719968 - Add hard faults to xperf talos output

Status: RESOLVED WONTFIX

(Testing :: Talos, defect)

Description

[Marco Bonardo [:mak]]

Increase in hard faults may bring to lot of additional reads. So would be really interesting to measure it.

Comment 1

[Joel Maher ( :jmaher ) (UTC -8)]

Currently we collect the HARD_FAULT provider:
http://hg.mozilla.org/build/talos/file/ca816866e975/talos/xperf.config#l106

I believe I need the stackwalk parameter "HardFault", as  found here:
http://msdn.microsoft.com/en-us/library/windows/desktop/ff191012%28v=vs.85%29.aspx

Let me know if this looks right and I can get a recent .etl file and run some stuff by hand.

Comment 2

[Joel Maher ( :jmaher ) (UTC -8)]

This bug has been around for almost 4 years, if we don't plan to work on it in the next couple of months, I would like to close it out. Please speak up if there are reasons to keep it open.

Comment 3

[Marco Bonardo [:mak]]

As I said in bug 719965, it's up to the Perf team. I think the same considerations apply here.

Hard Faults is an interesting metric, since it tells we are not being efficient in memory usage and thus the system has to fallback to disk. Though, in this case I fear the measurement may end up being too noisy and dependent on the tinderbox memory status. Especially when using cloud boxes, it may not be very meaningful.
It would be fare more expensive than flushes to make this measurement reliable, so this is more likely a wontfix.

Comment 4

[Vladan Djeric (:vladan)]

I'll let Aaron make the call here. This seems more difficult than bug 719965 and our focus is elsewhere for now, so I'm leaning toward closing it

Comment 5

[(No longer employed by Mozilla) Aaron Klotz]

Same as https://bugzilla.mozilla.org/show_bug.cgi?id=719965#c5