Page Cycler V2: one of the test case records data for both runs of the cache-warm case. |
|||||
Issue descriptionTesting URL: http://www.ticketmaster.com/JAY-Z-and-Justin-Timberlake-tickets/artist/1837448?brand=none&tm_link=tm_homeA_rc_name2 Page cycler V2 records two data points for the "cache_temperature: pcv1-warm" configuration. The other testing URLs only record one data point in this case. Please see attached picture for an example. IIUC, "cache_temperature: pcv1-warm" loads the page twice, so that the cache is warm for the second load. Recording data for both loads is probably wrong. Kouhei: I noticed that you were the original owner of PageCyclerV2. Are you the right person to look into this issue? If not, would you please help to find an appropriate owner? I also CCed some tools/perf owners. Thanks!
,
May 2 2017
Thanks for reply! I am guessing that this may indicate some bug in the logic how we collect/calculate numbers in the cache-warm case. Even using a new set of page doesn't reproduce the problem, I am a little concerned that the root issue may not go away. WDYT?
,
May 2 2017
I dont know much about that, this currently has the wrong email for ned, so fixing that and he may be able to say if it will be an issue. The page cycler v2 benchmarks run with pageset_repeat set to two: https://cs.chromium.org/chromium/src/tools/perf/benchmarks/page_cycler_v2.py?type=cs&q=page_cycler&l=43 Shouldn't htat mean that all the results should have 2 entries? Adding Ethan since he did a lot of work with results2. Could you upload the complete results.html file?
,
May 2 2017
I mean each pageset_repeat results in two data points for the cache-warm config. This only happens for the ticketmaster URL, not for the other URLs in typical_25. In the attached file, I ran with pageset_repeat=5, you could display data with both "storyset repeat" and "storyGroupingKey_cache_temperature" checkboxes checked. And then you could see that for ticketmaster, in the "cache_temprature:pcv-warm" category, each storyset repeat has two data points.
,
May 2 2017
+1 to Kouhei's comment in #1. We should only dig into this issue if it is also happening in the new loading.desktop benchmark. Randy: maybe you can upload the results.html file generated from loading.desktop here so yzshen@ can check if this is still the issue in the new benchmark?
,
May 2 2017
RE#6: it would be nice to have a clean run. My browser hung with the results.html file attached to #6.
,
May 2 2017
I'll get you a clean run tomorrow then, it takes about an hour and a half to run.
,
May 3 2017
,
Aug 2
,
Nov 12
|
|||||
►
Sign in to add a comment |
|||||
Comment 1 by kouhei@chromium.org
, May 2 2017Owner: rnep...@chromium.org