mozilla :: #e10s

19 May 2017
16:51mconleyelan: sorry, gotta drop!
16:57elanthank you!!
19:06mstangeis there a bug about decreasing the process priority of e10s content processes of background tabs?
19:09mstangeehsan: I filed
19:09firebotBug 1366356 NEW, Decrease the process priority of content processes that are not running a foreground tab
19:47erahmjimm: have you looked at the awsy numbers of beta vs central?
19:48erahmit's pretty bizzarro:
19:49RyanVMwonder what a 55-as-Beta Try push would look like
19:50RyanVMi.e. is it a 54 vs. 55 issue or Beta vs. Nightly
19:51erahmRyanVM: in this context I was thinking 1 vs 4 content processes would dominate
19:52RyanVMah, maybe - I had mixed up which was which there (was seeing Beta on top)
19:52erahmI think we get such a big win on tabs closed that it dominates the cumulative metric, digging into tabs opened we're behaving more like I expected:
19:52RyanVMwell, they're not that far off, are we?
19:52RyanVMare they*
19:53erahmRyanVM: yeah it was that nightly is better than beta that made me question the numbers
19:53RyanVMbeta w/ 4 CPs would be an interesting Try push too
19:54erahmIt's sad that the best way to deal with leaks is to just nuke a process
19:55erahmRyanVM: I *think* that's what this is:
20:08jimmerahm: does treeherder actually pick up valid values? for my beta pushes to try I had to dig into the sub test numbers
20:08jimmand manually pull those out
20:08jimmI never could figure out what a value like "Resident Memory summary opt" represented
20:09erahmjimm: yeah it's a little weird how they have it setup. Basically the summary is the geometric mean of the sub-tests. It makes slightly less sense with awsy
20:09erahmbut it does okay for larger regressions
20:11jimmerahm: yeah that was weird, for example Resident Memory opt doesn't regress -
20:11jimmclick on the subtests link though, and you find regressions
20:12jimmI just chose to ignore all the top level numbers and look at the subtest values instead.
20:13jimmerahm: fyi, beta default is 1 content process so perfherder compare of automation runs isn't going to tell you anything.
20:14erahmjimm: yeah, I just wanted to see how 1 vs 4 looked over time (so beta vs nightly), saw that nightly was lower and got confused
20:15erahmI think the problem is that expanded there's almost too much data (b/w explicit and rss), summed it's not a great view for real numbers (but is probably good for regression tracking)
20 May 2017
No messages
Last message: 96 days and 7 hours ago