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 If you are inside a corporate firewall, git:// may not always work for
20 you. In that case you can use the http protocol, path is the same:
22 http://git.kernel.dk/fio.git
24 Snapshots are frequently generated and they include the git meta data as
25 well. You can download them here:
27 http://brick.kernel.dk/snaps/
34 Starting with Debian "Squeeze", fio packages are part of the official
35 Debian repository. http://packages.debian.org/search?keywords=fio
38 Starting with Ubuntu 10.04 LTS (aka "Lucid Lynx"), fio packages are part
39 of the Ubuntu "universe" repository.
40 http://packages.ubuntu.com/search?keywords=fio
43 Dag Wieƫrs has RPMs for Red Hat related distros, find them here:
44 http://dag.wieers.com/rpm/packages/fio/
47 Mandriva has integrated fio into their package repository, so installing
48 on that distro should be as easy as typing 'urpmi fio'.
51 Packages for Solaris are available from OpenCSW. Install their pkgutil
52 tool (http://www.opencsw.org/get-it/pkgutil/) and then install fio via
56 Bruce Cran <bruce@cran.org.uk> has fio packages for Windows at
57 http://www.bluestop.org/fio/ .
63 There's a mailing list associated with fio. It's meant for general
64 discussion, bug reporting, questions, and development - basically anything
65 that has to do with fio. An automated mail detailing recent commits is
66 automatically sent to the list at most daily. The list address is
67 fio@vger.kernel.org, subscribe by sending an email to
68 majordomo@vger.kernel.org with
72 in the body of the email. Archives can be found here:
74 http://www.spinics.net/lists/fio/
76 and archives for the old list can be found here:
78 http://maillist.kernel.dk/fio-devel/
84 Just type 'make' and 'make install'.
86 Note that GNU make is required. On BSD it's available from devel/gmake;
87 on Solaris it's in the SUNWgmake package. On platforms where GNU make
88 isn't the default, type 'gmake' instead of 'make'.
90 If your compile fails with an error like this:
93 In file included from fio.h:23,
95 os/os.h:15:20: error: libaio.h: No such file or directory
96 In file included from gettime.c:8:
97 fio.h:119: error: field 'iocb' has incomplete type
98 make: *** [gettime.o] Error 1
100 Check that you have the libaio development package installed. On RPM
101 based distros, it's typically called libaio-devel.
107 On Windows Cygwin (http://www.cygwin.com/) is required in order to
108 build fio. To create an MSI installer package install WiX 3.7 from
109 http://wixtoolset.org and run dobuild.cmd from the
110 os/windows directory.
112 How to compile FIO on 64-bit Windows:
114 1. Install Cygwin (http://www.cygwin.com/setup.exe). Install 'make' and all
115 packages starting with 'mingw64-i686' and 'mingw64-x86_64'.
116 2. Download ftp://sourceware.org/pub/pthreads-win32/prebuilt-dll-2-9-1-release/dll/x64/pthreadGC2.dll
117 and copy to the fio source directory.
118 3. Open the Cygwin Terminal.
119 4. Go to the fio directory (source files).
123 To build fio on 32-bit Windows, download x86/pthreadGC2.dll instead and do
124 './configure --build-32bit-win=yes' before 'make'.
126 It's recommended that once built or installed, fio be run in a Command Prompt
127 or other 'native' console such as console2, since there are known to be display
128 and signal issues when running it under a Cygwin shell
129 (see http://code.google.com/p/mintty/issues/detail?id=56 for details).
136 --debug Enable some debugging options (see below)
137 --parse-only Parse options only, don't start any IO
138 --output Write output to file
139 --runtime Runtime in seconds
140 --latency-log Generate per-job latency logs
141 --bandwidth-log Generate per-job bandwidth logs
142 --minimal Minimal (terse) output
143 --output-format=type Output format (terse,json,normal)
144 --terse-version=type Terse version output format (default 3, or 2 or 4).
145 --version Print version info and exit
146 --help Print this page
147 --cpuclock-test Perform test/validation of CPU clock
148 --cmdhelp=cmd Print command help, "all" for all of them
149 --enghelp=engine Print ioengine help, or list available ioengines
150 --enghelp=engine,cmd Print help for an ioengine cmd
151 --showcmd Turn a job file into command line options
152 --readonly Turn on safety read-only checks, preventing
154 --eta=when When ETA estimate should be printed
155 May be "always", "never" or "auto"
156 --eta-newline=time Force a new line for every 'time' period passed
157 --section=name Only run specified section in job file.
158 Multiple sections can be specified.
159 --alloc-size=kb Set smalloc pool to this size in kb (def 1024)
160 --warnings-fatal Fio parser warnings are fatal
161 --max-jobs Maximum number of threads/processes to support
162 --server=args Start backend server. See Client/Server section.
163 --client=host Connect to specified backend.
164 --idle-prof=option Report cpu idleness on a system or percpu basis
165 (option=system,percpu) or run unit work
166 calibration only (option=calibrate).
169 Any parameters following the options will be assumed to be job files,
170 unless they match a job file parameter. You can add as many as you want,
171 each job file will be regarded as a separate group and fio will stonewall
174 The --readonly switch is an extra safety guard to prevent accidentally
175 turning on a write setting when that is not desired. Fio will only write
176 if rw=write/randwrite/rw/randrw is given, but this extra safety net can
177 be used as an extra precaution. It will also enable a write check in the
178 io engine core to prevent an accidental write due to a fio bug.
180 The debug switch allows adding options that trigger certain logging
181 options in fio. Currently the options are:
183 process Dump info related to processes
184 file Dump info related to file actions
185 io Dump info related to IO queuing
186 mem Dump info related to memory allocations
187 blktrace Dump info related to blktrace setup
188 verify Dump info related to IO verification
189 all Enable all debug options
190 random Dump info related to random offset generation
191 parse Dump info related to option matching and parsing
192 diskutil Dump info related to disk utilization updates
193 job:x Dump info only related to job number x
194 mutex Dump info only related to mutex up/down ops
195 profile Dump info related to profile extensions
196 time Dump info related to internal time keeping
197 ? or help Show available debug options.
199 You can specify as many as you want, eg --debug=file,mem will enable
200 file and memory debugging.
202 The section switch is meant to make it easier to ship a bigger job file
203 instead of several smaller ones. Say you define a job file with light,
204 moderate, and heavy parts. Then you can ask fio to run the given part
205 only by giving it a --section=heavy command line option. The section
206 option only applies to job sections, the reserved 'global' section is
207 always parsed and taken into account.
209 Fio has an internal allocator for shared memory called smalloc. It
210 allocates shared structures from this pool. The pool defaults to 1024k
211 in size, and can grow to 128 pools. If running large jobs with randommap
212 enabled it can run out of memory, in which case the --alloc-size switch
213 is handy for starting with a larger pool size. The backing store is
214 files in /tmp. Fio cleans up after itself, while it is running you
215 may see .fio_smalloc.* files in /tmp.
221 See the HOWTO file for a more detailed description of parameters and what
222 they mean. This file contains the terse version. You can describe big and
223 complex setups with the command line, but generally it's a lot easier to
224 just write a simple job file to describe the workload. The job file format
225 is in the ini style format, as that is easy to read and write for the user.
227 The job file parameters are:
229 name=x Use 'x' as the identifier for this job.
230 description=x 'x' is a text description of the job.
231 directory=x Use 'x' as the top level directory for storing files
232 filename=x Force the use of 'x' as the filename for all files
233 in this thread. If not given, fio will make up
234 a suitable filename based on the thread and file
236 rw=x 'x' may be: read, randread, write, randwrite,
237 rw (read-write mix), randrw (read-write random mix)
238 rwmixcycle=x Base cycle for switching between read and write
240 rwmixread=x 'x' percentage of rw mix ios will be reads. If
241 rwmixwrite is also given, the last of the two will
242 be used if they don't add up to 100%.
243 rwmixwrite=x 'x' percentage of rw mix ios will be writes. See
245 rand_repeatable=x The sequence of random io blocks can be repeatable
246 across runs, if 'x' is 1.
247 size=x Set file size to x bytes (x string can include k/m/g)
248 ioengine=x 'x' may be: aio/libaio/linuxaio for Linux aio,
249 posixaio for POSIX aio, solarisaio for Solaris
250 native async IO, windowsaio for Windows native async IO,
251 sync for regular read/write io,
252 psync for regular pread/pwrite io, vsync for regular
253 readv/writev (with queuing emulation) mmap for mmap'ed
254 io, syslet-rw for syslet driven read/write, splice for
255 using splice/vmsplice, sg for direct SG_IO io, net
256 for network io, rdma for RDMA io, or cpuio for a
257 cycler burner load. sg only works on Linux on
258 SCSI (or SCSI-like devices, such as usb-storage or
259 sata/libata driven) devices. Fio also has a null
260 io engine, which is mainly used for testing
263 iodepth=x For async io, allow 'x' ios in flight
264 overwrite=x If 'x', layout a write file first.
265 nrfiles=x Spread io load over 'x' number of files per job,
267 prio=x Run io at prio X, 0-7 is the kernel allowed range
268 prioclass=x Run io at prio class X
269 bs=x Use 'x' for thread blocksize. May include k/m postfix.
270 bsrange=x-y Mix thread block sizes randomly between x and y. May
271 also include k/m postfix.
272 direct=x 1 for direct IO, 0 for buffered IO
273 thinktime=x "Think" x usec after each io
274 rate=x Throttle rate to x KB/sec
275 ratemin=x Quit if rate of x KB/sec can't be met
276 ratecycle=x ratemin averaged over x msecs
277 cpumask=x Only allow job to run on CPUs defined by mask.
278 cpus_allowed=x Like 'cpumask', but allow text setting of CPU affinity.
279 numa_cpu_nodes=x,y-z Allow job to run on specified NUMA nodes' CPU.
280 numa_mem_policy=m:x,y-z Setup numa memory allocation policy.
281 'm' stands for policy, such as local, interleave,
282 bind, prefer, local. 'x, y-z' are numa node(s) for
283 memory allocation according to policy.
284 fsync=x If writing with buffered IO, fsync after every
285 'x' blocks have been written.
286 end_fsync=x If 'x', run fsync() after end-of-job.
287 startdelay=x Start this thread x seconds after startup
288 runtime=x Terminate x seconds after startup. Can include a
289 normal time suffix if not given in seconds, such as
290 'm' for minutes, 'h' for hours, and 'd' for days.
291 offset=x Start io at offset x (x string can include k/m/g)
292 invalidate=x Invalidate page cache for file prior to doing io
293 sync=x Use sync writes if x and writing buffered IO.
294 mem=x If x == malloc, use malloc for buffers. If x == shm,
295 use shared memory for buffers. If x == mmap, use
297 exitall When one thread quits, terminate the others
298 bwavgtime=x Average bandwidth stats over an x msec window.
299 create_serialize=x If 'x', serialize file creation.
300 create_fsync=x If 'x', run fsync() after file creation.
301 unlink If set, unlink files when done.
302 loops=x Run the job 'x' number of times.
303 verify=x If 'x' == md5, use md5 for verifies. If 'x' == crc32,
304 use crc32 for verifies. md5 is 'safer', but crc32 is
305 a lot faster. Only makes sense for writing to a file.
306 For other types of checksumming, see HOWTO.
307 stonewall Wait for preceeding jobs to end before running.
308 numjobs=x Create 'x' similar entries for this job
309 thread Use pthreads instead of forked jobs
311 zoneskip=y Zone options must be paired. If given, the job
312 will skip y bytes for every x read/written. This
313 can be used to gauge hard drive speed over the entire
314 platter, without reading everything. Both x/y can
315 include k/m/g suffix.
316 read_iolog=x Open and read io pattern from file 'x'. The file format
317 is described in the HOWTO.
318 write_iolog=x Write an iolog to file 'x' in the same format as iolog.
319 The iolog options are exclusive, if both given the
320 read iolog will be performed. Specify a separate file
321 for each job, otherwise the iologs will be interspersed
322 and the file may be corrupt.
323 write_bw_log Write a bandwidth log.
324 write_lat_log Write a latency log.
325 lockmem=x Lock down x amount of memory on the machine, to
326 simulate a machine with less memory available. x can
327 include k/m/g suffix.
328 nice=x Run job at given nice value.
329 exec_prerun=x Run 'x' before job io is begun.
330 exec_postrun=x Run 'x' after job io has finished.
331 ioscheduler=x Use ioscheduler 'x' for this job.
332 cpuload=x For a CPU io thread, percentage of CPU time to attempt
334 cpuchunks=x Split burn cycles into pieces of x usecs.
341 Normally you would run fio as a stand-alone application on the machine
342 where the IO workload should be generated. However, it is also possible to
343 run the frontend and backend of fio separately. This makes it possible to
344 have a fio server running on the machine(s) where the IO workload should
345 be running, while controlling it from another machine.
347 To start the server, you would do:
351 on that machine, where args defines what fio listens to. The arguments
352 are of the form 'type,hostname or IP,port'. 'type' is either 'ip' (or ip4)
353 for TCP/IP v4, 'ip6' for TCP/IP v6, or 'sock' for a local unix domain socket.
354 'hostname' is either a hostname or IP address, and 'port' is the port to
355 listen to (only valid for TCP/IP, not a local socket). Some examples:
359 Start a fio server, listening on all interfaces on the default port (8765).
361 2) fio --server=ip:hostname,4444
363 Start a fio server, listening on IP belonging to hostname and on port 4444.
365 3) fio --server=ip6:::1,4444
367 Start a fio server, listening on IPv6 localhost ::1 and on port 4444.
369 4) fio --server=,4444
371 Start a fio server, listening on all interfaces on port 4444.
373 5) fio --server=1.2.3.4
375 Start a fio server, listening on IP 1.2.3.4 on the default port.
377 6) fio --server=sock:/tmp/fio.sock
379 Start a fio server, listening on the local socket /tmp/fio.sock.
381 When a server is running, you can connect to it from a client. The client
384 fio --local-args --client=server --remote-args <job file(s)>
386 where --local-args are arguments that are local to the client where it is
387 running, 'server' is the connect string, and --remote-args and <job file(s)>
388 are sent to the server. The 'server' string follows the same format as it
389 does on the server side, to allow IP/hostname/socket and port strings.
390 You can connect to multiple clients as well, to do that you could run:
392 fio --client=server2 <job file(s)> --client=server2 <job file(s)>
398 Fio works on (at least) Linux, Solaris, AIX, HP-UX, OSX, NetBSD, Windows
399 and FreeBSD. Some features and/or options may only be available on some of
400 the platforms, typically because those features only apply to that platform
401 (like the solarisaio engine, or the splice engine on Linux).
403 Some features are not available on FreeBSD/Solaris even if they could be
404 implemented, I'd be happy to take patches for that. An example of that is
405 disk utility statistics and (I think) huge page support, support for that
406 does exist in FreeBSD/Solaris.
408 Fio uses pthread mutexes for signalling and locking and FreeBSD does not
409 support process shared pthread mutexes. As a result, only threads are
410 supported on FreeBSD. This could be fixed with sysv ipc locking or
411 other locking alternatives.
413 Other *BSD platforms are untested, but fio should work there almost out
414 of the box. Since I don't do test runs or even compiles on those platforms,
415 your mileage may vary. Sending me patches for other platforms is greatly
416 appreciated. There's a lot of value in having the same test/benchmark tool
417 available on all platforms.
419 Note that POSIX aio is not enabled by default on AIX. If you get messages like:
421 Symbol resolution failed for /usr/lib/libc.a(posix_aio.o) because:
422 Symbol _posix_kaio_rdwr (number 2) is not exported from dependent module /unix.
424 you need to enable POSIX aio. Run the following commands as root:
426 # lsdev -C -l posix_aio0
427 posix_aio0 Defined Posix Asynchronous I/O
428 # cfgmgr -l posix_aio0
429 # lsdev -C -l posix_aio0
430 posix_aio0 Available Posix Asynchronous I/O
432 POSIX aio should work now. To make the change permanent:
434 # chdev -l posix_aio0 -P -a autoconfig='available'
441 Fio was written by Jens Axboe <axboe@kernel.dk> to enable flexible testing
442 of the Linux IO subsystem and schedulers. He got tired of writing
443 specific test applications to simulate a given workload, and found that
444 the existing io benchmark/test tools out there weren't flexible enough
445 to do what he wanted.
447 Jens Axboe <axboe@kernel.dk> 20060905