Community
Participate
Working Groups
Changes to stop running tests on 1.5 described in bug 321142 should be applied to the performance baseline branch used to compare 3.7 and perf_36x stream builds.
This is a critical issue because it rings false alarm and causes useless investigation work (bugs got filed, time got spent to investigate, results that were looked at were wrong). The titles on the result pages are wrong too, they still say: SLED 10 Sun 1.5.0_10 (2 x 3.00GHz - 3GB RAM) Win XP Sun 1.5.0_10 (2 x 3.00GHz - 3GB RAM) >Changes to stop running tests on 1.5 described in bug 321142 Kim mentioned that in one of the arch calls but we probably didn't notice the impact on performance results. How bad/expensive would it be to keep the 1.5 tests at least for 3.6.x and 3.7? Note, that if we indeed switch everything to 1.6 then I don't see the point to run 2 times 1.6 on Win XP Sun (2 x 3.00GHz - 3GB RAM) Performance test results after v20100727-0800 until this is fixed are useless for those where 1.5 was used on the baseline. We have to remove all the data that corresponds to those tests, so that new performance result pages again show reliable data. If possible, old performance result pages should be adjusted as well or at lest get a big red note on it. I've filed bug 323439 to investigate why we run slower on 1.6.
> I've filed bug 323439 to investigate why we run slower on 1.6. It's bug 323461.
Created attachment 177334 [details] patch
patch released and started another baseline run. Also, I updated the labels for the machines in /home/users/releng/buildTools/eclipse.perf/ref37.properties.
Verified that performance baseline results are in the database. One of the machines is still running performance tests.