RFR: 8322732: ForkJoinPool may underutilize cores in async mode

Viktor Klang vklang at openjdk.org
Fri May 10 08:07:16 UTC 2024


On Tue, 7 May 2024 22:50:18 GMT, Doug Lea <dl at openjdk.org> wrote:

> This set of changes address causes of poor utilization with small numbers of cores due to overly aggressive contention avoidance. A number of further adjustments were needed to still avoid most contention effects in deployments with large numbers of cores

src/java.base/share/classes/java/util/concurrent/ForkJoinPool.java line 1405:

> 1403:                     U.loadFence();
> 1404:                     if (base != b)
> 1405:                         ;                              // inconsistent

@DougLea This seems like it could be a `continue;`, which might be easier to reason about?

src/java.base/share/classes/java/util/concurrent/ForkJoinPool.java line 1412:

> 1410:                         if (stalled)
> 1411:                             Thread.onSpinWait();
> 1412:                         stalled = true;

@DougLea Performance-wise, does it make any difference to make the write to `stalled` in an `else-branch` to avoid writing it every time beyond the initial?

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

PR Review Comment: https://git.openjdk.org/jdk/pull/19131#discussion_r1596410561
PR Review Comment: https://git.openjdk.org/jdk/pull/19131#discussion_r1596413050


More information about the core-libs-dev mailing list