The Computer Language Shootout

[Ulf Wiger]

Den 2006-03-26 12:47:52 skrev ke han <ke.han@REDACTED>:

> I would like to solicit a little further input from the erlang community  
> that can explain why erlang still performs slower than a few others.   
> The most interesting comparisons I would be interested in are to OCaml  
> and Java.  OCaml since it gets grouped with erlang as functional.  Java  
> since its a market leader and I have enough experience with Java to  
> better understand the rational.

OCaml has very strict static type system. This helps a lot, as no runtime
type checks are necessary.

Java is also statically typed, and allows for destructive update
of data structures. This can help in low-level benchmarks.
Furthermore, Erlang beats the socks off of both in the cheap-
concurrency benchmark:

cpu time    N    5,000  10,000  15,000
OCaml           252.56  515.45  757.96	
OCaml #2	    131.82  263.52  394.79	
JDK -server #2   65.74  95.71   194.60	
HiPE #2           1.84  3.59      5.18

memory      N    5,000  10,000     15,000
JDK -server #2  24,500  27,556     24,500	
OCaml           19,072  18,916     19,096	
HiPE #2          5,000   4,996      4,996	
OCaml #2         4,328   4,332      4,284

(And this is still for small N, as far as Erlang's
concerned...)

Java is not only slow in concurrency benchmarks.
It has a poor concurrency model as well. OCaml
relies on the concurrency support provided by the
underlying operating system, as far as I remember.

For an all-round contender within Erlang's domain,
you might look at Concurrent Haskell. I think Haskell
beats Erlang in expressiveness, and often in speed,
but esp. the support for concurrency and distribution
is somewhat less mature than in Erlang. You should find
more comparative data here:

http://www.cee.hw.ac.uk/~dsg/telecoms/

(at least eventually. Currently, is seems unreachable.)

BR,
Ulf W
-- 
Ulf Wiger