Closed Bug 1098683 Opened 10 years ago Closed 7 years ago

reftest-e10s takes nearly twice as long to run on linux32 as its non-e10s equivalent

Categories

(Release Engineering :: General, defect, P3)

Product:
Release Engineering
Component:
General
Type:
defect

Tracking

(e10s+)

Status:
RESOLVED INCOMPLETE
Tracking Flags:
Tracking Status
e10s + ---

People

(Reporter: RyanVM, Unassigned)

References

(Blocks 1 open bug)

Details

On Linux32 opt, a regular reftest run takes 45-50min to complete. Reftest-e10s takes 95-100min (and I even saw a 120min timeout get hit once). On Linux64, there's also a performance penalty, but nowhere near as extreme (~35min -> ~50min).

Are our 32bit AWS slaves really that much slower or are we doing something horrible in e10s mode? Given the other known perf issues we seem to be having lately with Linux32, I'm inclined to go with the former?

Linux32 opt non-e10s reftest log:
https://treeherder.mozilla.org/logviewer.html#?job_id=629543&repo=mozilla-central

Linux32 opt e10s reftest log:
https://treeherder.mozilla.org/logviewer.html#?job_id=629571&repo=mozilla-central
Maybe catlee or rail can answer the question if we're using slower instances for linux32 than we are for linux64.
Flags: needinfo?(rail)
Flags: needinfo?(catlee)
tst-linux32 are running on m1.medium instances, as are tst-linux64, so they should have equivalent CPUs
Flags: needinfo?(catlee)
Yeah, they are equivalent...
Flags: needinfo?(rail)
Blocks: e10s-tests
Priority: -- → P3
Status: NEW → RESOLVED
Closed: 7 years ago
Resolution: --- → INCOMPLETE
Component: General Automation → General
You need to log in before you can comment on or make changes to this bug.