[erlang-questions] Erlang performance on Windows

[Jesper Eskilson]

On Mon, Nov 24, 2008 at 10:55 PM, Kostis Sagonas <kostis@REDACTED> wrote:

>
>>> In summary, HiPE (native code generation) is great, if you use it with
>>> care you can get outstanding performance.
>>> You don't need HiPE in order to get good or competitive performance with
>>> Erlang.
>>
>> I'll run the corresponding program on my Linux-box and see what kind
>> of performance difference we're really talking about.
>
> I would be interested to see the numbers.  Especially since on that box you
> can factor out any possible "Windows surprises" and quantify the performance
> improvements that native code compilation provides for you.

Well, here's a surprise I was not expecting. I tried running a
slightly optimized (thanks to Richard C) version of my sieve-function
on my laptop with hipe (Ubuntu 8.10 on a dual-core running at 2.2GHz)
and compared it with my workstation (XP on a quad-core running at
2.4GHz), i.e. without hipe, and hipe did not give any speedup at all.
The total runtime was lower on the Windows box (1700 msecs vs. 1900
msecs on the linux machine), but then it does run at a higher clock
frequency.

Yet again benchmarking proves to be an area where you should never
assume *anything*.

(BTW: is there a way to switch off hipe? I couldn't find anything in
the erl man-page.)

-- 
/Jesper (a little wiser)