8 5. Detailed list of parameters
12 9. CPU idleness profiling
14 1.0 Overview and history
15 ------------------------
16 fio was originally written to save me the hassle of writing special test
17 case programs when I wanted to test a specific workload, either for
18 performance reasons or to find/reproduce a bug. The process of writing
19 such a test app can be tiresome, especially if you have to do it often.
20 Hence I needed a tool that would be able to simulate a given io workload
21 without resorting to writing a tailored test case again and again.
23 A test work load is difficult to define, though. There can be any number
24 of processes or threads involved, and they can each be using their own
25 way of generating io. You could have someone dirtying large amounts of
26 memory in an memory mapped file, or maybe several threads issuing
27 reads using asynchronous io. fio needed to be flexible enough to
28 simulate both of these cases, and many more.
32 The first step in getting fio to simulate a desired io workload, is
33 writing a job file describing that specific setup. A job file may contain
34 any number of threads and/or files - the typical contents of the job file
35 is a global section defining shared parameters, and one or more job
36 sections describing the jobs involved. When run, fio parses this file
37 and sets everything up as described. If we break down a job from top to
38 bottom, it contains the following basic parameters:
40 IO type Defines the io pattern issued to the file(s).
41 We may only be reading sequentially from this
42 file(s), or we may be writing randomly. Or even
43 mixing reads and writes, sequentially or randomly.
45 Block size In how large chunks are we issuing io? This may be
46 a single value, or it may describe a range of
49 IO size How much data are we going to be reading/writing.
51 IO engine How do we issue io? We could be memory mapping the
52 file, we could be using regular read/write, we
53 could be using splice, async io, syslet, or even
56 IO depth If the io engine is async, how large a queuing
57 depth do we want to maintain?
59 IO type Should we be doing buffered io, or direct/raw io?
61 Num files How many files are we spreading the workload over.
63 Num threads How many threads or processes should we spread
66 The above are the basic parameters defined for a workload, in addition
67 there's a multitude of parameters that modify other aspects of how this
73 See the README file for command line parameters, there are only a few
76 Running fio is normally the easiest part - you just give it the job file
77 (or job files) as parameters:
81 and it will start doing what the job_file tells it to do. You can give
82 more than one job file on the command line, fio will serialize the running
83 of those files. Internally that is the same as using the 'stonewall'
84 parameter described in the parameter section.
86 If the job file contains only one job, you may as well just give the
87 parameters on the command line. The command line parameters are identical
88 to the job parameters, with a few extra that control global parameters
89 (see README). For example, for the job file parameter iodepth=2, the
90 mirror command line option would be --iodepth 2 or --iodepth=2. You can
91 also use the command line for giving more than one job entry. For each
92 --name option that fio sees, it will start a new job with that name.
93 Command line entries following a --name entry will apply to that job,
94 until there are no more entries or a new --name entry is seen. This is
95 similar to the job file options, where each option applies to the current
96 job until a new [] job entry is seen.
98 fio does not need to run as root, except if the files or devices specified
99 in the job section requires that. Some other options may also be restricted,
100 such as memory locking, io scheduler switching, and decreasing the nice value.
105 As previously described, fio accepts one or more job files describing
106 what it is supposed to do. The job file format is the classic ini file,
107 where the names enclosed in [] brackets define the job name. You are free
108 to use any ascii name you want, except 'global' which has special meaning.
109 A global section sets defaults for the jobs described in that file. A job
110 may override a global section parameter, and a job file may even have
111 several global sections if so desired. A job is only affected by a global
112 section residing above it. If the first character in a line is a ';' or a
113 '#', the entire line is discarded as a comment.
115 So let's look at a really simple job file that defines two processes, each
116 randomly reading from a 128MB file.
118 ; -- start job file --
129 As you can see, the job file sections themselves are empty as all the
130 described parameters are shared. As no filename= option is given, fio
131 makes up a filename for each of the jobs as it sees fit. On the command
132 line, this job would look as follows:
134 $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2
137 Let's look at an example that has a number of processes writing randomly
140 ; -- start job file --
152 Here we have no global section, as we only have one job defined anyway.
153 We want to use async io here, with a depth of 4 for each file. We also
154 increased the buffer size used to 32KB and define numjobs to 4 to
155 fork 4 identical jobs. The result is 4 processes each randomly writing
156 to their own 64MB file. Instead of using the above job file, you could
157 have given the parameters on the command line. For this case, you would
160 $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4
162 When fio is utilized as a basis of any reasonably large test suite, it might be
163 desirable to share a set of standardized settings across multiple job files.
164 Instead of copy/pasting such settings, any section may pull in an external
165 .fio file with 'include filename' directive, as in the following example:
167 ; -- start job file including.fio --
171 include glob-include.fio
178 include test-include.fio
179 ; -- end job file including.fio --
181 ; -- start job file glob-include.fio --
184 ; -- end job file glob-include.fio --
186 ; -- start job file test-include.fio --
189 ; -- end job file test-include.fio --
191 Settings pulled into a section apply to that section only (except global
192 section). Include directives may be nested in that any included file may
193 contain further include directive(s). Include files may not contain []
197 4.1 Environment variables
198 -------------------------
200 fio also supports environment variable expansion in job files. Any
201 substring of the form "${VARNAME}" as part of an option value (in other
202 words, on the right of the `='), will be expanded to the value of the
203 environment variable called VARNAME. If no such environment variable
204 is defined, or VARNAME is the empty string, the empty string will be
207 As an example, let's look at a sample fio invocation and job file:
209 $ SIZE=64m NUMJOBS=4 fio jobfile.fio
211 ; -- start job file --
218 This will expand to the following equivalent job file at runtime:
220 ; -- start job file --
227 fio ships with a few example job files, you can also look there for
230 4.2 Reserved keywords
231 ---------------------
233 Additionally, fio has a set of reserved keywords that will be replaced
234 internally with the appropriate value. Those keywords are:
236 $pagesize The architecture page size of the running system
237 $mb_memory Megabytes of total memory in the system
238 $ncpus Number of online available CPUs
240 These can be used on the command line or in the job file, and will be
241 automatically substituted with the current system values when the job
242 is run. Simple math is also supported on these keywords, so you can
243 perform actions like:
247 and get that properly expanded to 8 times the size of memory in the
251 5.0 Detailed list of parameters
252 -------------------------------
254 This section describes in details each parameter associated with a job.
255 Some parameters take an option of a given type, such as an integer or
256 a string. Anywhere a numeric value is required, an arithmetic expression
257 may be used, provided it is surrounded by parentheses. Supported operators
267 For time values in expressions, units are microseconds by default. This is
268 different than for time values not in expressions (not enclosed in
269 parentheses). The following types are used:
271 str String. This is a sequence of alpha characters.
272 time Integer with possible time suffix. In seconds unless otherwise
273 specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
274 minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
275 and 'us' (or 'usec') for microseconds.
276 int SI integer. A whole number value, which may contain a suffix
277 describing the base of the number. Accepted suffixes are k/m/g/t/p,
278 meaning kilo, mega, giga, tera, and peta. The suffix is not case
279 sensitive, and you may also include trailing 'b' (eg 'kb' is the same
280 as 'k'). So if you want to specify 4096, you could either write
281 out '4096' or just give 4k. The suffixes signify base 2 values, so
282 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
283 set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
284 case, then 1000 is used as the multiplier. This can be handy for
285 disks, since manufacturers generally use base 10 values when listing
286 the capacity of a drive. If the option accepts an upper and lower
287 range, use a colon ':' or minus '-' to separate such values. May also
288 include a prefix to indicate numbers base. If 0x is used, the number
289 is assumed to be hexadecimal. See irange.
290 bool Boolean. Usually parsed as an integer, however only defined for
291 true and false (1 and 0).
292 irange Integer range with suffix. Allows value range to be given, such
293 as 1024-4096. A colon may also be used as the separator, eg
294 1k:4k. If the option allows two sets of ranges, they can be
295 specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
297 float_list A list of floating numbers, separated by a ':' character.
299 With the above in mind, here follows the complete list of fio job
302 name=str ASCII name of the job. This may be used to override the
303 name printed by fio for this job. Otherwise the job
304 name is used. On the command line this parameter has the
305 special purpose of also signaling the start of a new
308 description=str Text description of the job. Doesn't do anything except
309 dump this text description when this job is run. It's
312 directory=str Prefix filenames with this directory. Used to place files
313 in a different location than "./". See the 'filename' option
314 for escaping certain characters.
316 filename=str Fio normally makes up a filename based on the job name,
317 thread number, and file number. If you want to share
318 files between threads in a job or several jobs, specify
319 a filename for each of them to override the default. If
320 the ioengine used is 'net', the filename is the host, port,
321 and protocol to use in the format of =host,port,protocol.
322 See ioengine=net for more. If the ioengine is file based, you
323 can specify a number of files by separating the names with a
324 ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
325 as the two working files, you would use
326 filename=/dev/sda:/dev/sdb. On Windows, disk devices are
327 accessed as \\.\PhysicalDrive0 for the first device,
328 \\.\PhysicalDrive1 for the second etc. Note: Windows and
329 FreeBSD prevent write access to areas of the disk containing
330 in-use data (e.g. filesystems).
331 If the wanted filename does need to include a colon, then
332 escape that with a '\' character. For instance, if the filename
333 is "/dev/dsk/foo@3,0:c", then you would use
334 filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
335 stdin or stdout. Which of the two depends on the read/write
339 If sharing multiple files between jobs, it is usually necessary
340 to have fio generate the exact names that you want. By default,
341 fio will name a file based on the default file format
342 specification of jobname.jobnumber.filenumber. With this
343 option, that can be customized. Fio will recognize and replace
344 the following keywords in this string:
347 The name of the worker thread or process.
350 The incremental number of the worker thread or
354 The incremental number of the file for that worker
357 To have dependent jobs share a set of files, this option can
358 be set to have fio generate filenames that are shared between
359 the two. For instance, if testfiles.$filenum is specified,
360 file number 4 for any job will be named testfiles.4. The
361 default of $jobname.$jobnum.$filenum will be used if
362 no other format specifier is given.
364 opendir=str Tell fio to recursively add any file it can find in this
365 directory and down the file system tree.
367 lockfile=str Fio defaults to not locking any files before it does
368 IO to them. If a file or file descriptor is shared, fio
369 can serialize IO to that file to make the end result
370 consistent. This is usual for emulating real workloads that
371 share files. The lock modes are:
373 none No locking. The default.
374 exclusive Only one thread/process may do IO,
375 excluding all others.
376 readwrite Read-write locking on the file. Many
377 readers may access the file at the
378 same time, but writes get exclusive
382 rw=str Type of io pattern. Accepted values are:
384 read Sequential reads
385 write Sequential writes
386 randwrite Random writes
387 randread Random reads
388 rw,readwrite Sequential mixed reads and writes
389 randrw Random mixed reads and writes
391 For the mixed io types, the default is to split them 50/50.
392 For certain types of io the result may still be skewed a bit,
393 since the speed may be different. It is possible to specify
394 a number of IO's to do before getting a new offset, this is
395 done by appending a ':<nr>' to the end of the string given.
396 For a random read, it would look like 'rw=randread:8' for
397 passing in an offset modifier with a value of 8. If the
398 suffix is used with a sequential IO pattern, then the value
399 specified will be added to the generated offset for each IO.
400 For instance, using rw=write:4k will skip 4k for every
401 write. It turns sequential IO into sequential IO with holes.
402 See the 'rw_sequencer' option.
404 rw_sequencer=str If an offset modifier is given by appending a number to
405 the rw=<str> line, then this option controls how that
406 number modifies the IO offset being generated. Accepted
409 sequential Generate sequential offset
410 identical Generate the same offset
412 'sequential' is only useful for random IO, where fio would
413 normally generate a new random offset for every IO. If you
414 append eg 8 to randread, you would get a new random offset for
415 every 8 IO's. The result would be a seek for only every 8
416 IO's, instead of for every IO. Use rw=randread:8 to specify
417 that. As sequential IO is already sequential, setting
418 'sequential' for that would not result in any differences.
419 'identical' behaves in a similar fashion, except it sends
420 the same offset 8 number of times before generating a new
423 kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024.
424 Storage manufacturers like to use 10^3 or 1000 as a base
425 ten unit instead, for obvious reasons. Allow values are
426 1024 or 1000, with 1024 being the default.
428 unified_rw_reporting=bool Fio normally reports statistics on a per
429 data direction basis, meaning that read, write, and trim are
430 accounted and reported separately. If this option is set,
431 the fio will sum the results and report them as "mixed"
434 randrepeat=bool For random IO workloads, seed the generator in a predictable
435 way so that results are repeatable across repetitions.
437 randseed=int Seed the random number generators based on this seed value, to
438 be able to control what sequence of output is being generated.
439 If not set, the random sequence depends on the randrepeat
442 fallocate=str Whether pre-allocation is performed when laying down files.
445 none Do not pre-allocate space
446 posix Pre-allocate via posix_fallocate()
447 keep Pre-allocate via fallocate() with
448 FALLOC_FL_KEEP_SIZE set
449 0 Backward-compatible alias for 'none'
450 1 Backward-compatible alias for 'posix'
452 May not be available on all supported platforms. 'keep' is only
453 available on Linux.If using ZFS on Solaris this must be set to
454 'none' because ZFS doesn't support it. Default: 'posix'.
456 fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
457 on what IO patterns it is likely to issue. Sometimes you
458 want to test specific IO patterns without telling the
459 kernel about it, in which case you can disable this option.
460 If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
461 IO and POSIX_FADV_RANDOM for random IO.
463 fadvise_stream=int Notify the kernel what write stream ID to place these
464 writes under. Only supported on Linux. Note, this option
465 may change going forward.
467 size=int The total size of file io for this job. Fio will run until
468 this many bytes has been transferred, unless runtime is
469 limited by other options (such as 'runtime', for instance,
470 or increased/decreased by 'io_size'). Unless specific nrfiles
471 and filesize options are given, fio will divide this size
472 between the available files specified by the job. If not set,
473 fio will use the full size of the given files or devices.
474 If the files do not exist, size must be given. It is also
475 possible to give size as a percentage between 1 and 100. If
476 size=20% is given, fio will use 20% of the full size of the
477 given files or devices.
480 io_limit=int Normally fio operates within the region set by 'size', which
481 means that the 'size' option sets both the region and size of
482 IO to be performed. Sometimes that is not what you want. With
483 this option, it is possible to define just the amount of IO
484 that fio should do. For instance, if 'size' is set to 20G and
485 'io_size' is set to 5G, fio will perform IO within the first
486 20G but exit when 5G have been done. The opposite is also
487 possible - if 'size' is set to 20G, and 'io_size' is set to
488 40G, then fio will do 40G of IO within the 0..20G region.
490 filesize=int Individual file sizes. May be a range, in which case fio
491 will select sizes for files at random within the given range
492 and limited to 'size' in total (if that is given). If not
493 given, each created file is the same size.
495 file_append=bool Perform IO after the end of the file. Normally fio will
496 operate within the size of a file. If this option is set, then
497 fio will append to the file instead. This has identical
498 behavior to setting offset to the size of a file. This option
499 is ignored on non-regular files.
502 fill_fs=bool Sets size to something really large and waits for ENOSPC (no
503 space left on device) as the terminating condition. Only makes
504 sense with sequential write. For a read workload, the mount
505 point will be filled first then IO started on the result. This
506 option doesn't make sense if operating on a raw device node,
507 since the size of that is already known by the file system.
508 Additionally, writing beyond end-of-device will not return
512 bs=int The block size used for the io units. Defaults to 4k. Values
513 can be given for both read and writes. If a single int is
514 given, it will apply to both. If a second int is specified
515 after a comma, it will apply to writes only. In other words,
516 the format is either bs=read_and_write or bs=read,write,trim.
517 bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
518 writes, and 8k for trims. You can terminate the list with
519 a trailing comma. bs=4k,8k, would use the default value for
520 trims.. If you only wish to set the write size, you
521 can do so by passing an empty read size - bs=,8k will set
522 8k for writes and leave the read default value.
525 ba=int At what boundary to align random IO offsets. Defaults to
526 the same as 'blocksize' the minimum blocksize given.
527 Minimum alignment is typically 512b for using direct IO,
528 though it usually depends on the hardware block size. This
529 option is mutually exclusive with using a random map for
530 files, so it will turn off that option.
532 blocksize_range=irange
533 bsrange=irange Instead of giving a single block size, specify a range
534 and fio will mix the issued io block sizes. The issued
535 io unit will always be a multiple of the minimum value
536 given (also see bs_unaligned). Applies to both reads and
537 writes, however a second range can be given after a comma.
540 bssplit=str Sometimes you want even finer grained control of the
541 block sizes issued, not just an even split between them.
542 This option allows you to weight various block sizes,
543 so that you are able to define a specific amount of
544 block sizes issued. The format for this option is:
546 bssplit=blocksize/percentage:blocksize/percentage
548 for as many block sizes as needed. So if you want to define
549 a workload that has 50% 64k blocks, 10% 4k blocks, and
550 40% 32k blocks, you would write:
552 bssplit=4k/10:64k/50:32k/40
554 Ordering does not matter. If the percentage is left blank,
555 fio will fill in the remaining values evenly. So a bssplit
556 option like this one:
558 bssplit=4k/50:1k/:32k/
560 would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
561 always add up to 100, if bssplit is given a range that adds
562 up to more, it will error out.
564 bssplit also supports giving separate splits to reads and
565 writes. The format is identical to what bs= accepts. You
566 have to separate the read and write parts with a comma. So
567 if you want a workload that has 50% 2k reads and 50% 4k reads,
568 while having 90% 4k writes and 10% 8k writes, you would
571 bssplit=2k/50:4k/50,4k/90:8k/10
574 bs_unaligned If this option is given, any byte size value within bsrange
575 may be used as a block range. This typically wont work with
576 direct IO, as that normally requires sector alignment.
578 bs_is_seq_rand If this option is set, fio will use the normal read,write
579 blocksize settings as sequential,random instead. Any random
580 read or write will use the WRITE blocksize settings, and any
581 sequential read or write will use the READ blocksize setting.
583 zero_buffers If this option is given, fio will init the IO buffers to
584 all zeroes. The default is to fill them with random data.
586 refill_buffers If this option is given, fio will refill the IO buffers
587 on every submit. The default is to only fill it at init
588 time and reuse that data. Only makes sense if zero_buffers
589 isn't specified, naturally. If data verification is enabled,
590 refill_buffers is also automatically enabled.
592 scramble_buffers=bool If refill_buffers is too costly and the target is
593 using data deduplication, then setting this option will
594 slightly modify the IO buffer contents to defeat normal
595 de-dupe attempts. This is not enough to defeat more clever
596 block compression attempts, but it will stop naive dedupe of
597 blocks. Default: true.
599 buffer_compress_percentage=int If this is set, then fio will attempt to
600 provide IO buffer content (on WRITEs) that compress to
601 the specified level. Fio does this by providing a mix of
602 random data and a fixed pattern. The fixed pattern is either
603 zeroes, or the pattern specified by buffer_pattern. If the
604 pattern option is used, it might skew the compression ratio
605 slightly. Note that this is per block size unit, for file/disk
606 wide compression level that matches this setting, you'll also
607 want to set refill_buffers.
609 buffer_compress_chunk=int See buffer_compress_percentage. This
610 setting allows fio to manage how big the ranges of random
611 data and zeroed data is. Without this set, fio will
612 provide buffer_compress_percentage of blocksize random
613 data, followed by the remaining zeroed. With this set
614 to some chunk size smaller than the block size, fio can
615 alternate random and zeroed data throughout the IO
618 buffer_pattern=str If set, fio will fill the io buffers with this
619 pattern. If not set, the contents of io buffers is defined by
620 the other options related to buffer contents. The setting can
621 be any pattern of bytes, and can be prefixed with 0x for hex
622 values. It may also be a string, where the string must then
625 dedupe_percentage=int If set, fio will generate this percentage of
626 identical buffers when writing. These buffers will be
627 naturally dedupable. The contents of the buffers depend on
628 what other buffer compression settings have been set. It's
629 possible to have the individual buffers either fully
630 compressible, or not at all. This option only controls the
631 distribution of unique buffers.
633 nrfiles=int Number of files to use for this job. Defaults to 1.
635 openfiles=int Number of files to keep open at the same time. Defaults to
636 the same as nrfiles, can be set smaller to limit the number
639 file_service_type=str Defines how fio decides which file from a job to
640 service next. The following types are defined:
642 random Just choose a file at random.
644 roundrobin Round robin over open files. This
647 sequential Finish one file before moving on to
648 the next. Multiple files can still be
649 open depending on 'openfiles'.
651 The string can have a number appended, indicating how
652 often to switch to a new file. So if option random:4 is
653 given, fio will switch to a new random file after 4 ios
656 ioengine=str Defines how the job issues io to the file. The following
659 sync Basic read(2) or write(2) io. lseek(2) is
660 used to position the io location.
662 psync Basic pread(2) or pwrite(2) io.
664 vsync Basic readv(2) or writev(2) IO.
666 psyncv Basic preadv(2) or pwritev(2) IO.
668 libaio Linux native asynchronous io. Note that Linux
669 may only support queued behaviour with
670 non-buffered IO (set direct=1 or buffered=0).
671 This engine defines engine specific options.
673 posixaio glibc posix asynchronous io.
675 solarisaio Solaris native asynchronous io.
677 windowsaio Windows native asynchronous io.
679 mmap File is memory mapped and data copied
680 to/from using memcpy(3).
682 splice splice(2) is used to transfer the data and
683 vmsplice(2) to transfer data from user
686 syslet-rw Use the syslet system calls to make
687 regular read/write async.
689 sg SCSI generic sg v3 io. May either be
690 synchronous using the SG_IO ioctl, or if
691 the target is an sg character device
692 we use read(2) and write(2) for asynchronous
695 null Doesn't transfer any data, just pretends
696 to. This is mainly used to exercise fio
697 itself and for debugging/testing purposes.
699 net Transfer over the network to given host:port.
700 Depending on the protocol used, the hostname,
701 port, listen and filename options are used to
702 specify what sort of connection to make, while
703 the protocol option determines which protocol
705 This engine defines engine specific options.
707 netsplice Like net, but uses splice/vmsplice to
708 map data and send/receive.
709 This engine defines engine specific options.
711 cpuio Doesn't transfer any data, but burns CPU
712 cycles according to the cpuload= and
713 cpucycle= options. Setting cpuload=85
714 will cause that job to do nothing but burn
715 85% of the CPU. In case of SMP machines,
716 use numjobs=<no_of_cpu> to get desired CPU
717 usage, as the cpuload only loads a single
718 CPU at the desired rate.
720 guasi The GUASI IO engine is the Generic Userspace
721 Asyncronous Syscall Interface approach
724 http://www.xmailserver.org/guasi-lib.html
726 for more info on GUASI.
728 rdma The RDMA I/O engine supports both RDMA
729 memory semantics (RDMA_WRITE/RDMA_READ) and
730 channel semantics (Send/Recv) for the
731 InfiniBand, RoCE and iWARP protocols.
733 falloc IO engine that does regular fallocate to
734 simulate data transfer as fio ioengine.
735 DDIR_READ does fallocate(,mode = keep_size,)
736 DDIR_WRITE does fallocate(,mode = 0)
737 DDIR_TRIM does fallocate(,mode = punch_hole)
739 e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
740 ioctls to simulate defragment activity in
741 request to DDIR_WRITE event
743 rbd IO engine supporting direct access to Ceph
744 Rados Block Devices (RBD) via librbd without
745 the need to use the kernel rbd driver. This
746 ioengine defines engine specific options.
748 gfapi Using Glusterfs libgfapi sync interface to
749 direct access to Glusterfs volumes without
752 gfapi_async Using Glusterfs libgfapi async interface
753 to direct access to Glusterfs volumes without
754 having to go through FUSE. This ioengine
755 defines engine specific options.
757 libhdfs Read and write through Hadoop (HDFS).
758 The 'filename' option is used to specify host,
759 port of the hdfs name-node to connect. This
760 engine interprets offsets a little
761 differently. In HDFS, files once created
762 cannot be modified. So random writes are not
763 possible. To imitate this, libhdfs engine
764 expects bunch of small files to be created
765 over HDFS, and engine will randomly pick a
766 file out of those files based on the offset
767 generated by fio backend. (see the example
768 job file to create such files, use rw=write
769 option). Please note, you might want to set
770 necessary environment variables to work with
771 hdfs/libhdfs properly.
773 mtd Read, write and erase an MTD character device
774 (e.g., /dev/mtd0). Discards are treated as
775 erases. Depending on the underlying device
776 type, the I/O may have to go in a certain
777 pattern, e.g., on NAND, writing sequentially
778 to erase blocks and discarding before
779 overwriting. The writetrim mode works well
782 external Prefix to specify loading an external
783 IO engine object file. Append the engine
784 filename, eg ioengine=external:/tmp/foo.o
785 to load ioengine foo.o in /tmp.
787 iodepth=int This defines how many io units to keep in flight against
788 the file. The default is 1 for each file defined in this
789 job, can be overridden with a larger value for higher
790 concurrency. Note that increasing iodepth beyond 1 will not
791 affect synchronous ioengines (except for small degress when
792 verify_async is in use). Even async engines may impose OS
793 restrictions causing the desired depth not to be achieved.
794 This may happen on Linux when using libaio and not setting
795 direct=1, since buffered IO is not async on that OS. Keep an
796 eye on the IO depth distribution in the fio output to verify
797 that the achieved depth is as expected. Default: 1.
799 iodepth_batch_submit=int
800 iodepth_batch=int This defines how many pieces of IO to submit at once.
801 It defaults to 1 which means that we submit each IO
802 as soon as it is available, but can be raised to submit
803 bigger batches of IO at the time.
805 iodepth_batch_complete=int This defines how many pieces of IO to retrieve
806 at once. It defaults to 1 which means that we'll ask
807 for a minimum of 1 IO in the retrieval process from
808 the kernel. The IO retrieval will go on until we
809 hit the limit set by iodepth_low. If this variable is
810 set to 0, then fio will always check for completed
811 events before queuing more IO. This helps reduce
812 IO latency, at the cost of more retrieval system calls.
814 iodepth_low=int The low water mark indicating when to start filling
815 the queue again. Defaults to the same as iodepth, meaning
816 that fio will attempt to keep the queue full at all times.
817 If iodepth is set to eg 16 and iodepth_low is set to 4, then
818 after fio has filled the queue of 16 requests, it will let
819 the depth drain down to 4 before starting to fill it again.
821 direct=bool If value is true, use non-buffered io. This is usually
822 O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
823 On Windows the synchronous ioengines don't support direct io.
825 atomic=bool If value is true, attempt to use atomic direct IO. Atomic
826 writes are guaranteed to be stable once acknowledged by
827 the operating system. Only Linux supports O_ATOMIC right
830 buffered=bool If value is true, use buffered io. This is the opposite
831 of the 'direct' option. Defaults to true.
833 offset=int Start io at the given offset in the file. The data before
834 the given offset will not be touched. This effectively
835 caps the file size at real_size - offset.
837 offset_increment=int If this is provided, then the real offset becomes
838 offset + offset_increment * thread_number, where the thread
839 number is a counter that starts at 0 and is incremented for
840 each sub-job (i.e. when numjobs option is specified). This
841 option is useful if there are several jobs which are intended
842 to operate on a file in parallel disjoint segments, with
843 even spacing between the starting points.
845 number_ios=int Fio will normally perform IOs until it has exhausted the size
846 of the region set by size=, or if it exhaust the allocated
847 time (or hits an error condition). With this setting, the
848 range/size can be set independently of the number of IOs to
849 perform. When fio reaches this number, it will exit normally
850 and report status. Note that this does not extend the amount
851 of IO that will be done, it will only stop fio if this
852 condition is met before other end-of-job criteria.
854 fsync=int If writing to a file, issue a sync of the dirty data
855 for every number of blocks given. For example, if you give
856 32 as a parameter, fio will sync the file for every 32
857 writes issued. If fio is using non-buffered io, we may
858 not sync the file. The exception is the sg io engine, which
859 synchronizes the disk cache anyway.
861 fdatasync=int Like fsync= but uses fdatasync() to only sync data and not
863 In FreeBSD and Windows there is no fdatasync(), this falls back to
866 sync_file_range=str:val Use sync_file_range() for every 'val' number of
867 write operations. Fio will track range of writes that
868 have happened since the last sync_file_range() call. 'str'
869 can currently be one or more of:
871 wait_before SYNC_FILE_RANGE_WAIT_BEFORE
872 write SYNC_FILE_RANGE_WRITE
873 wait_after SYNC_FILE_RANGE_WAIT_AFTER
875 So if you do sync_file_range=wait_before,write:8, fio would
876 use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
877 every 8 writes. Also see the sync_file_range(2) man page.
878 This option is Linux specific.
880 overwrite=bool If true, writes to a file will always overwrite existing
881 data. If the file doesn't already exist, it will be
882 created before the write phase begins. If the file exists
883 and is large enough for the specified write phase, nothing
886 end_fsync=bool If true, fsync file contents when a write stage has completed.
888 fsync_on_close=bool If true, fio will fsync() a dirty file on close.
889 This differs from end_fsync in that it will happen on every
890 file close, not just at the end of the job.
892 rwmixread=int How large a percentage of the mix should be reads.
894 rwmixwrite=int How large a percentage of the mix should be writes. If both
895 rwmixread and rwmixwrite is given and the values do not add
896 up to 100%, the latter of the two will be used to override
897 the first. This may interfere with a given rate setting,
898 if fio is asked to limit reads or writes to a certain rate.
899 If that is the case, then the distribution may be skewed.
901 random_distribution=str:float By default, fio will use a completely uniform
902 random distribution when asked to perform random IO. Sometimes
903 it is useful to skew the distribution in specific ways,
904 ensuring that some parts of the data is more hot than others.
905 fio includes the following distribution models:
907 random Uniform random distribution
908 zipf Zipf distribution
909 pareto Pareto distribution
911 When using a zipf or pareto distribution, an input value
912 is also needed to define the access pattern. For zipf, this
913 is the zipf theta. For pareto, it's the pareto power. Fio
914 includes a test program, genzipf, that can be used visualize
915 what the given input values will yield in terms of hit rates.
916 If you wanted to use zipf with a theta of 1.2, you would use
917 random_distribution=zipf:1.2 as the option. If a non-uniform
918 model is used, fio will disable use of the random map.
920 percentage_random=int For a random workload, set how big a percentage should
921 be random. This defaults to 100%, in which case the workload
922 is fully random. It can be set from anywhere from 0 to 100.
923 Setting it to 0 would make the workload fully sequential. Any
924 setting in between will result in a random mix of sequential
925 and random IO, at the given percentages. It is possible to
926 set different values for reads, writes, and trim. To do so,
927 simply use a comma separated list. See blocksize.
929 norandommap Normally fio will cover every block of the file when doing
930 random IO. If this option is given, fio will just get a
931 new random offset without looking at past io history. This
932 means that some blocks may not be read or written, and that
933 some blocks may be read/written more than once. If this option
934 is used with verify= and multiple blocksizes (via bsrange=),
935 only intact blocks are verified, i.e., partially-overwritten
938 softrandommap=bool See norandommap. If fio runs with the random block map
939 enabled and it fails to allocate the map, if this option is
940 set it will continue without a random block map. As coverage
941 will not be as complete as with random maps, this option is
944 random_generator=str Fio supports the following engines for generating
945 IO offsets for random IO:
947 tausworthe Strong 2^88 cycle random number generator
948 lfsr Linear feedback shift register generator
950 Tausworthe is a strong random number generator, but it
951 requires tracking on the side if we want to ensure that
952 blocks are only read or written once. LFSR guarantees
953 that we never generate the same offset twice, and it's
954 also less computationally expensive. It's not a true
955 random generator, however, though for IO purposes it's
956 typically good enough. LFSR only works with single
957 block sizes, not with workloads that use multiple block
958 sizes. If used with such a workload, fio may read or write
959 some blocks multiple times.
961 nice=int Run the job with the given nice value. See man nice(2).
963 prio=int Set the io priority value of this job. Linux limits us to
964 a positive value between 0 and 7, with 0 being the highest.
967 prioclass=int Set the io priority class. See man ionice(1).
969 thinktime=int Stall the job x microseconds after an io has completed before
970 issuing the next. May be used to simulate processing being
971 done by an application. See thinktime_blocks and
975 Only valid if thinktime is set - pretend to spend CPU time
976 doing something with the data received, before falling back
977 to sleeping for the rest of the period specified by
981 Only valid if thinktime is set - control how many blocks
982 to issue, before waiting 'thinktime' usecs. If not set,
983 defaults to 1 which will make fio wait 'thinktime' usecs
984 after every block. This effectively makes any queue depth
985 setting redundant, since no more than 1 IO will be queued
986 before we have to complete it and do our thinktime. In
987 other words, this setting effectively caps the queue depth
988 if the latter is larger.
990 rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
991 the normal suffix rules apply. You can use rate=500k to limit
992 reads and writes to 500k each, or you can specify read and
993 writes separately. Using rate=1m,500k would limit reads to
994 1MB/sec and writes to 500KB/sec. Capping only reads or
995 writes can be done with rate=,500k or rate=500k,. The former
996 will only limit writes (to 500KB/sec), the latter will only
999 ratemin=int Tell fio to do whatever it can to maintain at least this
1000 bandwidth. Failing to meet this requirement, will cause
1001 the job to exit. The same format as rate is used for
1002 read vs write separation.
1004 rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same
1005 as rate, just specified independently of bandwidth. If the
1006 job is given a block size range instead of a fixed value,
1007 the smallest block size is used as the metric. The same format
1008 as rate is used for read vs write separation.
1010 rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
1011 the job to exit. The same format as rate is used for read vs
1014 latency_target=int If set, fio will attempt to find the max performance
1015 point that the given workload will run at while maintaining a
1016 latency below this target. The values is given in microseconds.
1017 See latency_window and latency_percentile
1019 latency_window=int Used with latency_target to specify the sample window
1020 that the job is run at varying queue depths to test the
1021 performance. The value is given in microseconds.
1023 latency_percentile=float The percentage of IOs that must fall within the
1024 criteria specified by latency_target and latency_window. If not
1025 set, this defaults to 100.0, meaning that all IOs must be equal
1026 or below to the value set by latency_target.
1028 max_latency=int If set, fio will exit the job if it exceeds this maximum
1029 latency. It will exit with an ETIME error.
1031 ratecycle=int Average bandwidth for 'rate' and 'ratemin' over this number
1034 cpumask=int Set the CPU affinity of this job. The parameter given is a
1035 bitmask of allowed CPU's the job may run on. So if you want
1036 the allowed CPUs to be 1 and 5, you would pass the decimal
1037 value of (1 << 1 | 1 << 5), or 34. See man
1038 sched_setaffinity(2). This may not work on all supported
1039 operating systems or kernel versions. This option doesn't
1040 work well for a higher CPU count than what you can store in
1041 an integer mask, so it can only control cpus 1-32. For
1042 boxes with larger CPU counts, use cpus_allowed.
1044 cpus_allowed=str Controls the same options as cpumask, but it allows a text
1045 setting of the permitted CPUs instead. So to use CPUs 1 and
1046 5, you would specify cpus_allowed=1,5. This options also
1047 allows a range of CPUs. Say you wanted a binding to CPUs
1048 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.
1050 cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
1051 specified by cpus_allowed or cpumask. Two policies are
1054 shared All jobs will share the CPU set specified.
1055 split Each job will get a unique CPU from the CPU set.
1057 'shared' is the default behaviour, if the option isn't
1058 specified. If split is specified, then fio will will assign
1059 one cpu per job. If not enough CPUs are given for the jobs
1060 listed, then fio will roundrobin the CPUs in the set.
1062 numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
1063 arguments allow comma delimited list of cpu numbers,
1064 A-B ranges, or 'all'. Note, to enable numa options support,
1065 fio must be built on a system with libnuma-dev(el) installed.
1067 numa_mem_policy=str Set this job's memory policy and corresponding NUMA
1068 nodes. Format of the argements:
1070 `mode' is one of the following memory policy:
1071 default, prefer, bind, interleave, local
1072 For `default' and `local' memory policy, no node is
1073 needed to be specified.
1074 For `prefer', only one node is allowed.
1075 For `bind' and `interleave', it allow comma delimited
1076 list of numbers, A-B ranges, or 'all'.
1078 startdelay=time Start this job the specified number of seconds after fio
1079 has started. Only useful if the job file contains several
1080 jobs, and you want to delay starting some jobs to a certain
1083 runtime=time Tell fio to terminate processing after the specified number
1084 of seconds. It can be quite hard to determine for how long
1085 a specified job will run, so this parameter is handy to
1086 cap the total runtime to a given time.
1088 time_based If set, fio will run for the duration of the runtime
1089 specified even if the file(s) are completely read or
1090 written. It will simply loop over the same workload
1091 as many times as the runtime allows.
1093 ramp_time=time If set, fio will run the specified workload for this amount
1094 of time before logging any performance numbers. Useful for
1095 letting performance settle before logging results, thus
1096 minimizing the runtime required for stable results. Note
1097 that the ramp_time is considered lead in time for a job,
1098 thus it will increase the total runtime if a special timeout
1099 or runtime is specified.
1101 invalidate=bool Invalidate the buffer/page cache parts for this file prior
1102 to starting io. Defaults to true.
1104 sync=bool Use sync io for buffered writes. For the majority of the
1105 io engines, this means using O_SYNC.
1108 mem=str Fio can use various types of memory as the io unit buffer.
1109 The allowed values are:
1111 malloc Use memory from malloc(3) as the buffers.
1113 shm Use shared memory as the buffers. Allocated
1116 shmhuge Same as shm, but use huge pages as backing.
1118 mmap Use mmap to allocate buffers. May either be
1119 anonymous memory, or can be file backed if
1120 a filename is given after the option. The
1121 format is mem=mmap:/path/to/file.
1123 mmaphuge Use a memory mapped huge file as the buffer
1124 backing. Append filename after mmaphuge, ala
1125 mem=mmaphuge:/hugetlbfs/file
1127 The area allocated is a function of the maximum allowed
1128 bs size for the job, multiplied by the io depth given. Note
1129 that for shmhuge and mmaphuge to work, the system must have
1130 free huge pages allocated. This can normally be checked
1131 and set by reading/writing /proc/sys/vm/nr_hugepages on a
1132 Linux system. Fio assumes a huge page is 4MB in size. So
1133 to calculate the number of huge pages you need for a given
1134 job file, add up the io depth of all jobs (normally one unless
1135 iodepth= is used) and multiply by the maximum bs set. Then
1136 divide that number by the huge page size. You can see the
1137 size of the huge pages in /proc/meminfo. If no huge pages
1138 are allocated by having a non-zero number in nr_hugepages,
1139 using mmaphuge or shmhuge will fail. Also see hugepage-size.
1141 mmaphuge also needs to have hugetlbfs mounted and the file
1142 location should point there. So if it's mounted in /huge,
1143 you would use mem=mmaphuge:/huge/somefile.
1145 iomem_align=int This indiciates the memory alignment of the IO memory buffers.
1146 Note that the given alignment is applied to the first IO unit
1147 buffer, if using iodepth the alignment of the following buffers
1148 are given by the bs used. In other words, if using a bs that is
1149 a multiple of the page sized in the system, all buffers will
1150 be aligned to this value. If using a bs that is not page
1151 aligned, the alignment of subsequent IO memory buffers is the
1152 sum of the iomem_align and bs used.
1155 Defines the size of a huge page. Must at least be equal
1156 to the system setting, see /proc/meminfo. Defaults to 4MB.
1157 Should probably always be a multiple of megabytes, so using
1158 hugepage-size=Xm is the preferred way to set this to avoid
1159 setting a non-pow-2 bad value.
1161 exitall When one job finishes, terminate the rest. The default is
1162 to wait for each job to finish, sometimes that is not the
1165 bwavgtime=int Average the calculated bandwidth over the given time. Value
1166 is specified in milliseconds.
1168 iopsavgtime=int Average the calculated IOPS over the given time. Value
1169 is specified in milliseconds.
1171 create_serialize=bool If true, serialize the file creating for the jobs.
1172 This may be handy to avoid interleaving of data
1173 files, which may greatly depend on the filesystem
1174 used and even the number of processors in the system.
1176 create_fsync=bool fsync the data file after creation. This is the
1179 create_on_open=bool Don't pre-setup the files for IO, just create open()
1180 when it's time to do IO to that file.
1182 create_only=bool If true, fio will only run the setup phase of the job.
1183 If files need to be laid out or updated on disk, only
1184 that will be done. The actual job contents are not
1187 pre_read=bool If this is given, files will be pre-read into memory before
1188 starting the given IO operation. This will also clear
1189 the 'invalidate' flag, since it is pointless to pre-read
1190 and then drop the cache. This will only work for IO engines
1191 that are seekable, since they allow you to read the same data
1192 multiple times. Thus it will not work on eg network or splice
1195 unlink=bool Unlink the job files when done. Not the default, as repeated
1196 runs of that job would then waste time recreating the file
1197 set again and again.
1199 loops=int Run the specified number of iterations of this job. Used
1200 to repeat the same workload a given number of times. Defaults
1203 verify_only Do not perform specified workload---only verify data still
1204 matches previous invocation of this workload. This option
1205 allows one to check data multiple times at a later date
1206 without overwriting it. This option makes sense only for
1207 workloads that write data, and does not support workloads
1208 with the time_based option set.
1210 do_verify=bool Run the verify phase after a write phase. Only makes sense if
1211 verify is set. Defaults to 1.
1213 verify=str If writing to a file, fio can verify the file contents
1214 after each iteration of the job. The allowed values are:
1216 md5 Use an md5 sum of the data area and store
1217 it in the header of each block.
1219 crc64 Use an experimental crc64 sum of the data
1220 area and store it in the header of each
1223 crc32c Use a crc32c sum of the data area and store
1224 it in the header of each block.
1226 crc32c-intel Use hardware assisted crc32c calcuation
1227 provided on SSE4.2 enabled processors. Falls
1228 back to regular software crc32c, if not
1229 supported by the system.
1231 crc32 Use a crc32 sum of the data area and store
1232 it in the header of each block.
1234 crc16 Use a crc16 sum of the data area and store
1235 it in the header of each block.
1237 crc7 Use a crc7 sum of the data area and store
1238 it in the header of each block.
1240 xxhash Use xxhash as the checksum function. Generally
1241 the fastest software checksum that fio
1244 sha512 Use sha512 as the checksum function.
1246 sha256 Use sha256 as the checksum function.
1248 sha1 Use optimized sha1 as the checksum function.
1250 meta Write extra information about each io
1251 (timestamp, block number etc.). The block
1252 number is verified. The io sequence number is
1253 verified for workloads that write data.
1254 See also verify_pattern.
1256 null Only pretend to verify. Useful for testing
1257 internals with ioengine=null, not for much
1260 This option can be used for repeated burn-in tests of a
1261 system to make sure that the written data is also
1262 correctly read back. If the data direction given is
1263 a read or random read, fio will assume that it should
1264 verify a previously written file. If the data direction
1265 includes any form of write, the verify will be of the
1268 verifysort=bool If set, fio will sort written verify blocks when it deems
1269 it faster to read them back in a sorted manner. This is
1270 often the case when overwriting an existing file, since
1271 the blocks are already laid out in the file system. You
1272 can ignore this option unless doing huge amounts of really
1273 fast IO where the red-black tree sorting CPU time becomes
1276 verify_offset=int Swap the verification header with data somewhere else
1277 in the block before writing. Its swapped back before
1280 verify_interval=int Write the verification header at a finer granularity
1281 than the blocksize. It will be written for chunks the
1282 size of header_interval. blocksize should divide this
1285 verify_pattern=str If set, fio will fill the io buffers with this
1286 pattern. Fio defaults to filling with totally random
1287 bytes, but sometimes it's interesting to fill with a known
1288 pattern for io verification purposes. Depending on the
1289 width of the pattern, fio will fill 1/2/3/4 bytes of the
1290 buffer at the time(it can be either a decimal or a hex number).
1291 The verify_pattern if larger than a 32-bit quantity has to
1292 be a hex number that starts with either "0x" or "0X". Use
1295 verify_fatal=bool Normally fio will keep checking the entire contents
1296 before quitting on a block verification failure. If this
1297 option is set, fio will exit the job on the first observed
1300 verify_dump=bool If set, dump the contents of both the original data
1301 block and the data block we read off disk to files. This
1302 allows later analysis to inspect just what kind of data
1303 corruption occurred. Off by default.
1305 verify_async=int Fio will normally verify IO inline from the submitting
1306 thread. This option takes an integer describing how many
1307 async offload threads to create for IO verification instead,
1308 causing fio to offload the duty of verifying IO contents
1309 to one or more separate threads. If using this offload
1310 option, even sync IO engines can benefit from using an
1311 iodepth setting higher than 1, as it allows them to have
1312 IO in flight while verifies are running.
1314 verify_async_cpus=str Tell fio to set the given CPU affinity on the
1315 async IO verification threads. See cpus_allowed for the
1318 verify_backlog=int Fio will normally verify the written contents of a
1319 job that utilizes verify once that job has completed. In
1320 other words, everything is written then everything is read
1321 back and verified. You may want to verify continually
1322 instead for a variety of reasons. Fio stores the meta data
1323 associated with an IO block in memory, so for large
1324 verify workloads, quite a bit of memory would be used up
1325 holding this meta data. If this option is enabled, fio
1326 will write only N blocks before verifying these blocks.
1328 verify_backlog_batch=int Control how many blocks fio will verify
1329 if verify_backlog is set. If not set, will default to
1330 the value of verify_backlog (meaning the entire queue
1331 is read back and verified). If verify_backlog_batch is
1332 less than verify_backlog then not all blocks will be verified,
1333 if verify_backlog_batch is larger than verify_backlog, some
1334 blocks will be verified more than once.
1336 verify_state_save=bool When a job exits during the write phase of a verify
1337 workload, save its current state. This allows fio to replay
1338 up until that point, if the verify state is loaded for the
1339 verify read phase. The format of the filename is, roughly,
1340 <type>-<jobname>-<jobindex>-verify.state. <type> is "local"
1341 for a local run, "sock" for a client/server socket connection,
1342 and "ip" (192.168.0.1, for instance) for a networked
1343 client/server connection.
1345 verify_state_load=bool If a verify termination trigger was used, fio stores
1346 the current write state of each thread. This can be used at
1347 verification time so that fio knows how far it should verify.
1348 Without this information, fio will run a full verification
1349 pass, according to the settings in the job file used.
1352 wait_for_previous Wait for preceding jobs in the job file to exit, before
1353 starting this one. Can be used to insert serialization
1354 points in the job file. A stone wall also implies starting
1355 a new reporting group.
1357 new_group Start a new reporting group. See: group_reporting.
1359 numjobs=int Create the specified number of clones of this job. May be
1360 used to setup a larger number of threads/processes doing
1361 the same thing. Each thread is reported separately; to see
1362 statistics for all clones as a whole, use group_reporting in
1363 conjunction with new_group.
1365 group_reporting It may sometimes be interesting to display statistics for
1366 groups of jobs as a whole instead of for each individual job.
1367 This is especially true if 'numjobs' is used; looking at
1368 individual thread/process output quickly becomes unwieldy.
1369 To see the final report per-group instead of per-job, use
1370 'group_reporting'. Jobs in a file will be part of the same
1371 reporting group, unless if separated by a stonewall, or by
1374 thread fio defaults to forking jobs, however if this option is
1375 given, fio will use pthread_create(3) to create threads
1378 zonesize=int Divide a file into zones of the specified size. See zoneskip.
1380 zoneskip=int Skip the specified number of bytes when zonesize data has
1381 been read. The two zone options can be used to only do
1382 io on zones of a file.
1384 write_iolog=str Write the issued io patterns to the specified file. See
1385 read_iolog. Specify a separate file for each job, otherwise
1386 the iologs will be interspersed and the file may be corrupt.
1388 read_iolog=str Open an iolog with the specified file name and replay the
1389 io patterns it contains. This can be used to store a
1390 workload and replay it sometime later. The iolog given
1391 may also be a blktrace binary file, which allows fio
1392 to replay a workload captured by blktrace. See blktrace
1393 for how to capture such logging data. For blktrace replay,
1394 the file needs to be turned into a blkparse binary data
1395 file first (blkparse <device> -o /dev/null -d file_for_fio.bin).
1397 replay_no_stall=int When replaying I/O with read_iolog the default behavior
1398 is to attempt to respect the time stamps within the log and
1399 replay them with the appropriate delay between IOPS. By
1400 setting this variable fio will not respect the timestamps and
1401 attempt to replay them as fast as possible while still
1402 respecting ordering. The result is the same I/O pattern to a
1403 given device, but different timings.
1405 replay_redirect=str While replaying I/O patterns using read_iolog the
1406 default behavior is to replay the IOPS onto the major/minor
1407 device that each IOP was recorded from. This is sometimes
1408 undesirable because on a different machine those major/minor
1409 numbers can map to a different device. Changing hardware on
1410 the same system can also result in a different major/minor
1411 mapping. Replay_redirect causes all IOPS to be replayed onto
1412 the single specified device regardless of the device it was
1413 recorded from. i.e. replay_redirect=/dev/sdc would cause all
1414 IO in the blktrace to be replayed onto /dev/sdc. This means
1415 multiple devices will be replayed onto a single, if the trace
1416 contains multiple devices. If you want multiple devices to be
1417 replayed concurrently to multiple redirected devices you must
1418 blkparse your trace into separate traces and replay them with
1419 independent fio invocations. Unfortuantely this also breaks
1420 the strict time ordering between multiple device accesses.
1422 write_bw_log=str If given, write a bandwidth log of the jobs in this job
1423 file. Can be used to store data of the bandwidth of the
1424 jobs in their lifetime. The included fio_generate_plots
1425 script uses gnuplot to turn these text files into nice
1426 graphs. See write_lat_log for behaviour of given
1427 filename. For this option, the suffix is _bw.x.log, where
1428 x is the index of the job (1..N, where N is the number of
1431 write_lat_log=str Same as write_bw_log, except that this option stores io
1432 submission, completion, and total latencies instead. If no
1433 filename is given with this option, the default filename of
1434 "jobname_type.log" is used. Even if the filename is given,
1435 fio will still append the type of log. So if one specifies
1439 The actual log names will be foo_slat.x.log, foo_clat.x.log,
1440 and foo_lat.x.log, where x is the index of the job (1..N,
1441 where N is the number of jobs). This helps fio_generate_plot
1442 fine the logs automatically.
1444 write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
1445 given with this option, the default filename of
1446 "jobname_type.x.log" is used,where x is the index of the job
1447 (1..N, where N is the number of jobs). Even if the filename
1448 is given, fio will still append the type of log.
1450 log_avg_msec=int By default, fio will log an entry in the iops, latency,
1451 or bw log for every IO that completes. When writing to the
1452 disk log, that can quickly grow to a very large size. Setting
1453 this option makes fio average the each log entry over the
1454 specified period of time, reducing the resolution of the log.
1457 log_offset=int If this is set, the iolog options will include the byte
1458 offset for the IO entry as well as the other data values.
1460 log_compression=int If this is set, fio will compress the IO logs as
1461 it goes, to keep the memory footprint lower. When a log
1462 reaches the specified size, that chunk is removed and
1463 compressed in the background. Given that IO logs are
1464 fairly highly compressible, this yields a nice memory
1465 savings for longer runs. The downside is that the
1466 compression will consume some background CPU cycles, so
1467 it may impact the run. This, however, is also true if
1468 the logging ends up consuming most of the system memory.
1469 So pick your poison. The IO logs are saved normally at the
1470 end of a run, by decompressing the chunks and storing them
1471 in the specified log file. This feature depends on the
1472 availability of zlib.
1474 log_store_compressed=bool If set, and log_compression is also set,
1475 fio will store the log files in a compressed format. They
1476 can be decompressed with fio, using the --inflate-log
1477 command line parameter. The files will be stored with a
1480 block_error_percentiles=bool If set, record errors in trim block-sized
1481 units from writes and trims and output a histogram of
1482 how many trims it took to get to errors, and what kind
1483 of error was encountered.
1485 lockmem=int Pin down the specified amount of memory with mlock(2). Can
1486 potentially be used instead of removing memory or booting
1487 with less memory to simulate a smaller amount of memory.
1488 The amount specified is per worker.
1490 exec_prerun=str Before running this job, issue the command specified
1491 through system(3). Output is redirected in a file called
1494 exec_postrun=str After the job completes, issue the command specified
1495 though system(3). Output is redirected in a file called
1496 jobname.postrun.txt.
1498 ioscheduler=str Attempt to switch the device hosting the file to the specified
1499 io scheduler before running.
1501 disk_util=bool Generate disk utilization statistics, if the platform
1502 supports it. Defaults to on.
1504 disable_lat=bool Disable measurements of total latency numbers. Useful
1505 only for cutting back the number of calls to gettimeofday,
1506 as that does impact performance at really high IOPS rates.
1507 Note that to really get rid of a large amount of these
1508 calls, this option must be used with disable_slat and
1511 disable_clat=bool Disable measurements of completion latency numbers. See
1514 disable_slat=bool Disable measurements of submission latency numbers. See
1517 disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
1520 clat_percentiles=bool Enable the reporting of percentiles of
1521 completion latencies.
1523 percentile_list=float_list Overwrite the default list of percentiles
1524 for completion latencies and the block error histogram.
1525 Each number is a floating number in the range (0,100],
1526 and the maximum length of the list is 20. Use ':'
1527 to separate the numbers, and list the numbers in ascending
1528 order. For example, --percentile_list=99.5:99.9 will cause
1529 fio to report the values of completion latency below which
1530 99.5% and 99.9% of the observed latencies fell, respectively.
1532 clocksource=str Use the given clocksource as the base of timing. The
1533 supported options are:
1535 gettimeofday gettimeofday(2)
1537 clock_gettime clock_gettime(2)
1539 cpu Internal CPU clock source
1541 cpu is the preferred clocksource if it is reliable, as it
1542 is very fast (and fio is heavy on time calls). Fio will
1543 automatically use this clocksource if it's supported and
1544 considered reliable on the system it is running on, unless
1545 another clocksource is specifically set. For x86/x86-64 CPUs,
1546 this means supporting TSC Invariant.
1548 gtod_reduce=bool Enable all of the gettimeofday() reducing options
1549 (disable_clat, disable_slat, disable_bw) plus reduce
1550 precision of the timeout somewhat to really shrink
1551 the gettimeofday() call count. With this option enabled,
1552 we only do about 0.4% of the gtod() calls we would have
1553 done if all time keeping was enabled.
1555 gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of
1556 execution to just getting the current time. Fio (and
1557 databases, for instance) are very intensive on gettimeofday()
1558 calls. With this option, you can set one CPU aside for
1559 doing nothing but logging current time to a shared memory
1560 location. Then the other threads/processes that run IO
1561 workloads need only copy that segment, instead of entering
1562 the kernel with a gettimeofday() call. The CPU set aside
1563 for doing these time calls will be excluded from other
1564 uses. Fio will manually clear it from the CPU mask of other
1567 continue_on_error=str Normally fio will exit the job on the first observed
1568 failure. If this option is set, fio will continue the job when
1569 there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
1570 is exceeded or the I/O size specified is completed. If this
1571 option is used, there are two more stats that are appended,
1572 the total error count and the first error. The error field
1573 given in the stats is the first error that was hit during the
1576 The allowed values are:
1578 none Exit on any IO or verify errors.
1580 read Continue on read errors, exit on all others.
1582 write Continue on write errors, exit on all others.
1584 io Continue on any IO error, exit on all others.
1586 verify Continue on verify errors, exit on all others.
1588 all Continue on all errors.
1590 0 Backward-compatible alias for 'none'.
1592 1 Backward-compatible alias for 'all'.
1594 ignore_error=str Sometimes you want to ignore some errors during test
1595 in that case you can specify error list for each error type.
1596 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
1597 errors for given error type is separated with ':'. Error
1598 may be symbol ('ENOSPC', 'ENOMEM') or integer.
1600 ignore_error=EAGAIN,ENOSPC:122
1601 This option will ignore EAGAIN from READ, and ENOSPC and
1602 122(EDQUOT) from WRITE.
1604 error_dump=bool If set dump every error even if it is non fatal, true
1605 by default. If disabled only fatal error will be dumped
1607 cgroup=str Add job to this control group. If it doesn't exist, it will
1608 be created. The system must have a mounted cgroup blkio
1609 mount point for this to work. If your system doesn't have it
1610 mounted, you can do so with:
1612 # mount -t cgroup -o blkio none /cgroup
1614 cgroup_weight=int Set the weight of the cgroup to this value. See
1615 the documentation that comes with the kernel, allowed values
1616 are in the range of 100..1000.
1618 cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
1619 the job completion. To override this behavior and to leave
1620 cgroups around after the job completion, set cgroup_nodelete=1.
1621 This can be useful if one wants to inspect various cgroup
1622 files after job completion. Default: false
1624 uid=int Instead of running as the invoking user, set the user ID to
1625 this value before the thread/process does any work.
1627 gid=int Set group ID, see uid.
1629 flow_id=int The ID of the flow. If not specified, it defaults to being a
1630 global flow. See flow.
1632 flow=int Weight in token-based flow control. If this value is used, then
1633 there is a 'flow counter' which is used to regulate the
1634 proportion of activity between two or more jobs. fio attempts
1635 to keep this flow counter near zero. The 'flow' parameter
1636 stands for how much should be added or subtracted to the flow
1637 counter on each iteration of the main I/O loop. That is, if
1638 one job has flow=8 and another job has flow=-1, then there
1639 will be a roughly 1:8 ratio in how much one runs vs the other.
1641 flow_watermark=int The maximum value that the absolute value of the flow
1642 counter is allowed to reach before the job must wait for a
1643 lower value of the counter.
1645 flow_sleep=int The period of time, in microseconds, to wait after the flow
1646 watermark has been exceeded before retrying operations
1648 In addition, there are some parameters which are only valid when a specific
1649 ioengine is in use. These are used identically to normal parameters, with the
1650 caveat that when used on the command line, they must come after the ioengine
1651 that defines them is selected.
1653 [libaio] userspace_reap Normally, with the libaio engine in use, fio will use
1654 the io_getevents system call to reap newly returned events.
1655 With this flag turned on, the AIO ring will be read directly
1656 from user-space to reap events. The reaping mode is only
1657 enabled when polling for a minimum of 0 events (eg when
1658 iodepth_batch_complete=0).
1660 [cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.
1662 [cpu] cpuchunks=int Split the load into cycles of the given time. In
1665 [cpu] exit_on_io_done=bool Detect when IO threads are done, then exit.
1667 [netsplice] hostname=str
1668 [net] hostname=str The host name or IP address to use for TCP or UDP based IO.
1669 If the job is a TCP listener or UDP reader, the hostname is not
1670 used and must be omitted unless it is a valid UDP multicast
1673 [netsplice] port=int
1674 [net] port=int The TCP or UDP port to bind to or connect to. If this is used
1675 with numjobs to spawn multiple instances of the same job type, then this will
1676 be the starting port number since fio will use a range of ports.
1678 [netsplice] interface=str
1679 [net] interface=str The IP address of the network interface used to send or
1680 receive UDP multicast
1683 [net] ttl=int Time-to-live value for outgoing UDP multicast packets.
1686 [netsplice] nodelay=bool
1687 [net] nodelay=bool Set TCP_NODELAY on TCP connections.
1689 [netsplice] protocol=str
1690 [netsplice] proto=str
1692 [net] proto=str The network protocol to use. Accepted values are:
1694 tcp Transmission control protocol
1695 tcpv6 Transmission control protocol V6
1696 udp User datagram protocol
1697 udpv6 User datagram protocol V6
1698 unix UNIX domain socket
1700 When the protocol is TCP or UDP, the port must also be given,
1701 as well as the hostname if the job is a TCP listener or UDP
1702 reader. For unix sockets, the normal filename option should be
1703 used and the port is invalid.
1705 [net] listen For TCP network connections, tell fio to listen for incoming
1706 connections rather than initiating an outgoing connection. The
1707 hostname must be omitted if this option is used.
1709 [net] pingpong Normaly a network writer will just continue writing data, and
1710 a network reader will just consume packages. If pingpong=1
1711 is set, a writer will send its normal payload to the reader,
1712 then wait for the reader to send the same payload back. This
1713 allows fio to measure network latencies. The submission
1714 and completion latencies then measure local time spent
1715 sending or receiving, and the completion latency measures
1716 how long it took for the other end to receive and send back.
1717 For UDP multicast traffic pingpong=1 should only be set for a
1718 single reader when multiple readers are listening to the same
1721 [net] window_size Set the desired socket buffer size for the connection.
1723 [net] mss Set the TCP maximum segment size (TCP_MAXSEG).
1725 [e4defrag] donorname=str
1726 File will be used as a block donor(swap extents between files)
1727 [e4defrag] inplace=int
1728 Configure donor file blocks allocation strategy
1729 0(default): Preallocate donor's file on init
1730 1 : allocate space immidietly inside defragment event,
1731 and free right after event
1733 [mtd] skip_bad=bool Skip operations against known bad blocks.
1736 6.0 Interpreting the output
1737 ---------------------------
1739 fio spits out a lot of output. While running, fio will display the
1740 status of the jobs created. An example of that would be:
1742 Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s]
1744 The characters inside the square brackets denote the current status of
1745 each thread. The possible values (in typical life cycle order) are:
1749 P Thread setup, but not started.
1751 I Thread initialized, waiting or generating necessary data.
1752 p Thread running pre-reading file(s).
1753 R Running, doing sequential reads.
1754 r Running, doing random reads.
1755 W Running, doing sequential writes.
1756 w Running, doing random writes.
1757 M Running, doing mixed sequential reads/writes.
1758 m Running, doing mixed random reads/writes.
1759 F Running, currently waiting for fsync()
1760 f Running, finishing up (writing IO logs, etc)
1761 V Running, doing verification of written data.
1762 E Thread exited, not reaped by main thread yet.
1764 X Thread reaped, exited with an error.
1765 K Thread reaped, exited due to signal.
1767 Fio will condense the thread string as not to take up more space on the
1768 command line as is needed. For instance, if you have 10 readers and 10
1769 writers running, the output would look like this:
1771 Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s]
1773 Fio will still maintain the ordering, though. So the above means that jobs
1774 1..10 are readers, and 11..20 are writers.
1776 The other values are fairly self explanatory - number of threads
1777 currently running and doing io, rate of io since last check (read speed
1778 listed first, then write speed), and the estimated completion percentage
1779 and time for the running group. It's impossible to estimate runtime of
1780 the following groups (if any). Note that the string is displayed in order,
1781 so it's possible to tell which of the jobs are currently doing what. The
1782 first character is the first job defined in the job file, and so forth.
1784 When fio is done (or interrupted by ctrl-c), it will show the data for
1785 each thread, group of threads, and disks in that order. For each data
1786 direction, the output looks like:
1788 Client1 (g=0): err= 0:
1789 write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec
1790 slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92
1791 clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82
1792 bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
1793 cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
1794 IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
1795 submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1796 complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
1797 issued r/w: total=0/32768, short=0/0
1798 lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
1799 lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%
1801 The client number is printed, along with the group id and error of that
1802 thread. Below is the io statistics, here for writes. In the order listed,
1805 io= Number of megabytes io performed
1806 bw= Average bandwidth rate
1807 iops= Average IOs performed per second
1808 runt= The runtime of that thread
1809 slat= Submission latency (avg being the average, stdev being the
1810 standard deviation). This is the time it took to submit
1811 the io. For sync io, the slat is really the completion
1812 latency, since queue/complete is one operation there. This
1813 value can be in milliseconds or microseconds, fio will choose
1814 the most appropriate base and print that. In the example
1815 above, milliseconds is the best scale. Note: in --minimal mode
1816 latencies are always expressed in microseconds.
1817 clat= Completion latency. Same names as slat, this denotes the
1818 time from submission to completion of the io pieces. For
1819 sync io, clat will usually be equal (or very close) to 0,
1820 as the time from submit to complete is basically just
1821 CPU time (io has already been done, see slat explanation).
1822 bw= Bandwidth. Same names as the xlat stats, but also includes
1823 an approximate percentage of total aggregate bandwidth
1824 this thread received in this group. This last value is
1825 only really useful if the threads in this group are on the
1826 same disk, since they are then competing for disk access.
1827 cpu= CPU usage. User and system time, along with the number
1828 of context switches this thread went through, usage of
1829 system and user time, and finally the number of major
1830 and minor page faults.
1831 IO depths= The distribution of io depths over the job life time. The
1832 numbers are divided into powers of 2, so for example the
1833 16= entries includes depths up to that value but higher
1834 than the previous entry. In other words, it covers the
1835 range from 16 to 31.
1836 IO submit= How many pieces of IO were submitting in a single submit
1837 call. Each entry denotes that amount and below, until
1838 the previous entry - eg, 8=100% mean that we submitted
1839 anywhere in between 5-8 ios per submit call.
1840 IO complete= Like the above submit number, but for completions instead.
1841 IO issued= The number of read/write requests issued, and how many
1843 IO latencies= The distribution of IO completion latencies. This is the
1844 time from when IO leaves fio and when it gets completed.
1845 The numbers follow the same pattern as the IO depths,
1846 meaning that 2=1.6% means that 1.6% of the IO completed
1847 within 2 msecs, 20=12.8% means that 12.8% of the IO
1848 took more than 10 msecs, but less than (or equal to) 20 msecs.
1850 After each client has been listed, the group statistics are printed. They
1851 will look like this:
1853 Run status group 0 (all jobs):
1854 READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
1855 WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
1857 For each data direction, it prints:
1859 io= Number of megabytes io performed.
1860 aggrb= Aggregate bandwidth of threads in this group.
1861 minb= The minimum average bandwidth a thread saw.
1862 maxb= The maximum average bandwidth a thread saw.
1863 mint= The smallest runtime of the threads in that group.
1864 maxt= The longest runtime of the threads in that group.
1866 And finally, the disk statistics are printed. They will look like this:
1868 Disk stats (read/write):
1869 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
1871 Each value is printed for both reads and writes, with reads first. The
1874 ios= Number of ios performed by all groups.
1875 merge= Number of merges io the io scheduler.
1876 ticks= Number of ticks we kept the disk busy.
1877 io_queue= Total time spent in the disk queue.
1878 util= The disk utilization. A value of 100% means we kept the disk
1879 busy constantly, 50% would be a disk idling half of the time.
1881 It is also possible to get fio to dump the current output while it is
1882 running, without terminating the job. To do that, send fio the USR1 signal.
1883 You can also get regularly timed dumps by using the --status-interval
1884 parameter, or by creating a file in /tmp named fio-dump-status. If fio
1885 sees this file, it will unlink it and dump the current output status.
1891 For scripted usage where you typically want to generate tables or graphs
1892 of the results, fio can output the results in a semicolon separated format.
1893 The format is one long line of values, such as:
1895 2;card0;0;0;7139336;121836;60004;1;10109;27.932460;116.933948;220;126861;3495.446807;1085.368601;226;126864;3523.635629;1089.012448;24063;99944;50.275485%;59818.274627;5540.657370;7155060;122104;60004;1;8338;29.086342;117.839068;388;128077;5032.488518;1234.785715;391;128085;5061.839412;1236.909129;23436;100928;50.287926%;59964.832030;5644.844189;14.595833%;19.394167%;123706;0;7313;0.1%;0.1%;0.1%;0.1%;0.1%;0.1%;100.0%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.01%;0.02%;0.05%;0.16%;6.04%;40.40%;52.68%;0.64%;0.01%;0.00%;0.01%;0.00%;0.00%;0.00%;0.00%;0.00%
1896 A description of this job goes here.
1898 The job description (if provided) follows on a second line.
1900 To enable terse output, use the --minimal command line option. The first
1901 value is the version of the terse output format. If the output has to
1902 be changed for some reason, this number will be incremented by 1 to
1903 signify that change.
1905 Split up, the format is as follows:
1907 terse version, fio version, jobname, groupid, error
1909 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1910 Submission latency: min, max, mean, deviation (usec)
1911 Completion latency: min, max, mean, deviation (usec)
1912 Completion latency percentiles: 20 fields (see below)
1913 Total latency: min, max, mean, deviation (usec)
1914 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1916 Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
1917 Submission latency: min, max, mean, deviation (usec)
1918 Completion latency: min, max, mean, deviation (usec)
1919 Completion latency percentiles: 20 fields (see below)
1920 Total latency: min, max, mean, deviation (usec)
1921 Bw (KB/s): min, max, aggregate percentage of total, mean, deviation
1922 CPU usage: user, system, context switches, major faults, minor faults
1923 IO depths: <=1, 2, 4, 8, 16, 32, >=64
1924 IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
1925 IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
1926 Disk utilization: Disk name, Read ios, write ios,
1927 Read merges, write merges,
1928 Read ticks, write ticks,
1929 Time spent in queue, disk utilization percentage
1930 Additional Info (dependent on continue_on_error, default off): total # errors, first error code
1932 Additional Info (dependent on description being set): Text description
1934 Completion latency percentiles can be a grouping of up to 20 sets, so
1935 for the terse output fio writes all of them. Each field will look like this:
1939 which is the Xth percentile, and the usec latency associated with it.
1941 For disk utilization, all disks used by fio are shown. So for each disk
1942 there will be a disk utilization section.
1945 8.0 Trace file format
1946 ---------------------
1947 There are two trace file format that you can encounter. The older (v1) format
1948 is unsupported since version 1.20-rc3 (March 2008). It will still be described
1949 below in case that you get an old trace and want to understand it.
1951 In any case the trace is a simple text file with a single action per line.
1954 8.1 Trace file format v1
1955 ------------------------
1956 Each line represents a single io action in the following format:
1960 where rw=0/1 for read/write, and the offset and length entries being in bytes.
1962 This format is not supported in Fio versions => 1.20-rc3.
1965 8.2 Trace file format v2
1966 ------------------------
1967 The second version of the trace file format was added in Fio version 1.17.
1968 It allows to access more then one file per trace and has a bigger set of
1969 possible file actions.
1971 The first line of the trace file has to be:
1975 Following this can be lines in two different formats, which are described below.
1977 The file management format:
1981 The filename is given as an absolute path. The action can be one of these:
1983 add Add the given filename to the trace
1984 open Open the file with the given filename. The filename has to have
1985 been added with the add action before.
1986 close Close the file with the given filename. The file has to have been
1990 The file io action format:
1992 filename action offset length
1994 The filename is given as an absolute path, and has to have been added and opened
1995 before it can be used with this format. The offset and length are given in
1996 bytes. The action can be one of these:
1998 wait Wait for 'offset' microseconds. Everything below 100 is discarded.
1999 read Read 'length' bytes beginning from 'offset'
2000 write Write 'length' bytes beginning from 'offset'
2001 sync fsync() the file
2002 datasync fdatasync() the file
2003 trim trim the given file from the given 'offset' for 'length' bytes
2006 9.0 CPU idleness profiling
2007 --------------------------
2008 In some cases, we want to understand CPU overhead in a test. For example,
2009 we test patches for the specific goodness of whether they reduce CPU usage.
2010 fio implements a balloon approach to create a thread per CPU that runs at
2011 idle priority, meaning that it only runs when nobody else needs the cpu.
2012 By measuring the amount of work completed by the thread, idleness of each
2013 CPU can be derived accordingly.
2015 An unit work is defined as touching a full page of unsigned characters. Mean
2016 and standard deviation of time to complete an unit work is reported in "unit
2017 work" section. Options can be chosen to report detailed percpu idleness or
2018 overall system idleness by aggregating percpu stats.
2021 10.0 Verification and triggers
2022 ------------------------------
2023 Fio is usually run in one of two ways, when data verification is done. The
2024 first is a normal write job of some sort with verify enabled. When the
2025 write phase has completed, fio switches to reads and verifies everything
2026 it wrote. The second model is running just the write phase, and then later
2027 on running the same job (but with reads instead of writes) to repeat the
2028 same IO patterns and verify the contents. Both of these methods depend
2029 on the write phase being completed, as fio otherwise has no idea how much
2032 With verification triggers, fio supports dumping the current write state
2033 to local files. Then a subsequent read verify workload can load this state
2034 and know exactly where to stop. This is useful for testing cases where
2035 power is cut to a server in a managed fashion, for instance.
2037 A verification trigger consists of two things:
2039 1) Storing the write state of each job
2040 2) Executing a trigger command
2042 The write state is relatively small, on the order of hundreds of bytes
2043 to single kilobytes. It contains information on the number of completions
2044 done, the last X completions, etc.
2046 A trigger is invoked either through creation ('touch') of a specified
2047 file in the system, or through a timeout setting. If fio is run with
2048 --trigger-file=/tmp/trigger-file, then it will continually check for
2049 the existence of /tmp/trigger-file. When it sees this file, it will
2050 fire off the trigger (thus saving state, and executing the trigger
2053 For client/server runs, there's both a local and remote trigger. If
2054 fio is running as a server backend, it will send the job states back
2055 to the client for safe storage, then execute the remote trigger, if
2056 specified. If a local trigger is specified, the server will still send
2057 back the write state, but the client will then execute the trigger.
2059 10.1 Verification trigger example
2060 ---------------------------------
2061 Lets say we want to run a powercut test on the remote machine 'server'.
2062 Our write workload is in write-test.fio. We want to cut power to 'server'
2063 at some point during the run, and we'll run this test from the safety
2064 or our local machine, 'localbox'. On the server, we'll start the fio
2067 server# fio --server
2069 and on the client, we'll fire off the workload:
2071 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""
2073 We set /tmp/my-trigger as the trigger file, and we tell fio to execute
2075 echo b > /proc/sysrq-trigger
2077 on the server once it has received the trigger and sent us the write
2078 state. This will work, but it's not _really_ cutting power to the server,
2079 it's merely abruptly rebooting it. If we have a remote way of cutting
2080 power to the server through IPMI or similar, we could do that through
2081 a local trigger command instead. Lets assume we have a script that does
2082 IPMI reboot of a given hostname, ipmi-reboot. On localbox, we could
2083 then have run fio with a local trigger instead:
2085 localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"
2087 For this case, fio would wait for the server to send us the write state,
2088 then execute 'ipmi-reboot server' when that happened.
2090 10.1 Loading verify state
2091 -------------------------
2092 To load store write state, read verification job file must contain
2093 the verify_state_load option. If that is set, fio will load the previously
2094 stored state. For a local fio run this is done by loading the files directly,
2095 and on a client/server run, the server backend will ask the client to send
2096 the files over and load them from there.