797964 - complete spdy ping experiment

Status: RESOLVED FIXED

(Core :: Networking: HTTP, enhancement)

Description

[Patrick McManus [:mcmanus]]

we've had an experiement for a while (bug 729736) to see if any different intervals of the spdy ping had higher failure rates.

It turns out there isn't much there. (graphs to be attached). values from 10 to 110 have similar pass and fail rates.

I'll remove the experiment and update the threshold to 58 (just a touch under 60) - this shows we don't need to be particularly concerned about a particular value, but it is still a battery vs latency tradeoff that we can feel better about eeking out a little battery and making that session last a little longer.

Comment 1

[Patrick McManus [:mcmanus]]

Attachment: patch 0

Comment 2

[Patrick McManus [:mcmanus]]

Attachment: fail distribution

Comment 3

[Patrick McManus [:mcmanus]]

Attachment: pass distribution

Comment 4

[Patrick McManus [:mcmanus]]

thanks honza.

fyi I've decided not to push this right away due to bug 798423 - the false fails caused by that bug could plausibly be swamping interesting values that this telemetry would show. We'll get some more data with that fixed. (I don't expect anything, but might as well find out)

Comment 5

[Patrick McManus [:mcmanus]]

Attachment: failure distribution

Comment 6

[Patrick McManus [:mcmanus]]

attachment in comment 5 is the updated distribution using only data on nightly that includes the 798423 fix.

It is much more in line with what I would expect to see - failure spikes at 60 and 110. So the approach of changing our default from 44 to 58 seems right in that it minimizes ping traffic without crossing the 60 boundary.

Comment 7

[Patrick McManus [:mcmanus]]

  https://hg.mozilla.org/integration/mozilla-inbound/rev/5db0e8ad0b7d

Comment 8

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/5db0e8ad0b7d