Auto-tools/Projects/Signal From Noise/Meetings/2012-02-16

From MozillaWiki
Jump to: navigation, search

Previous Action Items

  • bug 706912 - implement regression calculations - A*Team
  • collect a smaller regression that would not be caught or so obvious with 100 cycles then reboot
  • disable side by side staging - armenzg, jmaher

Status Reports

RelEng

  • bug 724561 disabled side by side for tp5r and we now use --ignore-first everywhere (except Linux/Windows)
  • bug 727823 enabling tp5r for Linux and Windows as well
    • we enabled tp5r for Linux/Windows last week
    • patches are ready to land

Metrics

  • Comparing current Talos data collection and aggregation methods to other proposed methods
    • Observed that the current data collection method has heavy carry-over effect (numbers increases over cycles)
    • Longitudinal method is better (more stable/tighter around the mean)
  • Propose new experiment
    • Similar to pagecycles.cvs method, except this time, reboot for every page (100 reboots total)
    • Make sure we run this experiment in the same environment as other previous experiments

Talos Changes

  • updated more configurations
  • moved pageloader into talos

Next steps:

  • get pageloader to stop calculating statistics
  • figure out what to do wrt sending all data to graphserver

Database Design

  • A static copy of the graphs database is setup on s4n4.qa.phx1.mozilla.com and is accessible over mozilla MPT. We can use this as a hosted development database.

Graph Server

jhammel - i will get a spec for graphserver-next db schema later today

UI Design

Round Table

Action Items