Check QuantumBar doesn't regress Talos tests
Categories
(Firefox :: Address Bar, defect, P1)
Tracking
()
People
(Reporter: standard8, Assigned: standard8)
References
(Blocks 1 open bug)
Details
(Whiteboard: [fxsearch])
As most of the QuantumBar architecture is currently stable, it seems a good time to do some Talos performance comparisons.
Bug 1523332 is likely the only one in the queue that I think may affect startup/window performance, however, it'd be good to get a comparison before and after that.
Assignee | ||
Comment 1•6 years ago
|
||
A first draft of this I did looked reasonable.
However I'm now doing one with pgo builds as well as opt:
Assignee | ||
Comment 2•6 years ago
|
||
Mike/Joel, could you have a look at the comparison results here, do you think we need to take further action?
There's a few "medium" confidence issues, I don't know if we should worry about those.
tsvgx seems a bit like it could be a false-positive, since I'd don't know how we'd affect svg performance with something like the address bar.
tart seems like a possible issue, but it is only bad on Windows?
Comment 3•6 years ago
|
||
nothing jumps out as me as really problematic. in looking at the data on try and m-c for these tests I see almost every one of them is a bi-modal distribution which means that you are seeing a concentration on one mode vs the other- these all fall into natural patterns.
The only risk is if you made a bi-modal test become a single mode, worse case in the slower scenario. That would make the browser more deterministic though.
Comment 4•6 years ago
|
||
Nothing jumps at me as anything worth being concerned about, to be honest. The only ones that are surfacing as "important" are the TART test, which... makes no sense. I don't think it's worth spending the time trying to figure out why, to be honest.
I also wouldn't really worry about the sessionrestore tests - those measurements aren't very helpful - see bug 1531520.
The others aren't confident enough to be discernible from noise.
Comment hidden (obsolete) |
Comment 6•6 years ago
|
||
it may be interesting also to check with both pref and attribute switched, that is the final situation we'll have once QB is enabled by default.
Assignee | ||
Comment 7•6 years ago
|
||
Yeah, that was my next step, I think comment 5 should be considered invalid (I've marked it obsolete).
New builds:
Assignee | ||
Comment 8•6 years ago
|
||
The additional results here look fine as well. I'm going to mark this as done, as I don't think there's anything more we need to do here.
Description
•