787245 - Integrate the GC log with the profiler

Status: RESOLVED FIXED

(Core :: Gecko Profiler, defect)

Description

[Benoit Girard (:BenWa)]

Currently the profiler only gathers which frames are executing during the GC which apparently isn't helpful. The GC logs is more helpful thus we should try to integrate it with the profiler:

MemChaser
http://people.mozilla.com/~wmccloskey/gc-collector/

Comment 1

[(no longer active)]

Should we do the same thing for CC as well?

Comment 2

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

The same API provides both GC and CC info, so that should be easy.

Comment 3

[Benoit Girard (:BenWa)]

Attachment: Mouse over info

Comment 4

[Benoit Girard (:BenWa)]

Attachment: Event details (click on icon)

Comment 5

[Benoit Girard (:BenWa)]

Comment on attachment 657963 [details]
Mouse over info

Let me know what you think of these two screenshot. Is the information presented in a useful way to help with profiling GC problems? If not please let me know what is lacking.

What is not clear is how the times reported by GC event logs appear to be shorter then the amount of time we spend executing the code according to our sampling. It's not a huge different so the errors may be acceptable. I'll look a bit more into why there is a different however.

Comment 6

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

Comment on attachment 657963 [details]
Mouse over info

This looks cool. What's the difference between the black and red sections?

Comment 7

[Benoit Girard (:BenWa)]

I interleave the colors between different sections kind of like the table rows are white/blue. What happens there is we detect there's a GC but it doesn't match the start time yet so I insert a generic 'Garbage Collection' then a few samples later it matches the start time so it inserts a 'Garbage Collection Slice' item.

I just need to implement the CC events and see how the events differ without incremental collection.