4 fio is a tool that will spawn a number of thread doing a particular
5 type of io action as specified by the user. fio takes a number of
6 global parameters, each inherited by the thread unless otherwise
7 parameters given to them overriding that setting is given.
13 fio resides in a git repo, the canonical place is:
15 git://brick.kernel.dk/data/git/fio.git
17 Snapshots are frequently generated as well and they include the git
18 meta data as well. You can download them here:
20 http://brick.kernel.dk/snaps/
22 Pascal Bleser <guru@unixtech.be> has fio RPMs in his repository, you
25 http://linux01.gwdg.de/~pbleser/rpm-navigation.php?cat=System/fio
31 Just type 'make' and 'make install'. If on FreeBSD, for now you have to
32 specify the FreeBSD Makefile with -f, eg:
34 $ make -f Makefile.Freebsd && make -f Makefile.FreeBSD install
36 Likewise with OpenSolaris, use the Makefile.solaris to compile there.
37 This might change in the future if I opt for an autoconf type setup.
45 -b block size in KiB for each io
46 -t <sec> Runtime in seconds
47 -r For random io, sequence must be repeatable
48 -R <on> If one thread fails to meet rate, quit all
49 -o <on> Use direct IO is 1, buffered if 0
50 -l Generate per-job latency logs
51 -w Generate per-job bandwidth logs
52 -f <file> Read <file> for job descriptions
54 -v Print version information and exit
56 The <jobs> format is as follows:
58 name=x Use 'x' as the identifier for this job.
59 directory=x Use 'x' as the top level directory for storing files
60 rw=x 'x' may be: read, randread, write, randwrite,
61 rw (read-write mix), randrw (read-write random mix)
62 rwmixcycle=x Base cycle for switching between read and write
64 rwmixread=x 'x' percentage of rw mix ios will be reads. If
65 rwmixwrite is also given, the last of the two will
66 be used if they don't add up to 100%.
67 rwmixwrite=x 'x' percentage of rw mix ios will be writes. See
69 size=x Set file size to x bytes (x string can include k/m/g)
70 ioengine=x 'x' may be: aio/libaio/linuxaio for Linux aio,
71 posixaio for POSIX aio, sync for regular read/write io,
72 mmap for mmap'ed io, splice for using splice/vmsplice,
73 or sgio for direct SG_IO io. The latter only works on
74 Linux on SCSI (or SCSI-like devices, such as
75 usb-storage or sata/libata driven) devices.
76 iodepth=x For async io, allow 'x' ios in flight
77 overwrite=x If 'x', layout a write file first.
78 prio=x Run io at prio X, 0-7 is the kernel allowed range
79 prioclass=x Run io at prio class X
80 bs=x Use 'x' for thread blocksize. May include k/m postfix.
81 bsrange=x-y Mix thread block sizes randomly between x and y. May
82 also include k/m postfix.
83 direct=x 1 for direct IO, 0 for buffered IO
84 thinktime=x "Think" x usec after each io
85 rate=x Throttle rate to x KiB/sec
86 ratemin=x Quit if rate of x KiB/sec can't be met
87 ratecycle=x ratemin averaged over x msecs
88 cpumask=x Only allow job to run on CPUs defined by mask.
89 fsync=x If writing, fsync after every x blocks have been written
90 startdelay=x Start this thread x seconds after startup
91 timeout=x Terminate x seconds after startup
92 offset=x Start io at offset x (x string can include k/m/g)
93 invalidate=x Invalidate page cache for file prior to doing io
94 sync=x Use sync writes if x and writing
95 mem=x If x == malloc, use malloc for buffers. If x == shm,
96 use shm for buffers. If x == mmap, use anon mmap.
97 exitall When one thread quits, terminate the others
98 bwavgtime=x Average bandwidth stats over an x msec window.
99 create_serialize=x If 'x', serialize file creation.
100 create_fsync=x If 'x', run fsync() after file creation.
101 end_fsync=x If 'x', run fsync() after end-of-job.
102 loops=x Run the job 'x' number of times.
103 verify=x If 'x' == md5, use md5 for verifies. If 'x' == crc32,
104 use crc32 for verifies. md5 is 'safer', but crc32 is
105 a lot faster. Only makes sense for writing to a file.
106 stonewall Wait for preceeding jobs to end before running.
107 numjobs=x Create 'x' similar entries for this job
108 thread Use pthreads instead of forked jobs
110 zoneskip=y Zone options must be paired. If given, the job
111 will skip y bytes for every x read/written. This
112 can be used to gauge hard drive speed over the entire
113 platter, without reading everything. Both x/y can
114 include k/m/g suffix.
115 iolog=x Open and read io pattern from file 'x'. The file must
116 contain one io action per line in the following format:
118 where with rw=0/1 for read/write, and the offset
119 and length entries being in bytes.
120 write_iolog=x Write an iolog to file 'x' in the same format as iolog.
121 The iolog options are exclusive, if both given the
122 read iolog will be performed.
123 lockmem=x Lock down x amount of memory on the machine, to
124 simulate a machine with less memory available. x can
125 include k/m/g suffix.
126 nice=x Run job at given nice value.
127 exec_prerun=x Run 'x' before job io is begun.
128 exec_postrun=x Run 'x' after job io has finished.
129 ioscheduler=x Use ioscheduler 'x' for this job.
131 Examples using a job file
132 -------------------------
134 A sample job file doing the same as above would look like this:
144 And fio would be invoked as:
146 $ fio -o1 -s -f file_with_above
148 The second example would look like this:
163 And fio would be invoked as:
165 $ fio -o0 -s -b4096 -f file_with_above
167 'global' is a reserved keyword. When used as the filename, it sets the
168 default options for the threads following that section. It is possible
169 to have more than one global section in the file, as it only affects
172 Also see the examples/ dir for sample job files.
175 Interpreting the output
176 -----------------------
178 fio spits out a lot of output. While running, fio will display the
179 status of the jobs created. An example of that would be:
181 Threads now running: 2 : [ww] [5.73% done]
183 The characters inside the square brackets denote the current status of
184 each thread. The possible values (in typical life cycle order) are:
188 P Thread setup, but not started.
189 C Thread created and running, but not doing anything yet
190 R Running, doing sequential reads.
191 r Running, doing random reads.
192 W Running, doing sequential writes.
193 w Running, doing random writes.
194 V Running, doing verification of written data.
195 E Thread exited, not reaped by main thread yet.
198 The other values are fairly self explanatory - number of thread currently
199 running and doing io, and the estimated completion percentage.
201 When fio is done (or interrupted by ctrl-c), it will show the data for
202 each thread, group of threads, and disks in that order. For each data
203 direction, the output looks like:
205 Client1 (g=0): err= 0:
206 write: io= 32MiB, bw= 666KiB/s, runt= 50320msec
207 slat (msec): min= 0, max= 136, avg= 0.03, dev= 1.92
208 clat (msec): min= 0, max= 631, avg=48.50, dev=86.82
209 bw (KiB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, dev=681.68
210 cpu : usr=1.49%, sys=0.25%, ctx=7969
212 The client number is printed, along with the group id and error of that
213 thread. Below is the io statistics, here for writes. In the order listed,
216 io= Number of megabytes io performed
217 bw= Average bandwidth rate
218 runt= The runtime of that thread
219 slat= Submission latency (avg being the average, dev being the
220 standard deviation). This is the time it took to submit
221 the io. For sync io, the slat is really the completion
222 latency, since queue/complete is one operation there.
223 clat= Completion latency. Same names as slat, this denotes the
224 time from submission to completion of the io pieces. For
225 sync io, clat will usually be equal (or very close) to 0,
226 as the time from submit to complete is basically just
227 CPU time (io has already been done, see slat explanation).
228 bw= Bandwidth. Same names as the xlat stats, but also includes
229 an approximate percentage of total aggregate bandwidth
230 this thread received in this group. This last value is
231 only really useful if the threads in this group are on the
232 same disk, since they are then competing for disk access.
233 cpu= CPU usage. User and system time, along with the number
234 of context switches this thread went through.
236 After each client has been listed, the group statistics are printed. They
239 Run status group 0 (all jobs):
240 READ: io=64MiB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
241 WRITE: io=64MiB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
243 For each data direction, it prints:
245 io= Number of megabytes io performed.
246 aggrb= Aggregate bandwidth of threads in this group.
247 minb= The minimum average bandwidth a thread saw.
248 maxb= The maximum average bandwidth a thread saw.
249 mint= The minimum runtime of a thread.
250 maxt= The maximum runtime of a thread.
252 And finally, the disk statistics are printed. They will look like this:
254 Disk stats (read/write):
255 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
257 Each value is printed for both reads and writes, with reads first. The
260 ios= Number of ios performed by all groups.
261 merge= Number of merges io the io scheduler.
262 ticks= Number of ticks we kept the disk busy.
263 io_queue= Total time spent in the disk queue.
264 util= The disk utilization. A value of 100% means we kept the disk
265 busy constantly, 50% would be a disk idling half of the time.