This talk was pretty interesting: Probably not at a point where we focus on optimization yet, but when we get there this causal profiler will likely be useful. https://github.com/plasma-umass/coz