of
472 block sizes issued. The format for this option is:
474 bssplit=blocksize/percentage:blocksize/percentage
476 for as many block sizes as needed. So if you want to define
477 a workload that has 50% 64k blocks, 10% 4k blocks, and
478 40% 32k blocks, you would write:
480 bssplit=4k/10:64k/50:32k/40
482 Ordering does not matter. If the percentage is left blank,
483 fio will fill in the remaining values evenly. So a bssplit
484 option like this one:
486 bssplit=4k/50:1k/:32k/
488 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
489 always add up to 100, if bssplit is given a range that adds
490 up to more, it will error out.
492 bssplit also supports giving separate splits to reads and
493 writes. The format is identical to what bs= accepts. You
494 have to separate the read and write parts with a comma. So
495 if you want a workload that has 50% 2k reads and 50% 4k reads,
496 while having 90% 4k writes and 10% 8k writes, you would
499 bssplit=2k/50:4k/50,4k/90,8k/10
502 bs_unaligned If this option is given, any byte size value within bsrange
503 may be used as a block range. This typically wont work with
504 direct IO, as that normally requires sector alignment.
506 zero_buffers If this option is given, fio will init the IO buffers to
507 all zeroes. The default is to fill them with random data.
509 refill_buffers If this option is given, fio will refill the IO buffers
510 on every submit. The default is to only fill it at init
511 time and reuse that data. Only makes sense if zero_buffers
512 isn't specified, naturally. If data verification is enabled,
513 refill_buffers is also automatically enabled.
515 scramble_buffers=bool If refill_buffers is too costly and the target is
516 using data deduplication, then setting this option will
517 slightly modify the IO buffer contents to defeat normal
518 de-dupe attempts. This is not enough to defeat more clever
519 block compression attempts, but it will stop naive dedupe of
520 blocks. Default: true.
522 buffer_compress_percentage=int If this is set, then fio will attempt to
523 provide IO buffer content (on WRITEs) that compress to
524 the specified level. Fio does this by providing a mix of
525 random data and zeroes. Note that this is per block size
526 unit, for file/disk wide compression level that matches
527 this setting, you'll also want to set refill_buffers.
529 buffer_compress_chunk=int See buffer_compress_percentage. This
530 setting allows fio to manage how big the ranges of random
531 data and zeroed data is. Without this set, fio will
532 provide buffer_compress_percentage of blocksize random
533 data, followed by the remaining zeroed. With this set
534 to some chunk size smaller than the block size, fio can
535 alternate random and zeroed data throughout the IO
538 nrfiles=int Number of files to use for this job. Defaults to 1.
540 openfiles=int Number of files to keep open at the same time. Defaults to
541 the same as nrfiles, can be set smaller to limit the number
544 file_service_type=str Defines how fio decides which file from a job to
545 service next. The following types are defined:
547 random Just choose a file at random.
549 roundrobin Round robin over open files. This
552 sequential Finish one file before moving on to
553 the next. Multiple files can still be
554 open depending on 'openfiles'.
556 The string can have a number appended, indicating how
557 often to switch to a new file. So if option random:4 is
558 given, fio will switch to a new random file after 4 ios
561 ioengine=str Defines how the job issues io to the file. The following
564 sync Basic read(2) or write(2) io. lseek(2) is
565 used to position the io location.
567 psync Basic pread(2) or pwrite(2) io.
569 vsync Basic readv(2) or writev(2) IO.
571 libaio Linux native asynchronous io. Note that Linux
572 may only support queued behaviour with
573 non-buffered IO (set direct=1 or buffered=0).
574 This engine defines engine specific options.
576 posixaio glibc posix asynchronous io.
578 solarisaio Solaris native asynchronous io.
580 windowsaio Windows native asynchronous io.
582 mmap File is memory mapped and data copied
583 to/from using memcpy(3).
585 splice splice(2) is used to transfer the data and
586 vmsplice(2) to transfer data from user
589 syslet-rw Use the syslet system calls to make
590 regular read/write async.
592 sg SCSI generic sg v3 io. May either be
593 synchronous using the SG_IO ioctl, or if
594 the target is an sg character device
595 we use read(2) and write(2) for asynchronous
598 null Doesn't transfer any data, just pretends
599 to. This is mainly used to exercise fio
600 itself and for debugging/testing purposes.
602 net Transfer over the network to given host:port.
603 Depending on the protocol used, the hostname,
604 port, listen and filename options are used to
605 specify what sort of connection to make, while
606 the protocol option determines which protocol
608 This engine defines engine specific options.
610 netsplice Like net, but uses splice/vmsplice to
611 map data and send/receive.
612 This engine defines engine specific options.
614 cpuio Doesn't transfer any data, but burns CPU
615 cycles according to the cpuload= and
616 cpucycle= options. Setting cpuload=85
617 will cause that job to do nothing but burn
618 85% of the CPU. In case of SMP machines,
619 use numjobs=<no_of_cpu> to get desired CPU
620 usage, as the cpuload only loads a single
621 CPU at the desired rate.
623 guasi The GUASI IO engine is the Generic Userspace
624 Asyncronous Syscall Interface approach
627 http://www.xmailserver.org/guasi-lib.html
629 for more info on GUASI.
631 rdma The RDMA I/O engine supports both RDMA
632 memory semantics (RDMA_WRITE/RDMA_READ) and
633 channel semantics (Send/Recv) for the
634 InfiniBand, RoCE and iWARP protocols.
636 falloc IO engine that does regular fallocate to
637 simulate data transfer as fio ioengine.
638 DDIR_READ does fallocate(,mode = keep_size,)
639 DDIR_WRITE does fallocate(,mode = 0)
640 DDIR_TRIM does fallocate(,mode = punch_hole)
642 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
643 ioctls to simulate defragment activity in
644 request to DDIR_WRITE event
646 external Prefix to specify loading an external
647 IO engine object file. Append the engine
648 filename, eg ioengine=external:/tmp/foo.o
649 to load ioengine foo.o in /tmp.
651 iodepth=int This defines how many io units to keep in flight against
652 the file. The default is 1 for each file defined in this
653 job, can be overridden with a larger value for higher
654 concurrency. Note that increasing iodepth beyond 1 will not
655 affect synchronous ioengines (except for small degress when
656 verify_async is in use). Even async engines may impose OS
657 restrictions causing the desired depth not to be achieved.
658 This may happen on Linux when using libaio and not setting
659 direct=1, since buffered IO is not async on that OS. Keep an
660 eye on the IO depth distribution in the fio output to verify
661 that the achieved depth is as expected. Default: 1.
663 iodepth_batch_submit=int
664 iodepth_batch=int This defines how many pieces of IO to submit at once.
665 It defaults to 1 which means that we submit each IO
666 as soon as it is available, but can be raised to submit
667 bigger batches of IO at the time.
669 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
670 at once. It defaults to 1 which means that we'll ask
671 for a minimum of 1 IO in the retrieval process from
672 the kernel. The IO retrieval will go on until we
673 hit the limit set by iodepth_low. If this variable is
674 set to 0, then fio will always check for completed
675 events before queuing more IO. This helps reduce
676 IO latency, at the cost of more retrieval system calls.
678 iodepth_low=int The low water mark indicating when to start filling
679 the queue again. Defaults to the same as iodepth, meaning
680 that fio will attempt to keep the queue full at all times.
681 If iodepth is set to eg 16 and iodepth_low is set to 4, then
682 after fio has filled the queue of 16 requests, it will let
683 the depth drain down to 4 before starting to fill it again.
685 direct=bool If value is true, use non-buffered io. This is usually
686 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
687 On Windows the synchronous ioengines don't support direct io.
689 buffered=bool If value is true, use buffered io. This is the opposite
690 of the 'direct' option. Defaults to true.
692 offset=int Start io at the given offset in the file. The data before
693 the given offset will not be touched. This effectively
694 caps the file size at real_size - offset.
696 offset_increment=int If this is provided, then the real offset becomes
697 the offset + offset_increment * thread_number, where the
698 thread number is a counter that starts at 0 and is incremented
699 for each job. This option is useful if there are several jobs
700 which are intended to operate on a file in parallel in disjoint
701 segments, with even spacing between the starting points.
703 fsync=int If writing to a file, issue a sync of the dirty data
704 for every number of blocks given. For example, if you give
705 32 as a parameter, fio will sync the file for every 32
706 writes issued. If fio is using non-buffered io, we may
707 not sync the file. The exception is the sg io engine, which
708 synchronizes the disk cache anyway.
710 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
712 In FreeBSD and Windows there is no fdatasync(), this falls back to
715 sync_file_range=str:val Use sync_file_range() for every 'val' number of
716 write operations. Fio will track range of writes that
717 have happened since the last sync_file_range() call. 'str'
718 can currently be one or more of:
720 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
721 write SYNC_FILE_RANGE_WRITE
722 wait_after SYNC_FILE_RANGE_WAIT_AFTER
724 So if you do sync_file_range=wait_before,write:8, fio would
725 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
726 every 8 writes. Also see the sync_file_range(2) man page.
727 This option is Linux specific.
729 overwrite=bool If true, writes to a file will always overwrite existing
730 data. If the file doesn't already exist, it will be
731 created before the write phase begins. If the file exists
732 and is large enough for the specified write phase, nothing
735 end_fsync=bool If true, fsync file contents when a write stage has completed.
737 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
738 This differs from end_fsync in that it will happen on every
739 file close, not just at the end of the job.
741 rwmixread=int How large a percentage of the mix should be reads.
743 rwmixwrite=int How large a percentage of the mix should be writes. If both
744 rwmixread and rwmixwrite is given and the values do not add
745 up to 100%, the latter of the two will be used to override
746 the first. This may interfere with a given rate setting,
747 if fio is asked to limit reads or writes to a certain rate.
748 If that is the case, then the distribution may be skewed.
750 random_distribution=str:float By default, fio will use a completely uniform
751 random distribution when asked to perform random IO. Sometimes
752 it is useful to skew the distribution in specific ways,
753 ensuring that some parts of the data is more hot than others.
754 fio includes the following distribution models:
756 random Uniform random distribution
757 zipf Zipf distribution
758 pareto Pareto distribution
760 When using a zipf or pareto distribution, an input value
761 is also needed to define the access pattern. For zipf, this
762 is the zipf theta. For pareto, it's the pareto power. Fio
763 includes a test program, genzipf, that can be used visualize
764 what the given input values will yield in terms of hit rates.
765 If you wanted to use zipf with a theta of 1.2, you would use
766 random_distribution=zipf:1.2 as the option. If a non-uniform
767 model is used, fio will disable use of the random map.
769 percentage_random=int For a random workload, set how big a percentage should
770 be random. This defaults to 100%, in which case the workload
771 is fully random. It can be set from anywhere from 0 to 100.
772 Setting it to 0 would make the workload fully sequential. Any
773 setting in between will result in a random mix of sequential
774 and random IO, at the given percentages.
776 percentage_sequential=int See percentage_random. It is guaranteed that
777 they add up to 100. The later setting has priority, each
778 will adjust the other.
780 norandommap Normally fio will cover every block of the file when doing
781 random IO. If this option is given, fio will just get a
782 new random offset without looking at past io history. This
783 means that some blocks may not be read or written, and that
784 some blocks may be read/written more than once. This option
785 is mutually exclusive with verify= if and only if multiple
786 blocksizes (via bsrange=) are used, since fio only tracks
787 complete rewrites of blocks.
789 softrandommap=bool See norandommap. If fio runs with the random block map
790 enabled and it fails to allocate the map, if this option is
791 set it will continue without a random block map. As coverage
792 will not be as complete as with random maps, this option is
795 random_generator=str Fio supports the following engines for generating
796 IO offsets for random IO:
798 tausworthe Strong 2^88 cycle random number generator
799 lfsr Linear feedback shift register generator
801 Tausworthe is a strong random number generator, but it
802 requires tracking on the side if we want to ensure that
803 blocks are only read or written once. LFSR guarantees
804 that we never generate the same offset twice, and it's
805 also less computationally expensive. It's not a true
806 random generator, however, though for IO purposes it's
807 typically good enough. LFSR only works with single
808 block sizes, not with workloads that use multiple block
809 sizes. If used with such a workload, fio may read or write
810 some blocks multiple times.
812 nice=int Run the job with the given nice value. See man nice(2).
814 prio=int Set the io priority value of this job. Linux limits us to
815 a positive value between 0 and 7, with 0 being the highest.
818 prioclass=int Set the io priority class. See man ionice(1).
820 thinktime=int Stall the job x microseconds after an io has completed before
821 issuing the next. May be used to simulate processing being
822 done by an application. See thinktime_blocks and
826 Only valid if thinktime is set - pretend to spend CPU time
827 doing something with the data received, before falling back
828 to sleeping for the rest of the period specified by
832 Only valid if thinktime is set - control how many blocks
833 to issue, before waiting 'thinktime' usecs. If not set,
834 defaults to 1 which will make fio wait 'thinktime' usecs
837 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
838 the normal suffix rules apply. You can use rate=500k to limit
839 reads and writes to 500k each, or you can specify read and
840 writes separately. Using rate=1m,500k would limit reads to
841 1MB/sec and writes to 500KB/sec. Capping only reads or
842 writes can be done with rate=,500k or rate=500k,. The former
843 will only limit writes (to 500KB/sec), the latter will only
846 ratemin=int Tell fio to do whatever it can to maintain at least this
847 bandwidth. Failing to meet this requirement, will cause
848 the job to exit. The same format as rate is used for
849 read vs write separation.
851 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
852 as rate, just specified independently of bandwidth. If the
853 job is given a block size range instead of a fixed value,
854 the smallest block size is used as the metric. The same format
855 as rate is used for read vs write seperation.
857 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
858 the job to exit. The same format as rate is used for read vs
861 max_latency=int If set, fio will exit the job if it exceeds this maximum
862 latency. It will exit with an ETIME error.
864 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
867 cpumask=int Set the CPU affinity of this job. The parameter given is a
868 bitmask of allowed CPU's the job may run on. So if you want
869 the allowed CPUs to be 1 and 5, you would pass the decimal
870 value of (1 << 1 | 1 << 5), or 34. See man
871 sched_setaffinity(2). This may not work on all supported
872 operating systems or kernel versions. This option doesn't
873 work well for a higher CPU count than what you can store in
874 an integer mask, so it can only control cpus 1-32. For
875 boxes with larger CPU counts, use cpus_allowed.
877 cpus_allowed=str Controls the same options as cpumask, but it allows a text
878 setting of the permitted CPUs instead. So to use CPUs 1 and
879 5, you would specify cpus_allowed=1,5. This options also
880 allows a range of CPUs. Say you wanted a binding to CPUs
881 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
883 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
884 arguments allow comma delimited list of cpu numbers,
885 A-B ranges, or 'all'. Note, to enable numa options support,
886 fio must be built on a system with libnuma-dev(el) installed.
888 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
889 nodes. Format of the argements:
891 `mode' is one of the following memory policy:
892 default, prefer, bind, interleave, local
893 For `default' and `local' memory policy, no node is
894 needed to be specified.
895 For `prefer', only one node is allowed.
896 For `bind' and `interleave', it allow comma delimited
897 list of numbers, A-B ranges, or 'all'.
899 startdelay=time Start this job the specified number of seconds after fio
900 has started. Only useful if the job file contains several
901 jobs, and you want to delay starting some jobs to a certain
904 runtime=time Tell fio to terminate processing after the specified number
905 of seconds. It can be quite hard to determine for how long
906 a specified job will run, so this parameter is handy to
907 cap the total runtime to a given time.
909 time_based If set, fio will run for the duration of the runtime
910 specified even if the file(s) are completely read or
911 written. It will simply loop over the same workload
912 as many times as the runtime allows.
914 ramp_time=time If set, fio will run the specified workload for this amount
915 of time before logging any performance numbers. Useful for
916 letting performance settle before logging results, thus
917 minimizing the runtime required for stable results. Note
918 that the ramp_time is considered lead in time for a job,
919 thus it will increase the total runtime if a special timeout
920 or runtime is specified.
922 invalidate=bool Invalidate the buffer/page cache parts for this file prior
923 to starting io. Defaults to true.
925 sync=bool Use sync io for buffered writes. For the majority of the
926 io engines, this means using O_SYNC.
929 mem=str Fio can use various types of memory as the io unit buffer.
930 The allowed values are:
932 malloc Use memory from malloc(3) as the buffers.
934 shm Use shared memory as the buffers. Allocated
937 shmhuge Same as shm, but use huge pages as backing.
939 mmap Use mmap to allocate buffers. May either be
940 anonymous memory, or can be file backed if
941 a filename is given after the option. The
942 format is mem=mmap:/path/to/file.
944 mmaphuge Use a memory mapped huge file as the buffer
945 backing. Append filename after mmaphuge, ala
946 mem=mmaphuge:/hugetlbfs/file
948 The area allocated is a function of the maximum allowed
949 bs size for the job, multiplied by the io depth given. Note
950 that for shmhuge and mmaphuge to work, the system must have
951 free huge pages allocated. This can normally be checked
952 and set by reading/writing /proc/sys/vm/nr_hugepages on a
953 Linux system. Fio assumes a huge page is 4MB in size. So
954 to calculate the number of huge pages you need for a given
955 job file, add up the io depth of all jobs (normally one unless
956 iodepth= is used) and multiply by the maximum bs set. Then
957 divide that number by the huge page size. You can see the
958 size of the huge pages in /proc/meminfo. If no huge pages
959 are allocated by having a non-zero number in nr_hugepages,
960 using mmaphuge or shmhuge will fail. Also see hugepage-size.
962 mmaphuge also needs to have hugetlbfs mounted and the file
963 location should point there. So if it's mounted in /huge,
964 you would use mem=mmaphuge:/huge/somefile.
966 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
967 Note that the given alignment is applied to the first IO unit
968 buffer, if using iodepth the alignment of the following buffers
969 are given by the bs used. In other words, if using a bs that is
970 a multiple of the page sized in the system, all buffers will
971 be aligned to this value. If using a bs that is not page
972 aligned, the alignment of subsequent IO memory buffers is the
973 sum of the iomem_align and bs used.
976 Defines the size of a huge page. Must at least be equal
977 to the system setting, see /proc/meminfo. Defaults to 4MB.
978 Should probably always be a multiple of megabytes, so using
979 hugepage-size=Xm is the preferred way to set this to avoid
980 setting a non-pow-2 bad value.
982 exitall When one job finishes, terminate the rest. The default is
983 to wait for each job to finish, sometimes that is not the
986 bwavgtime=int Average the calculated bandwidth over the given time. Value
987 is specified in milliseconds.
989 iopsavgtime=int Average the calculated IOPS over the given time. Value
990 is specified in milliseconds.
992 create_serialize=bool If true, serialize the file creating for the jobs.
993 This may be handy to avoid interleaving of data
994 files, which may greatly depend on the filesystem
995 used and even the number of processors in the system.
997 create_fsync=bool fsync the data file after creation. This is the
1000 create_on_open=bool Don't pre-setup the files for IO, just create open()
1001 when it's time to do IO to that file.
1003 create_only=bool If true, fio will only run the setup phase of the job.
1004 If files need to be laid out or updated on disk, only
1005 that will be done. The actual job contents are not
1008 pre_read=bool If this is given, files will be pre-read into memory before
1009 starting the given IO operation. This will also clear
1010 the 'invalidate' flag, since it is pointless to pre-read
1011 and then drop the cache. This will only work for IO engines
1012 that are seekable, since they allow you to read the same data
1013 multiple times. Thus it will not work on eg network or splice
1016 unlink=bool Unlink the job files when done. Not the default, as repeated
1017 runs of that job would then waste time recreating the file
1018 set again and again.
1020 loops=int Run the specified number of iterations of this job. Used
1021 to repeat the same workload a given number of times. Defaults
1024 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1025 verify is set. Defaults to 1.
1027 verify=str If writing to a file, fio can verify the file contents
1028 after each iteration of the job. The allowed values are:
1030 md5 Use an md5 sum of the data area and store
1031 it in the header of each block.
1033 crc64 Use an experimental crc64 sum of the data
1034 area and store it in the header of each
1037 crc32c Use a crc32c sum of the data area and store
1038 it in the header of each block.
1040 crc32c-intel Use hardware assisted crc32c calcuation
1041 provided on SSE4.2 enabled processors. Falls
1042 back to regular software crc32c, if not
1043 supported by the system.
1045 crc32 Use a crc32 sum of the data area and store
1046 it in the header of each block.
1048 crc16 Use a crc16 sum of the data area and store
1049 it in the header of each block.
1051 crc7 Use a crc7 sum of the data area and store
1052 it in the header of each block.
1054 sha512 Use sha512 as the checksum function.
1056 sha256 Use sha256 as the checksum function.
1058 sha1 Use optimized sha1 as the checksum function.
1060 meta Write extra information about each io
1061 (timestamp, block number etc.). The block
1062 number is verified. See also verify_pattern.
1064 null Only pretend to verify. Useful for testing
1065 internals with ioengine=null, not for much
1068 This option can be used for repeated burn-in tests of a
1069 system to make sure that the written data is also
1070 correctly read back. If the data direction given is
1071 a read or random read, fio will assume that it should
1072 verify a previously written file. If the data direction
1073 includes any form of write, the verify will be of the
1076 verifysort=bool If set, fio will sort written verify blocks when it deems
1077 it faster to read them back in a sorted manner. This is
1078 often the case when overwriting an existing file, since
1079 the blocks are already laid out in the file system. You
1080 can ignore this option unless doing huge amounts of really
1081 fast IO where the red-black tree sorting CPU time becomes
1084 verify_offset=int Swap the verification header with data somewhere else
1085 in the block before writing. Its swapped back before
1088 verify_interval=int Write the verification header at a finer granularity
1089 than the blocksize. It will be written for chunks the
1090 size of header_interval. blocksize should divide this
1093 verify_pattern=str If set, fio will fill the io buffers with this
1094 pattern. Fio defaults to filling with totally random
1095 bytes, but sometimes it's interesting to fill with a known
1096 pattern for io verification purposes. Depending on the
1097 width of the pattern, fio will fill 1/2/3/4 bytes of the
1098 buffer at the time(it can be either a decimal or a hex number).
1099 The verify_pattern if larger than a 32-bit quantity has to
1100 be a hex number that starts with either "0x" or "0X". Use
1103 verify_fatal=bool Normally fio will keep checking the entire contents
1104 before quitting on a block verification failure. If this
1105 option is set, fio will exit the job on the first observed
1108 verify_dump=bool If set, dump the contents of both the original data
1109 block and the data block we read off disk to files. This
1110 allows later analysis to inspect just what kind of data
1111 corruption occurred. Off by default.
1113 verify_async=int Fio will normally verify IO inline from the submitting
1114 thread. This option takes an integer describing how many
1115 async offload threads to create for IO verification instead,
1116 causing fio to offload the duty of verifying IO contents
1117 to one or more separate threads. If using this offload
1118 option, even sync IO engines can benefit from using an
1119 iodepth setting higher than 1, as it allows them to have
1120 IO in flight while verifies are running.
1122 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1123 async IO verification threads. See cpus_allowed for the
1126 verify_backlog=int Fio will normally verify the written contents of a
1127 job that utilizes verify once that job has completed. In
1128 other words, everything is written then everything is read
1129 back and verified. You may want to verify continually
1130 instead for a variety of reasons. Fio stores the meta data
1131 associated with an IO block in memory, so for large
1132 verify workloads, quite a bit of memory would be used up
1133 holding this meta data. If this option is enabled, fio
1134 will write only N blocks before verifying these blocks.
1136 will verify the previously written blocks before continuing
1139 verify_backlog_batch=int Control how many blocks fio will verify
1140 if verify_backlog is set. If not set, will default to
1141 the value of verify_backlog (meaning the entire queue
1142 is read back and verified). If verify_backlog_batch is
1143 less than verify_backlog then not all blocks will be verified,
1144 if verify_backlog_batch is larger than verify_backlog, some
1145 blocks will be verified more than once.
1148 wait_for_previous Wait for preceeding jobs in the job file to exit, before
1149 starting this one. Can be used to insert serialization
1150 points in the job file. A stone wall also implies starting
1151 a new reporting group.
1153 new_group Start a new reporting group. See: group_reporting.
1155 numjobs=int Create the specified number of clones of this job. May be
1156 used to setup a larger number of threads/processes doing
1157 the same thing. Each thread is reported separately; to see
1158 statistics for all clones as a whole, use group_reporting in
1159 conjunction with new_group.
1161 group_reporting It may sometimes be interesting to display statistics for
1162 groups of jobs as a whole instead of for each individual job.
1163 This is especially true if 'numjobs' is used; looking at
1164 individual thread/process output quickly becomes unwieldy.
1165 To see the final report per-group instead of per-job, use
1166 'group_reporting'. Jobs in a file will be part of the same
1167 reporting group, unless if separated by a stonewall, or by
1170 thread fio defaults to forking jobs, however if this option is
1171 given, fio will use pthread_create(3) to create threads
1174 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1176 zoneskip=int Skip the specified number of bytes when zonesize data has
1177 been read. The two zone options can be used to only do
1178 io on zones of a file.
1180 write_iolog=str Write the issued io patterns to the specified file. See
1181 read_iolog. Specify a separate file for each job, otherwise
1182 the iologs will be interspersed and the file may be corrupt.
1184 read_iolog=str Open an iolog with the specified file name and replay the
1185 io patterns it contains. This can be used to store a
1186 workload and replay it sometime later. The iolog given
1187 may also be a blktrace binary file, which allows fio
1188 to replay a workload captured by blktrace. See blktrace
1189 for how to capture such logging data. For blktrace replay,
1190 the file needs to be turned into a blkparse binary data
1191 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1193 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1194 is to attempt to respect the time stamps within the log and
1195 replay them with the appropriate delay between IOPS. By
1196 setting this variable fio will not respect the timestamps and
1197 attempt to replay them as fast as possible while still
1198 respecting ordering. The result is the same I/O pattern to a
1199 given device, but different timings.
1201 replay_redirect=str While replaying I/O patterns using read_iolog the
1202 default behavior is to replay the IOPS onto the major/minor
1203 device that each IOP was recorded from. This is sometimes
1204 undesireable because on a different machine those major/minor
1205 numbers can map to a different device. Changing hardware on
1206 the same system can also result in a different major/minor
1207 mapping. Replay_redirect causes all IOPS to be replayed onto
1208 the single specified device regardless of the device it was
1209 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1210 IO in the blktrace to be replayed onto /dev/sdc. This means
1211 multiple devices will be replayed onto a single, if the trace
1212 contains multiple devices. If you want multiple devices to be
1213 replayed concurrently to multiple redirected devices you must
1214 blkparse your trace into separate traces and replay them with
1215 independent fio invocations. Unfortuantely this also breaks
1216 the strict time ordering between multiple device accesses.
1218 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1219 file. Can be used to store data of the bandwidth of the
1220 jobs in their lifetime. The included fio_generate_plots
1221 script uses gnuplot to turn these text files into nice
1222 graphs. See write_lat_log for behaviour of given
1223 filename. For this option, the suffix is _bw.log.
1225 write_lat_log=str Same as write_bw_log, except that this option stores io
1226 submission, completion, and total latencies instead. If no
1227 filename is given with this option, the default filename of
1228 "jobname_type.log" is used. Even if the filename is given,
1229 fio will still append the type of log. So if one specifies
1233 The actual log names will be foo_slat.log, foo_slat.log,
1234 and foo_lat.log. This helps fio_generate_plot fine the logs
1237 write_bw_log=str If given, write an IOPS log of the jobs in this job
1238 file. See write_bw_log.
1240 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1241 given with this option, the default filename of
1242 "jobname_type.log" is used. Even if the filename is given,
1243 fio will still append the type of log.
1245 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1246 or bw log for every IO that completes. When writing to the
1247 disk log, that can quickly grow to a very large size. Setting
1248 this option makes fio average the each log entry over the
1249 specified period of time, reducing the resolution of the log.
1252 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1253 potentially be used instead of removing memory or booting
1254 with less memory to simulate a smaller amount of memory.
1255 The amount specified is per worker.
1257 exec_prerun=str Before running this job, issue the command specified
1260 exec_postrun=str After the job completes, issue the command specified
1263 ioscheduler=str Attempt to switch the device hosting the file to the specified
1264 io scheduler before running.
1266 disk_util=bool Generate disk utilization statistics, if the platform
1267 supports it. Defaults to on.
1269 disable_lat=bool Disable measurements of total latency numbers. Useful
1270 only for cutting back the number of calls to gettimeofday,
1271 as that does impact performance at really high IOPS rates.
1272 Note that to really get rid of a large amount of these
1273 calls, this option must be used with disable_slat and
1276 disable_clat=bool Disable measurements of completion latency numbers. See
1279 disable_slat=bool Disable measurements of submission latency numbers. See
1282 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1285 clat_percentiles=bool Enable the reporting of percentiles of
1286 completion latencies.
1288 percentile_list=float_list Overwrite the default list of percentiles
1289 for completion latencies. Each number is a floating
1290 number in the range (0,100], and the maximum length of
1291 the list is 20. Use ':' to separate the numbers, and
1292 list the numbers in ascending order. For example,
1293 --percentile_list=99.5:99.9 will cause fio to report
1294 the values of completion latency below which 99.5% and
1295 99.9% of the observed latencies fell, respectively.
1297 clocksource=str Use the given clocksource as the base of timing. The
1298 supported options are:
1300 gettimeofday gettimeofday(2)
1302 clock_gettime clock_gettime(2)
1304 cpu Internal CPU clock source
1306 cpu is the preferred clocksource if it is reliable, as it
1307 is very fast (and fio is heavy on time calls). Fio will
1308 automatically use this clocksource if it's supported and
1309 considered reliable on the system it is running on, unless
1310 another clocksource is specifically set. For x86/x86-64 CPUs,
1311 this means supporting TSC Invariant.
1313 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1314 (disable_clat, disable_slat, disable_bw) plus reduce
1315 precision of the timeout somewhat to really shrink
1316 the gettimeofday() call count. With this option enabled,
1317 we only do about 0.4% of the gtod() calls we would have
1318 done if all time keeping was enabled.
1320 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1321 execution to just getting the current time. Fio (and
1322 databases, for instance) are very intensive on gettimeofday()
1323 calls. With this option, you can set one CPU aside for
1324 doing nothing but logging current time to a shared memory
1325 location. Then the other threads/processes that run IO
1326 workloads need only copy that segment, instead of entering
1327 the kernel with a gettimeofday() call. The CPU set aside
1328 for doing these time calls will be excluded from other
1329 uses. Fio will manually clear it from the CPU mask of other
1332 continue_on_error=str Normally fio will exit the job on the first observed
1333 failure. If this option is set, fio will continue the job when
1334 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1335 is exceeded or the I/O size specified is completed. If this
1336 option is used, there are two more stats that are appended,
1337 the total error count and the first error. The error field
1338 given in the stats is the first error that was hit during the
1341 The allowed values are:
1343 none Exit on any IO or verify errors.
1345 read Continue on read errors, exit on all others.
1347 write Continue on write errors, exit on all others.
1349 io Continue on any IO error, exit on all others.
1351 verify Continue on verify errors, exit on all others.
1353 all Continue on all errors.
1355 0 Backward-compatible alias for 'none'.
1357 1 Backward-compatible alias for 'all'.
1359 ignore_error=str Sometimes you want to ignore some errors during test
1360 in that case you can specify error list for each error type.
1361 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1362 errors for given error type is separated with ':'. Error
1363 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1365 ignore_error=EAGAIN,ENOSPC:122
1366 This option will ignore EAGAIN from READ, and ENOSPC and
1367 122(EDQUOT) from WRITE.
1369 error_dump=bool If set dump every error even if it is non fatal, true
1370 by default. If disabled only fatal error will be dumped
1372 cgroup=str Add job to this control group. If it doesn't exist, it will
1373 be created. The system must have a mounted cgroup blkio
1374 mount point for this to work. If your system doesn't have it
1375 mounted, you can do so with:
1377 # mount -t cgroup -o blkio none /cgroup
1379 cgroup_weight=int Set the weight of the cgroup to this value. See
1380 the documentation that comes with the kernel, allowed values
1381 are in the range of 100..1000.
1383 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1384 the job completion. To override this behavior and to leave
1385 cgroups around after the job completion, set cgroup_nodelete=1.
1386 This can be useful if one wants to inspect various cgroup
1387 files after job completion. Default: false
1389 uid=int Instead of running as the invoking user, set the user ID to
1390 this value before the thread/process does any work.
1392 gid=int Set group ID, see uid.
1394 flow_id=int The ID of the flow. If not specified, it defaults to being a
1395 global flow. See flow.
1397 flow=int Weight in token-based flow control. If this value is used, then
1398 there is a 'flow counter' which is used to regulate the
1399 proportion of activity between two or more jobs. fio attempts
1400 to keep this flow counter near zero. The 'flow' parameter
1401 stands for how much should be added or subtracted to the flow
1402 counter on each iteration of the main I/O loop. That is, if
1403 one job has flow=8 and another job has flow=-1, then there
1404 will be a roughly 1:8 ratio in how much one runs vs the other.
1406 flow_watermark=int The maximum value that the absolute value of the flow
1407 counter is allowed to reach before the job must wait for a
1408 lower value of the counter.
1410 flow_sleep=int The period of time, in microseconds, to wait after the flow
1411 watermark has been exceeded before retrying operations
1413 In addition, there are some parameters which are only valid when a specific
1414 ioengine is in use. These are used identically to normal parameters, with the
1415 caveat that when used on the command line, they must come after the ioengine
1416 that defines them is selected.
1418 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1419 the io_getevents system call to reap newly returned events.
1420 With this flag turned on, the AIO ring will be read directly
1421 from user-space to reap events. The reaping mode is only
1422 enabled when polling for a minimum of 0 events (eg when
1423 iodepth_batch_complete=0).
1425 [cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1427 [cpu] cpuchunks=int Split the load into cycles of the given time. In
1430 [netsplice] hostname=str
1431 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1432 If the job is a TCP listener or UDP reader, the hostname is not
1433 used and must be omitted.
1435 [netsplice] port=int
1436 [net] port=int The TCP or UDP port to bind to or connect to.
1438 [netsplice] nodelay=bool
1439 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1441 [netsplice] protocol=str
1442 [netsplice] proto=str
1444 [net] proto=str The network protocol to use. Accepted values are:
1446 tcp Transmission control protocol
1447 udp User datagram protocol
1448 unix UNIX domain socket
1450 When the protocol is TCP or UDP, the port must also be given,
1451 as well as the hostname if the job is a TCP listener or UDP
1452 reader. For unix sockets, the normal filename option should be
1453 used and the port is invalid.
1455 [net] listen For TCP network connections, tell fio to listen for incoming
1456 connections rather than initiating an outgoing connection. The
1457 hostname must be omitted if this option is used.
1458 [net] pingpong Normal a network writer will just continue writing data, and
1459 a network reader will just consume packages. If pingpong=1
1460 is set, a writer will send its normal payload to the reader,
1461 then wait for the reader to send the same payload back. This
1462 allows fio to measure network latencies. The submission
1463 and completion latencies then measure local time spent
1464 sending or receiving, and the completion latency measures
1465 how long it took for the other end to receive and send back.
1467 [e4defrag] donorname=str
1468 File will be used as a block donor(swap extents between files)
1469 [e4defrag] inplace=int
1470 Configure donor file blocks allocation strategy
1471 0(default): Preallocate donor's file on init
1472 1 : allocate space immidietly inside defragment event,
1473 and free right after event
1477 6.0 Interpreting the output
1478 ---------------------------
1480 fio spits out a lot of output. While running, fio will display the
1481 status of the jobs created. An example of that would be:
1483 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1485 The characters inside the square brackets denote the current status of
1486 each thread. The possible values (in typical life cycle order) are:
1490 P Thread setup, but not started.
1492 I Thread initialized, waiting or generating necessary data.
1493 p Thread running pre-reading file(s).
1494 R Running, doing sequential reads.
1495 r Running, doing random reads.
1496 W Running, doing sequential writes.
1497 w Running, doing random writes.
1498 M Running, doing mixed sequential reads/writes.
1499 m Running, doing mixed random reads/writes.
1500 F Running, currently waiting for fsync()
1501 V Running, doing verification of written data.
1502 E Thread exited, not reaped by main thread yet.
1504 X Thread reaped, exited with an error.
1505 K Thread reaped, exited due to signal.
1507 The other values are fairly self explanatory - number of threads
1508 currently running and doing io, rate of io since last check (read speed
1509 listed first, then write speed), and the estimated completion percentage
1510 and time for the running group. It's impossible to estimate runtime of
1511 the following groups (if any). Note that the string is displayed in order,
1512 so it's possible to tell which of the jobs are currently doing what. The
1513 first character is the first job defined in the job file, and so forth.
1515 When fio is done (or interrupted by ctrl-c), it will show the data for
1516 each thread, group of threads, and disks in that order. For each data
1517 direction, the output looks like:
1519 Client1 (g=0): err= 0:
1520 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1521 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1522 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1523 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1524 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1525 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1526 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1527 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1528 issued r/w: total=0/32768, short=0/0
1529 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1530 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1532 The client number is printed, along with the group id and error of that
1533 thread. Below is the io statistics, here for writes. In the order listed,
1536 io= Number of megabytes io performed
1537 bw= Average bandwidth rate
1538 iops= Average IOs performed per second
1539 runt= The runtime of that thread
1540 slat= Submission latency (avg being the average, stdev being the
1541 standard deviation). This is the time it took to submit
1542 the io. For sync io, the slat is really the completion
1543 latency, since queue/complete is one operation there. This
1544 value can be in milliseconds or microseconds, fio will choose
1545 the most appropriate base and print that. In the example
1546 above, milliseconds is the best scale. Note: in --minimal mode
1547 latencies are always expressed in microseconds.
1548 clat= Completion latency. Same names as slat, this denotes the
1549 time from submission to completion of the io pieces. For
1550 sync io, clat will usually be equal (or very close) to 0,
1551 as the time from submit to complete is basically just
1552 CPU time (io has already been done, see slat explanation).
1553 bw= Bandwidth. Same names as the xlat stats, but also includes
1554 an approximate percentage of total aggregate bandwidth
1555 this thread received in this group. This last value is
1556 only really useful if the threads in this group are on the
1557 same disk, since they are then competing for disk access.
1558 cpu= CPU usage. User and system time, along with the number
1559 of context switches this thread went through, usage of
1560 system and user time, and finally the number of major
1561 and minor page faults.
1562 IO depths= The distribution of io depths over the job life time. The
1563 numbers are divided into powers of 2, so for example the
1564 16= entries includes depths up to that value but higher
1565 than the previous entry. In other words, it covers the
1566 range from 16 to 31.
1567 IO submit= How many pieces of IO were submitting in a single submit
1568 call. Each entry denotes that amount and below, until
1569 the previous entry - eg, 8=100% mean that we submitted
1570 anywhere in between 5-8 ios per submit call.
1571 IO complete= Like the above submit number, but for completions instead.
1572 IO issued= The number of read/write requests issued, and how many
1574 IO latencies= The distribution of IO completion latencies. This is the
1575 time from when IO leaves fio and when it gets completed.
1576 The numbers follow the same pattern as the IO depths,
1577 meaning that 2=1.6% means that 1.6% of the IO completed
1578 within 2 msecs, 20=12.8% means that 12.8% of the IO
1579 took more than 10 msecs, but less than (or equal to) 20 msecs.
1581 After each client has been listed, the group statistics are printed. They
1582 will look like this:
1584 Run status group 0 (all jobs):
1585 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1586 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1588 For each data direction, it prints:
1590 io= Number of megabytes io performed.
1591 aggrb= Aggregate bandwidth of threads in this group.
1592 minb= The minimum average bandwidth a thread saw.
1593 maxb= The maximum average bandwidth a thread saw.
1594 mint= The smallest runtime of the threads in that group.
1595 maxt= The longest runtime of the threads in that group.
1597 And finally, the disk statistics are printed. They will look like this:
1599 Disk stats (read/write):
1600 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1602 Each value is printed for both reads and writes, with reads first. The
1605 ios= Number of ios performed by all groups.
1606 merge= Number of merges io the io scheduler.
1607 ticks= Number of ticks we kept the disk busy.
1608 io_queue= Total time spent in the disk queue.
1609 util= The disk utilization. A value of 100% means we kept the disk
1610 busy constantly, 50% would be a disk idling half of the time.
1612 It is also possible to get fio to dump the current output while it is
1613 running, without terminating the job. To do that, send fio the USR1 signal.
1614 You can also get regularly timed dumps by using the --status-interval
1615 parameter, or by creating a file in /tmp named fio-dump-status. If fio
1616 sees this file, it will unlink it and dump the current output status.
1622 For scripted usage where you typically want to generate tables or graphs
1623 of the results, fio can output the results in a semicolon separated format.
1624 The format is one long line of values, such as:
1626 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%
1627 A description of this job goes here.
1629 The job description (if provided) follows on a second line.
1631 To enable terse output, use the --minimal command line option. The first
1632 value is the version of the terse output format. If the output has to
1633 be changed for some reason, this number will be incremented by 1 to
1634 signify that change.
1636 Split up, the format is as follows:
1638 terse version, fio version, jobname, groupid, error
1640 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1641 Submission latency: min, max, mean, deviation (usec)
1642 Completion latency: min, max, mean, deviation (usec)
1643 Completion latency percentiles: 20 fields (see below)
1644 Total latency: min, max, mean, deviation (usec)
1645 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1647 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1648 Submission latency: min, max, mean, deviation (usec)
1649 Completion latency: min, max, mean, deviation (usec)
1650 Completion latency percentiles: 20 fields (see below)
1651 Total latency: min, max, mean, deviation (usec)
1652 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1653 CPU usage: user, system, context switches, major faults, minor faults
1654 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1655 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1656 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1657 Disk utilization: Disk name, Read ios, write ios,
1658 Read merges, write merges,
1659 Read ticks, write ticks,
1660 Time spent in queue, disk utilization percentage
1661 Additional Info (dependant on continue_on_error, default off): total # errors, first error code
1663 Additional Info (dependant on description being set): Text description
1665 Completion latency percentiles can be a grouping of up to 20 sets, so
1666 for the terse output fio writes all of them. Each field will look like this:
1670 which is the Xth percentile, and the usec latency associated with it.
1672 For disk utilization, all disks used by fio are shown. So for each disk
1673 there will be a disk utilization section.
1676 8.0 Trace file format
1677 ---------------------
1678 There are two trace file format that you can encounter. The older (v1) format
1679 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1680 below in case that you get an old trace and want to understand it.
1682 In any case the trace is a simple text file with a single action per line.
1685 8.1 Trace file format v1
1686 ------------------------
1687 Each line represents a single io action in the following format:
1691 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1693 This format is not supported in Fio versions => 1.20-rc3.
1696 8.2 Trace file format v2
1697 ------------------------
1698 The second version of the trace file format was added in Fio version 1.17.
1699 It allows to access more then one file per trace and has a bigger set of
1700 possible file actions.
1702 The first line of the trace file has to be:
1706 Following this can be lines in two different formats, which are described below.
1708 The file management format:
1712 The filename is given as an absolute path. The action can be one of these:
1714 add Add the given filename to the trace
1715 open Open the file with the given filename. The filename has to have
1716 been added with the add action before.
1717 close Close the file with the given filename. The file has to have been
1721 The file io action format:
1723 filename action offset length
1725 The filename is given as an absolute path, and has to have been added and opened
1726 before it can be used with this format. The offset and length are given in
1727 bytes. The action can be one of these:
1729 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1730 read Read 'length' bytes beginning from 'offset'
1731 write Write 'length' bytes beginning from 'offset'
1732 sync fsync() the file
1733 datasync fdatasync() the file
1734 trim trim the given file from the given 'offset' for 'length' bytes
1737 9.0 CPU idleness profiling
1738 --------------------------
1739 In some cases, we want to understand CPU overhead in a test. For example,
1740 we test patches for the specific goodness of whether they reduce CPU usage.
1741 fio implements a balloon approach to create a thread per CPU that runs at
1742 idle priority, meaning that it only runs when nobody else needs the cpu.
1743 By measuring the amount of work completed by the thread, idleness of each
1744 CPU can be derived accordingly.
1746 An unit work is defined as touching a full page of unsigned characters. Mean
1747 and standard deviation of time to complete an unit work is reported in "unit
1748 work" section. Options can be chosen to report detailed percpu idleness or
1749 overall system idleness by aggregating percpu stats.