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. So if you want to specify 4096, you could either write
230 out '4096' or just give 4k. The suffixes signify base 2 values, so
231 1024 is 1k and 1024k is 1m and so on. If the option accepts an upper
232 and lower range, use a colon ':' or minus '-' to separate such values.
233 May also include a prefix to indicate numbers base. If 0x is used,
234 the number is assumed to be hexadecimal. See irange.
235 bool Boolean. Usually parsed as an integer, however only defined for
236 true and false (1 and 0).
237 irange Integer range with suffix. Allows value range to be given, such
238 as 1024-4096. A colon may also be used as the separator, eg
239 1k:4k. If the option allows two sets of ranges, they can be
240 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
243 With the above in mind, here follows the complete list of fio job
246 name=str ASCII name of the job. This may be used to override the
247 name printed by fio for this job. Otherwise the job
248 name is used. On the command line this parameter has the
249 special purpose of also signaling the start of a new
252 description=str Text description of the job. Doesn't do anything except
253 dump this text description when this job is run. It's
256 directory=str Prefix filenames with this directory. Used to place files
257 in a different location than "./".
259 filename=str Fio normally makes up a filename based on the job name,
260 thread number, and file number. If you want to share
261 files between threads in a job or several jobs, specify
262 a filename for each of them to override the default. If
263 the ioengine used is 'net', the filename is the host, port,
264 and protocol to use in the format of =host/port/protocol.
265 See ioengine=net for more. If the ioengine is file based, you
266 can specify a number of files by separating the names with a
267 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
268 as the two working files, you would use
269 filename=/dev/sda:/dev/sdb. If the wanted filename does need to
270 include a colon, then escape that with a '\' character. For
271 instance, if the filename is "/dev/dsk/foo@3,0:c", then you would
272 use filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name,
273 meaning stdin or stdout. Which of the two depends on the read/write
276 opendir=str Tell fio to recursively add any file it can find in this
277 directory and down the file system tree.
279 lockfile=str Fio defaults to not locking any files before it does
280 IO to them. If a file or file descriptor is shared, fio
281 can serialize IO to that file to make the end result
282 consistent. This is usual for emulating real workloads that
283 share files. The lock modes are:
285 none No locking. The default.
286 exclusive Only one thread/process may do IO,
287 excluding all others.
288 readwrite Read-write locking on the file. Many
289 readers may access the file at the
290 same time, but writes get exclusive
293 The option may be post-fixed with a lock batch number. If
294 set, then each thread/process may do that amount of IOs to
295 the file before giving up the lock. Since lock acquisition is
296 expensive, batching the lock/unlocks will speed up IO.
299 rw=str Type of io pattern. Accepted values are:
301 read Sequential reads
302 write Sequential writes
303 randwrite Random writes
304 randread Random reads
305 rw Sequential mixed reads and writes
306 randrw Random mixed reads and writes
308 For the mixed io types, the default is to split them 50/50.
309 For certain types of io the result may still be skewed a bit,
310 since the speed may be different. It is possible to specify
311 a number of IO's to do before getting a new offset - this
312 is only useful for random IO, where fio would normally
313 generate a new random offset for every IO. If you append
314 eg 8 to randread, you would get a new random offset for
315 every 8 IO's. The result would be a seek for only every 8
316 IO's, instead of for every IO. Use rw=randread:8 to specify
319 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
320 Storage manufacturers like to use 10^3 or 1000 as a base
321 ten unit instead, for obvious reasons. Allow values are
322 1024 or 1000, with 1024 being the default.
324 randrepeat=bool For random IO workloads, seed the generator in a predictable
325 way so that results are repeatable across repetitions.
327 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
328 on what IO patterns it is likely to issue. Sometimes you
329 want to test specific IO patterns without telling the
330 kernel about it, in which case you can disable this option.
331 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
332 IO and POSIX_FADV_RANDOM for random IO.
334 size=int The total size of file io for this job. Fio will run until
335 this many bytes has been transferred, unless runtime is
336 limited by other options (such as 'runtime', for instance).
337 Unless specific nrfiles and filesize options are given,
338 fio will divide this size between the available files
339 specified by the job.
341 filesize=int Individual file sizes. May be a range, in which case fio
342 will select sizes for files at random within the given range
343 and limited to 'size' in total (if that is given). If not
344 given, each created file is the same size.
346 fill_device=bool Sets size to something really large and waits for ENOSPC (no
347 space left on device) as the terminating condition. Only makes
348 sense with sequential write. For a read workload, the mount
349 point will be filled first then IO started on the result.
352 bs=int The block size used for the io units. Defaults to 4k. Values
353 can be given for both read and writes. If a single int is
354 given, it will apply to both. If a second int is specified
355 after a comma, it will apply to writes only. In other words,
356 the format is either bs=read_and_write or bs=read,write.
357 bs=4k,8k will thus use 4k blocks for reads, and 8k blocks
358 for writes. If you only wish to set the write size, you
359 can do so by passing an empty read size - bs=,8k will set
360 8k for writes and leave the read default value.
363 ba=int At what boundary to align random IO offsets. Defaults to
364 the same as 'blocksize' the minimum blocksize given.
365 Minimum alignment is typically 512b for using direct IO,
366 though it usually depends on the hardware block size. This
367 option is mutually exclusive with using a random map for
368 files, so it will turn off that option.
370 blocksize_range=irange
371 bsrange=irange Instead of giving a single block size, specify a range
372 and fio will mix the issued io block sizes. The issued
373 io unit will always be a multiple of the minimum value
374 given (also see bs_unaligned). Applies to both reads and
375 writes, however a second range can be given after a comma.
378 bssplit=str Sometimes you want even finer grained control of the
379 block sizes issued, not just an even split between them.
380 This option allows you to weight various block sizes,
381 so that you are able to define a specific amount of
382 block sizes issued. The format for this option is:
384 bssplit=blocksize/percentage:blocksize/percentage
386 for as many block sizes as needed. So if you want to define
387 a workload that has 50% 64k blocks, 10% 4k blocks, and
388 40% 32k blocks, you would write:
390 bssplit=4k/10:64k/50:32k/40
392 Ordering does not matter. If the percentage is left blank,
393 fio will fill in the remaining values evenly. So a bssplit
394 option like this one:
396 bssplit=4k/50:1k/:32k/
398 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
399 always add up to 100, if bssplit is given a range that adds
400 up to more, it will error out.
402 bssplit also supports giving separate splits to reads and
403 writes. The format is identical to what bs= accepts. You
404 have to separate the read and write parts with a comma. So
405 if you want a workload that has 50% 2k reads and 50% 4k reads,
406 while having 90% 4k writes and 10% 8k writes, you would
409 bssplit=2k/50:4k/50,4k/90,8k/10
412 bs_unaligned If this option is given, any byte size value within bsrange
413 may be used as a block range. This typically wont work with
414 direct IO, as that normally requires sector alignment.
416 zero_buffers If this option is given, fio will init the IO buffers to
417 all zeroes. The default is to fill them with random data.
419 refill_buffers If this option is given, fio will refill the IO buffers
420 on every submit. The default is to only fill it at init
421 time and reuse that data. Only makes sense if zero_buffers
422 isn't specified, naturally. If data verification is enabled,
423 refill_buffers is also automatically enabled.
425 nrfiles=int Number of files to use for this job. Defaults to 1.
427 openfiles=int Number of files to keep open at the same time. Defaults to
428 the same as nrfiles, can be set smaller to limit the number
431 file_service_type=str Defines how fio decides which file from a job to
432 service next. The following types are defined:
434 random Just choose a file at random.
436 roundrobin Round robin over open files. This
439 sequential Finish one file before moving on to
440 the next. Multiple files can still be
441 open depending on 'openfiles'.
443 The string can have a number appended, indicating how
444 often to switch to a new file. So if option random:4 is
445 given, fio will switch to a new random file after 4 ios
448 ioengine=str Defines how the job issues io to the file. The following
451 sync Basic read(2) or write(2) io. lseek(2) is
452 used to position the io location.
454 psync Basic pread(2) or pwrite(2) io.
456 vsync Basic readv(2) or writev(2) IO.
458 libaio Linux native asynchronous io. Note that Linux
459 may only support queued behaviour with
460 non-buffered IO (set direct=1 or buffered=0).
462 posixaio glibc posix asynchronous io.
464 solarisaio Solaris native asynchronous io.
466 mmap File is memory mapped and data copied
467 to/from using memcpy(3).
469 splice splice(2) is used to transfer the data and
470 vmsplice(2) to transfer data from user
473 syslet-rw Use the syslet system calls to make
474 regular read/write async.
476 sg SCSI generic sg v3 io. May either be
477 synchronous using the SG_IO ioctl, or if
478 the target is an sg character device
479 we use read(2) and write(2) for asynchronous
482 null Doesn't transfer any data, just pretends
483 to. This is mainly used to exercise fio
484 itself and for debugging/testing purposes.
486 net Transfer over the network to given host:port.
487 'filename' must be set appropriately to
488 filename=host/port/protocol regardless of send
489 or receive, if the latter only the port
490 argument is used. 'host' may be an IP address
491 or hostname, port is the port number to be used,
492 and protocol may be 'udp' or 'tcp'. If no
493 protocol is given, TCP is used.
495 netsplice Like net, but uses splice/vmsplice to
496 map data and send/receive.
498 cpuio Doesn't transfer any data, but burns CPU
499 cycles according to the cpuload= and
500 cpucycle= options. Setting cpuload=85
501 will cause that job to do nothing but burn
502 85% of the CPU. In case of SMP machines,
503 use numjobs=<no_of_cpu> to get desired CPU
504 usage, as the cpuload only loads a single
505 CPU at the desired rate.
507 guasi The GUASI IO engine is the Generic Userspace
508 Asyncronous Syscall Interface approach
511 http://www.xmailserver.org/guasi-lib.html
513 for more info on GUASI.
515 external Prefix to specify loading an external
516 IO engine object file. Append the engine
517 filename, eg ioengine=external:/tmp/foo.o
518 to load ioengine foo.o in /tmp.
520 iodepth=int This defines how many io units to keep in flight against
521 the file. The default is 1 for each file defined in this
522 job, can be overridden with a larger value for higher
525 iodepth_batch_submit=int
526 iodepth_batch=int This defines how many pieces of IO to submit at once.
527 It defaults to 1 which means that we submit each IO
528 as soon as it is available, but can be raised to submit
529 bigger batches of IO at the time.
531 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
532 at once. It defaults to 1 which means that we'll ask
533 for a minimum of 1 IO in the retrieval process from
534 the kernel. The IO retrieval will go on until we
535 hit the limit set by iodepth_low. If this variable is
536 set to 0, then fio will always check for completed
537 events before queuing more IO. This helps reduce
538 IO latency, at the cost of more retrieval system calls.
540 iodepth_low=int The low water mark indicating when to start filling
541 the queue again. Defaults to the same as iodepth, meaning
542 that fio will attempt to keep the queue full at all times.
543 If iodepth is set to eg 16 and iodepth_low is set to 4, then
544 after fio has filled the queue of 16 requests, it will let
545 the depth drain down to 4 before starting to fill it again.
547 direct=bool If value is true, use non-buffered io. This is usually
550 buffered=bool If value is true, use buffered io. This is the opposite
551 of the 'direct' option. Defaults to true.
553 offset=int Start io at the given offset in the file. The data before
554 the given offset will not be touched. This effectively
555 caps the file size at real_size - offset.
557 fsync=int If writing to a file, issue a sync of the dirty data
558 for every number of blocks given. For example, if you give
559 32 as a parameter, fio will sync the file for every 32
560 writes issued. If fio is using non-buffered io, we may
561 not sync the file. The exception is the sg io engine, which
562 synchronizes the disk cache anyway.
564 fsyncdata=int Like fsync= but uses fdatasync() to only sync data and not
567 overwrite=bool If true, writes to a file will always overwrite existing
568 data. If the file doesn't already exist, it will be
569 created before the write phase begins. If the file exists
570 and is large enough for the specified write phase, nothing
573 end_fsync=bool If true, fsync file contents when the job exits.
575 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
576 This differs from end_fsync in that it will happen on every
577 file close, not just at the end of the job.
579 rwmixread=int How large a percentage of the mix should be reads.
581 rwmixwrite=int How large a percentage of the mix should be writes. If both
582 rwmixread and rwmixwrite is given and the values do not add
583 up to 100%, the latter of the two will be used to override
584 the first. This may interfere with a given rate setting,
585 if fio is asked to limit reads or writes to a certain rate.
586 If that is the case, then the distribution may be skewed.
588 norandommap Normally fio will cover every block of the file when doing
589 random IO. If this option is given, fio will just get a
590 new random offset without looking at past io history. This
591 means that some blocks may not be read or written, and that
592 some blocks may be read/written more than once. This option
593 is mutually exclusive with verify= if and only if multiple
594 blocksizes (via bsrange=) are used, since fio only tracks
595 complete rewrites of blocks.
597 softrandommap See norandommap. If fio runs with the random block map enabled
598 and it fails to allocate the map, if this option is set it
599 will continue without a random block map. As coverage will
600 not be as complete as with random maps, this option is
603 nice=int Run the job with the given nice value. See man nice(2).
605 prio=int Set the io priority value of this job. Linux limits us to
606 a positive value between 0 and 7, with 0 being the highest.
609 prioclass=int Set the io priority class. See man ionice(1).
611 thinktime=int Stall the job x microseconds after an io has completed before
612 issuing the next. May be used to simulate processing being
613 done by an application. See thinktime_blocks and
617 Only valid if thinktime is set - pretend to spend CPU time
618 doing something with the data received, before falling back
619 to sleeping for the rest of the period specified by
623 Only valid if thinktime is set - control how many blocks
624 to issue, before waiting 'thinktime' usecs. If not set,
625 defaults to 1 which will make fio wait 'thinktime' usecs
628 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
629 the normal suffix rules apply. You can use rate=500k to limit
630 reads and writes to 500k each, or you can specify read and
631 writes separately. Using rate=1m,500k would limit reads to
632 1MB/sec and writes to 500KB/sec. Capping only reads or
633 writes can be done with rate=,500k or rate=500k,. The former
634 will only limit writes (to 500KB/sec), the latter will only
637 ratemin=int Tell fio to do whatever it can to maintain at least this
638 bandwidth. Failing to meet this requirement, will cause
639 the job to exit. The same format as rate is used for
640 read vs write separation.
642 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
643 as rate, just specified independently of bandwidth. If the
644 job is given a block size range instead of a fixed value,
645 the smallest block size is used as the metric. The same format
646 as rate is used for read vs write seperation.
648 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
649 the job to exit. The same format as rate is used for read vs
652 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
655 cpumask=int Set the CPU affinity of this job. The parameter given is a
656 bitmask of allowed CPU's the job may run on. So if you want
657 the allowed CPUs to be 1 and 5, you would pass the decimal
658 value of (1 << 1 | 1 << 5), or 34. See man
659 sched_setaffinity(2). This may not work on all supported
660 operating systems or kernel versions. This option doesn't
661 work well for a higher CPU count than what you can store in
662 an integer mask, so it can only control cpus 1-32. For
663 boxes with larger CPU counts, use cpus_allowed.
665 cpus_allowed=str Controls the same options as cpumask, but it allows a text
666 setting of the permitted CPUs instead. So to use CPUs 1 and
667 5, you would specify cpus_allowed=1,5. This options also
668 allows a range of CPUs. Say you wanted a binding to CPUs
669 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
671 startdelay=time Start this job the specified number of seconds after fio
672 has started. Only useful if the job file contains several
673 jobs, and you want to delay starting some jobs to a certain
676 runtime=time Tell fio to terminate processing after the specified number
677 of seconds. It can be quite hard to determine for how long
678 a specified job will run, so this parameter is handy to
679 cap the total runtime to a given time.
681 time_based If set, fio will run for the duration of the runtime
682 specified even if the file(s) are completely read or
683 written. It will simply loop over the same workload
684 as many times as the runtime allows.
686 ramp_time=time If set, fio will run the specified workload for this amount
687 of time before logging any performance numbers. Useful for
688 letting performance settle before logging results, thus
689 minimizing the runtime required for stable results. Note
690 that the ramp_time is considered lead in time for a job,
691 thus it will increase the total runtime if a special timeout
692 or runtime is specified.
694 invalidate=bool Invalidate the buffer/page cache parts for this file prior
695 to starting io. Defaults to true.
697 sync=bool Use sync io for buffered writes. For the majority of the
698 io engines, this means using O_SYNC.
701 mem=str Fio can use various types of memory as the io unit buffer.
702 The allowed values are:
704 malloc Use memory from malloc(3) as the buffers.
706 shm Use shared memory as the buffers. Allocated
709 shmhuge Same as shm, but use huge pages as backing.
711 mmap Use mmap to allocate buffers. May either be
712 anonymous memory, or can be file backed if
713 a filename is given after the option. The
714 format is mem=mmap:/path/to/file.
716 mmaphuge Use a memory mapped huge file as the buffer
717 backing. Append filename after mmaphuge, ala
718 mem=mmaphuge:/hugetlbfs/file
720 The area allocated is a function of the maximum allowed
721 bs size for the job, multiplied by the io depth given. Note
722 that for shmhuge and mmaphuge to work, the system must have
723 free huge pages allocated. This can normally be checked
724 and set by reading/writing /proc/sys/vm/nr_hugepages on a
725 Linux system. Fio assumes a huge page is 4MB in size. So
726 to calculate the number of huge pages you need for a given
727 job file, add up the io depth of all jobs (normally one unless
728 iodepth= is used) and multiply by the maximum bs set. Then
729 divide that number by the huge page size. You can see the
730 size of the huge pages in /proc/meminfo. If no huge pages
731 are allocated by having a non-zero number in nr_hugepages,
732 using mmaphuge or shmhuge will fail. Also see hugepage-size.
734 mmaphuge also needs to have hugetlbfs mounted and the file
735 location should point there. So if it's mounted in /huge,
736 you would use mem=mmaphuge:/huge/somefile.
738 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
739 Note that the given alignment is applied to the first IO unit
740 buffer, if using iodepth the alignment of the following buffers
741 are given by the bs used. In other words, if using a bs that is
742 a multiple of the page sized in the system, all buffers will
743 be aligned to this value. If using a bs that is not page
744 aligned, the alignment of subsequent IO memory buffers is the
745 sum of the iomem_align and bs used.
748 Defines the size of a huge page. Must at least be equal
749 to the system setting, see /proc/meminfo. Defaults to 4MB.
750 Should probably always be a multiple of megabytes, so using
751 hugepage-size=Xm is the preferred way to set this to avoid
752 setting a non-pow-2 bad value.
754 exitall When one job finishes, terminate the rest. The default is
755 to wait for each job to finish, sometimes that is not the
758 bwavgtime=int Average the calculated bandwidth over the given time. Value
759 is specified in milliseconds.
761 create_serialize=bool If true, serialize the file creating for the jobs.
762 This may be handy to avoid interleaving of data
763 files, which may greatly depend on the filesystem
764 used and even the number of processors in the system.
766 create_fsync=bool fsync the data file after creation. This is the
769 create_on_open=bool Don't pre-setup the files for IO, just create open()
770 when it's time to do IO to that file.
772 pre_read=bool If this is given, files will be pre-read into memory before
773 starting the given IO operation. This will also clear
774 the 'invalidate' flag, since it is pointless to pre-read
775 and then drop the cache. This will only work for IO engines
776 that are seekable, since they allow you to read the same data
777 multiple times. Thus it will not work on eg network or splice
780 unlink=bool Unlink the job files when done. Not the default, as repeated
781 runs of that job would then waste time recreating the file
784 loops=int Run the specified number of iterations of this job. Used
785 to repeat the same workload a given number of times. Defaults
788 do_verify=bool Run the verify phase after a write phase. Only makes sense if
789 verify is set. Defaults to 1.
791 verify=str If writing to a file, fio can verify the file contents
792 after each iteration of the job. The allowed values are:
794 md5 Use an md5 sum of the data area and store
795 it in the header of each block.
797 crc64 Use an experimental crc64 sum of the data
798 area and store it in the header of each
801 crc32c Use a crc32c sum of the data area and store
802 it in the header of each block.
804 crc32c-intel Use hardware assisted crc32c calcuation
805 provided on SSE4.2 enabled processors.
807 crc32 Use a crc32 sum of the data area and store
808 it in the header of each block.
810 crc16 Use a crc16 sum of the data area and store
811 it in the header of each block.
813 crc7 Use a crc7 sum of the data area and store
814 it in the header of each block.
816 sha512 Use sha512 as the checksum function.
818 sha256 Use sha256 as the checksum function.
820 sha1 Use optimized sha1 as the checksum function.
822 meta Write extra information about each io
823 (timestamp, block number etc.). The block
826 null Only pretend to verify. Useful for testing
827 internals with ioengine=null, not for much
830 This option can be used for repeated burn-in tests of a
831 system to make sure that the written data is also
832 correctly read back. If the data direction given is
833 a read or random read, fio will assume that it should
834 verify a previously written file. If the data direction
835 includes any form of write, the verify will be of the
838 verifysort=bool If set, fio will sort written verify blocks when it deems
839 it faster to read them back in a sorted manner. This is
840 often the case when overwriting an existing file, since
841 the blocks are already laid out in the file system. You
842 can ignore this option unless doing huge amounts of really
843 fast IO where the red-black tree sorting CPU time becomes
846 verify_offset=int Swap the verification header with data somewhere else
847 in the block before writing. Its swapped back before
850 verify_interval=int Write the verification header at a finer granularity
851 than the blocksize. It will be written for chunks the
852 size of header_interval. blocksize should divide this
855 verify_pattern=str If set, fio will fill the io buffers with this
856 pattern. Fio defaults to filling with totally random
857 bytes, but sometimes it's interesting to fill with a known
858 pattern for io verification purposes. Depending on the
859 width of the pattern, fio will fill 1/2/3/4 bytes of the
860 buffer at the time(it can be either a decimal or a hex number).
861 The verify_pattern if larger than a 32-bit quantity has to
862 be a hex number that starts with either "0x" or "0X".
864 verify_fatal=bool Normally fio will keep checking the entire contents
865 before quitting on a block verification failure. If this
866 option is set, fio will exit the job on the first observed
869 verify_async=int Fio will normally verify IO inline from the submitting
870 thread. This option takes an integer describing how many
871 async offload threads to create for IO verification instead,
872 causing fio to offload the duty of verifying IO contents
873 to one or more separate threads. If using this offload
874 option, even sync IO engines can benefit from using an
875 iodepth setting higher than 1, as it allows them to have
876 IO in flight while verifies are running.
878 verify_async_cpus=str Tell fio to set the given CPU affinity on the
879 async IO verification threads. See cpus_allowed for the
882 stonewall Wait for preceeding jobs in the job file to exit, before
883 starting this one. Can be used to insert serialization
884 points in the job file. A stone wall also implies starting
885 a new reporting group.
887 new_group Start a new reporting group. If this option isn't given,
888 jobs in a file will be part of the same reporting group
889 unless separated by a stone wall (or if it's a group
890 by itself, with the numjobs option).
892 numjobs=int Create the specified number of clones of this job. May be
893 used to setup a larger number of threads/processes doing
894 the same thing. We regard that grouping of jobs as a
897 group_reporting If 'numjobs' is set, it may be interesting to display
898 statistics for the group as a whole instead of for each
899 individual job. This is especially true of 'numjobs' is
900 large, looking at individual thread/process output quickly
901 becomes unwieldy. If 'group_reporting' is specified, fio
902 will show the final report per-group instead of per-job.
904 thread fio defaults to forking jobs, however if this option is
905 given, fio will use pthread_create(3) to create threads
908 zonesize=int Divide a file into zones of the specified size. See zoneskip.
910 zoneskip=int Skip the specified number of bytes when zonesize data has
911 been read. The two zone options can be used to only do
912 io on zones of a file.
914 write_iolog=str Write the issued io patterns to the specified file. See
917 read_iolog=str Open an iolog with the specified file name and replay the
918 io patterns it contains. This can be used to store a
919 workload and replay it sometime later. The iolog given
920 may also be a blktrace binary file, which allows fio
921 to replay a workload captured by blktrace. See blktrace
922 for how to capture such logging data. For blktrace replay,
923 the file needs to be turned into a blkparse binary data
924 file first (blktrace <device> -d file_for_fio.bin).
926 write_bw_log=str If given, write a bandwidth log of the jobs in this job
927 file. Can be used to store data of the bandwidth of the
928 jobs in their lifetime. The included fio_generate_plots
929 script uses gnuplot to turn these text files into nice
930 graphs. See write_log_log for behaviour of given
931 filename. For this option, the postfix is _bw.log.
933 write_lat_log=str Same as write_bw_log, except that this option stores io
934 completion latencies instead. If no filename is given
935 with this option, the default filename of "jobname_type.log"
936 is used. Even if the filename is given, fio will still
937 append the type of log. So if one specifies
941 The actual log names will be foo_clat.log and foo_slat.log.
942 This helps fio_generate_plot fine the logs automatically.
944 lockmem=int Pin down the specified amount of memory with mlock(2). Can
945 potentially be used instead of removing memory or booting
946 with less memory to simulate a smaller amount of memory.
948 exec_prerun=str Before running this job, issue the command specified
951 exec_postrun=str After the job completes, issue the command specified
954 ioscheduler=str Attempt to switch the device hosting the file to the specified
955 io scheduler before running.
957 cpuload=int If the job is a CPU cycle eater, attempt to use the specified
958 percentage of CPU cycles.
960 cpuchunks=int If the job is a CPU cycle eater, split the load into
961 cycles of the given time. In microseconds.
963 disk_util=bool Generate disk utilization statistics, if the platform
964 supports it. Defaults to on.
966 disable_clat=bool Disable measurements of completion latency numbers. Useful
967 only for cutting back the number of calls to gettimeofday,
968 as that does impact performance at really high IOPS rates.
969 Note that to really get rid of a large amount of these
970 calls, this option must be used with disable_slat and
973 disable_slat=bool Disable measurements of submission latency numbers. See
976 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
979 gtod_reduce=bool Enable all of the gettimeofday() reducing options
980 (disable_clat, disable_slat, disable_bw) plus reduce
981 precision of the timeout somewhat to really shrink
982 the gettimeofday() call count. With this option enabled,
983 we only do about 0.4% of the gtod() calls we would have
984 done if all time keeping was enabled.
986 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
987 execution to just getting the current time. Fio (and
988 databases, for instance) are very intensive on gettimeofday()
989 calls. With this option, you can set one CPU aside for
990 doing nothing but logging current time to a shared memory
991 location. Then the other threads/processes that run IO
992 workloads need only copy that segment, instead of entering
993 the kernel with a gettimeofday() call. The CPU set aside
994 for doing these time calls will be excluded from other
995 uses. Fio will manually clear it from the CPU mask of other
998 continue_on_error=bool Normally fio will exit the job on the first observed
999 failure. If this option is set, fio will continue the job when
1000 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1001 is exceeded or the I/O size specified is completed. If this
1002 option is used, there are two more stats that are appended,
1003 the total error count and the first error. The error field
1004 given in the stats is the first error that was hit during the
1007 cgroup=str Add job to this control group. If it doesn't exist, it will
1008 be created. The system must have a mounted cgroup blkio
1009 mount point for this to work. If your system doesn't have it
1010 mounted, you can do so with:
1012 # mount -t cgroup -o blkio none /cgroup
1014 cgroup_weight=int Set the weight of the cgroup to this value. See
1015 the documentation that comes with the kernel, allowed values
1016 are in the range of 100..1000.
1018 uid=int Instead of running as the invoking user, set the user ID to
1019 this value before the thread/process does any work.
1021 gid=int Set group ID, see uid.
1023 6.0 Interpreting the output
1024 ---------------------------
1026 fio spits out a lot of output. While running, fio will display the
1027 status of the jobs created. An example of that would be:
1029 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1031 The characters inside the square brackets denote the current status of
1032 each thread. The possible values (in typical life cycle order) are:
1036 P Thread setup, but not started.
1038 I Thread initialized, waiting.
1039 p Thread running pre-reading file(s).
1040 R Running, doing sequential reads.
1041 r Running, doing random reads.
1042 W Running, doing sequential writes.
1043 w Running, doing random writes.
1044 M Running, doing mixed sequential reads/writes.
1045 m Running, doing mixed random reads/writes.
1046 F Running, currently waiting for fsync()
1047 V Running, doing verification of written data.
1048 E Thread exited, not reaped by main thread yet.
1051 The other values are fairly self explanatory - number of threads
1052 currently running and doing io, rate of io since last check (read speed
1053 listed first, then write speed), and the estimated completion percentage
1054 and time for the running group. It's impossible to estimate runtime of
1055 the following groups (if any).
1057 When fio is done (or interrupted by ctrl-c), it will show the data for
1058 each thread, group of threads, and disks in that order. For each data
1059 direction, the output looks like:
1061 Client1 (g=0): err= 0:
1062 write: io= 32MB, bw= 666KB/s, runt= 50320msec
1063 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1064 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1065 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1066 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1067 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1068 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1069 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1070 issued r/w: total=0/32768, short=0/0
1071 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1072 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1074 The client number is printed, along with the group id and error of that
1075 thread. Below is the io statistics, here for writes. In the order listed,
1078 io= Number of megabytes io performed
1079 bw= Average bandwidth rate
1080 runt= The runtime of that thread
1081 slat= Submission latency (avg being the average, stdev being the
1082 standard deviation). This is the time it took to submit
1083 the io. For sync io, the slat is really the completion
1084 latency, since queue/complete is one operation there. This
1085 value can be in milliseconds or microseconds, fio will choose
1086 the most appropriate base and print that. In the example
1087 above, milliseconds is the best scale.
1088 clat= Completion latency. Same names as slat, this denotes the
1089 time from submission to completion of the io pieces. For
1090 sync io, clat will usually be equal (or very close) to 0,
1091 as the time from submit to complete is basically just
1092 CPU time (io has already been done, see slat explanation).
1093 bw= Bandwidth. Same names as the xlat stats, but also includes
1094 an approximate percentage of total aggregate bandwidth
1095 this thread received in this group. This last value is
1096 only really useful if the threads in this group are on the
1097 same disk, since they are then competing for disk access.
1098 cpu= CPU usage. User and system time, along with the number
1099 of context switches this thread went through, usage of
1100 system and user time, and finally the number of major
1101 and minor page faults.
1102 IO depths= The distribution of io depths over the job life time. The
1103 numbers are divided into powers of 2, so for example the
1104 16= entries includes depths up to that value but higher
1105 than the previous entry. In other words, it covers the
1106 range from 16 to 31.
1107 IO submit= How many pieces of IO were submitting in a single submit
1108 call. Each entry denotes that amount and below, until
1109 the previous entry - eg, 8=100% mean that we submitted
1110 anywhere in between 5-8 ios per submit call.
1111 IO complete= Like the above submit number, but for completions instead.
1112 IO issued= The number of read/write requests issued, and how many
1114 IO latencies= The distribution of IO completion latencies. This is the
1115 time from when IO leaves fio and when it gets completed.
1116 The numbers follow the same pattern as the IO depths,
1117 meaning that 2=1.6% means that 1.6% of the IO completed
1118 within 2 msecs, 20=12.8% means that 12.8% of the IO
1119 took more than 10 msecs, but less than (or equal to) 20 msecs.
1121 After each client has been listed, the group statistics are printed. They
1122 will look like this:
1124 Run status group 0 (all jobs):
1125 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1126 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1128 For each data direction, it prints:
1130 io= Number of megabytes io performed.
1131 aggrb= Aggregate bandwidth of threads in this group.
1132 minb= The minimum average bandwidth a thread saw.
1133 maxb= The maximum average bandwidth a thread saw.
1134 mint= The smallest runtime of the threads in that group.
1135 maxt= The longest runtime of the threads in that group.
1137 And finally, the disk statistics are printed. They will look like this:
1139 Disk stats (read/write):
1140 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1142 Each value is printed for both reads and writes, with reads first. The
1145 ios= Number of ios performed by all groups.
1146 merge= Number of merges io the io scheduler.
1147 ticks= Number of ticks we kept the disk busy.
1148 io_queue= Total time spent in the disk queue.
1149 util= The disk utilization. A value of 100% means we kept the disk
1150 busy constantly, 50% would be a disk idling half of the time.
1156 For scripted usage where you typically want to generate tables or graphs
1157 of the results, fio can output the results in a semicolon separated format.
1158 The format is one long line of values, such as:
1160 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%
1161 ;0.0%;0.0%;0.0%;0.0%;0.0%
1163 To enable terse output, use the --minimal command line option.
1165 Split up, the format is as follows:
1167 jobname, groupid, error
1169 KB IO, bandwidth (KB/sec), runtime (msec)
1170 Submission latency: min, max, mean, deviation
1171 Completion latency: min, max, mean, deviation
1172 Bw: min, max, aggregate percentage of total, mean, deviation
1174 KB IO, bandwidth (KB/sec), runtime (msec)
1175 Submission latency: min, max, mean, deviation
1176 Completion latency: min, max, mean, deviation
1177 Bw: min, max, aggregate percentage of total, mean, deviation
1178 CPU usage: user, system, context switches, major faults, minor faults
1179 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1180 IO latencies: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, >=2000