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 -f <file> Read <file> for job descriptions
50 -o <file> Log output to file
51 -m Minimal (terse) output
53 -v Print version information and exit
55 Any parameters following the options will be assumed to be job files.
56 You can add as many as you want, each job file will be regarded as a
57 separate group and fio will stonewall it's execution.
63 Only a few options can be controlled with command line parameters,
64 generally it's a lot easier to just write a simple job file to describe
65 the workload. The job file format is in the ini style format, as it's
66 easy to read and write for the user.
68 The job file parameters are:
70 name=x Use 'x' as the identifier for this job.
71 directory=x Use 'x' as the top level directory for storing files
72 rw=x 'x' may be: read, randread, write, randwrite,
73 rw (read-write mix), randrw (read-write random mix)
74 rwmixcycle=x Base cycle for switching between read and write
76 rwmixread=x 'x' percentage of rw mix ios will be reads. If
77 rwmixwrite is also given, the last of the two will
78 be used if they don't add up to 100%.
79 rwmixwrite=x 'x' percentage of rw mix ios will be writes. See
81 rand_repeatable=x The sequence of random io blocks can be repeatable
82 across runs, if 'x' is 1.
83 size=x Set file size to x bytes (x string can include k/m/g)
84 ioengine=x 'x' may be: aio/libaio/linuxaio for Linux aio,
85 posixaio for POSIX aio, sync for regular read/write io,
86 mmap for mmap'ed io, splice for using splice/vmsplice,
87 or sgio for direct SG_IO io. The latter only works on
88 Linux on SCSI (or SCSI-like devices, such as
89 usb-storage or sata/libata driven) devices.
90 iodepth=x For async io, allow 'x' ios in flight
91 overwrite=x If 'x', layout a write file first.
92 prio=x Run io at prio X, 0-7 is the kernel allowed range
93 prioclass=x Run io at prio class X
94 bs=x Use 'x' for thread blocksize. May include k/m postfix.
95 bsrange=x-y Mix thread block sizes randomly between x and y. May
96 also include k/m postfix.
97 direct=x 1 for direct IO, 0 for buffered IO
98 thinktime=x "Think" x usec after each io
99 rate=x Throttle rate to x KiB/sec
100 ratemin=x Quit if rate of x KiB/sec can't be met
101 ratecycle=x ratemin averaged over x msecs
102 cpumask=x Only allow job to run on CPUs defined by mask.
103 fsync=x If writing, fsync after every x blocks have been written
104 startdelay=x Start this thread x seconds after startup
105 timeout=x Terminate x seconds after startup
106 offset=x Start io at offset x (x string can include k/m/g)
107 invalidate=x Invalidate page cache for file prior to doing io
108 sync=x Use sync writes if x and writing
109 mem=x If x == malloc, use malloc for buffers. If x == shm,
110 use shm for buffers. If x == mmap, use anon mmap.
111 exitall When one thread quits, terminate the others
112 bwavgtime=x Average bandwidth stats over an x msec window.
113 create_serialize=x If 'x', serialize file creation.
114 create_fsync=x If 'x', run fsync() after file creation.
115 end_fsync=x If 'x', run fsync() after end-of-job.
116 loops=x Run the job 'x' number of times.
117 verify=x If 'x' == md5, use md5 for verifies. If 'x' == crc32,
118 use crc32 for verifies. md5 is 'safer', but crc32 is
119 a lot faster. Only makes sense for writing to a file.
120 stonewall Wait for preceeding jobs to end before running.
121 numjobs=x Create 'x' similar entries for this job
122 thread Use pthreads instead of forked jobs
124 zoneskip=y Zone options must be paired. If given, the job
125 will skip y bytes for every x read/written. This
126 can be used to gauge hard drive speed over the entire
127 platter, without reading everything. Both x/y can
128 include k/m/g suffix.
129 iolog=x Open and read io pattern from file 'x'. The file must
130 contain one io action per line in the following format:
132 where with rw=0/1 for read/write, and the offset
133 and length entries being in bytes.
134 write_iolog=x Write an iolog to file 'x' in the same format as iolog.
135 The iolog options are exclusive, if both given the
136 read iolog will be performed.
137 lockmem=x Lock down x amount of memory on the machine, to
138 simulate a machine with less memory available. x can
139 include k/m/g suffix.
140 nice=x Run job at given nice value.
141 exec_prerun=x Run 'x' before job io is begun.
142 exec_postrun=x Run 'x' after job io has finished.
143 ioscheduler=x Use ioscheduler 'x' for this job.
146 Examples using a job file
147 -------------------------
149 Example 1) Two random readers
151 Lets say we want to simulate two threads reading randomly from a file
152 each. They will be doing IO in 4KiB chunks, using raw (O_DIRECT) IO.
153 Since they share most parameters, we'll put those in the [global]
154 section. Job 1 will use a 128MiB file, job 2 will use a 256MiB file.
159 ioengine=sync ; regular read/write(2), the default
172 Generally the [] bracketed name specifies a file name, but the "global"
173 keyword is reserved for setting options that are inherited by each
174 subsequent job description. It's possible to have several [global]
175 sections in the job file, each one adds options that are inherited by
176 jobs defined below it. The name can also point to a block device, such
177 as /dev/sda. To run the above job file, simply do:
181 Example 2) Many random writers
183 Say we want to exercise the IO subsystem some more. We'll define 64
184 threads doing random buffered writes. We'll let each thread use async io
185 with a depth of 4 ios in flight. A job file would then look like this:
202 This will create files.[0-63] and perform the random writes to them.
204 There are endless ways to define jobs, the examples/ directory contains
208 Interpreting the output
209 -----------------------
211 fio spits out a lot of output. While running, fio will display the
212 status of the jobs created. An example of that would be:
214 Threads running: 1: [_r] [24.79% done] [eta 00h:01m:31s]
216 The characters inside the square brackets denote the current status of
217 each thread. The possible values (in typical life cycle order) are:
221 P Thread setup, but not started.
223 I Thread initialized, waiting.
224 R Running, doing sequential reads.
225 r Running, doing random reads.
226 W Running, doing sequential writes.
227 w Running, doing random writes.
228 M Running, doing mixed sequential reads/writes.
229 m Running, doing mixed random reads/writes.
230 F Running, currently waiting for fsync()
231 V Running, doing verification of written data.
232 E Thread exited, not reaped by main thread yet.
235 The other values are fairly self explanatory - number of threads
236 currently running and doing io, and the estimated completion percentage
237 and time for the running group. It's impossible to estimate runtime
238 of the following groups (if any).
240 When fio is done (or interrupted by ctrl-c), it will show the data for
241 each thread, group of threads, and disks in that order. For each data
242 direction, the output looks like:
244 Client1 (g=0): err= 0:
245 write: io= 32MiB, bw= 666KiB/s, runt= 50320msec
246 slat (msec): min= 0, max= 136, avg= 0.03, dev= 1.92
247 clat (msec): min= 0, max= 631, avg=48.50, dev=86.82
248 bw (KiB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, dev=681.68
249 cpu : usr=1.49%, sys=0.25%, ctx=7969
251 The client number is printed, along with the group id and error of that
252 thread. Below is the io statistics, here for writes. In the order listed,
255 io= Number of megabytes io performed
256 bw= Average bandwidth rate
257 runt= The runtime of that thread
258 slat= Submission latency (avg being the average, dev being the
259 standard deviation). This is the time it took to submit
260 the io. For sync io, the slat is really the completion
261 latency, since queue/complete is one operation there.
262 clat= Completion latency. Same names as slat, this denotes the
263 time from submission to completion of the io pieces. For
264 sync io, clat will usually be equal (or very close) to 0,
265 as the time from submit to complete is basically just
266 CPU time (io has already been done, see slat explanation).
267 bw= Bandwidth. Same names as the xlat stats, but also includes
268 an approximate percentage of total aggregate bandwidth
269 this thread received in this group. This last value is
270 only really useful if the threads in this group are on the
271 same disk, since they are then competing for disk access.
272 cpu= CPU usage. User and system time, along with the number
273 of context switches this thread went through.
275 After each client has been listed, the group statistics are printed. They
278 Run status group 0 (all jobs):
279 READ: io=64MiB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
280 WRITE: io=64MiB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
282 For each data direction, it prints:
284 io= Number of megabytes io performed.
285 aggrb= Aggregate bandwidth of threads in this group.
286 minb= The minimum average bandwidth a thread saw.
287 maxb= The maximum average bandwidth a thread saw.
288 mint= The smallest runtime of the threads in that group.
289 maxt= The longest runtime of the threads in that group.
291 And finally, the disk statistics are printed. They will look like this:
293 Disk stats (read/write):
294 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
296 Each value is printed for both reads and writes, with reads first. The
299 ios= Number of ios performed by all groups.
300 merge= Number of merges io the io scheduler.
301 ticks= Number of ticks we kept the disk busy.
302 io_queue= Total time spent in the disk queue.
303 util= The disk utilization. A value of 100% means we kept the disk
304 busy constantly, 50% would be a disk idling half of the time.
310 For scripted usage where you typically want to generate tables or graphs
311 of the results, fio can output the results in a comma seperated format.
312 The format is one long line of values, such as:
314 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
316 Split up, the format is as follows:
318 jobname, groupid, error
320 KiB IO, bandwidth (KiB/sec), runtime (msec)
321 Submission latency: min, max, mean, deviation
322 Completion latency: min, max, mean, deviation
323 Bw: min, max, aggreate percentage of total, mean, deviation
325 KiB IO, bandwidth (KiB/sec), runtime (msec)
326 Submission latency: min, max, mean, deviation
327 Completion latency: min, max, mean, deviation
328 Bw: min, max, aggreate percentage of total, mean, deviation
329 CPU usage: user, system, context switches
335 Fio was written by Jens Axboe <axboe@suse.de> to enable flexible testing
336 of the Linux IO subsystem and schedulers. He got tired of writing
337 specific test applications to simulate a given workload, and found that
338 the existing io benchmark/test tools out there weren't flexible enough
339 to do what he wanted.
341 Jens Axboe <axboe@suse.de> 20060609