[erlang-questions] Parallel Shootout & a style question

Mats Cronqvist <>
Wed Sep 3 15:59:25 CEST 2008


Benjamin Tolputt wrote:
> Mats Cronqvist wrote:
>   
>> in the near future, we'll be using CPU's with 100's or 1000's of 
>> cores. see e.g. this;
>>
>> http://www.sics.se/multicore08_abstracts_bios
>>
>>   on such hardware, it'll be (almost) completely uninteresting how much 
>> CPU time we spend. the important thing will be to parallelize 
>> everything, and we can take for granted that parallel execution will 
>> "pay off". software technologies that doesn't makes this happen under 
>> the hood will be the COBOL of 2010's.
>>   
>>     
> The issue is that taking the direction you suggest *right now* will have
> the effect of making Erlang slower for most tasks on most platforms it
> is currently deployed on. 
  Will it? I doubt it would make a realistic application (e.g. a phone 
switch) significantly slower even on single core HW. Perhaps you're 
aware of some measurement that would prove me wrong?
 
  regardless, the only way to surf on Moore's Law is to parallelize. 
parallelizing should be made under the hood. and, frankly, i'm 
unimpressed by people who care about 10% effects on CPU time but are 
unwilling to move to multicore HW.

 mats



More information about the erlang-questions mailing list