The Great Computer Language Shootout

Thomas Lindgren <>
Wed Aug 18 12:15:41 CEST 2004


--- Bengt Kleberg <> wrote:

> greetings,
> 
> i have written replacements/updated the erlang
> programs that did not 
> work in The Great Computer Language Shootout. as can
> be seen on the 
> erlang page
> http://shootout.alioth.debian.org/lang/erlang all my
> entries 
> (except one) are worse than the erlang average :-(
> 
> however, it might be of interest for someone to
> start making things 
> better now that everything works.

While the value of microbenchmarks can be questioned,
and a lot of the examples probably are not Erlang's
forte, I will fall for the temptation to make some
remarks:

- Erlang is often a factor 10-100 or worse(!) slower
than the best. A partial explanation: the test harness
starts a new VM for every benchmark. Still doesn't
explain it all. (Since runtime varies a bit :-)

- Ocaml, stalin and suchlike are often among the best.

- For some reason, Hipe is sometimes slower than
vanilla erlang. Due to loading native code at startup?

One factor is the source code, as Bengt notes. It
could perhaps be optimized further here and there.
Still, some of it seems straightforward enough.

Apart from that, on the cheerful side, it seems there
are still lots of tricks that an Erlang compiler could
use for speeding things up. At least on the micro
benchmark side :-)

Best,
Thomas



	
		
__________________________________
Do you Yahoo!?
New and Improved Yahoo! Mail - 100MB free storage!
http://promotions.yahoo.com/new_mail 



More information about the erlang-questions mailing list