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
248 With the above in mind, here follows the complete list of fio job
251 name=str ASCII name of the job. This may be used to override the
252 name printed by fio for this job. Otherwise the job
253 name is used. On the command line this parameter has the
254 special purpose of also signaling the start of a new
257 description=str Text description of the job. Doesn't do anything except
258 dump this text description when this job is run. It's
261 directory=str Prefix filenames with this directory. Used to place files
262 in a different location than "./".
264 filename=str Fio normally makes up a filename based on the job name,
265 thread number, and file number. If you want to share
266 files between threads in a job or several jobs, specify
267 a filename for each of them to override the default. If
268 the ioengine used is 'net', the filename is the host, port,
269 and protocol to use in the format of =host/port/protocol.
270 See ioengine=net for more. If the ioengine is file based, you
271 can specify a number of files by separating the names with a
272 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
273 as the two working files, you would use
274 filename=/dev/sda:/dev/sdb. On Windows, disk devices are accessed
275 as \\.\PhysicalDrive0 for the first device, \\.\PhysicalDrive1
276 for the second etc. If the wanted filename does need to
277 include a colon, then escape that with a '\' character.
278 For instance, if the filename is "/dev/dsk/foo@3,0:c",
279 then you would use filename="/dev/dsk/foo@3,0\:c".
280 '-' is a reserved name, meaning stdin or stdout. Which of the
281 two depends on the read/write direction set.
283 opendir=str Tell fio to recursively add any file it can find in this
284 directory and down the file system tree.
286 lockfile=str Fio defaults to not locking any files before it does
287 IO to them. If a file or file descriptor is shared, fio
288 can serialize IO to that file to make the end result
289 consistent. This is usual for emulating real workloads that
290 share files. The lock modes are:
292 none No locking. The default.
293 exclusive Only one thread/process may do IO,
294 excluding all others.
295 readwrite Read-write locking on the file. Many
296 readers may access the file at the
297 same time, but writes get exclusive
300 The option may be post-fixed with a lock batch number. If
301 set, then each thread/process may do that amount of IOs to
302 the file before giving up the lock. Since lock acquisition is
303 expensive, batching the lock/unlocks will speed up IO.
306 rw=str Type of io pattern. Accepted values are:
308 read Sequential reads
309 write Sequential writes
310 randwrite Random writes
311 randread Random reads
312 rw Sequential mixed reads and writes
313 randrw Random mixed reads and writes
315 For the mixed io types, the default is to split them 50/50.
316 For certain types of io the result may still be skewed a bit,
317 since the speed may be different. It is possible to specify
318 a number of IO's to do before getting a new offset, this is
319 one by appending a ':<nr>' to the end of the string given.
320 For a random read, it would look like 'rw=randread:8' for
321 passing in an offset modifier with a value of 8. See the
322 'rw_sequencer' option.
324 rw_sequencer=str If an offset modifier is given by appending a number to
325 the rw=<str> line, then this option controls how that
326 number modifies the IO offset being generated. Accepted
329 sequential Generate sequential offset
330 identical Generate the same offset
332 'sequential' is only useful for random IO, where fio would
333 normally generate a new random offset for every IO. If you
334 append eg 8 to randread, you would get a new random offset for
335 every 8 IO's. The result would be a seek for only every 8
336 IO's, instead of for every IO. Use rw=randread:8 to specify
337 that. As sequential IO is already sequential, setting
338 'sequential' for that would not result in any differences.
339 'identical' behaves in a similar fashion, except it sends
340 the same offset 8 number of times before generating a new
343 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
344 Storage manufacturers like to use 10^3 or 1000 as a base
345 ten unit instead, for obvious reasons. Allow values are
346 1024 or 1000, with 1024 being the default.
348 randrepeat=bool For random IO workloads, seed the generator in a predictable
349 way so that results are repeatable across repetitions.
351 fallocate=bool By default, fio will use fallocate() to advise the system
352 of the size of the file we are going to write. This can be
353 turned off with fallocate=0. May not be available on all
356 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
357 on what IO patterns it is likely to issue. Sometimes you
358 want to test specific IO patterns without telling the
359 kernel about it, in which case you can disable this option.
360 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
361 IO and POSIX_FADV_RANDOM for random IO.
363 size=int The total size of file io for this job. Fio will run until
364 this many bytes has been transferred, unless runtime is
365 limited by other options (such as 'runtime', for instance).
366 Unless specific nrfiles and filesize options are given,
367 fio will divide this size between the available files
368 specified by the job. If not set, fio will use the full
369 size of the given files or devices. If the the files
370 do not exist, size must be given.
372 filesize=int Individual file sizes. May be a range, in which case fio
373 will select sizes for files at random within the given range
374 and limited to 'size' in total (if that is given). If not
375 given, each created file is the same size.
377 fill_device=bool Sets size to something really large and waits for ENOSPC (no
378 space left on device) as the terminating condition. Only makes
379 sense with sequential write. For a read workload, the mount
380 point will be filled first then IO started on the result.
383 bs=int The block size used for the io units. Defaults to 4k. Values
384 can be given for both read and writes. If a single int is
385 given, it will apply to both. If a second int is specified
386 after a comma, it will apply to writes only. In other words,
387 the format is either bs=read_and_write or bs=read,write.
388 bs=4k,8k will thus use 4k blocks for reads, and 8k blocks
389 for writes. If you only wish to set the write size, you
390 can do so by passing an empty read size - bs=,8k will set
391 8k for writes and leave the read default value.
394 ba=int At what boundary to align random IO offsets. Defaults to
395 the same as 'blocksize' the minimum blocksize given.
396 Minimum alignment is typically 512b for using direct IO,
397 though it usually depends on the hardware block size. This
398 option is mutually exclusive with using a random map for
399 files, so it will turn off that option.
401 blocksize_range=irange
402 bsrange=irange Instead of giving a single block size, specify a range
403 and fio will mix the issued io block sizes. The issued
404 io unit will always be a multiple of the minimum value
405 given (also see bs_unaligned). Applies to both reads and
406 writes, however a second range can be given after a comma.
409 bssplit=str Sometimes you want even finer grained control of the
410 block sizes issued, not just an even split between them.
411 This option allows you to weight various block sizes,
412 so that you are able to define a specific amount of
413 block sizes issued. The format for this option is:
415 bssplit=blocksize/percentage:blocksize/percentage
417 for as many block sizes as needed. So if you want to define
418 a workload that has 50% 64k blocks, 10% 4k blocks, and
419 40% 32k blocks, you would write:
421 bssplit=4k/10:64k/50:32k/40
423 Ordering does not matter. If the percentage is left blank,
424 fio will fill in the remaining values evenly. So a bssplit
425 option like this one:
427 bssplit=4k/50:1k/:32k/
429 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
430 always add up to 100, if bssplit is given a range that adds
431 up to more, it will error out.
433 bssplit also supports giving separate splits to reads and
434 writes. The format is identical to what bs= accepts. You
435 have to separate the read and write parts with a comma. So
436 if you want a workload that has 50% 2k reads and 50% 4k reads,
437 while having 90% 4k writes and 10% 8k writes, you would
440 bssplit=2k/50:4k/50,4k/90,8k/10
443 bs_unaligned If this option is given, any byte size value within bsrange
444 may be used as a block range. This typically wont work with
445 direct IO, as that normally requires sector alignment.
447 zero_buffers If this option is given, fio will init the IO buffers to
448 all zeroes. The default is to fill them with random data.
450 refill_buffers If this option is given, fio will refill the IO buffers
451 on every submit. The default is to only fill it at init
452 time and reuse that data. Only makes sense if zero_buffers
453 isn't specified, naturally. If data verification is enabled,
454 refill_buffers is also automatically enabled.
456 nrfiles=int Number of files to use for this job. Defaults to 1.
458 openfiles=int Number of files to keep open at the same time. Defaults to
459 the same as nrfiles, can be set smaller to limit the number
462 file_service_type=str Defines how fio decides which file from a job to
463 service next. The following types are defined:
465 random Just choose a file at random.
467 roundrobin Round robin over open files. This
470 sequential Finish one file before moving on to
471 the next. Multiple files can still be
472 open depending on 'openfiles'.
474 The string can have a number appended, indicating how
475 often to switch to a new file. So if option random:4 is
476 given, fio will switch to a new random file after 4 ios
479 ioengine=str Defines how the job issues io to the file. The following
482 sync Basic read(2) or write(2) io. lseek(2) is
483 used to position the io location.
485 psync Basic pread(2) or pwrite(2) io.
487 vsync Basic readv(2) or writev(2) IO.
489 libaio Linux native asynchronous io. Note that Linux
490 may only support queued behaviour with
491 non-buffered IO (set direct=1 or buffered=0).
493 posixaio glibc posix asynchronous io.
495 solarisaio Solaris native asynchronous io.
497 windowsaio Windows native asynchronous io.
499 mmap File is memory mapped and data copied
500 to/from using memcpy(3).
502 splice splice(2) is used to transfer the data and
503 vmsplice(2) to transfer data from user
506 syslet-rw Use the syslet system calls to make
507 regular read/write async.
509 sg SCSI generic sg v3 io. May either be
510 synchronous using the SG_IO ioctl, or if
511 the target is an sg character device
512 we use read(2) and write(2) for asynchronous
515 null Doesn't transfer any data, just pretends
516 to. This is mainly used to exercise fio
517 itself and for debugging/testing purposes.
519 net Transfer over the network to given host:port.
520 'filename' must be set appropriately to
521 filename=host/port/protocol regardless of send
522 or receive, if the latter only the port
523 argument is used. 'host' may be an IP address
524 or hostname, port is the port number to be used,
525 and protocol may be 'udp' or 'tcp'. If no
526 protocol is given, TCP is used.
528 netsplice Like net, but uses splice/vmsplice to
529 map data and send/receive.
531 cpuio Doesn't transfer any data, but burns CPU
532 cycles according to the cpuload= and
533 cpucycle= options. Setting cpuload=85
534 will cause that job to do nothing but burn
535 85% of the CPU. In case of SMP machines,
536 use numjobs=<no_of_cpu> to get desired CPU
537 usage, as the cpuload only loads a single
538 CPU at the desired rate.
540 guasi The GUASI IO engine is the Generic Userspace
541 Asyncronous Syscall Interface approach
544 http://www.xmailserver.org/guasi-lib.html
546 for more info on GUASI.
548 external Prefix to specify loading an external
549 IO engine object file. Append the engine
550 filename, eg ioengine=external:/tmp/foo.o
551 to load ioengine foo.o in /tmp.
553 iodepth=int This defines how many io units to keep in flight against
554 the file. The default is 1 for each file defined in this
555 job, can be overridden with a larger value for higher
556 concurrency. Note that increasing iodepth beyond 1 will not
557 affect synchronous ioengines (except for small degress when
558 verify_async is in use). Even async engines my impose OS
559 restrictions causing the desired depth not to be achieved.
560 This may happen on Linux when using libaio and not setting
561 direct=1, since buffered IO is not async on that OS. Keep an
562 eye on the IO depth distribution in the fio output to verify
563 that the achieved depth is as expected. Default: 1.
565 iodepth_batch_submit=int
566 iodepth_batch=int This defines how many pieces of IO to submit at once.
567 It defaults to 1 which means that we submit each IO
568 as soon as it is available, but can be raised to submit
569 bigger batches of IO at the time.
571 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
572 at once. It defaults to 1 which means that we'll ask
573 for a minimum of 1 IO in the retrieval process from
574 the kernel. The IO retrieval will go on until we
575 hit the limit set by iodepth_low. If this variable is
576 set to 0, then fio will always check for completed
577 events before queuing more IO. This helps reduce
578 IO latency, at the cost of more retrieval system calls.
580 iodepth_low=int The low water mark indicating when to start filling
581 the queue again. Defaults to the same as iodepth, meaning
582 that fio will attempt to keep the queue full at all times.
583 If iodepth is set to eg 16 and iodepth_low is set to 4, then
584 after fio has filled the queue of 16 requests, it will let
585 the depth drain down to 4 before starting to fill it again.
587 direct=bool If value is true, use non-buffered io. This is usually
590 buffered=bool If value is true, use buffered io. This is the opposite
591 of the 'direct' option. Defaults to true.
593 offset=int Start io at the given offset in the file. The data before
594 the given offset will not be touched. This effectively
595 caps the file size at real_size - offset.
597 fsync=int If writing to a file, issue a sync of the dirty data
598 for every number of blocks given. For example, if you give
599 32 as a parameter, fio will sync the file for every 32
600 writes issued. If fio is using non-buffered io, we may
601 not sync the file. The exception is the sg io engine, which
602 synchronizes the disk cache anyway.
604 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
606 In FreeBSD there is no fdatasync(), this falls back to
609 sync_file_range=str:val Use sync_file_range() for every 'val' number of
610 write operations. Fio will track range of writes that
611 have happened since the last sync_file_range() call. 'str'
612 can currently be one or more of:
614 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
615 write SYNC_FILE_RANGE_WRITE
616 wait_after SYNC_FILE_RANGE_WAIT_AFTER
618 So if you do sync_file_range=wait_before,write:8, fio would
619 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
620 every 8 writes. Also see the sync_file_range(2) man page.
621 This option is Linux specific.
623 overwrite=bool If true, writes to a file will always overwrite existing
624 data. If the file doesn't already exist, it will be
625 created before the write phase begins. If the file exists
626 and is large enough for the specified write phase, nothing
629 end_fsync=bool If true, fsync file contents when the job exits.
631 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
632 This differs from end_fsync in that it will happen on every
633 file close, not just at the end of the job.
635 rwmixread=int How large a percentage of the mix should be reads.
637 rwmixwrite=int How large a percentage of the mix should be writes. If both
638 rwmixread and rwmixwrite is given and the values do not add
639 up to 100%, the latter of the two will be used to override
640 the first. This may interfere with a given rate setting,
641 if fio is asked to limit reads or writes to a certain rate.
642 If that is the case, then the distribution may be skewed.
644 norandommap Normally fio will cover every block of the file when doing
645 random IO. If this option is given, fio will just get a
646 new random offset without looking at past io history. This
647 means that some blocks may not be read or written, and that
648 some blocks may be read/written more than once. This option
649 is mutually exclusive with verify= if and only if multiple
650 blocksizes (via bsrange=) are used, since fio only tracks
651 complete rewrites of blocks.
653 softrandommap See norandommap. If fio runs with the random block map enabled
654 and it fails to allocate the map, if this option is set it
655 will continue without a random block map. As coverage will
656 not be as complete as with random maps, this option is
659 nice=int Run the job with the given nice value. See man nice(2).
661 prio=int Set the io priority value of this job. Linux limits us to
662 a positive value between 0 and 7, with 0 being the highest.
665 prioclass=int Set the io priority class. See man ionice(1).
667 thinktime=int Stall the job x microseconds after an io has completed before
668 issuing the next. May be used to simulate processing being
669 done by an application. See thinktime_blocks and
673 Only valid if thinktime is set - pretend to spend CPU time
674 doing something with the data received, before falling back
675 to sleeping for the rest of the period specified by
679 Only valid if thinktime is set - control how many blocks
680 to issue, before waiting 'thinktime' usecs. If not set,
681 defaults to 1 which will make fio wait 'thinktime' usecs
684 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
685 the normal suffix rules apply. You can use rate=500k to limit
686 reads and writes to 500k each, or you can specify read and
687 writes separately. Using rate=1m,500k would limit reads to
688 1MB/sec and writes to 500KB/sec. Capping only reads or
689 writes can be done with rate=,500k or rate=500k,. The former
690 will only limit writes (to 500KB/sec), the latter will only
693 ratemin=int Tell fio to do whatever it can to maintain at least this
694 bandwidth. Failing to meet this requirement, will cause
695 the job to exit. The same format as rate is used for
696 read vs write separation.
698 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
699 as rate, just specified independently of bandwidth. If the
700 job is given a block size range instead of a fixed value,
701 the smallest block size is used as the metric. The same format
702 as rate is used for read vs write seperation.
704 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
705 the job to exit. The same format as rate is used for read vs
708 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
711 cpumask=int Set the CPU affinity of this job. The parameter given is a
712 bitmask of allowed CPU's the job may run on. So if you want
713 the allowed CPUs to be 1 and 5, you would pass the decimal
714 value of (1 << 1 | 1 << 5), or 34. See man
715 sched_setaffinity(2). This may not work on all supported
716 operating systems or kernel versions. This option doesn't
717 work well for a higher CPU count than what you can store in
718 an integer mask, so it can only control cpus 1-32. For
719 boxes with larger CPU counts, use cpus_allowed.
721 cpus_allowed=str Controls the same options as cpumask, but it allows a text
722 setting of the permitted CPUs instead. So to use CPUs 1 and
723 5, you would specify cpus_allowed=1,5. This options also
724 allows a range of CPUs. Say you wanted a binding to CPUs
725 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
727 startdelay=time Start this job the specified number of seconds after fio
728 has started. Only useful if the job file contains several
729 jobs, and you want to delay starting some jobs to a certain
732 runtime=time Tell fio to terminate processing after the specified number
733 of seconds. It can be quite hard to determine for how long
734 a specified job will run, so this parameter is handy to
735 cap the total runtime to a given time.
737 time_based If set, fio will run for the duration of the runtime
738 specified even if the file(s) are completely read or
739 written. It will simply loop over the same workload
740 as many times as the runtime allows.
742 ramp_time=time If set, fio will run the specified workload for this amount
743 of time before logging any performance numbers. Useful for
744 letting performance settle before logging results, thus
745 minimizing the runtime required for stable results. Note
746 that the ramp_time is considered lead in time for a job,
747 thus it will increase the total runtime if a special timeout
748 or runtime is specified.
750 invalidate=bool Invalidate the buffer/page cache parts for this file prior
751 to starting io. Defaults to true.
753 sync=bool Use sync io for buffered writes. For the majority of the
754 io engines, this means using O_SYNC.
757 mem=str Fio can use various types of memory as the io unit buffer.
758 The allowed values are:
760 malloc Use memory from malloc(3) as the buffers.
762 shm Use shared memory as the buffers. Allocated
765 shmhuge Same as shm, but use huge pages as backing.
767 mmap Use mmap to allocate buffers. May either be
768 anonymous memory, or can be file backed if
769 a filename is given after the option. The
770 format is mem=mmap:/path/to/file.
772 mmaphuge Use a memory mapped huge file as the buffer
773 backing. Append filename after mmaphuge, ala
774 mem=mmaphuge:/hugetlbfs/file
776 The area allocated is a function of the maximum allowed
777 bs size for the job, multiplied by the io depth given. Note
778 that for shmhuge and mmaphuge to work, the system must have
779 free huge pages allocated. This can normally be checked
780 and set by reading/writing /proc/sys/vm/nr_hugepages on a
781 Linux system. Fio assumes a huge page is 4MB in size. So
782 to calculate the number of huge pages you need for a given
783 job file, add up the io depth of all jobs (normally one unless
784 iodepth= is used) and multiply by the maximum bs set. Then
785 divide that number by the huge page size. You can see the
786 size of the huge pages in /proc/meminfo. If no huge pages
787 are allocated by having a non-zero number in nr_hugepages,
788 using mmaphuge or shmhuge will fail. Also see hugepage-size.
790 mmaphuge also needs to have hugetlbfs mounted and the file
791 location should point there. So if it's mounted in /huge,
792 you would use mem=mmaphuge:/huge/somefile.
794 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
795 Note that the given alignment is applied to the first IO unit
796 buffer, if using iodepth the alignment of the following buffers
797 are given by the bs used. In other words, if using a bs that is
798 a multiple of the page sized in the system, all buffers will
799 be aligned to this value. If using a bs that is not page
800 aligned, the alignment of subsequent IO memory buffers is the
801 sum of the iomem_align and bs used.
804 Defines the size of a huge page. Must at least be equal
805 to the system setting, see /proc/meminfo. Defaults to 4MB.
806 Should probably always be a multiple of megabytes, so using
807 hugepage-size=Xm is the preferred way to set this to avoid
808 setting a non-pow-2 bad value.
810 exitall When one job finishes, terminate the rest. The default is
811 to wait for each job to finish, sometimes that is not the
814 bwavgtime=int Average the calculated bandwidth over the given time. Value
815 is specified in milliseconds.
817 create_serialize=bool If true, serialize the file creating for the jobs.
818 This may be handy to avoid interleaving of data
819 files, which may greatly depend on the filesystem
820 used and even the number of processors in the system.
822 create_fsync=bool fsync the data file after creation. This is the
825 create_on_open=bool Don't pre-setup the files for IO, just create open()
826 when it's time to do IO to that file.
828 pre_read=bool If this is given, files will be pre-read into memory before
829 starting the given IO operation. This will also clear
830 the 'invalidate' flag, since it is pointless to pre-read
831 and then drop the cache. This will only work for IO engines
832 that are seekable, since they allow you to read the same data
833 multiple times. Thus it will not work on eg network or splice
836 unlink=bool Unlink the job files when done. Not the default, as repeated
837 runs of that job would then waste time recreating the file
840 loops=int Run the specified number of iterations of this job. Used
841 to repeat the same workload a given number of times. Defaults
844 do_verify=bool Run the verify phase after a write phase. Only makes sense if
845 verify is set. Defaults to 1.
847 verify=str If writing to a file, fio can verify the file contents
848 after each iteration of the job. The allowed values are:
850 md5 Use an md5 sum of the data area and store
851 it in the header of each block.
853 crc64 Use an experimental crc64 sum of the data
854 area and store it in the header of each
857 crc32c Use a crc32c sum of the data area and store
858 it in the header of each block.
860 crc32c-intel Use hardware assisted crc32c calcuation
861 provided on SSE4.2 enabled processors. Falls
862 back to regular software crc32c, if not
863 supported by the system.
865 crc32 Use a crc32 sum of the data area and store
866 it in the header of each block.
868 crc16 Use a crc16 sum of the data area and store
869 it in the header of each block.
871 crc7 Use a crc7 sum of the data area and store
872 it in the header of each block.
874 sha512 Use sha512 as the checksum function.
876 sha256 Use sha256 as the checksum function.
878 sha1 Use optimized sha1 as the checksum function.
880 meta Write extra information about each io
881 (timestamp, block number etc.). The block
882 number is verified. See also verify_pattern.
884 null Only pretend to verify. Useful for testing
885 internals with ioengine=null, not for much
888 This option can be used for repeated burn-in tests of a
889 system to make sure that the written data is also
890 correctly read back. If the data direction given is
891 a read or random read, fio will assume that it should
892 verify a previously written file. If the data direction
893 includes any form of write, the verify will be of the
896 verifysort=bool If set, fio will sort written verify blocks when it deems
897 it faster to read them back in a sorted manner. This is
898 often the case when overwriting an existing file, since
899 the blocks are already laid out in the file system. You
900 can ignore this option unless doing huge amounts of really
901 fast IO where the red-black tree sorting CPU time becomes
904 verify_offset=int Swap the verification header with data somewhere else
905 in the block before writing. Its swapped back before
908 verify_interval=int Write the verification header at a finer granularity
909 than the blocksize. It will be written for chunks the
910 size of header_interval. blocksize should divide this
913 verify_pattern=str If set, fio will fill the io buffers with this
914 pattern. Fio defaults to filling with totally random
915 bytes, but sometimes it's interesting to fill with a known
916 pattern for io verification purposes. Depending on the
917 width of the pattern, fio will fill 1/2/3/4 bytes of the
918 buffer at the time(it can be either a decimal or a hex number).
919 The verify_pattern if larger than a 32-bit quantity has to
920 be a hex number that starts with either "0x" or "0X". Use
923 verify_fatal=bool Normally fio will keep checking the entire contents
924 before quitting on a block verification failure. If this
925 option is set, fio will exit the job on the first observed
928 verify_async=int Fio will normally verify IO inline from the submitting
929 thread. This option takes an integer describing how many
930 async offload threads to create for IO verification instead,
931 causing fio to offload the duty of verifying IO contents
932 to one or more separate threads. If using this offload
933 option, even sync IO engines can benefit from using an
934 iodepth setting higher than 1, as it allows them to have
935 IO in flight while verifies are running.
937 verify_async_cpus=str Tell fio to set the given CPU affinity on the
938 async IO verification threads. See cpus_allowed for the
941 verify_backlog=int Fio will normally verify the written contents of a
942 job that utilizes verify once that job has completed. In
943 other words, everything is written then everything is read
944 back and verified. You may want to verify continually
945 instead for a variety of reasons. Fio stores the meta data
946 associated with an IO block in memory, so for large
947 verify workloads, quite a bit of memory would be used up
948 holding this meta data. If this option is enabled, fio
949 will write only N blocks before verifying these blocks.
951 will verify the previously written blocks before continuing
954 verify_backlog_batch=int Control how many blocks fio will verify
955 if verify_backlog is set. If not set, will default to
956 the value of verify_backlog (meaning the entire queue
957 is read back and verified). If verify_backlog_batch is
958 less than verify_backlog then not all blocks will be verified,
959 if verify_backlog_batch is larger than verify_backlog, some
960 blocks will be verified more than once.
962 stonewall Wait for preceeding jobs in the job file to exit, before
963 starting this one. Can be used to insert serialization
964 points in the job file. A stone wall also implies starting
965 a new reporting group.
967 new_group Start a new reporting group. If this option isn't given,
968 jobs in a file will be part of the same reporting group
969 unless separated by a stone wall (or if it's a group
970 by itself, with the numjobs option).
972 numjobs=int Create the specified number of clones of this job. May be
973 used to setup a larger number of threads/processes doing
974 the same thing. We regard that grouping of jobs as a
977 group_reporting If 'numjobs' is set, it may be interesting to display
978 statistics for the group as a whole instead of for each
979 individual job. This is especially true of 'numjobs' is
980 large, looking at individual thread/process output quickly
981 becomes unwieldy. If 'group_reporting' is specified, fio
982 will show the final report per-group instead of per-job.
984 thread fio defaults to forking jobs, however if this option is
985 given, fio will use pthread_create(3) to create threads
988 zonesize=int Divide a file into zones of the specified size. See zoneskip.
990 zoneskip=int Skip the specified number of bytes when zonesize data has
991 been read. The two zone options can be used to only do
992 io on zones of a file.
994 write_iolog=str Write the issued io patterns to the specified file. See
997 read_iolog=str Open an iolog with the specified file name and replay the
998 io patterns it contains. This can be used to store a
999 workload and replay it sometime later. The iolog given
1000 may also be a blktrace binary file, which allows fio
1001 to replay a workload captured by blktrace. See blktrace
1002 for how to capture such logging data. For blktrace replay,
1003 the file needs to be turned into a blkparse binary data
1004 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1006 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1007 is to attempt to respect the time stamps within the log and
1008 replay them with the appropriate delay between IOPS. By
1009 setting this variable fio will not respect the timestamps and
1010 attempt to replay them as fast as possible while still
1011 respecting ordering. The result is the same I/O pattern to a
1012 given device, but different timings.
1014 replay_redirect=str While replaying I/O patterns using read_iolog the
1015 default behavior is to replay the IOPS onto the major/minor
1016 device that each IOP was recorded from. This is sometimes
1017 undesireable because on a different machine those major/minor
1018 numbers can map to a different device. Changing hardware on
1019 the same system can also result in a different major/minor
1020 mapping. Replay_redirect causes all IOPS to be replayed onto
1021 the single specified device regardless of the device it was
1022 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1023 IO in the blktrace to be replayed onto /dev/sdc. This means
1024 multiple devices will be replayed onto a single, if the trace
1025 contains multiple devices. If you want multiple devices to be
1026 replayed concurrently to multiple redirected devices you must
1027 blkparse your trace into separate traces and replay them with
1028 independent fio invocations. Unfortuantely this also breaks
1029 the strict time ordering between multiple device accesses.
1031 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1032 file. Can be used to store data of the bandwidth of the
1033 jobs in their lifetime. The included fio_generate_plots
1034 script uses gnuplot to turn these text files into nice
1035 graphs. See write_log_log for behaviour of given
1036 filename. For this option, the postfix is _bw.log.
1038 write_lat_log=str Same as write_bw_log, except that this option stores io
1039 submission, completion, and total latencies instead. If no
1040 filename is given with this option, the default filename of
1041 "jobname_type.log" is used. Even if the filename is given,
1042 fio will still append the type of log. So if one specifies
1046 The actual log names will be foo_slat.log, foo_slat.log,
1047 and foo_lat.log. This helps fio_generate_plot fine the logs
1050 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1051 potentially be used instead of removing memory or booting
1052 with less memory to simulate a smaller amount of memory.
1054 exec_prerun=str Before running this job, issue the command specified
1057 exec_postrun=str After the job completes, issue the command specified
1060 ioscheduler=str Attempt to switch the device hosting the file to the specified
1061 io scheduler before running.
1063 cpuload=int If the job is a CPU cycle eater, attempt to use the specified
1064 percentage of CPU cycles.
1066 cpuchunks=int If the job is a CPU cycle eater, split the load into
1067 cycles of the given time. In microseconds.
1069 disk_util=bool Generate disk utilization statistics, if the platform
1070 supports it. Defaults to on.
1072 disable_lat=bool Disable measurements of total latency numbers. Useful
1073 only for cutting back the number of calls to gettimeofday,
1074 as that does impact performance at really high IOPS rates.
1075 Note that to really get rid of a large amount of these
1076 calls, this option must be used with disable_slat and
1079 disable_clat=bool Disable measurements of completion latency numbers. See
1082 disable_slat=bool Disable measurements of submission latency numbers. See
1085 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1088 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1089 (disable_clat, disable_slat, disable_bw) plus reduce
1090 precision of the timeout somewhat to really shrink
1091 the gettimeofday() call count. With this option enabled,
1092 we only do about 0.4% of the gtod() calls we would have
1093 done if all time keeping was enabled.
1095 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1096 execution to just getting the current time. Fio (and
1097 databases, for instance) are very intensive on gettimeofday()
1098 calls. With this option, you can set one CPU aside for
1099 doing nothing but logging current time to a shared memory
1100 location. Then the other threads/processes that run IO
1101 workloads need only copy that segment, instead of entering
1102 the kernel with a gettimeofday() call. The CPU set aside
1103 for doing these time calls will be excluded from other
1104 uses. Fio will manually clear it from the CPU mask of other
1107 continue_on_error=bool Normally fio will exit the job on the first observed
1108 failure. If this option is set, fio will continue the job when
1109 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1110 is exceeded or the I/O size specified is completed. If this
1111 option is used, there are two more stats that are appended,
1112 the total error count and the first error. The error field
1113 given in the stats is the first error that was hit during the
1116 cgroup=str Add job to this control group. If it doesn't exist, it will
1117 be created. The system must have a mounted cgroup blkio
1118 mount point for this to work. If your system doesn't have it
1119 mounted, you can do so with:
1121 # mount -t cgroup -o blkio none /cgroup
1123 cgroup_weight=int Set the weight of the cgroup to this value. See
1124 the documentation that comes with the kernel, allowed values
1125 are in the range of 100..1000.
1127 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1128 the job completion. To override this behavior and to leave
1129 cgroups around after the job completion, set cgroup_nodelete=1.
1130 This can be useful if one wants to inspect various cgroup
1131 files after job completion. Default: false
1133 uid=int Instead of running as the invoking user, set the user ID to
1134 this value before the thread/process does any work.
1136 gid=int Set group ID, see uid.
1138 6.0 Interpreting the output
1139 ---------------------------
1141 fio spits out a lot of output. While running, fio will display the
1142 status of the jobs created. An example of that would be:
1144 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1146 The characters inside the square brackets denote the current status of
1147 each thread. The possible values (in typical life cycle order) are:
1151 P Thread setup, but not started.
1153 I Thread initialized, waiting.
1154 p Thread running pre-reading file(s).
1155 R Running, doing sequential reads.
1156 r Running, doing random reads.
1157 W Running, doing sequential writes.
1158 w Running, doing random writes.
1159 M Running, doing mixed sequential reads/writes.
1160 m Running, doing mixed random reads/writes.
1161 F Running, currently waiting for fsync()
1162 V Running, doing verification of written data.
1163 E Thread exited, not reaped by main thread yet.
1166 The other values are fairly self explanatory - number of threads
1167 currently running and doing io, rate of io since last check (read speed
1168 listed first, then write speed), and the estimated completion percentage
1169 and time for the running group. It's impossible to estimate runtime of
1170 the following groups (if any).
1172 When fio is done (or interrupted by ctrl-c), it will show the data for
1173 each thread, group of threads, and disks in that order. For each data
1174 direction, the output looks like:
1176 Client1 (g=0): err= 0:
1177 write: io= 32MB, bw= 666KB/s, runt= 50320msec
1178 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1179 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1180 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1181 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1182 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1183 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1184 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1185 issued r/w: total=0/32768, short=0/0
1186 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1187 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1189 The client number is printed, along with the group id and error of that
1190 thread. Below is the io statistics, here for writes. In the order listed,
1193 io= Number of megabytes io performed
1194 bw= Average bandwidth rate
1195 runt= The runtime of that thread
1196 slat= Submission latency (avg being the average, stdev being the
1197 standard deviation). This is the time it took to submit
1198 the io. For sync io, the slat is really the completion
1199 latency, since queue/complete is one operation there. This
1200 value can be in milliseconds or microseconds, fio will choose
1201 the most appropriate base and print that. In the example
1202 above, milliseconds is the best scale.
1203 clat= Completion latency. Same names as slat, this denotes the
1204 time from submission to completion of the io pieces. For
1205 sync io, clat will usually be equal (or very close) to 0,
1206 as the time from submit to complete is basically just
1207 CPU time (io has already been done, see slat explanation).
1208 bw= Bandwidth. Same names as the xlat stats, but also includes
1209 an approximate percentage of total aggregate bandwidth
1210 this thread received in this group. This last value is
1211 only really useful if the threads in this group are on the
1212 same disk, since they are then competing for disk access.
1213 cpu= CPU usage. User and system time, along with the number
1214 of context switches this thread went through, usage of
1215 system and user time, and finally the number of major
1216 and minor page faults.
1217 IO depths= The distribution of io depths over the job life time. The
1218 numbers are divided into powers of 2, so for example the
1219 16= entries includes depths up to that value but higher
1220 than the previous entry. In other words, it covers the
1221 range from 16 to 31.
1222 IO submit= How many pieces of IO were submitting in a single submit
1223 call. Each entry denotes that amount and below, until
1224 the previous entry - eg, 8=100% mean that we submitted
1225 anywhere in between 5-8 ios per submit call.
1226 IO complete= Like the above submit number, but for completions instead.
1227 IO issued= The number of read/write requests issued, and how many
1229 IO latencies= The distribution of IO completion latencies. This is the
1230 time from when IO leaves fio and when it gets completed.
1231 The numbers follow the same pattern as the IO depths,
1232 meaning that 2=1.6% means that 1.6% of the IO completed
1233 within 2 msecs, 20=12.8% means that 12.8% of the IO
1234 took more than 10 msecs, but less than (or equal to) 20 msecs.
1236 After each client has been listed, the group statistics are printed. They
1237 will look like this:
1239 Run status group 0 (all jobs):
1240 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1241 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1243 For each data direction, it prints:
1245 io= Number of megabytes io performed.
1246 aggrb= Aggregate bandwidth of threads in this group.
1247 minb= The minimum average bandwidth a thread saw.
1248 maxb= The maximum average bandwidth a thread saw.
1249 mint= The smallest runtime of the threads in that group.
1250 maxt= The longest runtime of the threads in that group.
1252 And finally, the disk statistics are printed. They will look like this:
1254 Disk stats (read/write):
1255 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1257 Each value is printed for both reads and writes, with reads first. The
1260 ios= Number of ios performed by all groups.
1261 merge= Number of merges io the io scheduler.
1262 ticks= Number of ticks we kept the disk busy.
1263 io_queue= Total time spent in the disk queue.
1264 util= The disk utilization. A value of 100% means we kept the disk
1265 busy constantly, 50% would be a disk idling half of the time.
1271 For scripted usage where you typically want to generate tables or graphs
1272 of the results, fio can output the results in a semicolon separated format.
1273 The format is one long line of values, such as:
1275 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%
1276 A description of this job goes here.
1278 The job description (if provided) follows on a second line.
1280 To enable terse output, use the --minimal command line option. The first
1281 value is the version of the terse output format. If the output has to
1282 be changed for some reason, this number will be incremented by 1 to
1283 signify that change.
1285 Split up, the format is as follows:
1287 version, jobname, groupid, error
1289 KB IO, bandwidth (KB/sec), runtime (msec)
1290 Submission latency: min, max, mean, deviation
1291 Completion latency: min, max, mean, deviation
1292 Total latency: min, max, mean, deviation
1293 Bw: min, max, aggregate percentage of total, mean, deviation
1295 KB IO, bandwidth (KB/sec), runtime (msec)
1296 Submission latency: min, max, mean, deviation
1297 Completion latency: min, max, mean, deviation
1298 Total latency: min, max, mean, deviation
1299 Bw: min, max, aggregate percentage of total, mean, deviation
1300 CPU usage: user, system, context switches, major faults, minor faults
1301 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1302 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1303 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1304 Additional Info (dependant on continue_on_error, default off): total # errors, first error code
1306 Additional Info (dependant on description being set): Text description