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/
26 Pascal Bleser <guru@unixtech.be> has fio RPMs in his repository for
27 SUSE variants, you can find them here:
29 http://linux01.gwdg.de/~pbleser/rpm-navigation.php?cat=System/fio
31 Dag Wieƫrs has RPMs for Red Hat related distros, find them here:
33 http://dag.wieers.com/rpm/packages/fio/
35 Mandriva has integrated fio into their package repository, so installing
36 on that distro should be as easy as typing 'urpmi fio'.
42 There's a mailing list associated with fio. It's meant for general
43 discussion, bug reporting, questions, and development - basically anything
44 that has to do with fio. An automated mail detailing recent commits is
45 automatically sent to the list at most daily. The list address is
46 fio@vger.kernel.org, subscribe by sending an email to
47 majordomo@vger.kernel.org with
51 in the body of the email. There is no archive for the new list yet,
52 archives for the old list can be found here:
54 http://maillist.kernel.dk/fio-devel/
60 Just type 'make' and 'make install'. If on FreeBSD, for now you have to
61 specify the FreeBSD Makefile with -f and use gmake (not make), eg:
63 $ gmake -f Makefile.Freebsd && gmake -f Makefile.FreeBSD install
65 Likewise with OpenSolaris, use the Makefile.solaris to compile there.
66 The OpenSolaris make should work fine. This might change in the
67 future if I opt for an autoconf type setup.
69 If your compile fails with an error like this:
72 In file included from fio.h:23,
74 os/os.h:15:20: error: libaio.h: No such file or directory
75 In file included from gettime.c:8:
76 fio.h:119: error: field 'iocb' has incomplete type
77 make: *** [gettime.o] Error 1
79 Check that you have the libaio development package installed. On RPM
80 based distros, it's typically called libaio-devel.
87 --debug Enable some debugging options (see below)
88 --output Write output to file
89 --timeout Runtime in seconds
90 --latency-log Generate per-job latency logs
91 --bandwidth-log Generate per-job bandwidth logs
92 --minimal Minimal (terse) output
93 --version Print version info and exit
94 --help Print this page
95 --cmdhelp=cmd Print command help, "all" for all of them
96 --showcmd Turn a job file into command line options
97 --readonly Turn on safety read-only checks, preventing writes
98 --eta=when When ETA estimate should be printed
99 May be "always", "never" or "auto"
100 --section=name Only run specified section in job file
101 --alloc-size=kb Set smalloc pool to this size in kb (def 1024)
104 Any parameters following the options will be assumed to be job files,
105 unless they match a job file parameter. You can add as many as you want,
106 each job file will be regarded as a separate group and fio will stonewall
109 The --readonly switch is an extra safety guard to prevent accidentically
110 turning on a write setting when that is not desired. Fio will only write
111 if rw=write/randwrite/rw/randrw is given, but this extra safety net can
112 be used as an extra precaution. It will also enable a write check in the
113 io engine core to prevent an accidental write due to a fio bug.
115 The debug switch allows adding options that trigger certain logging
116 options in fio. Currently the options are:
118 process Dump info related to processes
119 file Dump info related to file actions
120 io Dump info related to IO queuing
121 mem Dump info related to memory allocations
122 blktrace Dump info related to blktrace setup
123 verify Dump info related to IO verification
124 all Enable all debug options
125 random Dump info related to random offset generation
126 parse Dump info related to option matching and parsing
127 diskutil Dump info related to disk utilization updates
128 job:x Dump info only related to job number x
129 mutex Dump info only related to mutex up/down ops
130 profile Dump info related to profile extensions
131 time Dump info related to internal time keeping
132 ? or help Show available debug options.
134 You can specify as many as you want, eg --debug=file,mem will enable
135 file and memory debugging.
137 The section switch is meant to make it easier to ship a bigger job file
138 instead of several smaller ones. Say you define a job file with light,
139 moderate, and heavy parts. Then you can ask fio to run the given part
140 only by giving it a --section=heavy command line option. The section
141 option only applies to job sections, the reserved 'global' section is
142 always parsed and taken into account.
144 Fio has an internal allocator for shared memory called smalloc. It
145 allocates shared structures from this pool. The pool defaults to 1024k
146 in size, and can grow to 128 pools. If running large jobs with randommap
147 enabled it can run out of memory, in which case the --alloc-size switch
148 is handy for starting with a larger pool size. The backing store is
149 files in /tmp. Fio cleans up after itself, while it is running you
150 may see .fio_smalloc.* files in /tmp.
156 See the HOWTO file for a more detailed description of parameters and what
157 they mean. This file contains the terse version. You can describe big and
158 complex setups with the command line, but generally it's a lot easier to
159 just write a simple job file to describe the workload. The job file format
160 is in the ini style format, as that is easy to read and write for the user.
162 The job file parameters are:
164 name=x Use 'x' as the identifier for this job.
165 description=x 'x' is a text description of the job.
166 directory=x Use 'x' as the top level directory for storing files
167 filename=x Force the use of 'x' as the filename for all files
168 in this thread. If not given, fio will make up
169 a suitable filename based on the thread and file
171 rw=x 'x' may be: read, randread, write, randwrite,
172 rw (read-write mix), randrw (read-write random mix)
173 rwmixcycle=x Base cycle for switching between read and write
175 rwmixread=x 'x' percentage of rw mix ios will be reads. If
176 rwmixwrite is also given, the last of the two will
177 be used if they don't add up to 100%.
178 rwmixwrite=x 'x' percentage of rw mix ios will be writes. See
180 rand_repeatable=x The sequence of random io blocks can be repeatable
181 across runs, if 'x' is 1.
182 size=x Set file size to x bytes (x string can include k/m/g)
183 ioengine=x 'x' may be: aio/libaio/linuxaio for Linux aio,
184 posixaio for POSIX aio, solarisaio for Solaris
185 native async IO, sync for regular read/write io,
186 psync for regular pread/pwrite io, vsync for regular
187 readv/writev (with queuing emulation) mmap for mmap'ed
188 io, syslet-rw for syslet driven read/write, splice for
189 using splice/vmsplice, sg for direct SG_IO io, net
190 for network io, or cpuio for a cycler burner load. sg
191 only works on Linux on SCSI (or SCSI-like devices, such
192 as usb-storage or sata/libata driven) devices. Fio also
193 has a null io engine, which is mainly used for testing
196 iodepth=x For async io, allow 'x' ios in flight
197 overwrite=x If 'x', layout a write file first.
198 nrfiles=x Spread io load over 'x' number of files per job,
200 prio=x Run io at prio X, 0-7 is the kernel allowed range
201 prioclass=x Run io at prio class X
202 bs=x Use 'x' for thread blocksize. May include k/m postfix.
203 bsrange=x-y Mix thread block sizes randomly between x and y. May
204 also include k/m postfix.
205 direct=x 1 for direct IO, 0 for buffered IO
206 thinktime=x "Think" x usec after each io
207 rate=x Throttle rate to x KB/sec
208 ratemin=x Quit if rate of x KB/sec can't be met
209 ratecycle=x ratemin averaged over x msecs
210 cpumask=x Only allow job to run on CPUs defined by mask.
211 cpus_allowed=x Like 'cpumask', but allow text setting of CPU affinity.
212 fsync=x If writing with buffered IO, fsync after every
213 'x' blocks have been written.
214 end_fsync=x If 'x', run fsync() after end-of-job.
215 startdelay=x Start this thread x seconds after startup
216 runtime=x Terminate x seconds after startup. Can include a
217 normal time suffix if not given in seconds, such as
218 'm' for minutes, 'h' for hours, and 'd' for days.
219 offset=x Start io at offset x (x string can include k/m/g)
220 invalidate=x Invalidate page cache for file prior to doing io
221 sync=x Use sync writes if x and writing buffered IO.
222 mem=x If x == malloc, use malloc for buffers. If x == shm,
223 use shared memory for buffers. If x == mmap, use
225 exitall When one thread quits, terminate the others
226 bwavgtime=x Average bandwidth stats over an x msec window.
227 create_serialize=x If 'x', serialize file creation.
228 create_fsync=x If 'x', run fsync() after file creation.
229 unlink If set, unlink files when done.
230 loops=x Run the job 'x' number of times.
231 verify=x If 'x' == md5, use md5 for verifies. If 'x' == crc32,
232 use crc32 for verifies. md5 is 'safer', but crc32 is
233 a lot faster. Only makes sense for writing to a file.
234 For other types of checksumming, see HOWTO.
235 stonewall Wait for preceeding jobs to end before running.
236 numjobs=x Create 'x' similar entries for this job
237 thread Use pthreads instead of forked jobs
239 zoneskip=y Zone options must be paired. If given, the job
240 will skip y bytes for every x read/written. This
241 can be used to gauge hard drive speed over the entire
242 platter, without reading everything. Both x/y can
243 include k/m/g suffix.
244 iolog=x Open and read io pattern from file 'x'. The file must
245 contain one io action per line in the following format:
247 where with rw=0/1 for read/write, and the offset
248 and length entries being in bytes.
249 write_iolog=x Write an iolog to file 'x' in the same format as iolog.
250 The iolog options are exclusive, if both given the
251 read iolog will be performed.
252 write_bw_log Write a bandwidth log.
253 write_lat_log Write a latency log.
254 lockmem=x Lock down x amount of memory on the machine, to
255 simulate a machine with less memory available. x can
256 include k/m/g suffix.
257 nice=x Run job at given nice value.
258 exec_prerun=x Run 'x' before job io is begun.
259 exec_postrun=x Run 'x' after job io has finished.
260 ioscheduler=x Use ioscheduler 'x' for this job.
261 cpuload=x For a CPU io thread, percentage of CPU time to attempt
263 cpuchunks=x Split burn cycles into pieces of x usecs.
270 Fio works on (at least) Linux, Solaris, and FreeBSD. Some features and/or
271 options may only be available on some of the platforms, typically because
272 those features only apply to that platform (like the solarisaio engine, or
273 the splice engine on Linux).
275 Some features are not available on FreeBSD/Solaris even if they could be
276 implemented, I'd be happy to take patches for that. An example of that is
277 disk utility statistics and (I think) huge page support, support for that
278 does exist in FreeBSD/Solaris.
280 Fio uses pthread mutexes for signalling and locking and FreeBSD does not
281 support process shared pthread mutexes. As a result, only threads are
282 supported on FreeBSD. This could be fixed with sysv ipc locking or
283 other locking alternatives.
285 Other *BSD platforms are untested, but fio should work there almost out
286 of the box. Since I don't do test runs or even compiles on those platforms,
287 your mileage may vary. Sending me patches for other platforms is greatly
288 appreciated. There's a lot of value in having the same test/benchmark tool
289 available on all platforms.
296 Fio was written by Jens Axboe <axboe@kernel.dk> to enable flexible testing
297 of the Linux IO subsystem and schedulers. He got tired of writing
298 specific test applications to simulate a given workload, and found that
299 the existing io benchmark/test tools out there weren't flexible enough
300 to do what he wanted.
302 Jens Axboe <axboe@kernel.dk> 20060905