Chandler Engineering Meeting, Thursday April 26th 2007

This is a special meeting devoted to the Performance Tweaks period review Present: Grant, Aparna, Sheila, Katie, Heikki, Ted, Philippe

Agenda

• Status of the 3 weeks.
• Areas still in need of work
• Promissing directions left
• Decision: do we call this period completed?

Notes

• Perf much better now, close to target for most
• Half of the Apps team ready to move to debug, half need to continue on perf improvements
• We should review the scenarios and revise the baseline
• Actions:
  • organize a collaborative session of tests where we review the scenarios based on perception by users
  • lift the "Performance work only" limit so that some devs can go back to fixing bugs
  • continue for some devs to focus on perf, track those as bugs in Bugzilla
  • need to get data set to Aparna (using obfuscated dump and reload) and get the perf tests to use this

Heikki's notes

There has been very little progress with regards to performance this week and last week - the most productive time was the first week. This was somewhat expected, since easy problems tend to get fixed first. However, I wonder if continuing the performance period will mean we'll just be spinning our wheels without getting any noticeable results. I think we'd need to check with every engineer to see if they feel there are things that they could do if they had one extra week or if they feel there is nothing more they can think of.

The following primary test cases are still too slow:

• Overlay calendar with 3k
• Publish calendar with 3k
• Subscribe to calendar with 3k
• Triage in Dashboard with 3k

I think we've spent the least amount of time looking at the sharing performance, which is understandable given how new the new sharing code is. It is also a bit difficult to profile the sharing cases because they involve threads. I think some more time could be spent on improving sharing performance. The next least attacked case is probably the overlay test. The one that would help the users the most IMO is getting new event creation faster. Finally there are the other cases that have also received less scrutiny that we could probably get some improvements if we spent an additional week looking.

-- PhilippeBossut - 26 Apr 2007