6 perf-bench - General framework for benchmark suites
11 'perf bench' [<common options>] <subsystem> <suite> [<options>]
15 This 'perf bench' command is a general framework for benchmark suites.
21 Specify amount of times to repeat the run (default 10).
26 Current available format styles are:
29 Default style. This is mainly for human reading.
31 % perf bench sched pipe # with no style specified
32 (executing 1000000 pipe operations between two tasks)
39 This simple style is friendly for automated
40 processing by scripts.
42 % perf bench --format=simple sched pipe # specified simple
50 Scheduler and IPC mechanisms.
53 System call performance (throughput).
56 Memory access performance.
59 NUMA scheduling and MM benchmarks.
62 Futex stressing benchmarks.
65 Eventpoll (epoll) stressing benchmarks.
68 Benchmark internal perf functionality.
71 All benchmark subsystems.
76 Suite for evaluating performance of scheduler and IPC mechanisms.
77 Based on hackbench by Rusty Russell.
79 Options of *messaging*
80 ^^^^^^^^^^^^^^^^^^^^^^
83 Use pipe() instead of socketpair()
87 Be multi thread instead of multi process
91 Specify number of groups
95 Specify number of loops
97 Example of *messaging*
98 ^^^^^^^^^^^^^^^^^^^^^^
100 ---------------------
101 % perf bench sched messaging # run with default
102 options (20 sender and receiver processes per group)
103 (10 groups == 400 processes run)
107 % perf bench sched messaging -t -g 20 # be multi-thread, with 20 groups
108 (20 sender and receiver threads per group)
109 (20 groups == 800 threads run)
112 ---------------------
115 Suite for pipe() system call.
116 Based on pipe-test-1m.c by Ingo Molnar.
122 Specify number of loops.
127 ---------------------
128 % perf bench sched pipe
129 (executing 1000000 pipe operations between two tasks)
135 % perf bench sched pipe -l 1000 # loop 1000
136 (executing 1000 pipe operations between two tasks)
141 ---------------------
146 Suite for evaluating performance of core system call throughput (both usecs/op and ops/sec metrics).
147 This uses a single thread simply doing getppid(2), which is a simple syscall where the result is not
154 Suite for evaluating performance of simple memory copy in various ways.
160 Specify size of memory to copy (default: 1MB).
161 Available units are B, KB, MB, GB and TB (case insensitive).
165 Specify function to copy (default: default).
166 Available functions are depend on the architecture.
167 On x86-64, x86-64-unrolled, x86-64-movsq and x86-64-movsb are supported.
171 Repeat memcpy invocation this number of times.
175 Use perf's cpu-cycles event instead of gettimeofday syscall.
178 Suite for evaluating performance of simple memory set in various ways.
184 Specify size of memory to set (default: 1MB).
185 Available units are B, KB, MB, GB and TB (case insensitive).
189 Specify function to set (default: default).
190 Available functions are depend on the architecture.
191 On x86-64, x86-64-unrolled, x86-64-stosq and x86-64-stosb are supported.
195 Repeat memset invocation this number of times.
199 Use perf's cpu-cycles event instead of gettimeofday syscall.
204 Suite for evaluating NUMA workloads.
209 Suite for evaluating hash tables.
212 Suite for evaluating wake calls.
215 Suite for evaluating parallel wake calls.
218 Suite for evaluating requeue calls.
221 Suite for evaluating futex lock_pi calls.
226 Suite for evaluating concurrent epoll_wait calls.
229 Suite for evaluating multiple epoll_ctl calls.
231 SUITES FOR 'internals'
232 ~~~~~~~~~~~~~~~~~~~~~~
234 Suite for evaluating perf's event synthesis performance.