New issue
Advanced search Search tips

Issue 923772 link

Starred by 1 user
Status: Unconfirmed
Owner: ----
Cc:
hajimehoshi@chromium.org
altimin@chromium.org
EstimatedDays: ----
NextAction: ----
OS: Chrome
Pri: 3
Type: Bug



Sign in to add a comment

7% regression for Robohornet benchmark from Chrome R68 to Chrome R69

Project Member Reported by hong.zh...@intel.com, Jan 21 (2 days ago) 

We observe 7% regression for Robohornet(http://www.robohornet.org/) benchmark from CrOS version 10718.88.2(Chrome: 68.0.3440.118) to 10895.78.0(Chrome: 69.0.3497.120) on Electro.

Steps To Reproduce:
(1)run Robohornet on Electro, separately in version 10718.88.2 and 10895.78.0
(2)compare results

Expected Result: 
no gap

Actual Result:
7% regression

How frequently does this problem reproduce? (Always, sometimes, hard to
reproduce?)
Always

Comment 1 by hong.zh...@intel.com, Yesterday (28 hours ago)

Cc: hajimehoshi@chromium.org
we did further breakdown, and found the issue was caused by commit #570370 (https://chromium-review.googlesource.com/c/chromium/src/+/1114405/ ).
@Hajime, is it a known issue?

Comment 2 by hajimehoshi@chromium.org, Yesterday (24 hours ago) 

I don't think my CL caused this regression since my CL just avoids unnecessary allocations.

Comment 3 by hong.zh...@intel.com, Today (21 hours ago) 

Hi Hajime, we found only "Write to localStorage" and "Read from localStorage" subcases of Robohornet are obviously impacted, separately ~40% and ~20% regression. Could you please help to double check? Thanks in advance

Comment 4 by hajimehoshi@chromium.org, Today (20 hours ago)

Cc: altimin@chromium.org
+altimin

Comment 5 by altimin@chromium.org, Today (15 hours ago) 

This is very surprising. Could you record a trace: https://www.chromium.org/developers/how-tos/trace-event-profiling-tool/recording-tracing-runs?

The interesting categories are toplevel,benchmark,blink,renderer.scheduler and sequence_manager.
►

Sign in to add a comment