[Shootout-list] Re: ring of processes

Isaac Gouy igouy2@yahoo.com
Fri, 5 Nov 2004 07:02:09 -0800 (PST)


> > The deleted question was: So how do you propose to demonstrate that
> > these tests are workable across languages with different kinds of
> > concurrency - rather than being limited to a handfull of languages?
-snip-
> i think it is difficult/impossible for a test design to demonstrate
> how the claim "use less than 8k threads if it is too much" hold true.

I haven't asked for "a test design to demonstrate..." 

Timing results from programs in Erlang, Haskell?, Java, Perl would be
fine.

My concern is straightforward - I'd like to see timing measurements for
many languages rather than "death by concurrency" for many languages.

If we are to believe the current procinst measurements, we'd need
~5,000 processes to be created before the GHC program timings were
>0.01s

Using N=1500, 2000, 2500 on the current test machine would give
measurements below the timing resolution. 
Using N>x will give "death by concurrency" for many programs.



__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com