Closed Bug 1291349 Opened 8 years ago Closed 8 years ago

8.91 - 13.28% tp5o Private Bytes (linux64) regression on push 9cdec67ac091 (Mon Aug 1 2016)


(Firefox :: Untriaged, defect)

50 Branch
Not set





(Reporter: jmaher, Unassigned)



(Keywords: perf, regression, talos-regression)

Talos has detected a Firefox performance regression from push 9cdec67ac091. As author of one of the patches included in that push, we need your help to address this regression.

Summary of tests that regressed:

  tp5o Private Bytes linux64 opt - 8.91% worse
  tp5o Private Bytes linux64 opt e10s - 11.8% worse
  tp5o Private Bytes linux64 pgo - 9.89% worse
  tp5o Private Bytes linux64 pgo e10s - 13.28% worse

You can find links to graphs and comparison views for each of the above tests at:

On the page above you can see an alert for each affected platform as well as a link to a graph showing the history of scores for this test. There is also a link to a treeherder page showing the Talos jobs in a pushlog format.

To learn more about the regressing test(s), please see:

For information on reproducing and debugging the regression, either on try or locally, see:

*** Please let us know your plans within 3 business days, or the offending patch(es) will be backed out! ***

Our wiki page outlines the common responses and expectations:
there are a lot of patches which landed at the same time, all the same author/reviewer- I assume we don't need bisection, but if we do, please let us know!

:fitzgen, can you look into this?  I am verifying there are no other regressions, but so far we see linux64 private bytes regression on all configs (opt, pgo, e10s)
Flags: needinfo?(nfitzgerald)
Blocks: 1291351
Hi! I was on vacation, but I'm back now. Looking into this
Flags: needinfo?(nfitzgerald)
Joel, I'm 95% sure that bug 1290287 is the actual regression, but can you double check that via bisection? I don't want to start digging in too deep into the wrong thing. Thanks!
Flags: needinfo?(jmaher)
After some digging into what is actually being measured here and correlating these results with the RSS measurements, I think we can safely WONTFIX.

On linux, the private size is the size of the address space that we have reserved, regardless whether those pages actually get physically allocated. In contrast, RSS includes only the physically allocated pages. As you can see in this graph[0], RSS is flat across this jump in private bytes. That means that we are now reserving more of the address space, but we aren't actually allocating any more physical memory. This is slightly interesting, but not worrying.

Closed: 8 years ago
Flags: needinfo?(jmaher)
Resolution: --- → WONTFIX
thanks for looking into this Nick- the initial bisection had a problem on try and we were going to try again in a while- this saves more try pushes!
You need to log in before you can comment on or make changes to this bug.