4 fio is a tool that will spawn a number of threads or processes doing a
5 particular type of io action as specified by the user. fio takes a
6 number of global parameters, each inherited by the thread unless
7 otherwise parameters given to them overriding that setting is given.
8 The typical use of fio is to write a job file matching the io load
15 fio resides in a git repo, the canonical place is:
17 git://git.kernel.dk/fio.git
19 The http protocol also works, path is the same.
21 Snapshots are frequently generated and they include the git meta data as
22 well. You can download them here:
24 http://brick.kernel.dk/snaps/
31 Starting with Debian "Squeeze", fio packages are part of the official
32 Debian repository. http://packages.debian.org/search?keywords=fio
35 Starting with Ubuntu 10.04 LTS (aka "Lucid Lynx"), fio packages are part
36 of the Ubuntu "universe" repository.
37 http://packages.ubuntu.com/search?keywords=fio
40 Pascal Bleser <guru@unixtech.be> has fio RPMs in his repository for SUSE
41 variants, you can find them here:
42 http://linux01.gwdg.de/~pbleser/rpm-navigation.php?cat=System/fio
45 Dag Wieƫrs has RPMs for Red Hat related distros, find them here:
46 http://dag.wieers.com/rpm/packages/fio/
49 Mandriva has integrated fio into their package repository, so installing
50 on that distro should be as easy as typing 'urpmi fio'.
53 Packages for Solaris are available from OpenCSW. Install their pkgutil
54 tool (http://www.opencsw.org/get-it/pkgutil/) and then install fio via
58 Bruce Cran <bruce@cran.org.uk> has fio packages for Windows at
59 http://www.bluestop.org/fio .
65 There's a mailing list associated with fio. It's meant for general
66 discussion, bug reporting, questions, and development - basically anything
67 that has to do with fio. An automated mail detailing recent commits is
68 automatically sent to the list at most daily. The list address is
69 fio@vger.kernel.org, subscribe by sending an email to
70 majordomo@vger.kernel.org with
74 in the body of the email. Archives can be found here:
76 http://www.spinics.net/lists/fio/
78 and archives for the old list can be found here:
80 http://maillist.kernel.dk/fio-devel/
86 Just type 'make' and 'make install'.
88 Note that GNU make is required. On BSD it's available from devel/gmake;
89 on Solaris it's in the SUNWgmake package. On platforms where GNU make
90 isn't the default, type 'gmake' instead of 'make'.
92 If your compile fails with an error like this:
95 In file included from fio.h:23,
97 os/os.h:15:20: error: libaio.h: No such file or directory
98 In file included from gettime.c:8:
99 fio.h:119: error: field 'iocb' has incomplete type
100 make: *** [gettime.o] Error 1
102 Check that you have the libaio development package installed. On RPM
103 based distros, it's typically called libaio-devel.
109 On Windows Cygwin (http://www.cygwin.com) is required with at least
110 devel/gcc4 and devel/make installed in order to build fio, and
111 admin/cygrunsrv to run it. You can also install devel/git to fetch/update
112 the source files. To create an MSI installer package, install WiX 3.6 from
113 http://wix.sourceforge.net/releases/ and run dobuild.cmd from the
114 os/windows directory.
116 Before running fio you'll need to have a copy of cygserver running. Run
117 "/usr/bin/cygserver-config" from an elevated Cygwin shell (i.e. launch the
118 Cygwin shell under the Administrator account) to configure it. Once
119 configured, run "net start cygserver" to start it, or type
120 "/usr/sbin/cygserver &" in the Cygwin shell to start a local copy.
122 If fio exits with the message "Bad system call" it normally means that
123 Cygserver isn't running.
130 --debug Enable some debugging options (see below)
131 --output Write output to file
132 --timeout Runtime in seconds
133 --latency-log Generate per-job latency logs
134 --bandwidth-log Generate per-job bandwidth logs
135 --minimal Minimal (terse) output
136 --version Print version info and exit
137 --help Print this page
138 --cmdhelp=cmd Print command help, "all" for all of them
139 --showcmd Turn a job file into command line options
140 --readonly Turn on safety read-only checks, preventing writes
141 --eta=when When ETA estimate should be printed
142 May be "always", "never" or "auto"
143 --section=name Only run specified section in job file
144 --alloc-size=kb Set smalloc pool to this size in kb (def 1024)
145 --warnings-fatal Fio parser warnings are fatal
148 Any parameters following the options will be assumed to be job files,
149 unless they match a job file parameter. You can add as many as you want,
150 each job file will be regarded as a separate group and fio will stonewall
153 The --readonly switch is an extra safety guard to prevent accidentally
154 turning on a write setting when that is not desired. Fio will only write
155 if rw=write/randwrite/rw/randrw is given, but this extra safety net can
156 be used as an extra precaution. It will also enable a write check in the
157 io engine core to prevent an accidental write due to a fio bug.
159 The debug switch allows adding options that trigger certain logging
160 options in fio. Currently the options are:
162 process Dump info related to processes
163 file Dump info related to file actions
164 io Dump info related to IO queuing
165 mem Dump info related to memory allocations
166 blktrace Dump info related to blktrace setup
167 verify Dump info related to IO verification
168 all Enable all debug options
169 random Dump info related to random offset generation
170 parse Dump info related to option matching and parsing
171 diskutil Dump info related to disk utilization updates
172 job:x Dump info only related to job number x
173 mutex Dump info only related to mutex up/down ops
174 profile Dump info related to profile extensions
175 time Dump info related to internal time keeping
176 ? or help Show available debug options.
178 You can specify as many as you want, eg --debug=file,mem will enable
179 file and memory debugging.
181 The section switch is meant to make it easier to ship a bigger job file
182 instead of several smaller ones. Say you define a job file with light,
183 moderate, and heavy parts. Then you can ask fio to run the given part
184 only by giving it a --section=heavy command line option. The section
185 option only applies to job sections, the reserved 'global' section is
186 always parsed and taken into account.
188 Fio has an internal allocator for shared memory called smalloc. It
189 allocates shared structures from this pool. The pool defaults to 1024k
190 in size, and can grow to 128 pools. If running large jobs with randommap
191 enabled it can run out of memory, in which case the --alloc-size switch
192 is handy for starting with a larger pool size. The backing store is
193 files in /tmp. Fio cleans up after itself, while it is running you
194 may see .fio_smalloc.* files in /tmp.
200 See the HOWTO file for a more detailed description of parameters and what
201 they mean. This file contains the terse version. You can describe big and
202 complex setups with the command line, but generally it's a lot easier to
203 just write a simple job file to describe the workload. The job file format
204 is in the ini style format, as that is easy to read and write for the user.
206 The job file parameters are:
208 name=x Use 'x' as the identifier for this job.
209 description=x 'x' is a text description of the job.
210 directory=x Use 'x' as the top level directory for storing files
211 filename=x Force the use of 'x' as the filename for all files
212 in this thread. If not given, fio will make up
213 a suitable filename based on the thread and file
215 rw=x 'x' may be: read, randread, write, randwrite,
216 rw (read-write mix), randrw (read-write random mix)
217 rwmixcycle=x Base cycle for switching between read and write
219 rwmixread=x 'x' percentage of rw mix ios will be reads. If
220 rwmixwrite is also given, the last of the two will
221 be used if they don't add up to 100%.
222 rwmixwrite=x 'x' percentage of rw mix ios will be writes. See
224 rand_repeatable=x The sequence of random io blocks can be repeatable
225 across runs, if 'x' is 1.
226 size=x Set file size to x bytes (x string can include k/m/g)
227 ioengine=x 'x' may be: aio/libaio/linuxaio for Linux aio,
228 posixaio for POSIX aio, solarisaio for Solaris
229 native async IO, windowsaio for Windows native async IO,
230 sync for regular read/write io,
231 psync for regular pread/pwrite io, vsync for regular
232 readv/writev (with queuing emulation) mmap for mmap'ed
233 io, syslet-rw for syslet driven read/write, splice for
234 using splice/vmsplice, sg for direct SG_IO io, net
235 for network io, or cpuio for a cycler burner load. sg
236 only works on Linux on SCSI (or SCSI-like devices, such
237 as usb-storage or sata/libata driven) devices. Fio also
238 has a null io engine, which is mainly used for testing
241 iodepth=x For async io, allow 'x' ios in flight
242 overwrite=x If 'x', layout a write file first.
243 nrfiles=x Spread io load over 'x' number of files per job,
245 prio=x Run io at prio X, 0-7 is the kernel allowed range
246 prioclass=x Run io at prio class X
247 bs=x Use 'x' for thread blocksize. May include k/m postfix.
248 bsrange=x-y Mix thread block sizes randomly between x and y. May
249 also include k/m postfix.
250 direct=x 1 for direct IO, 0 for buffered IO
251 thinktime=x "Think" x usec after each io
252 rate=x Throttle rate to x KB/sec
253 ratemin=x Quit if rate of x KB/sec can't be met
254 ratecycle=x ratemin averaged over x msecs
255 cpumask=x Only allow job to run on CPUs defined by mask.
256 cpus_allowed=x Like 'cpumask', but allow text setting of CPU affinity.
257 fsync=x If writing with buffered IO, fsync after every
258 'x' blocks have been written.
259 end_fsync=x If 'x', run fsync() after end-of-job.
260 startdelay=x Start this thread x seconds after startup
261 runtime=x Terminate x seconds after startup. Can include a
262 normal time suffix if not given in seconds, such as
263 'm' for minutes, 'h' for hours, and 'd' for days.
264 offset=x Start io at offset x (x string can include k/m/g)
265 invalidate=x Invalidate page cache for file prior to doing io
266 sync=x Use sync writes if x and writing buffered IO.
267 mem=x If x == malloc, use malloc for buffers. If x == shm,
268 use shared memory for buffers. If x == mmap, use
270 exitall When one thread quits, terminate the others
271 bwavgtime=x Average bandwidth stats over an x msec window.
272 create_serialize=x If 'x', serialize file creation.
273 create_fsync=x If 'x', run fsync() after file creation.
274 unlink If set, unlink files when done.
275 loops=x Run the job 'x' number of times.
276 verify=x If 'x' == md5, use md5 for verifies. If 'x' == crc32,
277 use crc32 for verifies. md5 is 'safer', but crc32 is
278 a lot faster. Only makes sense for writing to a file.
279 For other types of checksumming, see HOWTO.
280 stonewall Wait for preceeding jobs to end before running.
281 numjobs=x Create 'x' similar entries for this job
282 thread Use pthreads instead of forked jobs
284 zoneskip=y Zone options must be paired. If given, the job
285 will skip y bytes for every x read/written. This
286 can be used to gauge hard drive speed over the entire
287 platter, without reading everything. Both x/y can
288 include k/m/g suffix.
289 iolog=x Open and read io pattern from file 'x'. The file must
290 contain one io action per line in the following format:
292 where with rw=0/1 for read/write, and the offset
293 and length entries being in bytes.
294 write_iolog=x Write an iolog to file 'x' in the same format as iolog.
295 The iolog options are exclusive, if both given the
296 read iolog will be performed. Specify a separate file
297 for each job, otherwise the iologs will be interspersed
298 and the file may be corrupt.
299 write_bw_log Write a bandwidth log.
300 write_lat_log Write a latency log.
301 lockmem=x Lock down x amount of memory on the machine, to
302 simulate a machine with less memory available. x can
303 include k/m/g suffix.
304 nice=x Run job at given nice value.
305 exec_prerun=x Run 'x' before job io is begun.
306 exec_postrun=x Run 'x' after job io has finished.
307 ioscheduler=x Use ioscheduler 'x' for this job.
308 cpuload=x For a CPU io thread, percentage of CPU time to attempt
310 cpuchunks=x Split burn cycles into pieces of x usecs.
317 Fio works on (at least) Linux, Solaris, AIX, OSX, NetBSD, Windows and FreeBSD.
318 Some features and/or options may only be available on some of the platforms,
319 typically because those features only apply to that platform (like the
320 solarisaio engine, or the splice engine on Linux).
322 Some features are not available on FreeBSD/Solaris even if they could be
323 implemented, I'd be happy to take patches for that. An example of that is
324 disk utility statistics and (I think) huge page support, support for that
325 does exist in FreeBSD/Solaris.
327 Fio uses pthread mutexes for signalling and locking and FreeBSD does not
328 support process shared pthread mutexes. As a result, only threads are
329 supported on FreeBSD. This could be fixed with sysv ipc locking or
330 other locking alternatives.
332 Other *BSD platforms are untested, but fio should work there almost out
333 of the box. Since I don't do test runs or even compiles on those platforms,
334 your mileage may vary. Sending me patches for other platforms is greatly
335 appreciated. There's a lot of value in having the same test/benchmark tool
336 available on all platforms.
338 Note that POSIX aio is not enabled by default on AIX. If you get messages like:
340 Symbol resolution failed for /usr/lib/libc.a(posix_aio.o) because:
341 Symbol _posix_kaio_rdwr (number 2) is not exported from dependent module /unix.
343 you need to enable POSIX aio. Run the following commands as root:
345 # lsdev -C -l posix_aio0
346 posix_aio0 Defined Posix Asynchronous I/O
347 # cfgmgr -l posix_aio0
348 # lsdev -C -l posix_aio0
349 posix_aio0 Available Posix Asynchronous I/O
351 POSIX aio should work now. To make the change permanent:
353 # chdev -l posix_aio0 -P -a autoconfig='available'
360 Fio was written by Jens Axboe <axboe@kernel.dk> to enable flexible testing
361 of the Linux IO subsystem and schedulers. He got tired of writing
362 specific test applications to simulate a given workload, and found that
363 the existing io benchmark/test tools out there weren't flexible enough
364 to do what he wanted.
366 Jens Axboe <axboe@kernel.dk> 20060905