Auto-tools/Projects/Signal From Noise/Meetings/2012-04-18

From MozillaWiki
Jump to: navigation, search

Graph Server In Person Meeting

  • Attendees:
    • Rhelmer
    • mcote
    • camd
    • rniwa
    • christina
    • jmaher
    • jeads
    • ctalbert

Overview of what we have so far

  • Consider different numerical treatments of the raw data
    • Allow people to adjust the y axis
  • Leave points highlighted when you click on the individual points
    • When you select the two points you should be able to get the blame list from hg when you select them.

Page Centric View

  • take mean across pages over time
  • be able to have a tree of tests instead of a list of tests
    • webkit has about 30 metrics per test
    • be nice to have a way to see the tree views per page and drill into specific data points/attributes on a per page per test basis.
  • Would be good to display page-centric variance

Page Regression View

  • Take baseline from last release and show the values from the last release
    • We have a way to store that in the schema and keep that metric
  • We also need a rolling mean on a per page basis to know when we regress per page performance.

Wishlist

  • The new UI should be able to help us test different statistical algorithms and see if the different algorithms detect/show different things.
    • We are writing AnalyzeThis which can help with that as a standalone statistical package which should be consumed by both the UI and the Developer running it locally use case.
    • This is a goal of the project
    • Need to be sure that the new algorithm and show that the new algorithm can detect all the previous regressions that we knew about. So, you may need a way to run historical data through the system.
  • We need some deep integration with try server so that people can try various ideas and builds to see and debug their attempts to fix performance issues.

UI Issues

  • Make the UI snappy and fast
    • Will be adding some caching
    • Current one uses zeus load balancing, images written to the file system and use expiration headers in the browser. That makes it pretty fast the first time and very fast next time you load the graph
  • Don't worry too much about it being complex as long as you can justify its usefulness.
  • Fine to reuse the graphs.m.o web ID, and might also use the graphs-new.m.o idea as well just as we did last time we upgraded graphsever.