Bug 1729890 Comment 11 Edit History

Note: The actual edited comment in the bug view page will always show the original commenter’s name and original timestamp.

(In reply to Jon Coppeard (:jonco) from comment #3)
> It's really strange that making more GCs full GCs would increase memory usage.  Just wondering if you have an idea of what's going on here Paul?

Thanks for lending a 2nd (& third Ted!) 

I was looking at the memory reports for and I see the difference appears to be in the bin-unused, which is the fragmentation of the jemalloc heap.  But for any cause I could think of I expected to see a corresponding negative amount for some memory being freed.  So I appriciate the extra eyes here.

I started with the Tabs Closed Settled memory reports and saw only this fragmentation.  I was unable to get perfherder to show me the subtests due to Bug 1729057, which might have helped direct my search.  So my best guess is like Ted said:

> We wonder if the full-GC is throwing off the CC heuristics and we are somehow keeping old windows around longer than before?

I want to try messing with the patch to see what extra clues I can get.  But I also want to try DMD like Ted did, that may be a more solid lead here.

Thanks everyone.
(In reply to Jon Coppeard (:jonco) from comment #3)
> It's really strange that making more GCs full GCs would increase memory usage.  Just wondering if you have an idea of what's going on here Paul?

Thanks for lending a 2nd (& 3rd Ted!) set of eyes.

I was looking at the memory reports for and I see the difference appears to be in the bin-unused, which is the fragmentation of the jemalloc heap.  But for any cause I could think of I expected to see a corresponding negative amount for some memory being freed.  So I appriciate the extra eyes here.

I started with the Tabs Closed Settled memory reports and saw only this fragmentation.  I was unable to get perfherder to show me the subtests due to Bug 1729057, which might have helped direct my search.  So my best guess is like Ted said:

> We wonder if the full-GC is throwing off the CC heuristics and we are somehow keeping old windows around longer than before?

I want to try messing with the patch to see what extra clues I can get.  But I also want to try DMD like Ted did, that may be a more solid lead here.

Thanks everyone.

Back to Bug 1729890 Comment 11