TestEngineering/Performance/Fenix: Difference between revisions

← Older edit

TestEngineering/Performance/Fenix (view source)

Revision as of 19:52, 17 September 2020

1,594 bytes added , 17 September 2020

Add mach perftest vs. normal startup section

Mcomella

Confirmed users

975

edits

@@ Line 2: / Line 2: @@
 * [[TestEngineering/Performance/NimbleDroid|Nimbledroid]]: no longer active, see page for details
+* mach perftest
+== mach perftest vs. normal startup ==
+This is a list of known differences between a normal VIEW startup of Fenix vs. a VIEW mach perftest run of Fenix. Many of these were found while investigating in https://github.com/mozilla-mobile/perf-frontend-issues/issues/141
+=== Configuration differences ===
+* set-debug-app (appears to have no perf impact by itself - [https://github.com/mozilla-mobile/perf-frontend-issues/issues/141#issuecomment-689161140 analysis] - but causes other code to run)
+* a new user profile is used
+* Browsertime disables a set of [https://github.com/sitespeedio/browsertime/blob/325d5389b4deb6b68bff0e5b8fa42c9d457bfdc5/lib/firefox/settings/firefoxPreferences.js firefox preferences]
+* Geckodriver also configures gecko via preferences
+=== Runtime differences ===
+* In response to set-debug-app, GV reads /data/local/tmp/org.mozilla.geckoview-config.yaml on startup; file is configured for marionette and crashreporter stuff
+* marionette JSMs start up (+155ms on P2 from 9/11/2020)
+* First android.graphics.HardwareRenderer.nSetStopped call is 89ms in perftest, 14ms in adb start (in profiles below)
+* Condprof (currently disabled) runs checkForChanges in add-ons DB – is not in normal use start path & increases start up time ([https://jira.mozilla.com/browse/FXP-1339 issue to fix])
+=== For later use ===
+Comparative profiles ([https://github.com/mozilla-mobile/perf-frontend-issues/issues/141#issuecomment-693692338 details here])
+Adb shell am start VIEW: https://share.firefox.dev/32C3FeX
+Mach perftest VIEW: https://share.firefox.dev/3c3q1cv