[erlang-questions] Multiple core processors

[Ulf Wiger (TN/EAB)]

The "Big bang" benchmark, which has been used to test
the scalability of SMP Erlang in fact starts 1000 processes,
which then communicate in a broadcast fashion, each process
pinging every other process. In other words, it's not a 
simple ring benchmark.(*)

This benchmark actually shows super-linear scalability,
in the sense that with N processors, you get better than
N times the speedup, compared to the single-CPU case.

This was discussed briefly in 
http://www.erlang.org/ml-archive/erlang-questions/200605/msg00196.html

BR,
Ulf W

(*) The standard ring benchmark is pretty serial, as Richark points out.
It also has the curious property that it is perfect for the current
"hybrid heap" implementation, since it copies the message only the first
time it's sent, and the ring benchmark typically passes the same message
around thousands of times.  (:

Benchmarks are a tricky business...

> -----Original Message-----
> From: erlang-questions-bounces@REDACTED 
> [mailto:erlang-questions-bounces@REDACTED] On Behalf Of 
> Richard Carlsson
> Sent: den 3 oktober 2006 09:52
> To: erlang-questions@REDACTED
> Cc: Gerd Flaig
> Subject: Re: [erlang-questions] Multiple core processors
> 
> Gerd Flaig wrote:
> > Cameron Kerr <ckerr@REDACTED> writes:
> > 
> >> And is there a benchmarking process/document/result for 
> multicore-etc 
> >> machines. I have an opportunity to use a Sun UltraSPARC 
> T2000 running 
> >> Solaris and Linux, and would like to get an idea of the likely 
> >> performance metrics compared to classically threaded programs.
> > 
> > there has been some discussion about this in June:
> > 
> >    
> > 
> http://thread.gmane.org/gmane.comp.lang.erlang.general/15639/focus=156
> > 43
> 
> There are also some benchmark results in the R11 release notes:
> http://www.erlang.org/doc/doc-5.5/doc/highlights.html
> 
> > If I had to make a wild guess, I'd expect that the process ring 
> > benchmark would scale almost linearly with the number of processors.
> 
> Probably not, since the typical ring benchmark only has one 
> active process at a time. If there were multiple messages 
> being circulated simultaneously, things would be different.
> 
> 	/Richard
> 
> _______________________________________________
> erlang-questions mailing list
> erlang-questions@REDACTED
> http://www.erlang.org/mailman/listinfo/erlang-questions
>