[erlang-questions] multicore performance fine grained concurrency

[Raimo Niskanen]

On Fri, May 07, 2010 at 11:41:27AM -0600, Tony Arcieri wrote:
> On Fri, May 7, 2010 at 1:27 AM, Raimo Niskanen <
> raimo+erlang-questions@REDACTED<raimo%2Berlang-questions@REDACTED>
> > wrote:
> 
> > There are many different reasons why SMP, especially SMP benchmarks
> > (as Kenneth explained in another mail in this thread) performs poorly.
> >
> 
> Wouldn't some kind of scheduler affinity for processes that send a lot of
> messages to each other and don't consume a lot of CPU be helpful in cases
> like this?

That is part of the much bigger problem of how to schedule processes,
and for the ring benchmark it is an obvious criterion, but for the
general problem there are more (criterions) and to make a perfect
in every situation scheduling algorighm is presumably impossible.

We are of course aiming to get scheduling good enough with as
little degenerated cases possible one day, but before that
day it will become possible to control some aspects, such
as processor affinity, by the application programmer, to
aid specific applications.

> 
> -- 
> Tony Arcieri
> Medioh! A Kudelski Brand

-- 

/ Raimo Niskanen, Erlang/OTP, Ericsson AB