1272403 - Hangs over 100ms per minute regressed in second week of Beta47 experiment

Status: RESOLVED INVALID

(Toolkit :: Telemetry, defect, P2)

Description

[Chris H-C :chutten]

Beta 47 second week analysis shows:
Median difference in hangs over 100ms per minute (parent) is -1.44, (0.59, 2.03).
The probability of this effect being purely by chance is 0.00.

Median difference in hangs over 100ms per minute (parent + child) is 0.37, (2.40, 2.03).
The probability of this effect being purely by chance is 0.00.

Whereas Beta 47 week 1 analysis had:
Median difference in hangs over 100ms per minute (parent) is -1.50, (0.50, 2.00).
The probability of this effect being purely by chance is 0.00.

Median difference in hangs over 100ms per minute (parent + child) is 0.23, (2.23, 2.00).
The probability of this effect being purely by chance is 0.00.


This amounts to a regression of 0.06 hangs per minute in the parent and another 0.08 hangs per minute in the child. This is accompanied by a very slight tightening of the gap between e10s and !e10s on the INPUT_EVENT_RESPONSE_MS (parent+children) measure (which doesn't really show much above hangs > 40ms, let alone 100ms).

Might be something, might be nothing.

Comment 1

[Chris H-C :chutten]

( Put this in Telemtry for the nonce because I don't know a good component to place it in )

Comment 2

[Chris H-C :chutten]

Beta47 week3: https://github.com/vitillo/e10s_analyses/blob/master/beta47-week3/e10s_experiment.ipynb

Median difference in hangs over 100ms per minute (parent) is -1.58, (0.62, 2.20).
The probability of this effect being purely by chance is 0.00.

Median difference in hangs over 100ms per minute (parent + child) is 0.30, (2.50, 2.20).
The probability of this effect being purely by chance is 0.00.

Comment 3

[Georg Fritzsche [:gfritzsche]]

Is this good to close?

Comment 4

[Chris H-C :chutten]

Yup. There were acknowledged problems with this approach of measuring hangs, and e10s has been released for over half a year.