Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(298)

Issue 196723004: Make the morejs page cycler measure cold times better. (Closed)

Created:
6 years, 9 months ago by marja
Modified:
6 years, 9 months ago
Reviewers:
dtu, fmeawad, tonyg
CC:
chromium-reviews, chrome-speed-team+watch_google.com
Visibility:
Public.

Description

Make the morejs page cycler measure cold times better. Currently it only does one cold run per page, so the results are noisy. BUG=350858 R=dtu@chromium.org, fmeawad@chromium.org Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=256595

Patch Set 1 #

Unified diffs Side-by-side diffs Delta from patch set Stats (+2 lines, -1 line) Patch
M tools/perf/benchmarks/page_cycler.py View 1 chunk +2 lines, -1 line 0 comments Download

Messages

Total messages: 7 (0 generated)
marja
fmeawad, ptal.
6 years, 9 months ago (2014-03-12 12:59:29 UTC) #1
fmeawad
LGTM, you will also need an owner.
6 years, 9 months ago (2014-03-12 17:18:25 UTC) #2
marja
I'm an owner :) But it would be good for tonyg@ to ACK this, since ...
6 years, 9 months ago (2014-03-12 17:28:12 UTC) #3
dtu
lgtm, no weird side effects here :)
6 years, 9 months ago (2014-03-12 18:13:40 UTC) #4
marja
Committed patchset #1 manually as r256595 (presubmit successful).
6 years, 9 months ago (2014-03-12 19:17:44 UTC) #5
tonyg
Oops, I was too slow, sorry. What I'd actually like to do is to change ...
6 years, 9 months ago (2014-03-14 01:50:55 UTC) #6
marja
6 years, 9 months ago (2014-03-14 11:04:46 UTC) #7
Message was sent while issue was closed.
On 2014/03/14 01:50:55, tonyg wrote:
> Oops, I was too slow, sorry.
> 
> What I'd actually like to do is to change the default cold-load-percent to 50
> for all page cyclers and leave the iterations at 10. Would that work for you?

In general yes, if the data we get from 5 runs is stable enough. I haven't
looked if the page cycler noise currently has is like "different runs take very
different times" or "all runs take generally the same, just for some reason
running the page set *now* versus 10 min ago takes different time". If it's the
latter kind of noise, dropping the runs to 5 should work.

Powered by Google App Engine
This is Rietveld 408576698