"big" benchmark performance (was Re: Erlang/OTP R11B has been released)

[Paul Brown]

Hi, Ulf --

> > Erlang/OTP R11B has been released as planned on May 17, 2006.
> Any explanation as to why the "big" benchmark runs 5 times faster on 4 schedulers than on one (AMD Opteron 2 dual core) and 2.8x faster on 2 cores (AMD Athlon 64 X2)?

I did some fiddling with the "big" benchmark on my four-core system
(2x2code, 2.5GHz PPC), and the results are here:

http://mult.ifario.us/articles/2006/05/11/more-on-erlang-performance-and-threading

(URL may get broken; here's a short one -- http://tinyurl.com/m75l3)

The answer is that you can't think purely about hardware, as the layer
below the operating system has a lot to do with the way that threads
(directly correlated to Erlang schedulers) are allocated and
scheduled.   So, what you're seeing is that the total throughput with
four schedulers is substantially higher than with two because the
operating system is able to efficiently schedule four threads for
execution.

My $0.02 from the cheap seats.

-- 
paulrbrown@REDACTED
http://mult.ifario.us/