< Performance | MemShrink
Most of the meeting was spent on triage.
- There are a lot of issues with large fixed size graphics things (images, frames) causing slop. Is there any way to solve this once and for all?
- jemalloc upgrade: still on hold. New version not supported on Windows, close to working on Mac, but need to upstream it.
- jlebar found that with a basic Google docs etc. workload the heap became 25% fragmented.
- Is it worth looking into? Can we reduce the number of malloc calls?
- jst said there was a patch that added an arena allocator for DOM, which was backed out because it didn't help. Would probably be a lot of work to add in, but might help this.
- jlebar suggested that we could try looking into which allocations are causing the fragmentation.
- khuey: probably intermixing short-lived and long-lived objects. Can we visualize the fragmentation? jst says that Stewart has some graphs about fragmentation. jlebar has been investigating.
- Would a new version of jemalloc help?
Dave Mandelin made a long post about mobile performance, but the testing was done with Firefox 6, so it isn't clear how relevant it is to the current state of affairs. The mobile people are going to do some retesting with Nightly for more up to date information.