General performance sessions:
- A session on linking that rambokid is planning
- Is this related to Michael Meeks's work on -bdirect for the linker? Maybe we can get him to show his work as well.
- Yes, mmeeks is involved already.
- Is this related to Michael Meeks's work on -bdirect for the linker? Maybe we can get him to show his work as well.
How To Write Shared Libraries (proposed)
Tools:
Performance-related sessions:
Proposed schedule
What should we talk about in each of the general sessions, and in what order should they happen? One idea is this:
- How much faster - Around the first day of the conference. Overview of what we have done. Point out interesting hotspots that remain to be fixed. Quick overview of the tools [sysprof, valgrind/massif, timeline script]
- Performance hackfest - Mostly during all of GUADEC. We can have a mini-BOF in the middle to get all the minds together. Form little teams of people to work on particular hotspots: AlexL and a nautilus gang, VUntz and a panel gang, Fejj and an Evolution gang, etc.
- Performance BOF - At the end of the conference, to recapitulate what we did during the hackfest.
General thoughts
We don't have much time, and doing the performance analysis is what takes time (fixing it is usually not that hard). It would be good to do our homework and arrive to the conference with analyses already done: we can then present pretty graphs to make it obvious to people where problems are, and then we can make them fix the problems.
Hard things to fix: Evolution, Mozilla. Need a lot of brainpower to do the analysis. So we can probably gather smart people to figure out how to analyze them.
We need a bulletin board (sticky notes on a blank sheet of paper at the conference site, or a wiki, or something) where the hackfesters can post questions:
- Does anyone have tips to analyze $random_problem?
- Someone help me get sysprof built on my machine.
- I need a Dtrace god to whip up a script.