[Chandler-dev] Performance work update
heikki at osafoundation.org
Fri Feb 16 12:41:01 PST 2007
Darn it, I wrote a long email but accidentally deleted without a trace...
Yesterday I checked in a change to start measuring the second time of
doing an action for the following tests: Switch to All View, Switch to
All View with 3k, Jump week with 3k, Overlay calendar with 3k, Switch
calendar with 3k. With these tests we now cover the slowest case (things
loaded from disk, constructing disk and in-memory caches) and the
fastest case (everything in memory, using all caches).
Previously calendar scrolling test was changed to scroll multiple units
instead of one, so historical data shows this as a big jump although
there was no real change in speed.
I also wrote a busy developers guide to Chandler performance
optimization. Feedback welcome. Doc at
Performance changes over the last 30 days, mostly regressions:
* New event regressed slightly
* Importing speed almost doubled
* Switching to All View about doubled
* Stamping regressed slightly
* Scrolling table on Windows got about 9 times slower
* Jumping a week regressed slightly
* Switching calendars regressed slightly
* Scrolling calendar speed improved by about 20%
* Resizing on Linux improved by about 50%
* Publish and Subscribe improves considerably when switching to Cosmo
0.6 (there is still some question as to how much switching away from SSL
affected this - still to be tested)
There have been reports that switching to mvcc (repository stuff)
improved performance considerably in some cases. However, our
performance tests don't seem to hit those scenarios.
Andi has also noticed considerable startup slowdown with certain
calendars - again this is not noticed in our tests.
The trend graphs are available at
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 254 bytes
Desc: OpenPGP digital signature
Url : http://lists.osafoundation.org/pipermail/chandler-dev/attachments/20070216/b2578a11/signature.pgp
More information about the chandler-dev