8 5. Detailed list of parameters
13 1.0 Overview and history
14 ------------------------
15 fio was originally written to save me the hassle of writing special test
16 case programs when I wanted to test a specific workload, either for
17 performance reasons or to find/reproduce a bug. The process of writing
18 such a test app can be tiresome, especially if you have to do it often.
19 Hence I needed a tool that would be able to simulate a given io workload
20 without resorting to writing a tailored test case again and again.
22 A test work load is difficult to define, though. There can be any number
23 of processes or threads involved, and they can each be using their own
24 way of generating io. You could have someone dirtying large amounts of
25 memory in an memory mapped file, or maybe several threads issuing
26 reads using asynchronous io. fio needed to be flexible enough to
27 simulate both of these cases, and many more.
31 The first step in getting fio to simulate a desired io workload, is
32 writing a job file describing that specific setup. A job file may contain
33 any number of threads and/or files - the typical contents of the job file
34 is a global section defining shared parameters, and one or more job
35 sections describing the jobs involved. When run, fio parses this file
36 and sets everything up as described. If we break down a job from top to
37 bottom, it contains the following basic parameters:
39 IO type Defines the io pattern issued to the file(s).
40 We may only be reading sequentially from this
41 file(s), or we may be writing randomly. Or even
42 mixing reads and writes, sequentially or randomly.
44 Block size In how large chunks are we issuing io? This may be
45 a single value, or it may describe a range of
48 IO size How much data are we going to be reading/writing.
50 IO engine How do we issue io? We could be memory mapping the
51 file, we could be using regular read/write, we
52 could be using splice, async io, syslet, or even
55 IO depth If the io engine is async, how large a queuing
56 depth do we want to maintain?
58 IO type Should we be doing buffered io, or direct/raw io?
60 Num files How many files are we spreading the workload over.
62 Num threads How many threads or processes should we spread
65 The above are the basic parameters defined for a workload, in addition
66 there's a multitude of parameters that modify other aspects of how this
72 See the README file for command line parameters, there are only a few
75 Running fio is normally the easiest part - you just give it the job file
76 (or job files) as parameters:
80 and it will start doing what the job_file tells it to do. You can give
81 more than one job file on the command line, fio will serialize the running
82 of those files. Internally that is the same as using the 'stonewall'
83 parameter described the the parameter section.
85 If the job file contains only one job, you may as well just give the
86 parameters on the command line. The command line parameters are identical
87 to the job parameters, with a few extra that control global parameters
88 (see README). For example, for the job file parameter iodepth=2, the
89 mirror command line option would be --iodepth 2 or --iodepth=2. You can
90 also use the command line for giving more than one job entry. For each
91 --name option that fio sees, it will start a new job with that name.
92 Command line entries following a --name entry will apply to that job,
93 until there are no more entries or a new --name entry is seen. This is
94 similar to the job file options, where each option applies to the current
95 job until a new [] job entry is seen.
97 fio does not need to run as root, except if the files or devices specified
98 in the job section requires that. Some other options may also be restricted,
99 such as memory locking, io scheduler switching, and decreasing the nice value.
104 As previously described, fio accepts one or more job files describing
105 what it is supposed to do. The job file format is the classic ini file,
106 where the names enclosed in [] brackets define the job name. You are free
107 to use any ascii name you want, except 'global' which has special meaning.
108 A global section sets defaults for the jobs described in that file. A job
109 may override a global section parameter, and a job file may even have
110 several global sections if so desired. A job is only affected by a global
111 section residing above it. If the first character in a line is a ';' or a
112 '#', the entire line is discarded as a comment.
114 So let's look at a really simple job file that defines two processes, each
115 randomly reading from a 128MB file.
117 ; -- start job file --
128 As you can see, the job file sections themselves are empty as all the
129 described parameters are shared. As no filename= option is given, fio
130 makes up a filename for each of the jobs as it sees fit. On the command
131 line, this job would look as follows:
133 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
136 Let's look at an example that has a number of processes writing randomly
139 ; -- start job file --
151 Here we have no global section, as we only have one job defined anyway.
152 We want to use async io here, with a depth of 4 for each file. We also
153 increased the buffer size used to 32KB and define numjobs to 4 to
154 fork 4 identical jobs. The result is 4 processes each randomly writing
155 to their own 64MB file. Instead of using the above job file, you could
156 have given the parameters on the command line. For this case, you would
159 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
161 4.1 Environment variables
162 -------------------------
164 fio also supports environment variable expansion in job files. Any
165 substring of the form "${VARNAME}" as part of an option value (in other
166 words, on the right of the `='), will be expanded to the value of the
167 environment variable called VARNAME. If no such environment variable
168 is defined, or VARNAME is the empty string, the empty string will be
171 As an example, let's look at a sample fio invocation and job file:
173 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
175 ; -- start job file --
182 This will expand to the following equivalent job file at runtime:
184 ; -- start job file --
191 fio ships with a few example job files, you can also look there for
194 4.2 Reserved keywords
195 ---------------------
197 Additionally, fio has a set of reserved keywords that will be replaced
198 internally with the appropriate value. Those keywords are:
200 $pagesize The architecture page size of the running system
201 $mb_memory Megabytes of total memory in the system
202 $ncpus Number of online available CPUs
204 These can be used on the command line or in the job file, and will be
205 automatically substituted with the current system values when the job
206 is run. Simple math is also supported on these keywords, so you can
207 perform actions like:
211 and get that properly expanded to 8 times the size of memory in the
215 5.0 Detailed list of parameters
216 -------------------------------
218 This section describes in details each parameter associated with a job.
219 Some parameters take an option of a given type, such as an integer or
220 a string. The following types are used:
222 str String. This is a sequence of alpha characters.
223 time Integer with possible time suffix. In seconds unless otherwise
224 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
226 int SI integer. A whole number value, which may contain a suffix
227 describing the base of the number. Accepted suffixes are k/m/g/t/p,
228 meaning kilo, mega, giga, tera, and peta. The suffix is not case
229 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
230 as 'k'). So if you want to specify 4096, you could either write
231 out '4096' or just give 4k. The suffixes signify base 2 values, so
232 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
233 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
234 case, then 1000 is used as the multiplier. This can be handy for
235 disks, since manufacturers generally use base 10 values when listing
236 the capacity of a drive. If the option accepts an upper and lower
237 range, use a colon ':' or minus '-' to separate such values. May also
238 include a prefix to indicate numbers base. If 0x is used, the number
239 is assumed to be hexadecimal. See irange.
240 bool Boolean. Usually parsed as an integer, however only defined for
241 true and false (1 and 0).
242 irange Integer range with suffix. Allows value range to be given, such
243 as 1024-4096. A colon may also be used as the separator, eg
244 1k:4k. If the option allows two sets of ranges, they can be
245 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
247 float_list A list of floating numbers, separated by a ':' character.
249 With the above in mind, here follows the complete list of fio job
252 name=str ASCII name of the job. This may be used to override the
253 name printed by fio for this job. Otherwise the job
254 name is used. On the command line this parameter has the
255 special purpose of also signaling the start of a new
258 description=str Text description of the job. Doesn't do anything except
259 dump this text description when this job is run. It's
262 directory=str Prefix filenames with this directory. Used to place files
263 in a different location than "./".
265 filename=str Fio normally makes up a filename based on the job name,
266 thread number, and file number. If you want to share
267 files between threads in a job or several jobs, specify
268 a filename for each of them to override the default. If
269 the ioengine used is 'net', the filename is the host, port,
270 and protocol to use in the format of =host,port,protocol.
271 See ioengine=net for more. If the ioengine is file based, you
272 can specify a number of files by separating the names with a
273 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
274 as the two working files, you would use
275 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
276 accessed as \\.\PhysicalDrive0 for the first device,
277 \\.\PhysicalDrive1 for the second etc. Note: Windows and
278 FreeBSD prevent write access to areas of the disk containing
279 in-use data (e.g. filesystems).
280 If the wanted filename does need to include a colon, then
281 escape that with a '\' character. For instance, if the filename
282 is "/dev/dsk/foo@3,0:c", then you would use
283 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
284 stdin or stdout. Which of the two depends on the read/write
287 opendir=str Tell fio to recursively add any file it can find in this
288 directory and down the file system tree.
290 lockfile=str Fio defaults to not locking any files before it does
291 IO to them. If a file or file descriptor is shared, fio
292 can serialize IO to that file to make the end result
293 consistent. This is usual for emulating real workloads that
294 share files. The lock modes are:
296 none No locking. The default.
297 exclusive Only one thread/process may do IO,
298 excluding all others.
299 readwrite Read-write locking on the file. Many
300 readers may access the file at the
301 same time, but writes get exclusive
304 The option may be post-fixed with a lock batch number. If
305 set, then each thread/process may do that amount of IOs to
306 the file before giving up the lock. Since lock acquisition is
307 expensive, batching the lock/unlocks will speed up IO.
310 rw=str Type of io pattern. Accepted values are:
312 read Sequential reads
313 write Sequential writes
314 randwrite Random writes
315 randread Random reads
316 rw,readwrite Sequential mixed reads and writes
317 randrw Random mixed reads and writes
319 For the mixed io types, the default is to split them 50/50.
320 For certain types of io the result may still be skewed a bit,
321 since the speed may be different. It is possible to specify
322 a number of IO's to do before getting a new offset, this is
323 one by appending a ':<nr>' to the end of the string given.
324 For a random read, it would look like 'rw=randread:8' for
325 passing in an offset modifier with a value of 8. If the
326 suffix is used with a sequential IO pattern, then the value
327 specified will be added to the generated offset for each IO.
328 For instance, using rw=write:4k will skip 4k for every
329 write. It turns sequential IO into sequential IO with holes.
330 See the 'rw_sequencer' option.
332 rw_sequencer=str If an offset modifier is given by appending a number to
333 the rw=<str> line, then this option controls how that
334 number modifies the IO offset being generated. Accepted
337 sequential Generate sequential offset
338 identical Generate the same offset
340 'sequential' is only useful for random IO, where fio would
341 normally generate a new random offset for every IO. If you
342 append eg 8 to randread, you would get a new random offset for
343 every 8 IO's. The result would be a seek for only every 8
344 IO's, instead of for every IO. Use rw=randread:8 to specify
345 that. As sequential IO is already sequential, setting
346 'sequential' for that would not result in any differences.
347 'identical' behaves in a similar fashion, except it sends
348 the same offset 8 number of times before generating a new
351 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
352 Storage manufacturers like to use 10^3 or 1000 as a base
353 ten unit instead, for obvious reasons. Allow values are
354 1024 or 1000, with 1024 being the default.
356 unified_rw_reporting=bool Fio normally reports statistics on a per
357 data direction basis, meaning that read, write, and trim are
358 accounted and reported separately. If this option is set,
359 the fio will sum the results and report them as "mixed"
362 randrepeat=bool For random IO workloads, seed the generator in a predictable
363 way so that results are repeatable across repetitions.
365 use_os_rand=bool Fio can either use the random generator supplied by the OS
366 to generator random offsets, or it can use it's own internal
367 generator (based on Tausworthe). Default is to use the
368 internal generator, which is often of better quality and
371 fallocate=str Whether pre-allocation is performed when laying down files.
374 none Do not pre-allocate space
375 posix Pre-allocate via posix_fallocate()
376 keep Pre-allocate via fallocate() with
377 FALLOC_FL_KEEP_SIZE set
378 0 Backward-compatible alias for 'none'
379 1 Backward-compatible alias for 'posix'
381 May not be available on all supported platforms. 'keep' is only
382 available on Linux.If using ZFS on Solaris this must be set to
383 'none' because ZFS doesn't support it. Default: 'posix'.
385 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
386 on what IO patterns it is likely to issue. Sometimes you
387 want to test specific IO patterns without telling the
388 kernel about it, in which case you can disable this option.
389 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
390 IO and POSIX_FADV_RANDOM for random IO.
392 size=int The total size of file io for this job. Fio will run until
393 this many bytes has been transferred, unless runtime is
394 limited by other options (such as 'runtime', for instance).
395 Unless specific nrfiles and filesize options are given,
396 fio will divide this size between the available files
397 specified by the job. If not set, fio will use the full
398 size of the given files or devices. If the the files
399 do not exist, size must be given. It is also possible to
400 give size as a percentage between 1 and 100. If size=20%
401 is given, fio will use 20% of the full size of the given
404 filesize=int Individual file sizes. May be a range, in which case fio
405 will select sizes for files at random within the given range
406 and limited to 'size' in total (if that is given). If not
407 given, each created file is the same size.
410 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
411 space left on device) as the terminating condition. Only makes
412 sense with sequential write. For a read workload, the mount
413 point will be filled first then IO started on the result. This
414 option doesn't make sense if operating on a raw device node,
415 since the size of that is already known by the file system.
416 Additionally, writing beyond end-of-device will not return
420 bs=int The block size used for the io units. Defaults to 4k. Values
421 can be given for both read and writes. If a single int is
422 given, it will apply to both. If a second int is specified
423 after a comma, it will apply to writes only. In other words,
424 the format is either bs=read_and_write or bs=read,write.
425 bs=4k,8k will thus use 4k blocks for reads, and 8k blocks
426 for writes. If you only wish to set the write size, you
427 can do so by passing an empty read size - bs=,8k will set
428 8k for writes and leave the read default value.
431 ba=int At what boundary to align random IO offsets. Defaults to
432 the same as 'blocksize' the minimum blocksize given.
433 Minimum alignment is typically 512b for using direct IO,
434 though it usually depends on the hardware block size. This
435 option is mutually exclusive with using a random map for
436 files, so it will turn off that option.
438 blocksize_range=irange
439 bsrange=irange Instead of giving a single block size, specify a range
440 and fio will mix the issued io block sizes. The issued
441 io unit will always be a multiple of the minimum value
442 given (also see bs_unaligned). Applies to both reads and
443 writes, however a second range can be given after a comma.
446 bssplit=str Sometimes you want even finer grained control of the
447 block sizes issued, not just an even split between them.
448 This option allows you to weight various block sizes,
449 so that you are able to define a specific amount of
450 block sizes issued. The format for this option is:
452 bssplit=blocksize/percentage:blocksize/percentage
454 for as many block sizes as needed. So if you want to define
455 a workload that has 50% 64k blocks, 10% 4k blocks, and
456 40% 32k blocks, you would write:
458 bssplit=4k/10:64k/50:32k/40
460 Ordering does not matter. If the percentage is left blank,
461 fio will fill in the remaining values evenly. So a bssplit
462 option like this one:
464 bssplit=4k/50:1k/:32k/
466 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
467 always add up to 100, if bssplit is given a range that adds
468 up to more, it will error out.
470 bssplit also supports giving separate splits to reads and
471 writes. The format is identical to what bs= accepts. You
472 have to separate the read and write parts with a comma. So
473 if you want a workload that has 50% 2k reads and 50% 4k reads,
474 while having 90% 4k writes and 10% 8k writes, you would
477 bssplit=2k/50:4k/50,4k/90,8k/10
480 bs_unaligned If this option is given, any byte size value within bsrange
481 may be used as a block range. This typically wont work with
482 direct IO, as that normally requires sector alignment.
484 zero_buffers If this option is given, fio will init the IO buffers to
485 all zeroes. The default is to fill them with random data.
487 refill_buffers If this option is given, fio will refill the IO buffers
488 on every submit. The default is to only fill it at init
489 time and reuse that data. Only makes sense if zero_buffers
490 isn't specified, naturally. If data verification is enabled,
491 refill_buffers is also automatically enabled.
493 scramble_buffers=bool If refill_buffers is too costly and the target is
494 using data deduplication, then setting this option will
495 slightly modify the IO buffer contents to defeat normal
496 de-dupe attempts. This is not enough to defeat more clever
497 block compression attempts, but it will stop naive dedupe of
498 blocks. Default: true.
500 buffer_compress_percentage=int If this is set, then fio will attempt to
501 provide IO buffer content (on WRITEs) that compress to
502 the specified level. Fio does this by providing a mix of
503 random data and zeroes. Note that this is per block size
504 unit, for file/disk wide compression level that matches
505 this setting, you'll also want to set refill_buffers.
507 buffer_compress_chunk=int See buffer_compress_percentage. This
508 setting allows fio to manage how big the ranges of random
509 data and zeroed data is. Without this set, fio will
510 provide buffer_compress_percentage of blocksize random
511 data, followed by the remaining zeroed. With this set
512 to some chunk size smaller than the block size, fio can
513 alternate random and zeroed data throughout the IO
516 nrfiles=int Number of files to use for this job. Defaults to 1.
518 openfiles=int Number of files to keep open at the same time. Defaults to
519 the same as nrfiles, can be set smaller to limit the number
522 file_service_type=str Defines how fio decides which file from a job to
523 service next. The following types are defined:
525 random Just choose a file at random.
527 roundrobin Round robin over open files. This
530 sequential Finish one file before moving on to
531 the next. Multiple files can still be
532 open depending on 'openfiles'.
534 The string can have a number appended, indicating how
535 often to switch to a new file. So if option random:4 is
536 given, fio will switch to a new random file after 4 ios
539 ioengine=str Defines how the job issues io to the file. The following
542 sync Basic read(2) or write(2) io. lseek(2) is
543 used to position the io location.
545 psync Basic pread(2) or pwrite(2) io.
547 vsync Basic readv(2) or writev(2) IO.
549 libaio Linux native asynchronous io. Note that Linux
550 may only support queued behaviour with
551 non-buffered IO (set direct=1 or buffered=0).
552 This engine defines engine specific options.
554 posixaio glibc posix asynchronous io.
556 solarisaio Solaris native asynchronous io.
558 windowsaio Windows native asynchronous io.
560 mmap File is memory mapped and data copied
561 to/from using memcpy(3).
563 splice splice(2) is used to transfer the data and
564 vmsplice(2) to transfer data from user
567 syslet-rw Use the syslet system calls to make
568 regular read/write async.
570 sg SCSI generic sg v3 io. May either be
571 synchronous using the SG_IO ioctl, or if
572 the target is an sg character device
573 we use read(2) and write(2) for asynchronous
576 null Doesn't transfer any data, just pretends
577 to. This is mainly used to exercise fio
578 itself and for debugging/testing purposes.
580 net Transfer over the network to given host:port.
581 Depending on the protocol used, the hostname,
582 port, listen and filename options are used to
583 specify what sort of connection to make, while
584 the protocol option determines which protocol
586 This engine defines engine specific options.
588 netsplice Like net, but uses splice/vmsplice to
589 map data and send/receive.
590 This engine defines engine specific options.
592 cpuio Doesn't transfer any data, but burns CPU
593 cycles according to the cpuload= and
594 cpucycle= options. Setting cpuload=85
595 will cause that job to do nothing but burn
596 85% of the CPU. In case of SMP machines,
597 use numjobs=<no_of_cpu> to get desired CPU
598 usage, as the cpuload only loads a single
599 CPU at the desired rate.
601 guasi The GUASI IO engine is the Generic Userspace
602 Asyncronous Syscall Interface approach
605 http://www.xmailserver.org/guasi-lib.html
607 for more info on GUASI.
609 rdma The RDMA I/O engine supports both RDMA
610 memory semantics (RDMA_WRITE/RDMA_READ) and
611 channel semantics (Send/Recv) for the
612 InfiniBand, RoCE and iWARP protocols.
614 falloc IO engine that does regular fallocate to
615 simulate data transfer as fio ioengine.
616 DDIR_READ does fallocate(,mode = keep_size,)
617 DDIR_WRITE does fallocate(,mode = 0)
618 DDIR_TRIM does fallocate(,mode = punch_hole)
620 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
621 ioctls to simulate defragment activity in
622 request to DDIR_WRITE event
624 external Prefix to specify loading an external
625 IO engine object file. Append the engine
626 filename, eg ioengine=external:/tmp/foo.o
627 to load ioengine foo.o in /tmp.
629 iodepth=int This defines how many io units to keep in flight against
630 the file. The default is 1 for each file defined in this
631 job, can be overridden with a larger value for higher
632 concurrency. Note that increasing iodepth beyond 1 will not
633 affect synchronous ioengines (except for small degress when
634 verify_async is in use). Even async engines may impose OS
635 restrictions causing the desired depth not to be achieved.
636 This may happen on Linux when using libaio and not setting
637 direct=1, since buffered IO is not async on that OS. Keep an
638 eye on the IO depth distribution in the fio output to verify
639 that the achieved depth is as expected. Default: 1.
641 iodepth_batch_submit=int
642 iodepth_batch=int This defines how many pieces of IO to submit at once.
643 It defaults to 1 which means that we submit each IO
644 as soon as it is available, but can be raised to submit
645 bigger batches of IO at the time.
647 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
648 at once. It defaults to 1 which means that we'll ask
649 for a minimum of 1 IO in the retrieval process from
650 the kernel. The IO retrieval will go on until we
651 hit the limit set by iodepth_low. If this variable is
652 set to 0, then fio will always check for completed
653 events before queuing more IO. This helps reduce
654 IO latency, at the cost of more retrieval system calls.
656 iodepth_low=int The low water mark indicating when to start filling
657 the queue again. Defaults to the same as iodepth, meaning
658 that fio will attempt to keep the queue full at all times.
659 If iodepth is set to eg 16 and iodepth_low is set to 4, then
660 after fio has filled the queue of 16 requests, it will let
661 the depth drain down to 4 before starting to fill it again.
663 direct=bool If value is true, use non-buffered io. This is usually
664 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
665 On Windows the synchronous ioengines don't support direct io.
667 buffered=bool If value is true, use buffered io. This is the opposite
668 of the 'direct' option. Defaults to true.
670 offset=int Start io at the given offset in the file. The data before
671 the given offset will not be touched. This effectively
672 caps the file size at real_size - offset.
674 offset_increment=int If this is provided, then the real offset becomes
675 the offset + offset_increment * thread_number, where the
676 thread number is a counter that starts at 0 and is incremented
677 for each job. This option is useful if there are several jobs
678 which are intended to operate on a file in parallel in disjoint
679 segments, with even spacing between the starting points.
681 fsync=int If writing to a file, issue a sync of the dirty data
682 for every number of blocks given. For example, if you give
683 32 as a parameter, fio will sync the file for every 32
684 writes issued. If fio is using non-buffered io, we may
685 not sync the file. The exception is the sg io engine, which
686 synchronizes the disk cache anyway.
688 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
690 In FreeBSD and Windows there is no fdatasync(), this falls back to
693 sync_file_range=str:val Use sync_file_range() for every 'val' number of
694 write operations. Fio will track range of writes that
695 have happened since the last sync_file_range() call. 'str'
696 can currently be one or more of:
698 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
699 write SYNC_FILE_RANGE_WRITE
700 wait_after SYNC_FILE_RANGE_WAIT_AFTER
702 So if you do sync_file_range=wait_before,write:8, fio would
703 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
704 every 8 writes. Also see the sync_file_range(2) man page.
705 This option is Linux specific.
707 overwrite=bool If true, writes to a file will always overwrite existing
708 data. If the file doesn't already exist, it will be
709 created before the write phase begins. If the file exists
710 and is large enough for the specified write phase, nothing
713 end_fsync=bool If true, fsync file contents when a write stage has completed.
715 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
716 This differs from end_fsync in that it will happen on every
717 file close, not just at the end of the job.
719 rwmixread=int How large a percentage of the mix should be reads.
721 rwmixwrite=int How large a percentage of the mix should be writes. If both
722 rwmixread and rwmixwrite is given and the values do not add
723 up to 100%, the latter of the two will be used to override
724 the first. This may interfere with a given rate setting,
725 if fio is asked to limit reads or writes to a certain rate.
726 If that is the case, then the distribution may be skewed.
728 random_distribution=str:float By default, fio will use a completely uniform
729 random distribution when asked to perform random IO. Sometimes
730 it is useful to skew the distribution in specific ways,
731 ensuring that some parts of the data is more hot than others.
732 fio includes the following distribution models:
734 random Uniform random distribution
735 zipf Zipf distribution
736 pareto Pareto distribution
738 When using a zipf or pareto distribution, an input value
739 is also needed to define the access pattern. For zipf, this
740 is the zipf theta. For pareto, it's the pareto power. Fio
741 includes a test program, genzipf, that can be used visualize
742 what the given input values will yield in terms of hit rates.
743 If you wanted to use zipf with a theta of 1.2, you would use
744 random_distribution=zipf:1.2 as the option. If a non-uniform
745 model is used, fio will disable use of the random map.
747 norandommap Normally fio will cover every block of the file when doing
748 random IO. If this option is given, fio will just get a
749 new random offset without looking at past io history. This
750 means that some blocks may not be read or written, and that
751 some blocks may be read/written more than once. This option
752 is mutually exclusive with verify= if and only if multiple
753 blocksizes (via bsrange=) are used, since fio only tracks
754 complete rewrites of blocks.
756 softrandommap=bool See norandommap. If fio runs with the random block map
757 enabled and it fails to allocate the map, if this option is
758 set it will continue without a random block map. As coverage
759 will not be as complete as with random maps, this option is
762 random_generator=str Fio supports the following engines for generating
763 IO offsets for random IO:
765 tausworthe Strong 2^88 cycle random number generator
766 lfsr Linear feedback shift register generator
768 Tausworthe is a strong random number generator, but it
769 requires tracking on the side if we want to ensure that
770 blocks are only read or written once. LFSR guarantees
771 that we never generate the same offset twice, and it's
772 also less computationally expensive. It's not a true
773 random generator, however, though for IO purposes it's
774 typically good enough. LFSR only works with single
775 block sizes, not with workloads that use multiple block
776 sizes. If used with such a workload, fio may read or write
777 some blocks multiple times.
779 nice=int Run the job with the given nice value. See man nice(2).
781 prio=int Set the io priority value of this job. Linux limits us to
782 a positive value between 0 and 7, with 0 being the highest.
785 prioclass=int Set the io priority class. See man ionice(1).
787 thinktime=int Stall the job x microseconds after an io has completed before
788 issuing the next. May be used to simulate processing being
789 done by an application. See thinktime_blocks and
793 Only valid if thinktime is set - pretend to spend CPU time
794 doing something with the data received, before falling back
795 to sleeping for the rest of the period specified by
799 Only valid if thinktime is set - control how many blocks
800 to issue, before waiting 'thinktime' usecs. If not set,
801 defaults to 1 which will make fio wait 'thinktime' usecs
804 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
805 the normal suffix rules apply. You can use rate=500k to limit
806 reads and writes to 500k each, or you can specify read and
807 writes separately. Using rate=1m,500k would limit reads to
808 1MB/sec and writes to 500KB/sec. Capping only reads or
809 writes can be done with rate=,500k or rate=500k,. The former
810 will only limit writes (to 500KB/sec), the latter will only
813 ratemin=int Tell fio to do whatever it can to maintain at least this
814 bandwidth. Failing to meet this requirement, will cause
815 the job to exit. The same format as rate is used for
816 read vs write separation.
818 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
819 as rate, just specified independently of bandwidth. If the
820 job is given a block size range instead of a fixed value,
821 the smallest block size is used as the metric. The same format
822 as rate is used for read vs write seperation.
824 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
825 the job to exit. The same format as rate is used for read vs
828 max_latency=int If set, fio will exit the job if it exceeds this maximum
829 latency. It will exit with an ETIME error.
831 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
834 cpumask=int Set the CPU affinity of this job. The parameter given is a
835 bitmask of allowed CPU's the job may run on. So if you want
836 the allowed CPUs to be 1 and 5, you would pass the decimal
837 value of (1 << 1 | 1 << 5), or 34. See man
838 sched_setaffinity(2). This may not work on all supported
839 operating systems or kernel versions. This option doesn't
840 work well for a higher CPU count than what you can store in
841 an integer mask, so it can only control cpus 1-32. For
842 boxes with larger CPU counts, use cpus_allowed.
844 cpus_allowed=str Controls the same options as cpumask, but it allows a text
845 setting of the permitted CPUs instead. So to use CPUs 1 and
846 5, you would specify cpus_allowed=1,5. This options also
847 allows a range of CPUs. Say you wanted a binding to CPUs
848 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
850 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
851 arguments allow comma delimited list of cpu numbers,
852 A-B ranges, or 'all'. Note, to enable numa options support,
853 fio must be built on a system with libnuma-dev(el) installed.
855 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
856 nodes. Format of the argements:
858 `mode' is one of the following memory policy:
859 default, prefer, bind, interleave, local
860 For `default' and `local' memory policy, no node is
861 needed to be specified.
862 For `prefer', only one node is allowed.
863 For `bind' and `interleave', it allow comma delimited
864 list of numbers, A-B ranges, or 'all'.
866 startdelay=time Start this job the specified number of seconds after fio
867 has started. Only useful if the job file contains several
868 jobs, and you want to delay starting some jobs to a certain
871 runtime=time Tell fio to terminate processing after the specified number
872 of seconds. It can be quite hard to determine for how long
873 a specified job will run, so this parameter is handy to
874 cap the total runtime to a given time.
876 time_based If set, fio will run for the duration of the runtime
877 specified even if the file(s) are completely read or
878 written. It will simply loop over the same workload
879 as many times as the runtime allows.
881 ramp_time=time If set, fio will run the specified workload for this amount
882 of time before logging any performance numbers. Useful for
883 letting performance settle before logging results, thus
884 minimizing the runtime required for stable results. Note
885 that the ramp_time is considered lead in time for a job,
886 thus it will increase the total runtime if a special timeout
887 or runtime is specified.
889 invalidate=bool Invalidate the buffer/page cache parts for this file prior
890 to starting io. Defaults to true.
892 sync=bool Use sync io for buffered writes. For the majority of the
893 io engines, this means using O_SYNC.
896 mem=str Fio can use various types of memory as the io unit buffer.
897 The allowed values are:
899 malloc Use memory from malloc(3) as the buffers.
901 shm Use shared memory as the buffers. Allocated
904 shmhuge Same as shm, but use huge pages as backing.
906 mmap Use mmap to allocate buffers. May either be
907 anonymous memory, or can be file backed if
908 a filename is given after the option. The
909 format is mem=mmap:/path/to/file.
911 mmaphuge Use a memory mapped huge file as the buffer
912 backing. Append filename after mmaphuge, ala
913 mem=mmaphuge:/hugetlbfs/file
915 The area allocated is a function of the maximum allowed
916 bs size for the job, multiplied by the io depth given. Note
917 that for shmhuge and mmaphuge to work, the system must have
918 free huge pages allocated. This can normally be checked
919 and set by reading/writing /proc/sys/vm/nr_hugepages on a
920 Linux system. Fio assumes a huge page is 4MB in size. So
921 to calculate the number of huge pages you need for a given
922 job file, add up the io depth of all jobs (normally one unless
923 iodepth= is used) and multiply by the maximum bs set. Then
924 divide that number by the huge page size. You can see the
925 size of the huge pages in /proc/meminfo. If no huge pages
926 are allocated by having a non-zero number in nr_hugepages,
927 using mmaphuge or shmhuge will fail. Also see hugepage-size.
929 mmaphuge also needs to have hugetlbfs mounted and the file
930 location should point there. So if it's mounted in /huge,
931 you would use mem=mmaphuge:/huge/somefile.
933 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
934 Note that the given alignment is applied to the first IO unit
935 buffer, if using iodepth the alignment of the following buffers
936 are given by the bs used. In other words, if using a bs that is
937 a multiple of the page sized in the system, all buffers will
938 be aligned to this value. If using a bs that is not page
939 aligned, the alignment of subsequent IO memory buffers is the
940 sum of the iomem_align and bs used.
943 Defines the size of a huge page. Must at least be equal
944 to the system setting, see /proc/meminfo. Defaults to 4MB.
945 Should probably always be a multiple of megabytes, so using
946 hugepage-size=Xm is the preferred way to set this to avoid
947 setting a non-pow-2 bad value.
949 exitall When one job finishes, terminate the rest. The default is
950 to wait for each job to finish, sometimes that is not the
953 bwavgtime=int Average the calculated bandwidth over the given time. Value
954 is specified in milliseconds.
956 iopsavgtime=int Average the calculated IOPS over the given time. Value
957 is specified in milliseconds.
959 create_serialize=bool If true, serialize the file creating for the jobs.
960 This may be handy to avoid interleaving of data
961 files, which may greatly depend on the filesystem
962 used and even the number of processors in the system.
964 create_fsync=bool fsync the data file after creation. This is the
967 create_on_open=bool Don't pre-setup the files for IO, just create open()
968 when it's time to do IO to that file.
970 create_only=bool If true, fio will only run the setup phase of the job.
971 If files need to be laid out or updated on disk, only
972 that will be done. The actual job contents are not
975 pre_read=bool If this is given, files will be pre-read into memory before
976 starting the given IO operation. This will also clear
977 the 'invalidate' flag, since it is pointless to pre-read
978 and then drop the cache. This will only work for IO engines
979 that are seekable, since they allow you to read the same data
980 multiple times. Thus it will not work on eg network or splice
983 unlink=bool Unlink the job files when done. Not the default, as repeated
984 runs of that job would then waste time recreating the file
987 loops=int Run the specified number of iterations of this job. Used
988 to repeat the same workload a given number of times. Defaults
991 do_verify=bool Run the verify phase after a write phase. Only makes sense if
992 verify is set. Defaults to 1.
994 verify=str If writing to a file, fio can verify the file contents
995 after each iteration of the job. The allowed values are:
997 md5 Use an md5 sum of the data area and store
998 it in the header of each block.
1000 crc64 Use an experimental crc64 sum of the data
1001 area and store it in the header of each
1004 crc32c Use a crc32c sum of the data area and store
1005 it in the header of each block.
1007 crc32c-intel Use hardware assisted crc32c calcuation
1008 provided on SSE4.2 enabled processors. Falls
1009 back to regular software crc32c, if not
1010 supported by the system.
1012 crc32 Use a crc32 sum of the data area and store
1013 it in the header of each block.
1015 crc16 Use a crc16 sum of the data area and store
1016 it in the header of each block.
1018 crc7 Use a crc7 sum of the data area and store
1019 it in the header of each block.
1021 sha512 Use sha512 as the checksum function.
1023 sha256 Use sha256 as the checksum function.
1025 sha1 Use optimized sha1 as the checksum function.
1027 meta Write extra information about each io
1028 (timestamp, block number etc.). The block
1029 number is verified. See also verify_pattern.
1031 null Only pretend to verify. Useful for testing
1032 internals with ioengine=null, not for much
1035 This option can be used for repeated burn-in tests of a
1036 system to make sure that the written data is also
1037 correctly read back. If the data direction given is
1038 a read or random read, fio will assume that it should
1039 verify a previously written file. If the data direction
1040 includes any form of write, the verify will be of the
1043 verifysort=bool If set, fio will sort written verify blocks when it deems
1044 it faster to read them back in a sorted manner. This is
1045 often the case when overwriting an existing file, since
1046 the blocks are already laid out in the file system. You
1047 can ignore this option unless doing huge amounts of really
1048 fast IO where the red-black tree sorting CPU time becomes
1051 verify_offset=int Swap the verification header with data somewhere else
1052 in the block before writing. Its swapped back before
1055 verify_interval=int Write the verification header at a finer granularity
1056 than the blocksize. It will be written for chunks the
1057 size of header_interval. blocksize should divide this
1060 verify_pattern=str If set, fio will fill the io buffers with this
1061 pattern. Fio defaults to filling with totally random
1062 bytes, but sometimes it's interesting to fill with a known
1063 pattern for io verification purposes. Depending on the
1064 width of the pattern, fio will fill 1/2/3/4 bytes of the
1065 buffer at the time(it can be either a decimal or a hex number).
1066 The verify_pattern if larger than a 32-bit quantity has to
1067 be a hex number that starts with either "0x" or "0X". Use
1070 verify_fatal=bool Normally fio will keep checking the entire contents
1071 before quitting on a block verification failure. If this
1072 option is set, fio will exit the job on the first observed
1075 verify_dump=bool If set, dump the contents of both the original data
1076 block and the data block we read off disk to files. This
1077 allows later analysis to inspect just what kind of data
1078 corruption occurred. Off by default.
1080 verify_async=int Fio will normally verify IO inline from the submitting
1081 thread. This option takes an integer describing how many
1082 async offload threads to create for IO verification instead,
1083 causing fio to offload the duty of verifying IO contents
1084 to one or more separate threads. If using this offload
1085 option, even sync IO engines can benefit from using an
1086 iodepth setting higher than 1, as it allows them to have
1087 IO in flight while verifies are running.
1089 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1090 async IO verification threads. See cpus_allowed for the
1093 verify_backlog=int Fio will normally verify the written contents of a
1094 job that utilizes verify once that job has completed. In
1095 other words, everything is written then everything is read
1096 back and verified. You may want to verify continually
1097 instead for a variety of reasons. Fio stores the meta data
1098 associated with an IO block in memory, so for large
1099 verify workloads, quite a bit of memory would be used up
1100 holding this meta data. If this option is enabled, fio
1101 will write only N blocks before verifying these blocks.
1103 will verify the previously written blocks before continuing
1106 verify_backlog_batch=int Control how many blocks fio will verify
1107 if verify_backlog is set. If not set, will default to
1108 the value of verify_backlog (meaning the entire queue
1109 is read back and verified). If verify_backlog_batch is
1110 less than verify_backlog then not all blocks will be verified,
1111 if verify_backlog_batch is larger than verify_backlog, some
1112 blocks will be verified more than once.
1115 wait_for_previous Wait for preceeding jobs in the job file to exit, before
1116 starting this one. Can be used to insert serialization
1117 points in the job file. A stone wall also implies starting
1118 a new reporting group.
1120 new_group Start a new reporting group. See: group_reporting.
1122 numjobs=int Create the specified number of clones of this job. May be
1123 used to setup a larger number of threads/processes doing
1124 the same thing. Each thread is reported separately; to see
1125 statistics for all clones as a whole, use group_reporting in
1126 conjunction with new_group.
1128 group_reporting It may sometimes be interesting to display statistics for
1129 groups of jobs as a whole instead of for each individual job.
1130 This is especially true if 'numjobs' is used; looking at
1131 individual thread/process output quickly becomes unwieldy.
1132 To see the final report per-group instead of per-job, use
1133 'group_reporting'. Jobs in a file will be part of the same
1134 reporting group, unless if separated by a stonewall, or by
1137 thread fio defaults to forking jobs, however if this option is
1138 given, fio will use pthread_create(3) to create threads
1141 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1143 zoneskip=int Skip the specified number of bytes when zonesize data has
1144 been read. The two zone options can be used to only do
1145 io on zones of a file.
1147 write_iolog=str Write the issued io patterns to the specified file. See
1148 read_iolog. Specify a separate file for each job, otherwise
1149 the iologs will be interspersed and the file may be corrupt.
1151 read_iolog=str Open an iolog with the specified file name and replay the
1152 io patterns it contains. This can be used to store a
1153 workload and replay it sometime later. The iolog given
1154 may also be a blktrace binary file, which allows fio
1155 to replay a workload captured by blktrace. See blktrace
1156 for how to capture such logging data. For blktrace replay,
1157 the file needs to be turned into a blkparse binary data
1158 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1160 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1161 is to attempt to respect the time stamps within the log and
1162 replay them with the appropriate delay between IOPS. By
1163 setting this variable fio will not respect the timestamps and
1164 attempt to replay them as fast as possible while still
1165 respecting ordering. The result is the same I/O pattern to a
1166 given device, but different timings.
1168 replay_redirect=str While replaying I/O patterns using read_iolog the
1169 default behavior is to replay the IOPS onto the major/minor
1170 device that each IOP was recorded from. This is sometimes
1171 undesireable because on a different machine those major/minor
1172 numbers can map to a different device. Changing hardware on
1173 the same system can also result in a different major/minor
1174 mapping. Replay_redirect causes all IOPS to be replayed onto
1175 the single specified device regardless of the device it was
1176 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1177 IO in the blktrace to be replayed onto /dev/sdc. This means
1178 multiple devices will be replayed onto a single, if the trace
1179 contains multiple devices. If you want multiple devices to be
1180 replayed concurrently to multiple redirected devices you must
1181 blkparse your trace into separate traces and replay them with
1182 independent fio invocations. Unfortuantely this also breaks
1183 the strict time ordering between multiple device accesses.
1185 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1186 file. Can be used to store data of the bandwidth of the
1187 jobs in their lifetime. The included fio_generate_plots
1188 script uses gnuplot to turn these text files into nice
1189 graphs. See write_lat_log for behaviour of given
1190 filename. For this option, the suffix is _bw.log.
1192 write_lat_log=str Same as write_bw_log, except that this option stores io
1193 submission, completion, and total latencies instead. If no
1194 filename is given with this option, the default filename of
1195 "jobname_type.log" is used. Even if the filename is given,
1196 fio will still append the type of log. So if one specifies
1200 The actual log names will be foo_slat.log, foo_slat.log,
1201 and foo_lat.log. This helps fio_generate_plot fine the logs
1204 write_bw_log=str If given, write an IOPS log of the jobs in this job
1205 file. See write_bw_log.
1207 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1208 given with this option, the default filename of
1209 "jobname_type.log" is used. Even if the filename is given,
1210 fio will still append the type of log.
1212 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1213 or bw log for every IO that completes. When writing to the
1214 disk log, that can quickly grow to a very large size. Setting
1215 this option makes fio average the each log entry over the
1216 specified period of time, reducing the resolution of the log.
1219 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1220 potentially be used instead of removing memory or booting
1221 with less memory to simulate a smaller amount of memory.
1223 exec_prerun=str Before running this job, issue the command specified
1226 exec_postrun=str After the job completes, issue the command specified
1229 ioscheduler=str Attempt to switch the device hosting the file to the specified
1230 io scheduler before running.
1232 cpuload=int If the job is a CPU cycle eater, attempt to use the specified
1233 percentage of CPU cycles.
1235 cpuchunks=int If the job is a CPU cycle eater, split the load into
1236 cycles of the given time. In microseconds.
1238 disk_util=bool Generate disk utilization statistics, if the platform
1239 supports it. Defaults to on.
1241 disable_lat=bool Disable measurements of total latency numbers. Useful
1242 only for cutting back the number of calls to gettimeofday,
1243 as that does impact performance at really high IOPS rates.
1244 Note that to really get rid of a large amount of these
1245 calls, this option must be used with disable_slat and
1248 disable_clat=bool Disable measurements of completion latency numbers. See
1251 disable_slat=bool Disable measurements of submission latency numbers. See
1254 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1257 clat_percentiles=bool Enable the reporting of percentiles of
1258 completion latencies.
1260 percentile_list=float_list Overwrite the default list of percentiles
1261 for completion latencies. Each number is a floating
1262 number in the range (0,100], and the maximum length of
1263 the list is 20. Use ':' to separate the numbers, and
1264 list the numbers in ascending order. For example,
1265 --percentile_list=99.5:99.9 will cause fio to report
1266 the values of completion latency below which 99.5% and
1267 99.9% of the observed latencies fell, respectively.
1269 clocksource=str Use the given clocksource as the base of timing. The
1270 supported options are:
1272 gettimeofday gettimeofday(2)
1274 clock_gettime clock_gettime(2)
1276 cpu Internal CPU clock source
1278 cpu is the preferred clocksource if it is reliable, as it
1279 is very fast (and fio is heavy on time calls). Fio will
1280 automatically use this clocksource if it's supported and
1281 considered reliable on the system it is running on, unless
1282 another clocksource is specifically set. For x86/x86-64 CPUs,
1283 this means supporting TSC Invariant.
1285 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1286 (disable_clat, disable_slat, disable_bw) plus reduce
1287 precision of the timeout somewhat to really shrink
1288 the gettimeofday() call count. With this option enabled,
1289 we only do about 0.4% of the gtod() calls we would have
1290 done if all time keeping was enabled.
1292 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1293 execution to just getting the current time. Fio (and
1294 databases, for instance) are very intensive on gettimeofday()
1295 calls. With this option, you can set one CPU aside for
1296 doing nothing but logging current time to a shared memory
1297 location. Then the other threads/processes that run IO
1298 workloads need only copy that segment, instead of entering
1299 the kernel with a gettimeofday() call. The CPU set aside
1300 for doing these time calls will be excluded from other
1301 uses. Fio will manually clear it from the CPU mask of other
1304 continue_on_error=str Normally fio will exit the job on the first observed
1305 failure. If this option is set, fio will continue the job when
1306 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1307 is exceeded or the I/O size specified is completed. If this
1308 option is used, there are two more stats that are appended,
1309 the total error count and the first error. The error field
1310 given in the stats is the first error that was hit during the
1313 The allowed values are:
1315 none Exit on any IO or verify errors.
1317 read Continue on read errors, exit on all others.
1319 write Continue on write errors, exit on all others.
1321 io Continue on any IO error, exit on all others.
1323 verify Continue on verify errors, exit on all others.
1325 all Continue on all errors.
1327 0 Backward-compatible alias for 'none'.
1329 1 Backward-compatible alias for 'all'.
1331 ignore_error=str Sometimes you want to ignore some errors during test
1332 in that case you can specify error list for each error type.
1333 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1334 errors for given error type is separated with ':'. Error
1335 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1337 ignore_error=EAGAIN,ENOSPC:122
1338 This option will ignore EAGAIN from READ, and ENOSPC and
1339 122(EDQUOT) from WRITE.
1341 error_dump=bool If set dump every error even if it is non fatal, true
1342 by default. If disabled only fatal error will be dumped
1344 cgroup=str Add job to this control group. If it doesn't exist, it will
1345 be created. The system must have a mounted cgroup blkio
1346 mount point for this to work. If your system doesn't have it
1347 mounted, you can do so with:
1349 # mount -t cgroup -o blkio none /cgroup
1351 cgroup_weight=int Set the weight of the cgroup to this value. See
1352 the documentation that comes with the kernel, allowed values
1353 are in the range of 100..1000.
1355 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1356 the job completion. To override this behavior and to leave
1357 cgroups around after the job completion, set cgroup_nodelete=1.
1358 This can be useful if one wants to inspect various cgroup
1359 files after job completion. Default: false
1361 uid=int Instead of running as the invoking user, set the user ID to
1362 this value before the thread/process does any work.
1364 gid=int Set group ID, see uid.
1366 flow_id=int The ID of the flow. If not specified, it defaults to being a
1367 global flow. See flow.
1369 flow=int Weight in token-based flow control. If this value is used, then
1370 there is a 'flow counter' which is used to regulate the
1371 proportion of activity between two or more jobs. fio attempts
1372 to keep this flow counter near zero. The 'flow' parameter
1373 stands for how much should be added or subtracted to the flow
1374 counter on each iteration of the main I/O loop. That is, if
1375 one job has flow=8 and another job has flow=-1, then there
1376 will be a roughly 1:8 ratio in how much one runs vs the other.
1378 flow_watermark=int The maximum value that the absolute value of the flow
1379 counter is allowed to reach before the job must wait for a
1380 lower value of the counter.
1382 flow_sleep=int The period of time, in microseconds, to wait after the flow
1383 watermark has been exceeded before retrying operations
1385 In addition, there are some parameters which are only valid when a specific
1386 ioengine is in use. These are used identically to normal parameters, with the
1387 caveat that when used on the command line, they must come after the ioengine
1388 that defines them is selected.
1390 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1391 the io_getevents system call to reap newly returned events.
1392 With this flag turned on, the AIO ring will be read directly
1393 from user-space to reap events. The reaping mode is only
1394 enabled when polling for a minimum of 0 events (eg when
1395 iodepth_batch_complete=0).
1397 [netsplice] hostname=str
1398 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1399 If the job is a TCP listener or UDP reader, the hostname is not
1400 used and must be omitted.
1402 [netsplice] port=int
1403 [net] port=int The TCP or UDP port to bind to or connect to.
1405 [netsplice] nodelay=bool
1406 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1408 [netsplice] protocol=str
1409 [netsplice] proto=str
1411 [net] proto=str The network protocol to use. Accepted values are:
1413 tcp Transmission control protocol
1414 udp User datagram protocol
1415 unix UNIX domain socket
1417 When the protocol is TCP or UDP, the port must also be given,
1418 as well as the hostname if the job is a TCP listener or UDP
1419 reader. For unix sockets, the normal filename option should be
1420 used and the port is invalid.
1422 [net] listen For TCP network connections, tell fio to listen for incoming
1423 connections rather than initiating an outgoing connection. The
1424 hostname must be omitted if this option is used.
1425 [net] pingpong Normal a network writer will just continue writing data, and
1426 a network reader will just consume packages. If pingpong=1
1427 is set, a writer will send its normal payload to the reader,
1428 then wait for the reader to send the same payload back. This
1429 allows fio to measure network latencies. The submission
1430 and completion latencies then measure local time spent
1431 sending or receiving, and the completion latency measures
1432 how long it took for the other end to receive and send back.
1434 [e4defrag] donorname=str
1435 File will be used as a block donor(swap extents between files)
1436 [e4defrag] inplace=int
1437 Configure donor file blocks allocation strategy
1438 0(default): Preallocate donor's file on init
1439 1 : allocate space immidietly inside defragment event,
1440 and free right after event
1444 6.0 Interpreting the output
1445 ---------------------------
1447 fio spits out a lot of output. While running, fio will display the
1448 status of the jobs created. An example of that would be:
1450 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1452 The characters inside the square brackets denote the current status of
1453 each thread. The possible values (in typical life cycle order) are:
1457 P Thread setup, but not started.
1459 I Thread initialized, waiting or generating necessary data.
1460 p Thread running pre-reading file(s).
1461 R Running, doing sequential reads.
1462 r Running, doing random reads.
1463 W Running, doing sequential writes.
1464 w Running, doing random writes.
1465 M Running, doing mixed sequential reads/writes.
1466 m Running, doing mixed random reads/writes.
1467 F Running, currently waiting for fsync()
1468 V Running, doing verification of written data.
1469 E Thread exited, not reaped by main thread yet.
1471 X Thread reaped, exited with an error.
1472 K Thread reaped, exited due to signal.
1474 The other values are fairly self explanatory - number of threads
1475 currently running and doing io, rate of io since last check (read speed
1476 listed first, then write speed), and the estimated completion percentage
1477 and time for the running group. It's impossible to estimate runtime of
1478 the following groups (if any). Note that the string is displayed in order,
1479 so it's possible to tell which of the jobs are currently doing what. The
1480 first character is the first job defined in the job file, and so forth.
1482 When fio is done (or interrupted by ctrl-c), it will show the data for
1483 each thread, group of threads, and disks in that order. For each data
1484 direction, the output looks like:
1486 Client1 (g=0): err= 0:
1487 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1488 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1489 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1490 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1491 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1492 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1493 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1494 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1495 issued r/w: total=0/32768, short=0/0
1496 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1497 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1499 The client number is printed, along with the group id and error of that
1500 thread. Below is the io statistics, here for writes. In the order listed,
1503 io= Number of megabytes io performed
1504 bw= Average bandwidth rate
1505 iops= Average IOs performed per second
1506 runt= The runtime of that thread
1507 slat= Submission latency (avg being the average, stdev being the
1508 standard deviation). This is the time it took to submit
1509 the io. For sync io, the slat is really the completion
1510 latency, since queue/complete is one operation there. This
1511 value can be in milliseconds or microseconds, fio will choose
1512 the most appropriate base and print that. In the example
1513 above, milliseconds is the best scale. Note: in --minimal mode
1514 latencies are always expressed in microseconds.
1515 clat= Completion latency. Same names as slat, this denotes the
1516 time from submission to completion of the io pieces. For
1517 sync io, clat will usually be equal (or very close) to 0,
1518 as the time from submit to complete is basically just
1519 CPU time (io has already been done, see slat explanation).
1520 bw= Bandwidth. Same names as the xlat stats, but also includes
1521 an approximate percentage of total aggregate bandwidth
1522 this thread received in this group. This last value is
1523 only really useful if the threads in this group are on the
1524 same disk, since they are then competing for disk access.
1525 cpu= CPU usage. User and system time, along with the number
1526 of context switches this thread went through, usage of
1527 system and user time, and finally the number of major
1528 and minor page faults.
1529 IO depths= The distribution of io depths over the job life time. The
1530 numbers are divided into powers of 2, so for example the
1531 16= entries includes depths up to that value but higher
1532 than the previous entry. In other words, it covers the
1533 range from 16 to 31.
1534 IO submit= How many pieces of IO were submitting in a single submit
1535 call. Each entry denotes that amount and below, until
1536 the previous entry - eg, 8=100% mean that we submitted
1537 anywhere in between 5-8 ios per submit call.
1538 IO complete= Like the above submit number, but for completions instead.
1539 IO issued= The number of read/write requests issued, and how many
1541 IO latencies= The distribution of IO completion latencies. This is the
1542 time from when IO leaves fio and when it gets completed.
1543 The numbers follow the same pattern as the IO depths,
1544 meaning that 2=1.6% means that 1.6% of the IO completed
1545 within 2 msecs, 20=12.8% means that 12.8% of the IO
1546 took more than 10 msecs, but less than (or equal to) 20 msecs.
1548 After each client has been listed, the group statistics are printed. They
1549 will look like this:
1551 Run status group 0 (all jobs):
1552 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1553 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1555 For each data direction, it prints:
1557 io= Number of megabytes io performed.
1558 aggrb= Aggregate bandwidth of threads in this group.
1559 minb= The minimum average bandwidth a thread saw.
1560 maxb= The maximum average bandwidth a thread saw.
1561 mint= The smallest runtime of the threads in that group.
1562 maxt= The longest runtime of the threads in that group.
1564 And finally, the disk statistics are printed. They will look like this:
1566 Disk stats (read/write):
1567 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1569 Each value is printed for both reads and writes, with reads first. The
1572 ios= Number of ios performed by all groups.
1573 merge= Number of merges io the io scheduler.
1574 ticks= Number of ticks we kept the disk busy.
1575 io_queue= Total time spent in the disk queue.
1576 util= The disk utilization. A value of 100% means we kept the disk
1577 busy constantly, 50% would be a disk idling half of the time.
1579 It is also possible to get fio to dump the current output while it is
1580 running, without terminating the job. To do that, send fio the USR1 signal.
1586 For scripted usage where you typically want to generate tables or graphs
1587 of the results, fio can output the results in a semicolon separated format.
1588 The format is one long line of values, such as:
1590 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%
1591 A description of this job goes here.
1593 The job description (if provided) follows on a second line.
1595 To enable terse output, use the --minimal command line option. The first
1596 value is the version of the terse output format. If the output has to
1597 be changed for some reason, this number will be incremented by 1 to
1598 signify that change.
1600 Split up, the format is as follows:
1602 terse version, fio version, jobname, groupid, error
1604 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1605 Submission latency: min, max, mean, deviation (usec)
1606 Completion latency: min, max, mean, deviation (usec)
1607 Completion latency percentiles: 20 fields (see below)
1608 Total latency: min, max, mean, deviation (usec)
1609 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1611 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1612 Submission latency: min, max, mean, deviation (usec)
1613 Completion latency: min, max, mean, deviation (usec)
1614 Completion latency percentiles: 20 fields (see below)
1615 Total latency: min, max, mean, deviation (usec)
1616 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1617 CPU usage: user, system, context switches, major faults, minor faults
1618 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1619 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1620 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1621 Disk utilization: Disk name, Read ios, write ios,
1622 Read merges, write merges,
1623 Read ticks, write ticks,
1624 Time spent in queue, disk utilization percentage
1625 Additional Info (dependant on continue_on_error, default off): total # errors, first error code
1627 Additional Info (dependant on description being set): Text description
1629 Completion latency percentiles can be a grouping of up to 20 sets, so
1630 for the terse output fio writes all of them. Each field will look like this:
1634 which is the Xth percentile, and the usec latency associated with it.
1636 For disk utilization, all disks used by fio are shown. So for each disk
1637 there will be a disk utilization section.
1640 8.0 Trace file format
1641 ---------------------
1642 There are two trace file format that you can encounter. The older (v1) format
1643 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1644 below in case that you get an old trace and want to understand it.
1646 In any case the trace is a simple text file with a single action per line.
1649 8.1 Trace file format v1
1650 ------------------------
1651 Each line represents a single io action in the following format:
1655 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1657 This format is not supported in Fio versions => 1.20-rc3.
1660 8.2 Trace file format v2
1661 ------------------------
1662 The second version of the trace file format was added in Fio version 1.17.
1663 It allows to access more then one file per trace and has a bigger set of
1664 possible file actions.
1666 The first line of the trace file has to be:
1670 Following this can be lines in two different formats, which are described below.
1672 The file management format:
1676 The filename is given as an absolute path. The action can be one of these:
1678 add Add the given filename to the trace
1679 open Open the file with the given filename. The filename has to have
1680 been added with the add action before.
1681 close Close the file with the given filename. The file has to have been
1685 The file io action format:
1687 filename action offset length
1689 The filename is given as an absolute path, and has to have been added and opened
1690 before it can be used with this format. The offset and length are given in
1691 bytes. The action can be one of these:
1693 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1694 read Read 'length' bytes beginning from 'offset'
1695 write Write 'length' bytes beginning from 'offset'
1696 sync fsync() the file
1697 datasync fdatasync() the file
1698 trim trim the given file from the given 'offset' for 'length' bytes
1701 9.0 CPU idleness profiling
1703 In some cases, we want to understand CPU overhead in a test. For example,
1704 we test patches for the specific goodness of whether they reduce CPU usage.
1705 fio implements a balloon approach to create a thread per CPU that runs at
1706 idle priority, meaning that it only runs when nobody else needs the cpu.
1707 By measuring the amount of work completed by the thread, idleness of each
1708 CPU can be derived accordingly.
1710 An unit work is defined as touching a full page of unsigned characters. Mean
1711 and standard deviation of time to complete an unit work is reported in "unit
1712 work" section. Options can be chosen to report detailed percpu idleness or
1713 overall system idleness by aggregating percpu stats.