How is the Time to Start Firefox affected by EARLY_GLUESTARTUP_READ_OPS?

Supposedly the time for Firefox to startup is slower for larger values of EARLY_GLUESTARTUP_READ_OPS . This is a number reported by Windows that measures how many read operations were made before Firefox started. I'me given to understand the number does not correspond to actual read operations and the hypothesis of interest is that time to start is slower when EARLY_GLUESTARTUP_READ_OPS is 0 vs numbers greater than 0.

The data source was 100K random sample of Firefox 12 telemetry packets between 2012-05-01 till 2012-05-14.

The values of EARLY_GLUESTARTUP_READ_OPS were clubbed as

Value of EARLY_GLUESTARTUP_READ_OPSno_prefetch(0)12-88+
%11.4673.545.949.06

The following display plots the percentiles of the \(log_{10}\) of main time to start. The vertical axis is the time for cases where EARLY_GLUESTARTUP_READ_OPS is 0 and the horizontal axis is for the case when EARLY_GLUESTARTUP_READ_OPS is (from left to right) 1,/2-8/ and 8+. Each point is a percentile. The blue points are (left to right) the 25%,50% and 75%. What is displayed but can hardly be seen (that narrow) are the 95% CI bands.

early gluestartup read ops

1 Interpretation

The two panels on the left demonstrate that startup time is slower when EARLY_GLUESTARTUP_READ_OPS is 0 (since the curve lies above the dotted red line which implies \(y>x\)). Only the bottom 10% of start up time is faster for the case EARLY_GLUESTARTUP_READ_OPS equal to 0 (probably makes sense).

The last panel is a surprise and shows for large values of EARLY_GLUESTARTUP_READ_OPS , the start up time is faster than when EARLY_GLUESTARTUP_READ_OPS is 0.

For more information (and to comment) see https://bugzilla.mozilla.org/showbug.cgi?id=757215

Author: Saptarshi Guha

Date: 2012-05-24 12:10:17 PDT

HTML generated by org-mode 7.3 in emacs 24