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
209 5.0 Detailed list of parameters
210 -------------------------------
212 This section describes in details each parameter associated with a job.
213 Some parameters take an option of a given type, such as an integer or
214 a string. The following types are used:
216 str String. This is a sequence of alpha characters.
217 time Integer with possible time suffix. In seconds unless otherwise
218 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
220 int SI integer. A whole number value, which may contain a suffix
221 describing the base of the number. Accepted suffixes are k/m/g/t/p,
222 meaning kilo, mega, giga, tera, and peta. The suffix is not case
223 sensitive. So if you want to specify 4096, you could either write
224 out '4096' or just give 4k. The suffixes signify base 2 values, so
225 1024 is 1k and 1024k is 1m and so on. If the option accepts an upper
226 and lower range, use a colon ':' or minus '-' to separate such values.
227 May also include a prefix to indicate numbers base. If 0x is used,
228 the number is assumed to be hexadecimal. See irange.
229 bool Boolean. Usually parsed as an integer, however only defined for
230 true and false (1 and 0).
231 irange Integer range with suffix. Allows value range to be given, such
232 as 1024-4096. A colon may also be used as the separator, eg
233 1k:4k. If the option allows two sets of ranges, they can be
234 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
237 With the above in mind, here follows the complete list of fio job
240 name=str ASCII name of the job. This may be used to override the
241 name printed by fio for this job. Otherwise the job
242 name is used. On the command line this parameter has the
243 special purpose of also signaling the start of a new
246 description=str Text description of the job. Doesn't do anything except
247 dump this text description when this job is run. It's
250 directory=str Prefix filenames with this directory. Used to place files
251 in a different location than "./".
253 filename=str Fio normally makes up a filename based on the job name,
254 thread number, and file number. If you want to share
255 files between threads in a job or several jobs, specify
256 a filename for each of them to override the default. If
257 the ioengine used is 'net', the filename is the host, port,
258 and protocol to use in the format of =host/port/protocol.
259 See ioengine=net for more. If the ioengine is file based, you
260 can specify a number of files by separating the names with a
261 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
262 as the two working files, you would use
263 filename=/dev/sda:/dev/sdb. If the wanted filename does need to
264 include a colon, then escape that with a '\' character. For
265 instance, if the filename is "/dev/dsk/foo@3,0:c", then you would
266 use filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name,
267 meaning stdin or stdout. Which of the two depends on the read/write
270 opendir=str Tell fio to recursively add any file it can find in this
271 directory and down the file system tree.
273 lockfile=str Fio defaults to not locking any files before it does
274 IO to them. If a file or file descriptor is shared, fio
275 can serialize IO to that file to make the end result
276 consistent. This is usual for emulating real workloads that
277 share files. The lock modes are:
279 none No locking. The default.
280 exclusive Only one thread/process may do IO,
281 excluding all others.
282 readwrite Read-write locking on the file. Many
283 readers may access the file at the
284 same time, but writes get exclusive
287 The option may be post-fixed with a lock batch number. If
288 set, then each thread/process may do that amount of IOs to
289 the file before giving up the lock. Since lock acquisition is
290 expensive, batching the lock/unlocks will speed up IO.
293 rw=str Type of io pattern. Accepted values are:
295 read Sequential reads
296 write Sequential writes
297 randwrite Random writes
298 randread Random reads
299 rw Sequential mixed reads and writes
300 randrw Random mixed reads and writes
302 For the mixed io types, the default is to split them 50/50.
303 For certain types of io the result may still be skewed a bit,
304 since the speed may be different. It is possible to specify
305 a number of IO's to do before getting a new offset - this
306 is only useful for random IO, where fio would normally
307 generate a new random offset for every IO. If you append
308 eg 8 to randread, you would get a new random offset for
309 every 8 IO's. The result would be a seek for only every 8
310 IO's, instead of for every IO. Use rw=randread:8 to specify
313 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
314 Storage manufacturers like to use 10^3 or 1000 as a base
315 ten unit instead, for obvious reasons. Allow values are
316 1024 or 1000, with 1024 being the default.
318 randrepeat=bool For random IO workloads, seed the generator in a predictable
319 way so that results are repeatable across repetitions.
321 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
322 on what IO patterns it is likely to issue. Sometimes you
323 want to test specific IO patterns without telling the
324 kernel about it, in which case you can disable this option.
325 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
326 IO and POSIX_FADV_RANDOM for random IO.
328 size=int The total size of file io for this job. Fio will run until
329 this many bytes has been transferred, unless runtime is
330 limited by other options (such as 'runtime', for instance).
331 Unless specific nrfiles and filesize options are given,
332 fio will divide this size between the available files
333 specified by the job.
335 filesize=int Individual file sizes. May be a range, in which case fio
336 will select sizes for files at random within the given range
337 and limited to 'size' in total (if that is given). If not
338 given, each created file is the same size.
340 fill_device=bool Sets size to something really large and waits for ENOSPC (no
341 space left on device) as the terminating condition. Only makes
342 sense with sequential write. For a read workload, the mount
343 point will be filled first then IO started on the result.
346 bs=int The block size used for the io units. Defaults to 4k. Values
347 can be given for both read and writes. If a single int is
348 given, it will apply to both. If a second int is specified
349 after a comma, it will apply to writes only. In other words,
350 the format is either bs=read_and_write or bs=read,write.
351 bs=4k,8k will thus use 4k blocks for reads, and 8k blocks
352 for writes. If you only wish to set the write size, you
353 can do so by passing an empty read size - bs=,8k will set
354 8k for writes and leave the read default value.
357 ba=int At what boundary to align random IO offsets. Defaults to
358 the same as 'blocksize' the minimum blocksize given.
359 Minimum alignment is typically 512b for using direct IO,
360 though it usually depends on the hardware block size. This
361 option is mutually exclusive with using a random map for
362 files, so it will turn off that option.
364 blocksize_range=irange
365 bsrange=irange Instead of giving a single block size, specify a range
366 and fio will mix the issued io block sizes. The issued
367 io unit will always be a multiple of the minimum value
368 given (also see bs_unaligned). Applies to both reads and
369 writes, however a second range can be given after a comma.
372 bssplit=str Sometimes you want even finer grained control of the
373 block sizes issued, not just an even split between them.
374 This option allows you to weight various block sizes,
375 so that you are able to define a specific amount of
376 block sizes issued. The format for this option is:
378 bssplit=blocksize/percentage:blocksize/percentage
380 for as many block sizes as needed. So if you want to define
381 a workload that has 50% 64k blocks, 10% 4k blocks, and
382 40% 32k blocks, you would write:
384 bssplit=4k/10:64k/50:32k/40
386 Ordering does not matter. If the percentage is left blank,
387 fio will fill in the remaining values evenly. So a bssplit
388 option like this one:
390 bssplit=4k/50:1k/:32k/
392 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
393 always add up to 100, if bssplit is given a range that adds
394 up to more, it will error out.
396 bssplit also supports giving separate splits to reads and
397 writes. The format is identical to what bs= accepts. You
398 have to separate the read and write parts with a comma. So
399 if you want a workload that has 50% 2k reads and 50% 4k reads,
400 while having 90% 4k writes and 10% 8k writes, you would
403 bssplit=2k/50:4k/50,4k/90,8k/10
406 bs_unaligned If this option is given, any byte size value within bsrange
407 may be used as a block range. This typically wont work with
408 direct IO, as that normally requires sector alignment.
410 zero_buffers If this option is given, fio will init the IO buffers to
411 all zeroes. The default is to fill them with random data.
413 refill_buffers If this option is given, fio will refill the IO buffers
414 on every submit. The default is to only fill it at init
415 time and reuse that data. Only makes sense if zero_buffers
416 isn't specified, naturally. If data verification is enabled,
417 refill_buffers is also automatically enabled.
419 nrfiles=int Number of files to use for this job. Defaults to 1.
421 openfiles=int Number of files to keep open at the same time. Defaults to
422 the same as nrfiles, can be set smaller to limit the number
425 file_service_type=str Defines how fio decides which file from a job to
426 service next. The following types are defined:
428 random Just choose a file at random.
430 roundrobin Round robin over open files. This
433 sequential Finish one file before moving on to
434 the next. Multiple files can still be
435 open depending on 'openfiles'.
437 The string can have a number appended, indicating how
438 often to switch to a new file. So if option random:4 is
439 given, fio will switch to a new random file after 4 ios
442 ioengine=str Defines how the job issues io to the file. The following
445 sync Basic read(2) or write(2) io. lseek(2) is
446 used to position the io location.
448 psync Basic pread(2) or pwrite(2) io.
450 vsync Basic readv(2) or writev(2) IO.
452 libaio Linux native asynchronous io. Note that Linux
453 may only support queued behaviour with
454 non-buffered IO (set direct=1 or buffered=0).
456 posixaio glibc posix asynchronous io.
458 solarisaio Solaris native asynchronous io.
460 mmap File is memory mapped and data copied
461 to/from using memcpy(3).
463 splice splice(2) is used to transfer the data and
464 vmsplice(2) to transfer data from user
467 syslet-rw Use the syslet system calls to make
468 regular read/write async.
470 sg SCSI generic sg v3 io. May either be
471 synchronous using the SG_IO ioctl, or if
472 the target is an sg character device
473 we use read(2) and write(2) for asynchronous
476 null Doesn't transfer any data, just pretends
477 to. This is mainly used to exercise fio
478 itself and for debugging/testing purposes.
480 net Transfer over the network to given host:port.
481 'filename' must be set appropriately to
482 filename=host/port/protocol regardless of send
483 or receive, if the latter only the port
484 argument is used. 'host' may be an IP address
485 or hostname, port is the port number to be used,
486 and protocol may be 'udp' or 'tcp'. If no
487 protocol is given, TCP is used.
489 netsplice Like net, but uses splice/vmsplice to
490 map data and send/receive.
492 cpuio Doesn't transfer any data, but burns CPU
493 cycles according to the cpuload= and
494 cpucycle= options. Setting cpuload=85
495 will cause that job to do nothing but burn
496 85% of the CPU. In case of SMP machines,
497 use numjobs=<no_of_cpu> to get desired CPU
498 usage, as the cpuload only loads a single
499 CPU at the desired rate.
501 guasi The GUASI IO engine is the Generic Userspace
502 Asyncronous Syscall Interface approach
505 http://www.xmailserver.org/guasi-lib.html
507 for more info on GUASI.
509 external Prefix to specify loading an external
510 IO engine object file. Append the engine
511 filename, eg ioengine=external:/tmp/foo.o
512 to load ioengine foo.o in /tmp.
514 iodepth=int This defines how many io units to keep in flight against
515 the file. The default is 1 for each file defined in this
516 job, can be overridden with a larger value for higher
519 iodepth_batch_submit=int
520 iodepth_batch=int This defines how many pieces of IO to submit at once.
521 It defaults to 1 which means that we submit each IO
522 as soon as it is available, but can be raised to submit
523 bigger batches of IO at the time.
525 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
526 at once. It defaults to 1 which means that we'll ask
527 for a minimum of 1 IO in the retrieval process from
528 the kernel. The IO retrieval will go on until we
529 hit the limit set by iodepth_low. If this variable is
530 set to 0, then fio will always check for completed
531 events before queuing more IO. This helps reduce
532 IO latency, at the cost of more retrieval system calls.
534 iodepth_low=int The low water mark indicating when to start filling
535 the queue again. Defaults to the same as iodepth, meaning
536 that fio will attempt to keep the queue full at all times.
537 If iodepth is set to eg 16 and iodepth_low is set to 4, then
538 after fio has filled the queue of 16 requests, it will let
539 the depth drain down to 4 before starting to fill it again.
541 direct=bool If value is true, use non-buffered io. This is usually
544 buffered=bool If value is true, use buffered io. This is the opposite
545 of the 'direct' option. Defaults to true.
547 offset=int Start io at the given offset in the file. The data before
548 the given offset will not be touched. This effectively
549 caps the file size at real_size - offset.
551 fsync=int If writing to a file, issue a sync of the dirty data
552 for every number of blocks given. For example, if you give
553 32 as a parameter, fio will sync the file for every 32
554 writes issued. If fio is using non-buffered io, we may
555 not sync the file. The exception is the sg io engine, which
556 synchronizes the disk cache anyway.
558 fsyncdata=int Like fsync= but uses fdatasync() to only sync data and not
561 overwrite=bool If true, writes to a file will always overwrite existing
562 data. If the file doesn't already exist, it will be
563 created before the write phase begins. If the file exists
564 and is large enough for the specified write phase, nothing
567 end_fsync=bool If true, fsync file contents when the job exits.
569 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
570 This differs from end_fsync in that it will happen on every
571 file close, not just at the end of the job.
573 rwmixread=int How large a percentage of the mix should be reads.
575 rwmixwrite=int How large a percentage of the mix should be writes. If both
576 rwmixread and rwmixwrite is given and the values do not add
577 up to 100%, the latter of the two will be used to override
578 the first. This may interfere with a given rate setting,
579 if fio is asked to limit reads or writes to a certain rate.
580 If that is the case, then the distribution may be skewed.
582 norandommap Normally fio will cover every block of the file when doing
583 random IO. If this option is given, fio will just get a
584 new random offset without looking at past io history. This
585 means that some blocks may not be read or written, and that
586 some blocks may be read/written more than once. This option
587 is mutually exclusive with verify= if and only if multiple
588 blocksizes (via bsrange=) are used, since fio only tracks
589 complete rewrites of blocks.
591 softrandommap See norandommap. If fio runs with the random block map enabled
592 and it fails to allocate the map, if this option is set it
593 will continue without a random block map. As coverage will
594 not be as complete as with random maps, this option is
597 nice=int Run the job with the given nice value. See man nice(2).
599 prio=int Set the io priority value of this job. Linux limits us to
600 a positive value between 0 and 7, with 0 being the highest.
603 prioclass=int Set the io priority class. See man ionice(1).
605 thinktime=int Stall the job x microseconds after an io has completed before
606 issuing the next. May be used to simulate processing being
607 done by an application. See thinktime_blocks and
611 Only valid if thinktime is set - pretend to spend CPU time
612 doing something with the data received, before falling back
613 to sleeping for the rest of the period specified by
617 Only valid if thinktime is set - control how many blocks
618 to issue, before waiting 'thinktime' usecs. If not set,
619 defaults to 1 which will make fio wait 'thinktime' usecs
622 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
623 the normal suffix rules apply. You can use rate=500k to limit
624 reads and writes to 500k each, or you can specify read and
625 writes separately. Using rate=1m,500k would limit reads to
626 1MB/sec and writes to 500KB/sec. Capping only reads or
627 writes can be done with rate=,500k or rate=500k,. The former
628 will only limit writes (to 500KB/sec), the latter will only
631 ratemin=int Tell fio to do whatever it can to maintain at least this
632 bandwidth. Failing to meet this requirement, will cause
633 the job to exit. The same format as rate is used for
634 read vs write separation.
636 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
637 as rate, just specified independently of bandwidth. If the
638 job is given a block size range instead of a fixed value,
639 the smallest block size is used as the metric. The same format
640 as rate is used for read vs write seperation.
642 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
643 the job to exit. The same format as rate is used for read vs
646 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
649 cpumask=int Set the CPU affinity of this job. The parameter given is a
650 bitmask of allowed CPU's the job may run on. So if you want
651 the allowed CPUs to be 1 and 5, you would pass the decimal
652 value of (1 << 1 | 1 << 5), or 34. See man
653 sched_setaffinity(2). This may not work on all supported
654 operating systems or kernel versions. This option doesn't
655 work well for a higher CPU count than what you can store in
656 an integer mask, so it can only control cpus 1-32. For
657 boxes with larger CPU counts, use cpus_allowed.
659 cpus_allowed=str Controls the same options as cpumask, but it allows a text
660 setting of the permitted CPUs instead. So to use CPUs 1 and
661 5, you would specify cpus_allowed=1,5. This options also
662 allows a range of CPUs. Say you wanted a binding to CPUs
663 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
665 startdelay=time Start this job the specified number of seconds after fio
666 has started. Only useful if the job file contains several
667 jobs, and you want to delay starting some jobs to a certain
670 runtime=time Tell fio to terminate processing after the specified number
671 of seconds. It can be quite hard to determine for how long
672 a specified job will run, so this parameter is handy to
673 cap the total runtime to a given time.
675 time_based If set, fio will run for the duration of the runtime
676 specified even if the file(s) are completely read or
677 written. It will simply loop over the same workload
678 as many times as the runtime allows.
680 ramp_time=time If set, fio will run the specified workload for this amount
681 of time before logging any performance numbers. Useful for
682 letting performance settle before logging results, thus
683 minimizing the runtime required for stable results. Note
684 that the ramp_time is considered lead in time for a job,
685 thus it will increase the total runtime if a special timeout
686 or runtime is specified.
688 invalidate=bool Invalidate the buffer/page cache parts for this file prior
689 to starting io. Defaults to true.
691 sync=bool Use sync io for buffered writes. For the majority of the
692 io engines, this means using O_SYNC.
695 mem=str Fio can use various types of memory as the io unit buffer.
696 The allowed values are:
698 malloc Use memory from malloc(3) as the buffers.
700 shm Use shared memory as the buffers. Allocated
703 shmhuge Same as shm, but use huge pages as backing.
705 mmap Use mmap to allocate buffers. May either be
706 anonymous memory, or can be file backed if
707 a filename is given after the option. The
708 format is mem=mmap:/path/to/file.
710 mmaphuge Use a memory mapped huge file as the buffer
711 backing. Append filename after mmaphuge, ala
712 mem=mmaphuge:/hugetlbfs/file
714 The area allocated is a function of the maximum allowed
715 bs size for the job, multiplied by the io depth given. Note
716 that for shmhuge and mmaphuge to work, the system must have
717 free huge pages allocated. This can normally be checked
718 and set by reading/writing /proc/sys/vm/nr_hugepages on a
719 Linux system. Fio assumes a huge page is 4MB in size. So
720 to calculate the number of huge pages you need for a given
721 job file, add up the io depth of all jobs (normally one unless
722 iodepth= is used) and multiply by the maximum bs set. Then
723 divide that number by the huge page size. You can see the
724 size of the huge pages in /proc/meminfo. If no huge pages
725 are allocated by having a non-zero number in nr_hugepages,
726 using mmaphuge or shmhuge will fail. Also see hugepage-size.
728 mmaphuge also needs to have hugetlbfs mounted and the file
729 location should point there. So if it's mounted in /huge,
730 you would use mem=mmaphuge:/huge/somefile.
732 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
733 Note that the given alignment is applied to the first IO unit
734 buffer, if using iodepth the alignment of the following buffers
735 are given by the bs used. In other words, if using a bs that is
736 a multiple of the page sized in the system, all buffers will
737 be aligned to this value. If using a bs that is not page
738 aligned, the alignment of subsequent IO memory buffers is the
739 sum of the iomem_align and bs used.
742 Defines the size of a huge page. Must at least be equal
743 to the system setting, see /proc/meminfo. Defaults to 4MB.
744 Should probably always be a multiple of megabytes, so using
745 hugepage-size=Xm is the preferred way to set this to avoid
746 setting a non-pow-2 bad value.
748 exitall When one job finishes, terminate the rest. The default is
749 to wait for each job to finish, sometimes that is not the
752 bwavgtime=int Average the calculated bandwidth over the given time. Value
753 is specified in milliseconds.
755 create_serialize=bool If true, serialize the file creating for the jobs.
756 This may be handy to avoid interleaving of data
757 files, which may greatly depend on the filesystem
758 used and even the number of processors in the system.
760 create_fsync=bool fsync the data file after creation. This is the
763 create_on_open=bool Don't pre-setup the files for IO, just create open()
764 when it's time to do IO to that file.
766 pre_read=bool If this is given, files will be pre-read into memory before
767 starting the given IO operation. This will also clear
768 the 'invalidate' flag, since it is pointless to pre-read
769 and then drop the cache. This will only work for IO engines
770 that are seekable, since they allow you to read the same data
771 multiple times. Thus it will not work on eg network or splice
774 unlink=bool Unlink the job files when done. Not the default, as repeated
775 runs of that job would then waste time recreating the file
778 loops=int Run the specified number of iterations of this job. Used
779 to repeat the same workload a given number of times. Defaults
782 do_verify=bool Run the verify phase after a write phase. Only makes sense if
783 verify is set. Defaults to 1.
785 verify=str If writing to a file, fio can verify the file contents
786 after each iteration of the job. The allowed values are:
788 md5 Use an md5 sum of the data area and store
789 it in the header of each block.
791 crc64 Use an experimental crc64 sum of the data
792 area and store it in the header of each
795 crc32c Use a crc32c sum of the data area and store
796 it in the header of each block.
798 crc32c-intel Use hardware assisted crc32c calcuation
799 provided on SSE4.2 enabled processors.
801 crc32 Use a crc32 sum of the data area and store
802 it in the header of each block.
804 crc16 Use a crc16 sum of the data area and store
805 it in the header of each block.
807 crc7 Use a crc7 sum of the data area and store
808 it in the header of each block.
810 sha512 Use sha512 as the checksum function.
812 sha256 Use sha256 as the checksum function.
814 sha1 Use optimized sha1 as the checksum function.
816 meta Write extra information about each io
817 (timestamp, block number etc.). The block
820 null Only pretend to verify. Useful for testing
821 internals with ioengine=null, not for much
824 This option can be used for repeated burn-in tests of a
825 system to make sure that the written data is also
826 correctly read back. If the data direction given is
827 a read or random read, fio will assume that it should
828 verify a previously written file. If the data direction
829 includes any form of write, the verify will be of the
832 verifysort=bool If set, fio will sort written verify blocks when it deems
833 it faster to read them back in a sorted manner. This is
834 often the case when overwriting an existing file, since
835 the blocks are already laid out in the file system. You
836 can ignore this option unless doing huge amounts of really
837 fast IO where the red-black tree sorting CPU time becomes
840 verify_offset=int Swap the verification header with data somewhere else
841 in the block before writing. Its swapped back before
844 verify_interval=int Write the verification header at a finer granularity
845 than the blocksize. It will be written for chunks the
846 size of header_interval. blocksize should divide this
849 verify_pattern=int If set, fio will fill the io buffers with this
850 pattern. Fio defaults to filling with totally random
851 bytes, but sometimes it's interesting to fill with a known
852 pattern for io verification purposes. Depending on the
853 width of the pattern, fio will fill 1/2/3/4 bytes of the
854 buffer at the time. The verify_pattern cannot be larger than
857 verify_fatal=bool Normally fio will keep checking the entire contents
858 before quitting on a block verification failure. If this
859 option is set, fio will exit the job on the first observed
862 verify_async=int Fio will normally verify IO inline from the submitting
863 thread. This option takes an integer describing how many
864 async offload threads to create for IO verification instead,
865 causing fio to offload the duty of verifying IO contents
866 to one or more separate threads. If using this offload
867 option, even sync IO engines can benefit from using an
868 iodepth setting higher than 1, as it allows them to have
869 IO in flight while verifies are running.
871 verify_async_cpus=str Tell fio to set the given CPU affinity on the
872 async IO verification threads. See cpus_allowed for the
875 stonewall Wait for preceeding jobs in the job file to exit, before
876 starting this one. Can be used to insert serialization
877 points in the job file. A stone wall also implies starting
878 a new reporting group.
880 new_group Start a new reporting group. If this option isn't given,
881 jobs in a file will be part of the same reporting group
882 unless separated by a stone wall (or if it's a group
883 by itself, with the numjobs option).
885 numjobs=int Create the specified number of clones of this job. May be
886 used to setup a larger number of threads/processes doing
887 the same thing. We regard that grouping of jobs as a
890 group_reporting If 'numjobs' is set, it may be interesting to display
891 statistics for the group as a whole instead of for each
892 individual job. This is especially true of 'numjobs' is
893 large, looking at individual thread/process output quickly
894 becomes unwieldy. If 'group_reporting' is specified, fio
895 will show the final report per-group instead of per-job.
897 thread fio defaults to forking jobs, however if this option is
898 given, fio will use pthread_create(3) to create threads
901 zonesize=int Divide a file into zones of the specified size. See zoneskip.
903 zoneskip=int Skip the specified number of bytes when zonesize data has
904 been read. The two zone options can be used to only do
905 io on zones of a file.
907 write_iolog=str Write the issued io patterns to the specified file. See
910 read_iolog=str Open an iolog with the specified file name and replay the
911 io patterns it contains. This can be used to store a
912 workload and replay it sometime later. The iolog given
913 may also be a blktrace binary file, which allows fio
914 to replay a workload captured by blktrace. See blktrace
915 for how to capture such logging data. For blktrace replay,
916 the file needs to be turned into a blkparse binary data
917 file first (blktrace <device> -d file_for_fio.bin).
919 write_bw_log=str If given, write a bandwidth log of the jobs in this job
920 file. Can be used to store data of the bandwidth of the
921 jobs in their lifetime. The included fio_generate_plots
922 script uses gnuplot to turn these text files into nice
923 graphs. See write_log_log for behaviour of given
924 filename. For this option, the postfix is _bw.log.
926 write_lat_log=str Same as write_bw_log, except that this option stores io
927 completion latencies instead. If no filename is given
928 with this option, the default filename of "jobname_type.log"
929 is used. Even if the filename is given, fio will still
930 append the type of log. So if one specifies
934 The actual log names will be foo_clat.log and foo_slat.log.
935 This helps fio_generate_plot fine the logs automatically.
937 lockmem=int Pin down the specified amount of memory with mlock(2). Can
938 potentially be used instead of removing memory or booting
939 with less memory to simulate a smaller amount of memory.
941 exec_prerun=str Before running this job, issue the command specified
944 exec_postrun=str After the job completes, issue the command specified
947 ioscheduler=str Attempt to switch the device hosting the file to the specified
948 io scheduler before running.
950 cpuload=int If the job is a CPU cycle eater, attempt to use the specified
951 percentage of CPU cycles.
953 cpuchunks=int If the job is a CPU cycle eater, split the load into
954 cycles of the given time. In microseconds.
956 disk_util=bool Generate disk utilization statistics, if the platform
957 supports it. Defaults to on.
959 disable_clat=bool Disable measurements of completion latency numbers. Useful
960 only for cutting back the number of calls to gettimeofday,
961 as that does impact performance at really high IOPS rates.
962 Note that to really get rid of a large amount of these
963 calls, this option must be used with disable_slat and
966 disable_slat=bool Disable measurements of submission latency numbers. See
969 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
972 gtod_reduce=bool Enable all of the gettimeofday() reducing options
973 (disable_clat, disable_slat, disable_bw) plus reduce
974 precision of the timeout somewhat to really shrink
975 the gettimeofday() call count. With this option enabled,
976 we only do about 0.4% of the gtod() calls we would have
977 done if all time keeping was enabled.
979 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
980 execution to just getting the current time. Fio (and
981 databases, for instance) are very intensive on gettimeofday()
982 calls. With this option, you can set one CPU aside for
983 doing nothing but logging current time to a shared memory
984 location. Then the other threads/processes that run IO
985 workloads need only copy that segment, instead of entering
986 the kernel with a gettimeofday() call. The CPU set aside
987 for doing these time calls will be excluded from other
988 uses. Fio will manually clear it from the CPU mask of other
990 continue_on_error=bool Normally fio will exit the job on the first observed
991 failure. If this option is set, fio will continue the job when
992 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
993 is exceeded or the I/O size specified is completed. If this
994 option is used, there are two more stats that are appended,
995 the total error count and the first error. The error field
996 given in the stats is the first error that was hit during the
1000 6.0 Interpreting the output
1001 ---------------------------
1003 fio spits out a lot of output. While running, fio will display the
1004 status of the jobs created. An example of that would be:
1006 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1008 The characters inside the square brackets denote the current status of
1009 each thread. The possible values (in typical life cycle order) are:
1013 P Thread setup, but not started.
1015 I Thread initialized, waiting.
1016 p Thread running pre-reading file(s).
1017 R Running, doing sequential reads.
1018 r Running, doing random reads.
1019 W Running, doing sequential writes.
1020 w Running, doing random writes.
1021 M Running, doing mixed sequential reads/writes.
1022 m Running, doing mixed random reads/writes.
1023 F Running, currently waiting for fsync()
1024 V Running, doing verification of written data.
1025 E Thread exited, not reaped by main thread yet.
1028 The other values are fairly self explanatory - number of threads
1029 currently running and doing io, rate of io since last check (read speed
1030 listed first, then write speed), and the estimated completion percentage
1031 and time for the running group. It's impossible to estimate runtime of
1032 the following groups (if any).
1034 When fio is done (or interrupted by ctrl-c), it will show the data for
1035 each thread, group of threads, and disks in that order. For each data
1036 direction, the output looks like:
1038 Client1 (g=0): err= 0:
1039 write: io= 32MB, bw= 666KB/s, runt= 50320msec
1040 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1041 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1042 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1043 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1044 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1045 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1046 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1047 issued r/w: total=0/32768, short=0/0
1048 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1049 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1051 The client number is printed, along with the group id and error of that
1052 thread. Below is the io statistics, here for writes. In the order listed,
1055 io= Number of megabytes io performed
1056 bw= Average bandwidth rate
1057 runt= The runtime of that thread
1058 slat= Submission latency (avg being the average, stdev being the
1059 standard deviation). This is the time it took to submit
1060 the io. For sync io, the slat is really the completion
1061 latency, since queue/complete is one operation there. This
1062 value can be in milliseconds or microseconds, fio will choose
1063 the most appropriate base and print that. In the example
1064 above, milliseconds is the best scale.
1065 clat= Completion latency. Same names as slat, this denotes the
1066 time from submission to completion of the io pieces. For
1067 sync io, clat will usually be equal (or very close) to 0,
1068 as the time from submit to complete is basically just
1069 CPU time (io has already been done, see slat explanation).
1070 bw= Bandwidth. Same names as the xlat stats, but also includes
1071 an approximate percentage of total aggregate bandwidth
1072 this thread received in this group. This last value is
1073 only really useful if the threads in this group are on the
1074 same disk, since they are then competing for disk access.
1075 cpu= CPU usage. User and system time, along with the number
1076 of context switches this thread went through, usage of
1077 system and user time, and finally the number of major
1078 and minor page faults.
1079 IO depths= The distribution of io depths over the job life time. The
1080 numbers are divided into powers of 2, so for example the
1081 16= entries includes depths up to that value but higher
1082 than the previous entry. In other words, it covers the
1083 range from 16 to 31.
1084 IO submit= How many pieces of IO were submitting in a single submit
1085 call. Each entry denotes that amount and below, until
1086 the previous entry - eg, 8=100% mean that we submitted
1087 anywhere in between 5-8 ios per submit call.
1088 IO complete= Like the above submit number, but for completions instead.
1089 IO issued= The number of read/write requests issued, and how many
1091 IO latencies= The distribution of IO completion latencies. This is the
1092 time from when IO leaves fio and when it gets completed.
1093 The numbers follow the same pattern as the IO depths,
1094 meaning that 2=1.6% means that 1.6% of the IO completed
1095 within 2 msecs, 20=12.8% means that 12.8% of the IO
1096 took more than 10 msecs, but less than (or equal to) 20 msecs.
1098 After each client has been listed, the group statistics are printed. They
1099 will look like this:
1101 Run status group 0 (all jobs):
1102 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1103 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1105 For each data direction, it prints:
1107 io= Number of megabytes io performed.
1108 aggrb= Aggregate bandwidth of threads in this group.
1109 minb= The minimum average bandwidth a thread saw.
1110 maxb= The maximum average bandwidth a thread saw.
1111 mint= The smallest runtime of the threads in that group.
1112 maxt= The longest runtime of the threads in that group.
1114 And finally, the disk statistics are printed. They will look like this:
1116 Disk stats (read/write):
1117 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1119 Each value is printed for both reads and writes, with reads first. The
1122 ios= Number of ios performed by all groups.
1123 merge= Number of merges io the io scheduler.
1124 ticks= Number of ticks we kept the disk busy.
1125 io_queue= Total time spent in the disk queue.
1126 util= The disk utilization. A value of 100% means we kept the disk
1127 busy constantly, 50% would be a disk idling half of the time.
1133 For scripted usage where you typically want to generate tables or graphs
1134 of the results, fio can output the results in a semicolon separated format.
1135 The format is one long line of values, such as:
1137 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%
1138 ;0.0%;0.0%;0.0%;0.0%;0.0%
1140 To enable terse output, use the --minimal command line option.
1142 Split up, the format is as follows:
1144 jobname, groupid, error
1146 KB IO, bandwidth (KB/sec), runtime (msec)
1147 Submission latency: min, max, mean, deviation
1148 Completion latency: min, max, mean, deviation
1149 Bw: min, max, aggregate percentage of total, mean, deviation
1151 KB IO, bandwidth (KB/sec), runtime (msec)
1152 Submission latency: min, max, mean, deviation
1153 Completion latency: min, max, mean, deviation
1154 Bw: min, max, aggregate percentage of total, mean, deviation
1155 CPU usage: user, system, context switches, major faults, minor faults
1156 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1157 IO latencies: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, >=2000