1 perf-bench(1)
2 =============
3
4 NAME
5 ----
6 perf-bench - General framework for benchmark s
7
8 SYNOPSIS
9 --------
10 [verse]
11 'perf bench' [<common options>] <subsystem> <s
12
13 DESCRIPTION
14 -----------
15 This 'perf bench' command is a general framewo
16
17 COMMON OPTIONS
18 --------------
19 -r::
20 --repeat=::
21 Specify number of times to repeat the run (def
22
23 -f::
24 --format=::
25 Specify format style.
26 Current available format styles are:
27
28 'default'::
29 Default style. This is mainly for human readin
30 ---------------------
31 % perf bench sched pipe #
32 (executing 1000000 pipe operations between two
33 Total time:5.855 sec
34 5.855061 usecs/op
35 170792 ops/sec
36 ---------------------
37
38 'simple'::
39 This simple style is friendly for automated
40 processing by scripts.
41 ---------------------
42 % perf bench --format=simple sched pipe #
43 5.988
44 ---------------------
45
46 SUBSYSTEM
47 ---------
48
49 'sched'::
50 Scheduler and IPC mechanisms.
51
52 'syscall'::
53 System call performance (throughput).
54
55 'mem'::
56 Memory access performance.
57
58 'numa'::
59 NUMA scheduling and MM benchmarks.
60
61 'futex'::
62 Futex stressing benchmarks.
63
64 'epoll'::
65 Eventpoll (epoll) stressing benchmarks
66
67 'internals'::
68 Benchmark internal perf functionality.
69
70 'uprobe'::
71 Benchmark overhead of uprobe + BPF.
72
73 'all'::
74 All benchmark subsystems.
75
76 SUITES FOR 'sched'
77 ~~~~~~~~~~~~~~~~~~
78 *messaging*::
79 Suite for evaluating performance of scheduler
80 Based on hackbench by Rusty Russell.
81
82 Options of *messaging*
83 ^^^^^^^^^^^^^^^^^^^^^^
84 -p::
85 --pipe::
86 Use pipe() instead of socketpair()
87
88 -t::
89 --thread::
90 Be multi thread instead of multi process
91
92 -g::
93 --group=::
94 Specify number of groups
95
96 -l::
97 --nr_loops=::
98 Specify number of loops
99
100 Example of *messaging*
101 ^^^^^^^^^^^^^^^^^^^^^^
102
103 ---------------------
104 % perf bench sched messaging #
105 options (20 sender and receiver processes per
106 (10 groups == 400 processes run)
107
108 Total time:0.308 sec
109
110 % perf bench sched messaging -t -g 20 #
111 (20 sender and receiver threads per group)
112 (20 groups == 800 threads run)
113
114 Total time:0.582 sec
115 ---------------------
116
117 *pipe*::
118 Suite for pipe() system call.
119 Based on pipe-test-1m.c by Ingo Molnar.
120
121 Options of *pipe*
122 ^^^^^^^^^^^^^^^^^
123 -l::
124 --loop=::
125 Specify number of loops.
126
127 -G::
128 --cgroups=::
129 Names of cgroups for sender and receiver, sepa
130 This is useful to check cgroup context switchi
131 Note that perf doesn't create nor delete the c
132 make sure that the cgroups exist and are acces
133
134
135 Example of *pipe*
136 ^^^^^^^^^^^^^^^^^
137
138 ---------------------
139 % perf bench sched pipe
140 (executing 1000000 pipe operations between two
141
142 Total time:8.091 sec
143 8.091833 usecs/op
144 123581 ops/sec
145
146 % perf bench sched pipe -l 1000 #
147 (executing 1000 pipe operations between two ta
148
149 Total time:0.016 sec
150 16.948000 usecs/op
151 59004 ops/sec
152
153 % perf bench sched pipe -G AAA,BBB
154 (executing 1000000 pipe operations between cgr
155 # Running 'sched/pipe' benchmark:
156 # Executed 1000000 pipe operations between two
157
158 Total time: 6.886 [sec]
159
160 6.886208 usecs/op
161 145217 ops/sec
162
163 ---------------------
164
165 SUITES FOR 'syscall'
166 ~~~~~~~~~~~~~~~~~~
167 *basic*::
168 Suite for evaluating performance of core syste
169 This uses a single thread simply doing getppid
170 cached by glibc.
171
172
173 SUITES FOR 'mem'
174 ~~~~~~~~~~~~~~~~
175 *memcpy*::
176 Suite for evaluating performance of simple mem
177
178 Options of *memcpy*
179 ^^^^^^^^^^^^^^^^^^^
180 -l::
181 --size::
182 Specify size of memory to copy (default: 1MB).
183 Available units are B, KB, MB, GB and TB (case
184
185 -f::
186 --function::
187 Specify function to copy (default: default).
188 Available functions are depend on the architec
189 On x86-64, x86-64-unrolled, x86-64-movsq and x
190
191 -l::
192 --nr_loops::
193 Repeat memcpy invocation this number of times.
194
195 -c::
196 --cycles::
197 Use perf's cpu-cycles event instead of gettime
198
199 *memset*::
200 Suite for evaluating performance of simple mem
201
202 Options of *memset*
203 ^^^^^^^^^^^^^^^^^^^
204 -l::
205 --size::
206 Specify size of memory to set (default: 1MB).
207 Available units are B, KB, MB, GB and TB (case
208
209 -f::
210 --function::
211 Specify function to set (default: default).
212 Available functions are depend on the architec
213 On x86-64, x86-64-unrolled, x86-64-stosq and x
214
215 -l::
216 --nr_loops::
217 Repeat memset invocation this number of times.
218
219 -c::
220 --cycles::
221 Use perf's cpu-cycles event instead of gettime
222
223 SUITES FOR 'numa'
224 ~~~~~~~~~~~~~~~~~
225 *mem*::
226 Suite for evaluating NUMA workloads.
227
228 SUITES FOR 'futex'
229 ~~~~~~~~~~~~~~~~~~
230 *hash*::
231 Suite for evaluating hash tables.
232
233 *wake*::
234 Suite for evaluating wake calls.
235
236 *wake-parallel*::
237 Suite for evaluating parallel wake calls.
238
239 *requeue*::
240 Suite for evaluating requeue calls.
241
242 *lock-pi*::
243 Suite for evaluating futex lock_pi calls.
244
245 SUITES FOR 'epoll'
246 ~~~~~~~~~~~~~~~~~~
247 *wait*::
248 Suite for evaluating concurrent epoll_wait cal
249
250 *ctl*::
251 Suite for evaluating multiple epoll_ctl calls.
252
253 SUITES FOR 'internals'
254 ~~~~~~~~~~~~~~~~~~~~~~
255 *synthesize*::
256 Suite for evaluating perf's event synthesis pe
257
258 SEE ALSO
259 --------
260 linkperf:perf[1]
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.