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/
28 -b block size in KiB for each io
29 -t <sec> Runtime in seconds
30 -r For random io, sequence must be repeatable
31 -R <on> If one thread fails to meet rate, quit all
32 -o <on> Use direct IO is 1, buffered if 0
33 -l Generate per-job latency logs
34 -w Generate per-job bandwidth logs
35 -f <file> Read <file> for job descriptions
37 -v Print version information and exit
39 The <jobs> format is as follows:
41 directory=x Use 'x' as the top level directory for storing files
42 rw=x 'x' may be: read, randread, write, randwrite,
43 rw (read-write mix), randrw (read-write random mix)
44 size=x Set file size to x bytes (x string can include k/m/g)
45 ioengine=x 'x' may be: aio/libaio/linuxaio for Linux aio,
46 posixaio for POSIX aio, sync for regular read/write io,
47 mmap for mmap'ed io, splice for using splice/vmsplice,
48 or sgio for direct SG_IO io. The latter only works on
49 Linux on SCSI (or SCSI-like devices, such as
50 usb-storage or sata/libata driven) devices.
51 iodepth=x For async io, allow 'x' ios in flight
52 overwrite=x If 'x', layout a write file first.
53 prio=x Run io at prio X, 0-7 is the kernel allowed range
54 prioclass=x Run io at prio class X
55 bs=x Use 'x' for thread blocksize. May include k/m postfix.
56 bsrange=x-y Mix thread block sizes randomly between x and y. May
57 also include k/m postfix.
58 direct=x 1 for direct IO, 0 for buffered IO
59 thinktime=x "Think" x usec after each io
60 rate=x Throttle rate to x KiB/sec
61 ratemin=x Quit if rate of x KiB/sec can't be met
62 ratecycle=x ratemin averaged over x msecs
63 cpumask=x Only allow job to run on CPUs defined by mask.
64 fsync=x If writing, fsync after every x blocks have been written
65 startdelay=x Start this thread x seconds after startup
66 timeout=x Terminate x seconds after startup
67 offset=x Start io at offset x (x string can include k/m/g)
68 invalidate=x Invalidate page cache for file prior to doing io
69 sync=x Use sync writes if x and writing
70 mem=x If x == malloc, use malloc for buffers. If x == shm,
71 use shm for buffers. If x == mmap, use anon mmap.
72 exitall When one thread quits, terminate the others
73 bwavgtime=x Average bandwidth stats over an x msec window.
74 create_serialize=x If 'x', serialize file creation.
75 create_fsync=x If 'x', run fsync() after file creation.
76 end_fsync=x If 'x', run fsync() after end-of-job.
77 loops=x Run the job 'x' number of times.
78 verify=x If 'x' == md5, use md5 for verifies. If 'x' == crc32,
79 use crc32 for verifies. md5 is 'safer', but crc32 is
80 a lot faster. Only makes sense for writing to a file.
81 stonewall Wait for preceeding jobs to end before running.
82 numjobs=x Create 'x' similar entries for this job
83 thread Use pthreads instead of forked jobs
85 zoneskip=y Zone options must be paired. If given, the job
86 will skip y bytes for every x read/written. This
87 can be used to gauge hard drive speed over the entire
88 platter, without reading everything. Both x/y can
90 iolog=x Open and read io pattern from file 'x'. The file must
91 contain one io action per line in the following format:
93 where with rw=0/1 for read/write, and the offset
94 and length entries being in bytes.
95 lockmem=x Lock down x amount of memory on the machine, to
96 simulate a machine with less memory available. x can
99 Examples using a job file
100 -------------------------
102 A sample job file doing the same as above would look like this:
112 And fio would be invoked as:
114 $ fio -o1 -s -f file_with_above
116 The second example would look like this:
131 And fio would be invoked as:
133 $ fio -o0 -s -b4096 -f file_with_above
135 'global' is a reserved keyword. When used as the filename, it sets the
136 default options for the threads following that section. It is possible
137 to have more than one global section in the file, as it only affects
140 Also see the examples/ dir for sample job files.
143 Interpreting the output
144 -----------------------
146 fio spits out a lot of output. While running, fio will display the
147 status of the jobs created. An example of that would be:
149 Threads now running: 2 : [ww] [5.73% done]
151 The characters inside the square brackets denote the current status of
152 each thread. The possible values (in typical life cycle order) are:
156 P Thread setup, but not started.
157 C Thread created and running, but not doing anything yet
158 R Running, doing sequential reads.
159 r Running, doing random reads.
160 W Running, doing sequential writes.
161 w Running, doing random writes.
162 V Running, doing verification of written data.
163 E Thread exited, not reaped by main thread yet.
166 The other values are fairly self explanatory - number of thread currently
167 running and doing io, and the estimated completion percentage.
169 When fio is done (or interrupted by ctrl-c), it will show the data for
170 each thread, group of threads, and disks in that order. For each data
171 direction, the output looks like:
173 Client1 (g=0): err= 0:
174 write: io= 32MiB, bw= 666KiB/s, runt= 50320msec
175 slat (msec): min= 0, max= 136, avg= 0.03, dev= 1.92
176 clat (msec): min= 0, max= 631, avg=48.50, dev=86.82
177 bw (KiB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, dev=681.68
178 cpu : usr=1.49%, sys=0.25%, ctx=7969
180 The client number is printed, along with the group id and error of that
181 thread. Below is the io statistics, here for writes. In the order listed,
184 io= Number of megabytes io performed
185 bw= Average bandwidth rate
186 runt= The runtime of that thread
187 slat= Submission latency (avg being the average, dev being the
188 standard deviation). This is the time it took to submit
189 the io. For sync io, the slat is really the completion
190 latency, since queue/complete is one operation there.
191 clat= Completion latency. Same names as slat, this denotes the
192 time from submission to completion of the io pieces. For
193 sync io, clat will usually be equal (or very close) to 0,
194 as the time from submit to complete is basically just
195 CPU time (io has already been done, see slat explanation).
196 bw= Bandwidth. Same names as the xlat stats, but also includes
197 an approximate percentage of total aggregate bandwidth
198 this thread received in this group. This last value is
199 only really useful if the threads in this group are on the
200 same disk, since they are then competing for disk access.
201 cpu= CPU usage. User and system time, along with the number
202 of context switches this thread went through.
204 After each client has been listed, the group statistics are printed. They
207 Run status group 0 (all jobs):
208 READ: io=64MiB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
209 WRITE: io=64MiB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec
211 For each data direction, it prints:
213 io= Number of megabytes io performed.
214 aggrb= Aggregate bandwidth of threads in this group.
215 minb= The minimum average bandwidth a thread saw.
216 maxb= The maximum average bandwidth a thread saw.
217 mint= The minimum runtime of a thread.
218 maxt= The maximum runtime of a thread.
220 And finally, the disk statistics are printed. They will look like this:
222 Disk stats (read/write):
223 sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%
225 Each value is printed for both reads and writes, with reads first. The
228 ios= Number of ios performed by all groups.
229 merge= Number of merges io the io scheduler.
230 ticks= Number of ticks we kept the disk busy.
231 io_queue= Total time spent in the disk queue.
232 util= The disk utilization. A value of 100% means we kept the disk
233 busy constantly, 50% would be a disk idling half of the time.