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://brick.kernel.dk/data/git/fio.git
19 Snapshots are frequently generated and they include the git meta data as
20 well. You can download them here:
22 http://brick.kernel.dk/snaps/
24 Pascal Bleser <guru@unixtech.be> has fio RPMs in his repository, you
27 http://linux01.gwdg.de/~pbleser/rpm-navigation.php?cat=System/fio
33 Just type 'make' and 'make install'. If on FreeBSD, for now you have to
34 specify the FreeBSD Makefile with -f, eg:
36 $ make -f Makefile.Freebsd && make -f Makefile.FreeBSD install
38 Likewise with OpenSolaris, use the Makefile.solaris to compile there.
39 This might change in the future if I opt for an autoconf type setup.
46 -t <sec> Runtime in seconds
47 -l Generate per-job latency logs
48 -w Generate per-job bandwidth logs
49 -o <file> Log output to file
50 -m Minimal (terse) output
52 -v Print version information and exit
54 Any parameters following the options will be assumed to be job files.
55 You can add as many as you want, each job file will be regarded as a
56 separate group and fio will stonewall it's execution.
62 Only a few options can be controlled with command line parameters,
63 generally it's a lot easier to just write a simple job file to describe
64 the workload. The job file format is in the ini style format, as it's
65 easy to read and write for the user.
67 The job file parameters are:
69 name=x Use 'x' as the identifier for this job.
70 directory=x Use 'x' as the top level directory for storing files
71 rw=x 'x' may be: read, randread, write, randwrite,
72 rw (read-write mix), randrw (read-write random mix)
73 rwmixcycle=x Base cycle for switching between read and write
75 rwmixread=x 'x' percentage of rw mix ios will be reads. If
76 rwmixwrite is also given, the last of the two will
77 be used if they don't add up to 100%.
78 rwmixwrite=x 'x' percentage of rw mix ios will be writes. See
80 rand_repeatable=x The sequence of random io blocks can be repeatable
81 across runs, if 'x' is 1.
82 size=x Set file size to x bytes (x string can include k/m/g)
83 ioengine=x 'x' may be: aio/libaio/linuxaio for Linux aio,
84 posixaio for POSIX aio, sync for regular read/write io,
85 mmap for mmap'ed io, splice for using splice/vmsplice,
86 or sgio for direct SG_IO io. The latter only works on
87 Linux on SCSI (or SCSI-like devices, such as
88 usb-storage or sata/libata driven) devices.
89 iodepth=x For async io, allow 'x' ios in flight
90 overwrite=x If 'x', layout a write file first.
91 prio=x Run io at prio X, 0-7 is the kernel allowed range
92 prioclass=x Run io at prio class X
93 bs=x Use 'x' for thread blocksize. May include k/m postfix.
94 bsrange=x-y Mix thread block sizes randomly between x and y. May
95 also include k/m postfix.
96 direct=x 1 for direct IO, 0 for buffered IO
97 thinktime=x "Think" x usec after each io
98 rate=x Throttle rate to x KiB/sec
99 ratemin=x Quit if rate of x KiB/sec can't be met
100 ratecycle=x ratemin averaged over x msecs
101 cpumask=x Only allow job to run on CPUs defined by mask.
102 fsync=x If writing, fsync after every x blocks have been written
103 startdelay=x Start this thread x seconds after startup
104 timeout=x Terminate x seconds after startup. Can include a
105 normal time suffix if not given in seconds, such as
106 'm' for minutes, 'h' for hours, and 'd' for days.
107 offset=x Start io at offset x (x string can include k/m/g)
108 invalidate=x Invalidate page cache for file prior to doing io
109 sync=x Use sync writes if x and writing
110 mem=x If x == malloc, use malloc for buffers. If x == shm,
111 use shm for buffers. If x == mmap, use anon mmap.
112 exitall When one thread quits, terminate the others
113 bwavgtime=x Average bandwidth stats over an x msec window.
114 create_serialize=x If 'x', serialize file creation.
115 create_fsync=x If 'x', run fsync() after file creation.
116 end_fsync=x If 'x', run fsync() after end-of-job.
117 loops=x Run the job 'x' number of times.
118 verify=x If 'x' == md5, use md5 for verifies. If 'x' == crc32,
119 use crc32 for verifies. md5 is 'safer', but crc32 is
120 a lot faster. Only makes sense for writing to a file.
121 stonewall Wait for preceeding jobs to end before running.
122 numjobs=x Create 'x' similar entries for this job
123 thread Use pthreads instead of forked jobs
125 zoneskip=y Zone options must be paired. If given, the job
126 will skip y bytes for every x read/written. This
127 can be used to gauge hard drive speed over the entire
128 platter, without reading everything. Both x/y can
129 include k/m/g suffix.
130 iolog=x Open and read io pattern from file 'x'. The file must
131 contain one io action per line in the following format:
133 where with rw=0/1 for read/write, and the offset
134 and length entries being in bytes.
135 write_iolog=x Write an iolog to file 'x' in the same format as iolog.
136 The iolog options are exclusive, if both given the
137 read iolog will be performed.
138 lockmem=x Lock down x amount of memory on the machine, to
139 simulate a machine with less memory available. x can
140 include k/m/g suffix.
141 nice=x Run job at given nice value.
142 exec_prerun=x Run 'x' before job io is begun.
143 exec_postrun=x Run 'x' after job io has finished.
144 ioscheduler=x Use ioscheduler 'x' for this job.
147 Examples using a job file
148 -------------------------
150 Example 1) Two random readers
152 Lets say we want to simulate two threads reading randomly from a file
153 each. They will be doing IO in 4KiB chunks, using raw (O_DIRECT) IO.
154 Since they share most parameters, we'll put those in the [global]
155 section. Job 1 will use a 128MiB file, job 2 will use a 256MiB file.
160 ioengine=sync ; regular read/write(2), the default
173 Generally the [] bracketed name specifies a file name, but the "global"
174 keyword is reserved for setting options that are inherited by each
175 subsequent job description. It's possible to have several [global]
176 sections in the job file, each one adds options that are inherited by
177 jobs defined below it. The name can also point to a block device, such
178 as /dev/sda. To run the above job file, simply do:
182 Example 2) Many random writers
184 Say we want to exercise the IO subsystem some more. We'll define 64
185 threads doing random buffered writes. We'll let each thread use async io
186 with a depth of 4 ios in flight. A job file would then look like this:
203 This will create files.[0-63] and perform the random writes to them.
205 There are endless ways to define jobs, the examples/ directory contains
209 Interpreting the output
210 -----------------------
212 fio spits out a lot of output. While running, fio will display the
213 status of the jobs created. An example of that would be:
215 Threads running: 1: [_r] [24.79% done] [eta 00h:01m:31s]
217 The characters inside the square brackets denote the current status of
218 each thread. The possible values (in typical life cycle order) are:
222 P Thread setup, but not started.
224 I Thread initialized, waiting.
225 R Running, doing sequential reads.
226 r Running, doing random reads.
227 W Running, doing sequential writes.
228 w Running, doing random writes.
229 M Running, doing mixed sequential reads/writes.
230 m Running, doing mixed random reads/writes.
231 F Running, currently waiting for fsync()
232 V Running, doing verification of written data.
233 E Thread exited, not reaped by main thread yet.
236 The other values are fairly self explanatory - number of threads
237 currently running and doing io, and the estimated completion percentage
238 and time for the running group. It's impossible to estimate runtime
239 of the following groups (if any).
241 When fio is done (or interrupted by ctrl-c), it will show the data for
242 each thread, group of threads, and disks in that order. For each data
243 direction, the output looks like:
245 Client1 (g=0): err= 0:
246 write: io= 32MiB, bw= 666KiB/s, runt= 50320msec
247 slat (msec): min= 0, max= 136, avg= 0.03, dev= 1.92
248 clat (msec): min= 0, max= 631, avg=48.50, dev=86.82
249 bw (KiB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, dev=681.68
250 cpu : usr=1.49%, sys=0.25%, ctx=7969
252 The client number is printed, along with the group id and error of that
253 thread. Below is the io statistics, here for writes. In the order listed,
256 io= Number of megabytes io performed
257 bw= Average bandwidth rate
258 runt= The runtime of that thread
259 slat= Submission latency (avg being the average, dev being the
260 standard deviation). This is the time it took to submit
261 the io. For sync io, the slat is really the completion
262 latency, since queue/complete is one operation there.
263 clat= Completion latency. Same names as slat, this denotes the
264 time from submission to completion of the io pieces. For
265 sync io, clat will usually be equal (or very close) to 0,
266 as the time from submit to complete is basically just
267 CPU time (io has already been done, see slat explanation).
268 bw= Bandwidth. Same names as the xlat stats, but also includes
269 an approximate percentage of total aggregate bandwidth
270 this thread received in this group. This last value is
271 only really useful if the threads in this group are on the
272 same disk, since they are then competing for disk access.
273 cpu= CPU usage. User and system time, along with the number
274 of context switches this thread went through.
276 After each client has been listed, the group statistics are printed. They
279 Run status group 0 (all jobs):
280 READ: io=64MiB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
281 WRITE: io=64MiB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
283 For each data direction, it prints:
285 io= Number of megabytes io performed.
286 aggrb= Aggregate bandwidth of threads in this group.
287 minb= The minimum average bandwidth a thread saw.
288 maxb= The maximum average bandwidth a thread saw.
289 mint= The smallest runtime of the threads in that group.
290 maxt= The longest runtime of the threads in that group.
292 And finally, the disk statistics are printed. They will look like this:
294 Disk stats (read/write):
295 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
297 Each value is printed for both reads and writes, with reads first. The
300 ios= Number of ios performed by all groups.
301 merge= Number of merges io the io scheduler.
302 ticks= Number of ticks we kept the disk busy.
303 io_queue= Total time spent in the disk queue.
304 util= The disk utilization. A value of 100% means we kept the disk
305 busy constantly, 50% would be a disk idling half of the time.
311 For scripted usage where you typically want to generate tables or graphs
312 of the results, fio can output the results in a comma seperated format.
313 The format is one long line of values, such as:
315 client1,0,0,936,331,2894,0,0,0.000000,0.000000,1,170,22.115385,34.290410,16,714,84.252874%,366.500000,566.417819,3496,1237,2894,0,0,0.000000,0.000000,0,246,6.671625,21.436952,0,2534,55.465300%,1406.600000,2008.044216,0.000000%,0.431928%,1109
317 Split up, the format is as follows:
319 jobname, groupid, error
321 KiB IO, bandwidth (KiB/sec), runtime (msec)
322 Submission latency: min, max, mean, deviation
323 Completion latency: min, max, mean, deviation
324 Bw: min, max, aggreate percentage of total, mean, deviation
326 KiB IO, bandwidth (KiB/sec), runtime (msec)
327 Submission latency: min, max, mean, deviation
328 Completion latency: min, max, mean, deviation
329 Bw: min, max, aggreate percentage of total, mean, deviation
330 CPU usage: user, system, context switches
336 Fio was written by Jens Axboe <axboe@suse.de> to enable flexible testing
337 of the Linux IO subsystem and schedulers. He got tired of writing
338 specific test applications to simulate a given workload, and found that
339 the existing io benchmark/test tools out there weren't flexible enough
340 to do what he wanted.
342 Jens Axboe <axboe@suse.de> 20060609