diff options
author | hunt <hunt> | 2005-06-28 21:20:10 +0000 |
---|---|---|
committer | hunt <hunt> | 2005-06-28 21:20:10 +0000 |
commit | 2a511e91a8631589c62ba95a3974cc5ae4f7530f (patch) | |
tree | 99a95a4500491fda25011560aa58af78fca9d7e1 /runtime | |
parent | 4d9dc675488169e9605222f969a32108df8f29b8 (diff) | |
download | systemtap-steved-2a511e91a8631589c62ba95a3974cc5ae4f7530f.tar.gz systemtap-steved-2a511e91a8631589c62ba95a3974cc5ae4f7530f.tar.xz systemtap-steved-2a511e91a8631589c62ba95a3974cc5ae4f7530f.zip |
*** empty log message ***
Diffstat (limited to 'runtime')
-rw-r--r-- | runtime/probes/bench/README | 22 |
1 files changed, 22 insertions, 0 deletions
diff --git a/runtime/probes/bench/README b/runtime/probes/bench/README new file mode 100644 index 00000000..5eaded78 --- /dev/null +++ b/runtime/probes/bench/README @@ -0,0 +1,22 @@ +This is a benchmark program for the SystemTap Runtime. + +It works by instrumenting sys_read() and sys_write(). It calls each a million +times and times that. Then it puts an empty kprobe on one and jprobe on the +other and times that. Subtracting the difference between the two runs gives +the kprobe and jprobe overhead. The process is then repeated for more +complicated probes. + +Any system has some background activity going on that will generate sys_read +and sys_write calls. However, unless some major activity is going on, it +should be statistically insignificant. + +To Start: + +1. Build the test program. +> gcc -O3 -o time time.c + +2. Run the benchmarks +> ./run_bench + + + |