change in SIMPLE_MEASURES_FIRSTPAINT in correlation to 3 values in STARTUP_USING_PRELOAD

RESOLVED WORKSFORME

Status

defect
RESOLVED WORKSFORME
7 years ago
7 years ago

People

(Reporter: taras.mozilla, Unassigned)

Tracking

unspecified
Backlogged - BZ
x86_64
Windows 7
Dependency tree / graph

Details

(Whiteboard: Telemetry -- needs PM project priority )

Attachments

(2 attachments)

We recently landed bug 692255. We need to measure how different values of STARTUP_USING_PRELOAD influence firstpaint speed. This should be done on the nightly channel, starting from builds from may 29th.
Hi,
I've added Joseph Kelly, a Metrics intern for the summer. He'll be working with us to answer this.

Does this need a version filter e.g. FF n  for some n? or just buildid should be greater than May 29th?
(In reply to Saptarshi Guha from comment #1)
> Hi,
> I've added Joseph Kelly, a Metrics intern for the summer. He'll be working
> with us to answer this.
> 
> Does this need a version filter e.g. FF n  for some n? or just buildid
> should be greater than May 29th?

channel should be nightly.
What I meant was, when channel == "nightly" build id is sufficient
Assignee: nobody → jkelly
Hey all,

I was just wondering what other variables may have an effect on firstpaint speed?

By taking into account other variables that may influence firstpaint we are better able to understand the relationship between STARTUP_USING_PRELOAD and firstpaint (particularly if STARTUP_USING_PRELOAD is correlated with these variables). 

Cheers!
(In reply to Joseph Kelly from comment #4)
> Hey all,
> 
> I was just wondering what other variables may have an effect on firstpaint
> speed?

firstpaint is a result of a complicated startup process, just about anything can influence it and some of the things that influence it are also measured by histograms. However, it's not easy to tell variables influence it for us, perhaps you guys can figure that one out better.

We hope that STARTUP_USING_PRELOAD is of biggest influence on firstpaint :)
There are 227 variables already present - many of which have high % of missing values. 
That size itself makes it tough to begin with.

These variables are histograms, some of which are categorical e.g. STARTUP_USING_PRELOAD and some continuous. Depending on what type they need to be handled differently (one doesnt take a mean of a white/blue/red variable). So its really tough to churn unprocessed variables through some variable reduction scheme.

Any help you guys provide would improve this analysis.
So final update: anyone with domain knowledge willing to help out ?


Q: Which of the 445 variables would possibly affect  SIMPLE_MEASURES_FIRSTPAINT (in addition to STARTUP_USING_PRELOAD) ? Attached a CSV file of the variable names.
Awesome. Thanks Saptarshi this definitely helps.
Status: NEW → ASSIGNED
Posted file Data analysis
Here is the analysis. Let me know if you have any questions. Cheers!
Let me know if we should carry this bug over to Q3 or if we can mark it as resolved.
Whiteboard: Telemetry -- needs PM project priority
Triaged.
Assignee: jkelly → nobody
Target Milestone: Unreviewed → Backlogged - BZ
Status: ASSIGNED → RESOLVED
Closed: 7 years ago
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.