Community
Participate
Working Groups
As per bug 323514 comment 42, opening a new bug to track the issue with the performance test...
Created attachment 184329 [details] Proposed draft patch I think the test initially proposed lasted too long to open all projects in the workspace. Hence the gain of the fix was diluted in the measure error of the global time... This new proposal, first close all Eclipse project but leave the BIG_PROJECT open which becomes the last project opened in the workspace... It also reopen them all at the end to have no impact on following test suites. Then this configuration becomes similar to Adrian's one and here are the number I get while running it which confirms this fact: 1) Without bug 323514 fix: ------------------------- Measures (~Elapsed Process time): - n° 1: 4080ms - n° 2: 4016ms - n° 3: 4046ms - n° 4: 4281ms - n° 5: 4438ms - n° 6: 4078ms - n° 7: 3469ms - n° 8: 4062ms - n° 9: 3906ms - n° 10: 3938ms Test duration = 40314ms Time average = 4031ms 2) With bug 323514 fix: ---------------------- Measures (~Elapsed Process time): - n° 1: 1189ms - n° 2: 1187ms - n° 3: 1250ms - n° 4: 1187ms - n° 5: 1250ms - n° 6: 1250ms - n° 7: 1250ms - n° 8: 1187ms - n° 9: 1187ms - n° 10: 1250ms Test duration = 12187ms Time average = 1218ms So, this test shows similar improvement than Adrian's test :-) Of course, this patch needs some finalization before releasing, but I think it now can be done for M4... Ayush, please do not forget to release it also into perf_36x stream to have a baseline reference for this test, thx
*** Bug 331631 has been marked as a duplicate of this bug. ***
Created attachment 184344 [details] cleaned up patch Here's the cleaned up patch. I get the following numbers with this (after fix for bug 323514): - n° 1: 1047ms - n° 2: 1047ms - n° 3: 1047ms - n° 4: 1063ms - n° 5: 1062ms - n° 6: 1047ms - n° 7: 1047ms - n° 8: 1063ms - n° 9: 1031ms - n° 10: 1062ms Commit measures: Test duration = 10516ms Time average = 1051 Frederic, with your go ahead, I'll release the patch to both streams.
(In reply to comment #3) > Created an attachment (id=184344) [details] > cleaned up patch > > > Frederic, with your go ahead, I'll release the patch to both streams. +1 for me :-) Furthermore, as the 3.6 baseline runs on Friday (6pm), we should have a result for this new test in the Sunday warm-up build (8pm)
Released in perf_36x stream. Will release in HEAD tomorrow once the baseline run is done.
(In reply to comment #5) > Released in perf_36x stream. > FYI, release the code in perf_36x stream is not enough to be taken into account by the builder when running the baseline. You also need to version the org.eclipse.jdt.core.tests.performance project and update the jdtcore.map file in perf_36x with the version... I've done it, the version is v_A58_perf36xb
Released in HEAD and perf_36x for 3.7M4.
(In reply to comment #4) > (In reply to comment #3) > > Created an attachment (id=184344) [details] [details] > > cleaned up patch > > > > > > Frederic, with your go ahead, I'll release the patch to both streams. > > +1 for me :-) > > Furthermore, as the 3.6 baseline runs on Friday (6pm), we should have a result > for this new test in the Sunday warm-up build (8pm) I don't see this in the performance results for I20101205-2000. Ayush any idea why ?
(In reply to comment #8) > > I don't see this in the performance results for I20101205-2000. > Ayush any idea why ? It seems that Ayush released the patch after v_B27 build input, hence the result could not be in Sunday's warmup build. As v_B28 was put in I20101206-1800 build, the result will be available only in the next build with performances, which should be I20101207-2000 according to the build schedule...
The performance result for this test was available for the N20101204-2000 build and yes there is some 75% performance improvement.
Verified using I20101207-0250 (4.1 I-build)
