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 fio also supports environment variable expansion in job files. Any
162 substring of the form "${VARNAME}" as part of an option value (in other
163 words, on the right of the `='), will be expanded to the value of the
164 environment variable called VARNAME. If no such environment variable
165 is defined, or VARNAME is the empty string, the empty string will be
168 As an example, let's look at a sample fio invocation and job file:
170 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
172 ; -- start job file --
179 This will expand to the following equivalent job file at runtime:
181 ; -- start job file --
188 fio ships with a few example job files, you can also look there for
192 5.0 Detailed list of parameters
193 -------------------------------
195 This section describes in details each parameter associated with a job.
196 Some parameters take an option of a given type, such as an integer or
197 a string. The following types are used:
199 str String. This is a sequence of alpha characters.
200 time Integer with possible time suffix. In seconds unless otherwise
201 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
203 int SI integer. A whole number value, which may contain a suffix
204 describing the base of the number. Accepted suffixes are k/m/g/t/p,
205 meaning kilo, mega, giga, tera, and peta. The suffix is not case
206 sensitive. So if you want to specify 4096, you could either write
207 out '4096' or just give 4k. The suffixes signify base 2 values, so
208 1024 is 1k and 1024k is 1m and so on. If the option accepts an upper
209 and lower range, use a colon ':' or minus '-' to separate such values.
210 May also include a prefix to indicate numbers base. If 0x is used,
211 the number is assumed to be hexadecimal. See irange.
212 bool Boolean. Usually parsed as an integer, however only defined for
213 true and false (1 and 0).
214 irange Integer range with suffix. Allows value range to be given, such
215 as 1024-4096. A colon may also be used as the separator, eg
216 1k:4k. If the option allows two sets of ranges, they can be
217 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
220 With the above in mind, here follows the complete list of fio job
223 name=str ASCII name of the job. This may be used to override the
224 name printed by fio for this job. Otherwise the job
225 name is used. On the command line this parameter has the
226 special purpose of also signaling the start of a new
229 description=str Text description of the job. Doesn't do anything except
230 dump this text description when this job is run. It's
233 directory=str Prefix filenames with this directory. Used to place files
234 in a different location than "./".
236 filename=str Fio normally makes up a filename based on the job name,
237 thread number, and file number. If you want to share
238 files between threads in a job or several jobs, specify
239 a filename for each of them to override the default. If
240 the ioengine used is 'net', the filename is the host, port,
241 and protocol to use in the format of =host/port/protocol.
242 See ioengine=net for more. If the ioengine is file based, you
243 can specify a number of files by separating the names with a
244 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
245 as the two working files, you would use
246 filename=/dev/sda:/dev/sdb. If the wanted filename does need to
247 include a colon, then escape that with a '\' character. For
248 instance, if the filename is "/dev/dsk/foo@3,0:c", then you would
249 use filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name,
250 meaning stdin or stdout. Which of the two depends on the read/write
253 opendir=str Tell fio to recursively add any file it can find in this
254 directory and down the file system tree.
256 lockfile=str Fio defaults to not locking any files before it does
257 IO to them. If a file or file descriptor is shared, fio
258 can serialize IO to that file to make the end result
259 consistent. This is usual for emulating real workloads that
260 share files. The lock modes are:
262 none No locking. The default.
263 exclusive Only one thread/process may do IO,
264 excluding all others.
265 readwrite Read-write locking on the file. Many
266 readers may access the file at the
267 same time, but writes get exclusive
270 The option may be post-fixed with a lock batch number. If
271 set, then each thread/process may do that amount of IOs to
272 the file before giving up the lock. Since lock acquisition is
273 expensive, batching the lock/unlocks will speed up IO.
276 rw=str Type of io pattern. Accepted values are:
278 read Sequential reads
279 write Sequential writes
280 randwrite Random writes
281 randread Random reads
282 rw Sequential mixed reads and writes
283 randrw Random mixed reads and writes
285 For the mixed io types, the default is to split them 50/50.
286 For certain types of io the result may still be skewed a bit,
287 since the speed may be different. It is possible to specify
288 a number of IO's to do before getting a new offset - this
289 is only useful for random IO, where fio would normally
290 generate a new random offset for every IO. If you append
291 eg 8 to randread, you would get a new random offset for
292 every 8 IO's. The result would be a seek for only every 8
293 IO's, instead of for every IO. Use rw=randread:8 to specify
296 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
297 Storage manufacturers like to use 10^3 or 1000 as a base
298 ten unit instead, for obvious reasons. Allow values are
299 1024 or 1000, with 1024 being the default.
301 randrepeat=bool For random IO workloads, seed the generator in a predictable
302 way so that results are repeatable across repetitions.
304 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
305 on what IO patterns it is likely to issue. Sometimes you
306 want to test specific IO patterns without telling the
307 kernel about it, in which case you can disable this option.
308 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
309 IO and POSIX_FADV_RANDOM for random IO.
311 size=int The total size of file io for this job. Fio will run until
312 this many bytes has been transferred, unless runtime is
313 limited by other options (such as 'runtime', for instance).
314 Unless specific nrfiles and filesize options are given,
315 fio will divide this size between the available files
316 specified by the job.
318 filesize=int Individual file sizes. May be a range, in which case fio
319 will select sizes for files at random within the given range
320 and limited to 'size' in total (if that is given). If not
321 given, each created file is the same size.
323 fill_device=bool Sets size to something really large and waits for ENOSPC (no
324 space left on device) as the terminating condition. Only makes
325 sense with sequential write. For a read workload, the mount
326 point will be filled first then IO started on the result.
329 bs=int The block size used for the io units. Defaults to 4k. Values
330 can be given for both read and writes. If a single int is
331 given, it will apply to both. If a second int is specified
332 after a comma, it will apply to writes only. In other words,
333 the format is either bs=read_and_write or bs=read,write.
334 bs=4k,8k will thus use 4k blocks for reads, and 8k blocks
335 for writes. If you only wish to set the write size, you
336 can do so by passing an empty read size - bs=,8k will set
337 8k for writes and leave the read default value.
340 ba=int At what boundary to align random IO offsets. Defaults to
341 the same as 'blocksize' the minimum blocksize given.
342 Minimum alignment is typically 512b for using direct IO,
343 though it usually depends on the hardware block size. This
344 option is mutually exclusive with using a random map for
345 files, so it will turn off that option.
347 blocksize_range=irange
348 bsrange=irange Instead of giving a single block size, specify a range
349 and fio will mix the issued io block sizes. The issued
350 io unit will always be a multiple of the minimum value
351 given (also see bs_unaligned). Applies to both reads and
352 writes, however a second range can be given after a comma.
355 bssplit=str Sometimes you want even finer grained control of the
356 block sizes issued, not just an even split between them.
357 This option allows you to weight various block sizes,
358 so that you are able to define a specific amount of
359 block sizes issued. The format for this option is:
361 bssplit=blocksize/percentage:blocksize/percentage
363 for as many block sizes as needed. So if you want to define
364 a workload that has 50% 64k blocks, 10% 4k blocks, and
365 40% 32k blocks, you would write:
367 bssplit=4k/10:64k/50:32k/40
369 Ordering does not matter. If the percentage is left blank,
370 fio will fill in the remaining values evenly. So a bssplit
371 option like this one:
373 bssplit=4k/50:1k/:32k/
375 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
376 always add up to 100, if bssplit is given a range that adds
377 up to more, it will error out.
379 bssplit also supports giving separate splits to reads and
380 writes. The format is identical to what bs= accepts. You
381 have to separate the read and write parts with a comma. So
382 if you want a workload that has 50% 2k reads and 50% 4k reads,
383 while having 90% 4k writes and 10% 8k writes, you would
386 bssplit=2k/50:4k/50,4k/90,8k/10
389 bs_unaligned If this option is given, any byte size value within bsrange
390 may be used as a block range. This typically wont work with
391 direct IO, as that normally requires sector alignment.
393 zero_buffers If this option is given, fio will init the IO buffers to
394 all zeroes. The default is to fill them with random data.
396 refill_buffers If this option is given, fio will refill the IO buffers
397 on every submit. The default is to only fill it at init
398 time and reuse that data. Only makes sense if zero_buffers
399 isn't specified, naturally. If data verification is enabled,
400 refill_buffers is also automatically enabled.
402 nrfiles=int Number of files to use for this job. Defaults to 1.
404 openfiles=int Number of files to keep open at the same time. Defaults to
405 the same as nrfiles, can be set smaller to limit the number
408 file_service_type=str Defines how fio decides which file from a job to
409 service next. The following types are defined:
411 random Just choose a file at random.
413 roundrobin Round robin over open files. This
416 sequential Finish one file before moving on to
417 the next. Multiple files can still be
418 open depending on 'openfiles'.
420 The string can have a number appended, indicating how
421 often to switch to a new file. So if option random:4 is
422 given, fio will switch to a new random file after 4 ios
425 ioengine=str Defines how the job issues io to the file. The following
428 sync Basic read(2) or write(2) io. lseek(2) is
429 used to position the io location.
431 psync Basic pread(2) or pwrite(2) io.
433 vsync Basic readv(2) or writev(2) IO.
435 libaio Linux native asynchronous io. Note that Linux
436 may only support queued behaviour with
437 non-buffered IO (set direct=1 or buffered=0).
439 posixaio glibc posix asynchronous io.
441 solarisaio Solaris native asynchronous io.
443 mmap File is memory mapped and data copied
444 to/from using memcpy(3).
446 splice splice(2) is used to transfer the data and
447 vmsplice(2) to transfer data from user
450 syslet-rw Use the syslet system calls to make
451 regular read/write async.
453 sg SCSI generic sg v3 io. May either be
454 synchronous using the SG_IO ioctl, or if
455 the target is an sg character device
456 we use read(2) and write(2) for asynchronous
459 null Doesn't transfer any data, just pretends
460 to. This is mainly used to exercise fio
461 itself and for debugging/testing purposes.
463 net Transfer over the network to given host:port.
464 'filename' must be set appropriately to
465 filename=host/port/protocol regardless of send
466 or receive, if the latter only the port
467 argument is used. 'host' may be an IP address
468 or hostname, port is the port number to be used,
469 and protocol may be 'udp' or 'tcp'. If no
470 protocol is given, TCP is used.
472 netsplice Like net, but uses splice/vmsplice to
473 map data and send/receive.
475 cpuio Doesn't transfer any data, but burns CPU
476 cycles according to the cpuload= and
477 cpucycle= options. Setting cpuload=85
478 will cause that job to do nothing but burn
479 85% of the CPU. In case of SMP machines,
480 use numjobs=<no_of_cpu> to get desired CPU
481 usage, as the cpuload only loads a single
482 CPU at the desired rate.
484 guasi The GUASI IO engine is the Generic Userspace
485 Asyncronous Syscall Interface approach
488 http://www.xmailserver.org/guasi-lib.html
490 for more info on GUASI.
492 external Prefix to specify loading an external
493 IO engine object file. Append the engine
494 filename, eg ioengine=external:/tmp/foo.o
495 to load ioengine foo.o in /tmp.
497 iodepth=int This defines how many io units to keep in flight against
498 the file. The default is 1 for each file defined in this
499 job, can be overridden with a larger value for higher
502 iodepth_batch_submit=int
503 iodepth_batch=int This defines how many pieces of IO to submit at once.
504 It defaults to 1 which means that we submit each IO
505 as soon as it is available, but can be raised to submit
506 bigger batches of IO at the time.
508 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
509 at once. It defaults to 1 which means that we'll ask
510 for a minimum of 1 IO in the retrieval process from
511 the kernel. The IO retrieval will go on until we
512 hit the limit set by iodepth_low. If this variable is
513 set to 0, then fio will always check for completed
514 events before queuing more IO. This helps reduce
515 IO latency, at the cost of more retrieval system calls.
517 iodepth_low=int The low water mark indicating when to start filling
518 the queue again. Defaults to the same as iodepth, meaning
519 that fio will attempt to keep the queue full at all times.
520 If iodepth is set to eg 16 and iodepth_low is set to 4, then
521 after fio has filled the queue of 16 requests, it will let
522 the depth drain down to 4 before starting to fill it again.
524 direct=bool If value is true, use non-buffered io. This is usually
527 buffered=bool If value is true, use buffered io. This is the opposite
528 of the 'direct' option. Defaults to true.
530 offset=int Start io at the given offset in the file. The data before
531 the given offset will not be touched. This effectively
532 caps the file size at real_size - offset.
534 fsync=int If writing to a file, issue a sync of the dirty data
535 for every number of blocks given. For example, if you give
536 32 as a parameter, fio will sync the file for every 32
537 writes issued. If fio is using non-buffered io, we may
538 not sync the file. The exception is the sg io engine, which
539 synchronizes the disk cache anyway.
541 fsyncdata=int Like fsync= but uses fdatasync() to only sync data and not
544 overwrite=bool If true, writes to a file will always overwrite existing
545 data. If the file doesn't already exist, it will be
546 created before the write phase begins. If the file exists
547 and is large enough for the specified write phase, nothing
550 end_fsync=bool If true, fsync file contents when the job exits.
552 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
553 This differs from end_fsync in that it will happen on every
554 file close, not just at the end of the job.
556 rwmixread=int How large a percentage of the mix should be reads.
558 rwmixwrite=int How large a percentage of the mix should be writes. If both
559 rwmixread and rwmixwrite is given and the values do not add
560 up to 100%, the latter of the two will be used to override
561 the first. This may interfere with a given rate setting,
562 if fio is asked to limit reads or writes to a certain rate.
563 If that is the case, then the distribution may be skewed.
565 norandommap Normally fio will cover every block of the file when doing
566 random IO. If this option is given, fio will just get a
567 new random offset without looking at past io history. This
568 means that some blocks may not be read or written, and that
569 some blocks may be read/written more than once. This option
570 is mutually exclusive with verify= if and only if multiple
571 blocksizes (via bsrange=) are used, since fio only tracks
572 complete rewrites of blocks.
574 softrandommap See norandommap. If fio runs with the random block map enabled
575 and it fails to allocate the map, if this option is set it
576 will continue without a random block map. As coverage will
577 not be as complete as with random maps, this option is
580 nice=int Run the job with the given nice value. See man nice(2).
582 prio=int Set the io priority value of this job. Linux limits us to
583 a positive value between 0 and 7, with 0 being the highest.
586 prioclass=int Set the io priority class. See man ionice(1).
588 thinktime=int Stall the job x microseconds after an io has completed before
589 issuing the next. May be used to simulate processing being
590 done by an application. See thinktime_blocks and
594 Only valid if thinktime is set - pretend to spend CPU time
595 doing something with the data received, before falling back
596 to sleeping for the rest of the period specified by
600 Only valid if thinktime is set - control how many blocks
601 to issue, before waiting 'thinktime' usecs. If not set,
602 defaults to 1 which will make fio wait 'thinktime' usecs
605 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
606 the normal suffix rules apply. You can use rate=500k to limit
607 reads and writes to 500k each, or you can specify read and
608 writes separately. Using rate=1m,500k would limit reads to
609 1MB/sec and writes to 500KB/sec. Capping only reads or
610 writes can be done with rate=,500k or rate=500k,. The former
611 will only limit writes (to 500KB/sec), the latter will only
614 ratemin=int Tell fio to do whatever it can to maintain at least this
615 bandwidth. Failing to meet this requirement, will cause
616 the job to exit. The same format as rate is used for
617 read vs write separation.
619 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
620 as rate, just specified independently of bandwidth. If the
621 job is given a block size range instead of a fixed value,
622 the smallest block size is used as the metric. The same format
623 as rate is used for read vs write seperation.
625 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
626 the job to exit. The same format as rate is used for read vs
629 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
632 cpumask=int Set the CPU affinity of this job. The parameter given is a
633 bitmask of allowed CPU's the job may run on. So if you want
634 the allowed CPUs to be 1 and 5, you would pass the decimal
635 value of (1 << 1 | 1 << 5), or 34. See man
636 sched_setaffinity(2). This may not work on all supported
637 operating systems or kernel versions. This option doesn't
638 work well for a higher CPU count than what you can store in
639 an integer mask, so it can only control cpus 1-32. For
640 boxes with larger CPU counts, use cpus_allowed.
642 cpus_allowed=str Controls the same options as cpumask, but it allows a text
643 setting of the permitted CPUs instead. So to use CPUs 1 and
644 5, you would specify cpus_allowed=1,5. This options also
645 allows a range of CPUs. Say you wanted a binding to CPUs
646 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
648 startdelay=time Start this job the specified number of seconds after fio
649 has started. Only useful if the job file contains several
650 jobs, and you want to delay starting some jobs to a certain
653 runtime=time Tell fio to terminate processing after the specified number
654 of seconds. It can be quite hard to determine for how long
655 a specified job will run, so this parameter is handy to
656 cap the total runtime to a given time.
658 time_based If set, fio will run for the duration of the runtime
659 specified even if the file(s) are completely read or
660 written. It will simply loop over the same workload
661 as many times as the runtime allows.
663 ramp_time=time If set, fio will run the specified workload for this amount
664 of time before logging any performance numbers. Useful for
665 letting performance settle before logging results, thus
666 minimizing the runtime required for stable results. Note
667 that the ramp_time is considered lead in time for a job,
668 thus it will increase the total runtime if a special timeout
669 or runtime is specified.
671 invalidate=bool Invalidate the buffer/page cache parts for this file prior
672 to starting io. Defaults to true.
674 sync=bool Use sync io for buffered writes. For the majority of the
675 io engines, this means using O_SYNC.
678 mem=str Fio can use various types of memory as the io unit buffer.
679 The allowed values are:
681 malloc Use memory from malloc(3) as the buffers.
683 shm Use shared memory as the buffers. Allocated
686 shmhuge Same as shm, but use huge pages as backing.
688 mmap Use mmap to allocate buffers. May either be
689 anonymous memory, or can be file backed if
690 a filename is given after the option. The
691 format is mem=mmap:/path/to/file.
693 mmaphuge Use a memory mapped huge file as the buffer
694 backing. Append filename after mmaphuge, ala
695 mem=mmaphuge:/hugetlbfs/file
697 The area allocated is a function of the maximum allowed
698 bs size for the job, multiplied by the io depth given. Note
699 that for shmhuge and mmaphuge to work, the system must have
700 free huge pages allocated. This can normally be checked
701 and set by reading/writing /proc/sys/vm/nr_hugepages on a
702 Linux system. Fio assumes a huge page is 4MB in size. So
703 to calculate the number of huge pages you need for a given
704 job file, add up the io depth of all jobs (normally one unless
705 iodepth= is used) and multiply by the maximum bs set. Then
706 divide that number by the huge page size. You can see the
707 size of the huge pages in /proc/meminfo. If no huge pages
708 are allocated by having a non-zero number in nr_hugepages,
709 using mmaphuge or shmhuge will fail. Also see hugepage-size.
711 mmaphuge also needs to have hugetlbfs mounted and the file
712 location should point there. So if it's mounted in /huge,
713 you would use mem=mmaphuge:/huge/somefile.
715 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
716 Note that the given alignment is applied to the first IO unit
717 buffer, if using iodepth the alignment of the following buffers
718 are given by the bs used. In other words, if using a bs that is
719 a multiple of the page sized in the system, all buffers will
720 be aligned to this value. If using a bs that is not page
721 aligned, the alignment of subsequent IO memory buffers is the
722 sum of the iomem_align and bs used.
725 Defines the size of a huge page. Must at least be equal
726 to the system setting, see /proc/meminfo. Defaults to 4MB.
727 Should probably always be a multiple of megabytes, so using
728 hugepage-size=Xm is the preferred way to set this to avoid
729 setting a non-pow-2 bad value.
731 exitall When one job finishes, terminate the rest. The default is
732 to wait for each job to finish, sometimes that is not the
735 bwavgtime=int Average the calculated bandwidth over the given time. Value
736 is specified in milliseconds.
738 create_serialize=bool If true, serialize the file creating for the jobs.
739 This may be handy to avoid interleaving of data
740 files, which may greatly depend on the filesystem
741 used and even the number of processors in the system.
743 create_fsync=bool fsync the data file after creation. This is the
746 create_on_open=bool Don't pre-setup the files for IO, just create open()
747 when it's time to do IO to that file.
749 pre_read=bool If this is given, files will be pre-read into memory before
750 starting the given IO operation. This will also clear
751 the 'invalidate' flag, since it is pointless to pre-read
752 and then drop the cache. This will only work for IO engines
753 that are seekable, since they allow you to read the same data
754 multiple times. Thus it will not work on eg network or splice
757 unlink=bool Unlink the job files when done. Not the default, as repeated
758 runs of that job would then waste time recreating the file
761 loops=int Run the specified number of iterations of this job. Used
762 to repeat the same workload a given number of times. Defaults
765 do_verify=bool Run the verify phase after a write phase. Only makes sense if
766 verify is set. Defaults to 1.
768 verify=str If writing to a file, fio can verify the file contents
769 after each iteration of the job. The allowed values are:
771 md5 Use an md5 sum of the data area and store
772 it in the header of each block.
774 crc64 Use an experimental crc64 sum of the data
775 area and store it in the header of each
778 crc32c Use a crc32c sum of the data area and store
779 it in the header of each block.
781 crc32c-intel Use hardware assisted crc32c calcuation
782 provided on SSE4.2 enabled processors.
784 crc32 Use a crc32 sum of the data area and store
785 it in the header of each block.
787 crc16 Use a crc16 sum of the data area and store
788 it in the header of each block.
790 crc7 Use a crc7 sum of the data area and store
791 it in the header of each block.
793 sha512 Use sha512 as the checksum function.
795 sha256 Use sha256 as the checksum function.
797 meta Write extra information about each io
798 (timestamp, block number etc.). The block
801 null Only pretend to verify. Useful for testing
802 internals with ioengine=null, not for much
805 This option can be used for repeated burn-in tests of a
806 system to make sure that the written data is also
809 verifysort=bool If set, fio will sort written verify blocks when it deems
810 it faster to read them back in a sorted manner. This is
811 often the case when overwriting an existing file, since
812 the blocks are already laid out in the file system. You
813 can ignore this option unless doing huge amounts of really
814 fast IO where the red-black tree sorting CPU time becomes
817 verify_offset=int Swap the verification header with data somewhere else
818 in the block before writing. Its swapped back before
821 verify_interval=int Write the verification header at a finer granularity
822 than the blocksize. It will be written for chunks the
823 size of header_interval. blocksize should divide this
826 verify_pattern=int If set, fio will fill the io buffers with this
827 pattern. Fio defaults to filling with totally random
828 bytes, but sometimes it's interesting to fill with a known
829 pattern for io verification purposes. Depending on the
830 width of the pattern, fio will fill 1/2/3/4 bytes of the
831 buffer at the time. The verify_pattern cannot be larger than
834 verify_fatal=bool Normally fio will keep checking the entire contents
835 before quitting on a block verification failure. If this
836 option is set, fio will exit the job on the first observed
839 verify_async=int Fio will normally verify IO inline from the submitting
840 thread. This option takes an integer describing how many
841 async offload threads to create for IO verification instead,
842 causing fio to offload the duty of verifying IO contents
843 to one or more separate threads. If using this offload
844 option, even sync IO engines can benefit from using an
845 iodepth setting higher than 1, as it allows them to have
846 IO in flight while verifies are running.
848 verify_async_cpus=str Tell fio to set the given CPU affinity on the
849 async IO verification threads. See cpus_allowed for the
852 stonewall Wait for preceeding jobs in the job file to exit, before
853 starting this one. Can be used to insert serialization
854 points in the job file. A stone wall also implies starting
855 a new reporting group.
857 new_group Start a new reporting group. If this option isn't given,
858 jobs in a file will be part of the same reporting group
859 unless separated by a stone wall (or if it's a group
860 by itself, with the numjobs option).
862 numjobs=int Create the specified number of clones of this job. May be
863 used to setup a larger number of threads/processes doing
864 the same thing. We regard that grouping of jobs as a
867 group_reporting If 'numjobs' is set, it may be interesting to display
868 statistics for the group as a whole instead of for each
869 individual job. This is especially true of 'numjobs' is
870 large, looking at individual thread/process output quickly
871 becomes unwieldy. If 'group_reporting' is specified, fio
872 will show the final report per-group instead of per-job.
874 thread fio defaults to forking jobs, however if this option is
875 given, fio will use pthread_create(3) to create threads
878 zonesize=int Divide a file into zones of the specified size. See zoneskip.
880 zoneskip=int Skip the specified number of bytes when zonesize data has
881 been read. The two zone options can be used to only do
882 io on zones of a file.
884 write_iolog=str Write the issued io patterns to the specified file. See
887 read_iolog=str Open an iolog with the specified file name and replay the
888 io patterns it contains. This can be used to store a
889 workload and replay it sometime later. The iolog given
890 may also be a blktrace binary file, which allows fio
891 to replay a workload captured by blktrace. See blktrace
892 for how to capture such logging data. For blktrace replay,
893 the file needs to be turned into a blkparse binary data
894 file first (blktrace <device> -d file_for_fio.bin).
896 write_bw_log=str If given, write a bandwidth log of the jobs in this job
897 file. Can be used to store data of the bandwidth of the
898 jobs in their lifetime. The included fio_generate_plots
899 script uses gnuplot to turn these text files into nice
900 graphs. See write_log_log for behaviour of given
901 filename. For this option, the postfix is _bw.log.
903 write_lat_log=str Same as write_bw_log, except that this option stores io
904 completion latencies instead. If no filename is given
905 with this option, the default filename of "jobname_type.log"
906 is used. Even if the filename is given, fio will still
907 append the type of log. So if one specifies
911 The actual log names will be foo_clat.log and foo_slat.log.
912 This helps fio_generate_plot fine the logs automatically.
914 lockmem=int Pin down the specified amount of memory with mlock(2). Can
915 potentially be used instead of removing memory or booting
916 with less memory to simulate a smaller amount of memory.
918 exec_prerun=str Before running this job, issue the command specified
921 exec_postrun=str After the job completes, issue the command specified
924 ioscheduler=str Attempt to switch the device hosting the file to the specified
925 io scheduler before running.
927 cpuload=int If the job is a CPU cycle eater, attempt to use the specified
928 percentage of CPU cycles.
930 cpuchunks=int If the job is a CPU cycle eater, split the load into
931 cycles of the given time. In microseconds.
933 disk_util=bool Generate disk utilization statistics, if the platform
934 supports it. Defaults to on.
936 disable_clat=bool Disable measurements of completion latency numbers. Useful
937 only for cutting back the number of calls to gettimeofday,
938 as that does impact performance at really high IOPS rates.
939 Note that to really get rid of a large amount of these
940 calls, this option must be used with disable_slat and
943 disable_slat=bool Disable measurements of submission latency numbers. See
946 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
949 gtod_reduce=bool Enable all of the gettimeofday() reducing options
950 (disable_clat, disable_slat, disable_bw) plus reduce
951 precision of the timeout somewhat to really shrink
952 the gettimeofday() call count. With this option enabled,
953 we only do about 0.4% of the gtod() calls we would have
954 done if all time keeping was enabled.
956 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
957 execution to just getting the current time. Fio (and
958 databases, for instance) are very intensive on gettimeofday()
959 calls. With this option, you can set one CPU aside for
960 doing nothing but logging current time to a shared memory
961 location. Then the other threads/processes that run IO
962 workloads need only copy that segment, instead of entering
963 the kernel with a gettimeofday() call. The CPU set aside
964 for doing these time calls will be excluded from other
965 uses. Fio will manually clear it from the CPU mask of other
967 continue_on_error=bool Normally fio will exit the job on the first observed
968 failure. If this option is set, fio will continue the job when
969 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
970 is exceeded or the I/O size specified is completed. If this
971 option is used, there are two more stats that are appended,
972 the total error count and the first error. The error field
973 given in the stats is the first error that was hit during the
977 6.0 Interpreting the output
978 ---------------------------
980 fio spits out a lot of output. While running, fio will display the
981 status of the jobs created. An example of that would be:
983 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
985 The characters inside the square brackets denote the current status of
986 each thread. The possible values (in typical life cycle order) are:
990 P Thread setup, but not started.
992 I Thread initialized, waiting.
993 p Thread running pre-reading file(s).
994 R Running, doing sequential reads.
995 r Running, doing random reads.
996 W Running, doing sequential writes.
997 w Running, doing random writes.
998 M Running, doing mixed sequential reads/writes.
999 m Running, doing mixed random reads/writes.
1000 F Running, currently waiting for fsync()
1001 V Running, doing verification of written data.
1002 E Thread exited, not reaped by main thread yet.
1005 The other values are fairly self explanatory - number of threads
1006 currently running and doing io, rate of io since last check (read speed
1007 listed first, then write speed), and the estimated completion percentage
1008 and time for the running group. It's impossible to estimate runtime of
1009 the following groups (if any).
1011 When fio is done (or interrupted by ctrl-c), it will show the data for
1012 each thread, group of threads, and disks in that order. For each data
1013 direction, the output looks like:
1015 Client1 (g=0): err= 0:
1016 write: io= 32MB, bw= 666KB/s, runt= 50320msec
1017 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1018 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1019 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1020 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1021 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1022 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1023 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1024 issued r/w: total=0/32768, short=0/0
1025 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1026 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1028 The client number is printed, along with the group id and error of that
1029 thread. Below is the io statistics, here for writes. In the order listed,
1032 io= Number of megabytes io performed
1033 bw= Average bandwidth rate
1034 runt= The runtime of that thread
1035 slat= Submission latency (avg being the average, stdev being the
1036 standard deviation). This is the time it took to submit
1037 the io. For sync io, the slat is really the completion
1038 latency, since queue/complete is one operation there. This
1039 value can be in milliseconds or microseconds, fio will choose
1040 the most appropriate base and print that. In the example
1041 above, milliseconds is the best scale.
1042 clat= Completion latency. Same names as slat, this denotes the
1043 time from submission to completion of the io pieces. For
1044 sync io, clat will usually be equal (or very close) to 0,
1045 as the time from submit to complete is basically just
1046 CPU time (io has already been done, see slat explanation).
1047 bw= Bandwidth. Same names as the xlat stats, but also includes
1048 an approximate percentage of total aggregate bandwidth
1049 this thread received in this group. This last value is
1050 only really useful if the threads in this group are on the
1051 same disk, since they are then competing for disk access.
1052 cpu= CPU usage. User and system time, along with the number
1053 of context switches this thread went through, usage of
1054 system and user time, and finally the number of major
1055 and minor page faults.
1056 IO depths= The distribution of io depths over the job life time. The
1057 numbers are divided into powers of 2, so for example the
1058 16= entries includes depths up to that value but higher
1059 than the previous entry. In other words, it covers the
1060 range from 16 to 31.
1061 IO submit= How many pieces of IO were submitting in a single submit
1062 call. Each entry denotes that amount and below, until
1063 the previous entry - eg, 8=100% mean that we submitted
1064 anywhere in between 5-8 ios per submit call.
1065 IO complete= Like the above submit number, but for completions instead.
1066 IO issued= The number of read/write requests issued, and how many
1068 IO latencies= The distribution of IO completion latencies. This is the
1069 time from when IO leaves fio and when it gets completed.
1070 The numbers follow the same pattern as the IO depths,
1071 meaning that 2=1.6% means that 1.6% of the IO completed
1072 within 2 msecs, 20=12.8% means that 12.8% of the IO
1073 took more than 10 msecs, but less than (or equal to) 20 msecs.
1075 After each client has been listed, the group statistics are printed. They
1076 will look like this:
1078 Run status group 0 (all jobs):
1079 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1080 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1082 For each data direction, it prints:
1084 io= Number of megabytes io performed.
1085 aggrb= Aggregate bandwidth of threads in this group.
1086 minb= The minimum average bandwidth a thread saw.
1087 maxb= The maximum average bandwidth a thread saw.
1088 mint= The smallest runtime of the threads in that group.
1089 maxt= The longest runtime of the threads in that group.
1091 And finally, the disk statistics are printed. They will look like this:
1093 Disk stats (read/write):
1094 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1096 Each value is printed for both reads and writes, with reads first. The
1099 ios= Number of ios performed by all groups.
1100 merge= Number of merges io the io scheduler.
1101 ticks= Number of ticks we kept the disk busy.
1102 io_queue= Total time spent in the disk queue.
1103 util= The disk utilization. A value of 100% means we kept the disk
1104 busy constantly, 50% would be a disk idling half of the time.
1110 For scripted usage where you typically want to generate tables or graphs
1111 of the results, fio can output the results in a semicolon separated format.
1112 The format is one long line of values, such as:
1114 client1;0;0;1906777;1090804;1790;0;0;0.000000;0.000000;0;0;0.000000;0.000000;929380;1152890;25.510151%;1078276.333333;128948.113404;0;0;0;0;0;0.000000;0.000000;0;0;0.000000;0.000000;0;0;0.000000%;0.000000;0.000000;100.000000%;0.000000%;324;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%;0.0%;100.0%;0.0%;0.0%;0.0%;0.0%;0.0%
1115 ;0.0%;0.0%;0.0%;0.0%;0.0%
1117 To enable terse output, use the --minimal command line option.
1119 Split up, the format is as follows:
1121 jobname, groupid, error
1123 KB IO, bandwidth (KB/sec), runtime (msec)
1124 Submission latency: min, max, mean, deviation
1125 Completion latency: min, max, mean, deviation
1126 Bw: min, max, aggregate percentage of total, mean, deviation
1128 KB IO, bandwidth (KB/sec), runtime (msec)
1129 Submission latency: min, max, mean, deviation
1130 Completion latency: min, max, mean, deviation
1131 Bw: min, max, aggregate percentage of total, mean, deviation
1132 CPU usage: user, system, context switches, major faults, minor faults
1133 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1134 IO latencies: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, >=2000