Electrolysis/Release Criteria/Stability: Difference between revisions

← Older edit

Electrolysis/Release Criteria/Stability (view source)

Revision as of 17:51, 23 March 2016

249 bytes added , 23 March 2016

updated to new experiment

Benjamin Smedberg

Confirmed users, Bureaucrats and Sysops emeriti

1,217

edits

@@ Line 14: / Line 14: @@
 A/B testing on beta should indicate that the rate of browser+content crashes is no larger than the non-e10s browser crash rate. Chrome process crash rates will be measured with crash ping counts per 1000 hours of subsessionLength. Content process crash rates will be measured with SUBPROCESS_CRASHES_WITH_DUMP['content'] per 1000 hours of subsessionLength.
-Latest Spark analysis: [https://github.com/vitillo/e10s_analyses/blob/master/beta45-withoutaddons/e10s-stability-analysis.ipynb e10s-stability-analysis.ipynb] (reviewed by rvitillo)
+Latest Spark analysis, 46 experiment #1: [https://github.com/vitillo/e10s_analyses/blob/master/beta46-noapz/e10s-stability-analysis.ipynb e10s-stability-analysis.ipynb] (reviewed by rvitillo)
+SUMMARY: e10s is now 62% worse in overall stability (chrome+content crashes). 17.1 crashes per 1000 usage hours for non-e10s, 27.8 for e10s. That regression is entirely in the content process: chrome-process stability has improved dramatically from 15.2 to 7.93.
 [https://docs.google.com/document/d/1af-3D7OmiOog4q51UGwaC-nSBNfcLzzvJQWAPzjigQk/ New UT/e10s Crash Reporting proposal]
-SUMMARY: e10s is 33% worse in overall stability. That is entirely in the content process.
 == plugin process crash rates ==
 A/B testing on beta should indicate that the rate of plugin crashes is no longer in e10s than non-e10s. Plugin process crash rates will be measured with SUBPROCESS_CRASHES_WITH_DUMP['plugin'] per 1000 hours of subsessionLength.
+e10s is worse: crash rate of 8.1 for non-e10s and 12.1 for e10s.
 = Bugs =