Loom and high performance networking

[Viktor Klang]

For non-CPU-bound workloads using ForkJoinPool, my own personal experience is that setting the parallelism to around a factor of 0.7-0.8x the number of physical cores tends to provide a decent starting point for tuning.


Cheers,
√


Viktor Klang
Software Architect, Java Platform Group
Oracle
________________________________
From: loom-dev <loom-dev-retn at openjdk.org> on behalf of Alan Bateman <alan.bateman at oracle.com>
Sent: Wednesday, 14 August 2024 21:16
To: Robert Engels <robaho at icloud.com>
Cc: loom-dev <loom-dev at openjdk.org>
Subject: Re: Loom and high performance networking

On 14/08/2024 16:47, Robert Engels wrote:
> :
>
> Which leads me to the question does Java need a better (or dynamic)
> way of determining the optimum parallelism based on load? It seems the
> only way to set the parallelism is at startup, and it is constant.
>
VirtualThreadSchedulerMXBean provides a monitoring and management
interface to the virtual thread scheduler. It exposes the current
parallelism and allows it be changed. There are also methods to get the
thread count, and estimates for the current number of carriers and
number of virtual threads queued to execute. It may potentially be used
to dynamically adjust parallelism but nothing is using it at this time.

-Alan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/loom-dev/attachments/20240815/dc537bbb/attachment.htm>