Computer Language Shootout - concurrency

[Brent Fulgham]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


On Dec 6, 2005, at 11:19 PM, Nic Linker wrote:

> Mikael Pettersson wrote:
>
>> Another data point: they list HiPE as being about 18% slower than  
>> BEAM
>> on the random benchmark. On our P4 Xeon, not using the performance
>> counters just the regular time command, HiPE is instead 18% faster  
>> than
>> BEAM. (Which still is disappointing.)
>>
>> So I'm not sure I have much confidence in their benchmarking  
>> methodology.

Hi,

I am one of the people (culprits) responsible for the revived  
shootout.  I think the basics of the shootout are sound, though I  
must admit that many lively debates have been had on the subject.

Just to be on the safe side, I reran all Erlang benchmarks, and the  
results are posted.  Apparently, the Erlang compiled output was being  
used in preference to the Hipe compile, resulting in the bad results  
you reported (see http://shootout.alioth.debian.org/sandbox/ 
benchmark.php?test=all&lang=hipe&lang2=erlang).  In most cases, Hipe  
is a clear win, but there do seem to be a few edge cases (such as  
Random).

I can't speak to why you see such a large difference from my  
measurements.  The measurement scripts launch a child process and  
measure the CPU times (system, user, and "real") using the standard  
Perl module (BSD::Resource).

I invite your comments (and criticism).

Thanks,

- -Brent


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (Darwin)

iD8DBQFDl9EUzGDdrzfvUpURAqhcAKCAQfpz6QyFCcq8qduGOELG1YkGwACcCCXY
6CxIYAvFCMPdQ+2ZdEaALo0=
=V0g/
-----END PGP SIGNATURE-----