This is spun off from bug 806279. My understanding from bug 806279 comment 45 is that we don't currently have CC performance tests that could help measure the performance impact of changes in the cycle collector. That would be nice to have.
We have telemetry data. It shows surprisingly well how even some small changes affect to mean or median CC times. If we want to create some perf test for CC, it would probably have to be xpcshell test where we manually create fixed size graph and run CC several times. (I doubt bug 806279 affects to performance in any measurable way.)
We might be able to record CC times on areweslimyet runs, though I don't know how much that would correspond to actual telemetry performance that we see.
You need to log in before you can comment on or make changes to this bug.