RFR: 8306738: Select num workers for safepoint ParallelCleanupTask [v2]

[Aleksey Shipilev]

On Thu, 27 Apr 2023 12:36:54 GMT, Axel Boldt-Christmas <aboldtch at openjdk.org> wrote:

> Alright, my only fear is underestimating the work for some workload, and providing a change where we increase the safepoint times due to selecting to few workers.

I feel pretty confident that we would not underestimate. The improvements I am seeing in local experiments provide enough cushion for little additional local costs, I think. But if we do underestimate in a significant way, that would be a simple fix to add more cases into `estimated_workers`, conditioned on the same "has work" function like the actual subtask :)

Running benchmarks before/after this fix with `-Xlog:safepoint` logging would be enough to see if we improve/regress the cleanup times in most cases.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/13616#issuecomment-1525633122