Jetty and Loom

[Ron Pressler]

Got it.
We’ll try to reproduce the problem and fix it when we get back to fixing performance issues,
hopefully in a few weeks.

— Ron


On 5 January 2021 at 20:40:17, Greg Wilkins (gregw at webtide.com) wrote:



On Tue, 5 Jan 2021 at 17:45, Ron Pressler <ron.pressler at oracle.com> wrote:
So, just to make sure I understand, for the same thread count, you don’t see a SLOW for either kind of thread,
and you only see it for virtual threads at a higher count?

Not exactly.

Kernel threads ran without significant pause up until 32K threads, which it took 55s to get to

Virtual threads ran up to 10K threads in 29s, at which time there was a SLOW thread start, taking 1.2s from spawn to max stack.  It then ran to 32K in 51s, which demonstrates that even with the pause it was faster to start threads. This lines up nicely with the 120ms spawn advantage I measured as *32K is approx 4s.    It then runs to 43K before a 30s SLOW dispatch just before it hits max heap.

cheers

--  
Greg Wilkins <gregw at webtide.com> CTO http://webtide.com