xref: /linux/tools/perf/Documentation/perf-bench.txt (revision 2a4b51666af8bf0b67ccc2e53120bad27351917c)
1perf-bench(1)
2=============
3
4NAME
5----
6perf-bench - General framework for benchmark suites
7
8SYNOPSIS
9--------
10[verse]
11'perf bench' [<common options>] <subsystem> <suite> [<options>]
12
13DESCRIPTION
14-----------
15This 'perf bench' command is a general framework for benchmark suites.
16
17COMMON OPTIONS
18--------------
19-r::
20--repeat=::
21Specify amount of times to repeat the run (default 10).
22
23-f::
24--format=::
25Specify format style.
26Current available format styles are:
27
28'default'::
29Default style. This is mainly for human reading.
30---------------------
31% perf bench sched pipe                      # with no style specified
32(executing 1000000 pipe operations between two tasks)
33        Total time:5.855 sec
34                5.855061 usecs/op
35		170792 ops/sec
36---------------------
37
38'simple'::
39This simple style is friendly for automated
40processing by scripts.
41---------------------
42% perf bench --format=simple sched pipe      # specified simple
435.988
44---------------------
45
46SUBSYSTEM
47---------
48
49'sched'::
50	Scheduler and IPC mechanisms.
51
52'mem'::
53	Memory access performance.
54
55'numa'::
56	NUMA scheduling and MM benchmarks.
57
58'futex'::
59	Futex stressing benchmarks.
60
61'epoll'::
62	Eventpoll (epoll) stressing benchmarks.
63
64'internals'::
65	Benchmark internal perf functionality.
66
67'all'::
68	All benchmark subsystems.
69
70SUITES FOR 'sched'
71~~~~~~~~~~~~~~~~~~
72*messaging*::
73Suite for evaluating performance of scheduler and IPC mechanisms.
74Based on hackbench by Rusty Russell.
75
76Options of *messaging*
77^^^^^^^^^^^^^^^^^^^^^^
78-p::
79--pipe::
80Use pipe() instead of socketpair()
81
82-t::
83--thread::
84Be multi thread instead of multi process
85
86-g::
87--group=::
88Specify number of groups
89
90-l::
91--nr_loops=::
92Specify number of loops
93
94Example of *messaging*
95^^^^^^^^^^^^^^^^^^^^^^
96
97---------------------
98% perf bench sched messaging                 # run with default
99options (20 sender and receiver processes per group)
100(10 groups == 400 processes run)
101
102      Total time:0.308 sec
103
104% perf bench sched messaging -t -g 20        # be multi-thread, with 20 groups
105(20 sender and receiver threads per group)
106(20 groups == 800 threads run)
107
108      Total time:0.582 sec
109---------------------
110
111*pipe*::
112Suite for pipe() system call.
113Based on pipe-test-1m.c by Ingo Molnar.
114
115Options of *pipe*
116^^^^^^^^^^^^^^^^^
117-l::
118--loop=::
119Specify number of loops.
120
121Example of *pipe*
122^^^^^^^^^^^^^^^^^
123
124---------------------
125% perf bench sched pipe
126(executing 1000000 pipe operations between two tasks)
127
128        Total time:8.091 sec
129                8.091833 usecs/op
130                123581 ops/sec
131
132% perf bench sched pipe -l 1000              # loop 1000
133(executing 1000 pipe operations between two tasks)
134
135        Total time:0.016 sec
136                16.948000 usecs/op
137                59004 ops/sec
138---------------------
139
140SUITES FOR 'mem'
141~~~~~~~~~~~~~~~~
142*memcpy*::
143Suite for evaluating performance of simple memory copy in various ways.
144
145Options of *memcpy*
146^^^^^^^^^^^^^^^^^^^
147-l::
148--size::
149Specify size of memory to copy (default: 1MB).
150Available units are B, KB, MB, GB and TB (case insensitive).
151
152-f::
153--function::
154Specify function to copy (default: default).
155Available functions are depend on the architecture.
156On x86-64, x86-64-unrolled, x86-64-movsq and x86-64-movsb are supported.
157
158-l::
159--nr_loops::
160Repeat memcpy invocation this number of times.
161
162-c::
163--cycles::
164Use perf's cpu-cycles event instead of gettimeofday syscall.
165
166*memset*::
167Suite for evaluating performance of simple memory set in various ways.
168
169Options of *memset*
170^^^^^^^^^^^^^^^^^^^
171-l::
172--size::
173Specify size of memory to set (default: 1MB).
174Available units are B, KB, MB, GB and TB (case insensitive).
175
176-f::
177--function::
178Specify function to set (default: default).
179Available functions are depend on the architecture.
180On x86-64, x86-64-unrolled, x86-64-stosq and x86-64-stosb are supported.
181
182-l::
183--nr_loops::
184Repeat memset invocation this number of times.
185
186-c::
187--cycles::
188Use perf's cpu-cycles event instead of gettimeofday syscall.
189
190SUITES FOR 'numa'
191~~~~~~~~~~~~~~~~~
192*mem*::
193Suite for evaluating NUMA workloads.
194
195SUITES FOR 'futex'
196~~~~~~~~~~~~~~~~~~
197*hash*::
198Suite for evaluating hash tables.
199
200*wake*::
201Suite for evaluating wake calls.
202
203*wake-parallel*::
204Suite for evaluating parallel wake calls.
205
206*requeue*::
207Suite for evaluating requeue calls.
208
209*lock-pi*::
210Suite for evaluating futex lock_pi calls.
211
212SUITES FOR 'epoll'
213~~~~~~~~~~~~~~~~~~
214*wait*::
215Suite for evaluating concurrent epoll_wait calls.
216
217*ctl*::
218Suite for evaluating multiple epoll_ctl calls.
219
220SUITES FOR 'internals'
221~~~~~~~~~~~~~~~~~~~~~~
222*synthesize*::
223Suite for evaluating perf's event synthesis performance.
224
225SEE ALSO
226--------
227linkperf:perf[1]
228