Created by: olefirenque
I would like to argue about the note "possible data-races can make it even more random":
- data-races are likely to disrupt the random distribution
- I suppose single random object still can cause contention by cache line collisions
I would also like to propose some heuristic for stealing single task. I think since it barely contends with other threads, it might be useful to do a few more iterations.
P.S: Are there any convenient deterministic benchmarks for comparing performance with such minor changes in xray?