diff options
author | Lars Wirzenius <liw@liw.fi> | 2011-08-05 17:30:54 +0100 |
---|---|---|
committer | Lars Wirzenius <liw@liw.fi> | 2011-08-05 17:30:54 +0100 |
commit | 3d9bd7bd336e375c6e11047df072bc4269bfae2a (patch) | |
tree | 177c03be0ecbca7b4ba447eaee8cf860685d6cc5 /README | |
parent | a8f133337486f6b98d57aa15c0cc84e2e00622ef (diff) | |
download | obnam-3d9bd7bd336e375c6e11047df072bc4269bfae2a.tar.gz |
Update README on running benchmarks and looking at results.
Diffstat (limited to 'README')
-rw-r--r-- | README | 28 |
1 files changed, 23 insertions, 5 deletions
@@ -34,8 +34,9 @@ and tools, which you can get from: * <http://liw.fi/larch/> * <http://liw.fi/ttystatus/> -* <http://liw.fi/coverage-test-runner/> +* <http://liw.fi/coverage-test-runner/> (for automatic tests) * <http://liw.fi/tracing/> +* <http://liw.fi/seivot/> (for benchmarks) You also need third party libraries: @@ -79,11 +80,28 @@ To run automatic tests: You need my CoverageTestRunner to run tests, see above for where to get it. A couple of scripts exist to run benchmarks and profiles: - ./run-benchmark - viewprof obnam.prof cumulative | less -S + ./metadata-speed 10000 + ./obnam-benchmark --size=1m/100k --results /tmp/benchmark-results + viewprof /tmp/benchmark-results/*/*backup-0.prof + seivots-summary /tmp/benchmark-results/*/*.seivot | less -S -viewprof is a little helper script I wrote, around the Python pstats module. -You can use your own, or get mine from extrautils (see above). +There are two kinds of results: Python profiling output, and `.seivot` +files. + +For the former, `viewprof` is a little helper script I wrote, +around the Python pstats module. +You can use your own, or get mine from extrautils +(<http://liw.fi/extrautils/>). Running the benchmarks under profiling +makes them a little slower (typically around 10% for me, when I've +compared), but that's OK: the absolute numbers of the benchmarks are +less important than the relative ones. It's nice to be able to look at +the profiler output, if a benchmark is surprisingly slow, without +having to re-run it. + +`seivots-summary` is a tool to display summaries of the measurements +made during a benchmark run. `seivot` is the tool that makes the +measurements. I typically save a number of benchmark results, so that +I can see how my changes affect performance over time. If you make any changes, I welcome patches, either as plain diffs, bzr bundles, or public repositories I can merge from. |