710943 - track CYCLE_COLLECTOR telemetry statistics

Status: RESOLVED WONTFIX

(Mozilla Metrics :: Frontend Reports, defect)

Description

[Andrew McCreight [:mccr8]]

CYCLE_COLLECTOR is the length of time the browser is paused while the cycle collector is running, in ms.  This may be a major source of pauses, so it would be good to track what is happening to this over time using telemetry.

Comment 1

[(dormant account)]

Saptashi,
I think we need pretty fine resolution on this in order to catch regressions. Can we track this one on a daily basis for nightly builds(using platform build id)?

Comment 2

["Saptarshi Guha[:joy]"]

Yes we would. Also, can we not bucket? That loses a **lot** of the granularity! 

Raw time series per user e.g. User1: [300,120,300,800], User2:[100,2404,20302,...]

is the form of the data we want.

Also, as with everything telemetry (and Taras knows this), ideally we want, for a given new feature (that hypothetically improves CYCLE_COLLECTOR) - before and after or on/off switches for users.
Cheers
Saptarshi

Comment 3

[Pedro Alves]

What are you mostly interested in? 

average / median / percentiles? And what is the comparison base? Same architecture?

Comment 4

[Pedro Alves]

Hum - not bucket is not an option, but currently the info we have has the same bucket configuration

Comment 5

[(dormant account)]

(In reply to Saptarshi Guha from comment #2)
> Yes we would. Also, can we not bucket? That loses a **lot** of the
> granularity! 
> 
> Raw time series per user e.g. User1: [300,120,300,800],
> User2:[100,2404,20302,...]
> 
> is the form of the data we want.
> 
> Also, as with everything telemetry (and Taras knows this), ideally we want,
> for a given new feature (that hypothetically improves CYCLE_COLLECTOR) -
> before and after or on/off switches for users.
> Cheers
> Saptarshi

Not bucketing adds book-keeping overhead. If you like we can rebucket it or give you more buckets.

Comment 6

["Saptarshi Guha[:joy]"]

You need more buckets. Do we already have a histogram of CYCLE_COLLECTOR? Can we see it here? We could further subdivide the buckets in cases where there are too many.

Comment 7

[(dormant account)]

(In reply to Saptarshi Guha from comment #6)
> You need more buckets. Do we already have a histogram of CYCLE_COLLECTOR?
> Can we see it here? We could further subdivide the buckets in cases where
> there are too many.

how many buckets do you want?

Comment 8

["Saptarshi Guha[:joy]"]

(In reply to Andrew McCreight [:mccr8] from comment #0)
> CYCLE_COLLECTOR is the length of time the browser is paused while the cycle
> collector is running, in ms.  This may be a major source of pauses, so it
> would be good to track what is happening to this over time using telemetry.

Btw, i recently did these charts: http://people.mozilla.org/~sguha/cyccollector.uptime.html

Comment 9

[Annie Elliott]

Closing 2011 ext still marked as NEW req. 

Not in current scope of work.