"stopping" misbehaving fibers

[Ron Pressler]

The plan is to provide schedulers that are efficient for common use-cases, and
let people who so wish write their own custom schedulers. This means that
The built-in schedulers don’t respect priorities.

Now, I am not aware of any scheduler that can support priorities of millions of
threads efficiently and well (let alone when preemption can only occur at a
safepoint). Schedulers have trouble doing that even for thousands of threads.
Real-time OSs implement priorities well (i.e. with inheritance) but aren’t
super-efficient, while non-realtime OSs are quite efficient but don’t implement
priorities well.

If the rather partial way non-realtime OSs have priorities is good enough
for you, a simple custom scheduler can schedule virtual threads with different
priorities onto kernel threads with different priorities.

As to more sophisticated usages, if someone shows us it can be done well
and is useful enough, we’ll consider what kind of support we can have.


Ron




On 10 May 2020 at 23:33:06, Alex Otenko (oleksandr.otenko at gmail.com(mailto:oleksandr.otenko at gmail.com)) wrote:

> Ok. Is there a plan to support thread priorities?
>  
> Alex
>  
> On Sun, 10 May 2020, 22:52 Martin Buchholz, wrote:
>  
> >
> >
> > On Sun, May 10, 2020 at 2:44 PM Alex Otenko  
> > wrote:
> >
> >> But that's because the JVM was using resources the OS was aware of. The
> >> OS is not aware of one thread being more than one thread.
> >>
> >
> > The JVM certainly provides an OS-like abstraction that can be superior to
> > the native OS in some ways.
> > But it's a very hard job. Hosting multiple independent "applications"
> > within a single JVM process remains problematic.
> >