[Shootout-list] X per second scoring system, resume

Einar Karttunen ekarttun@cs.helsinki.fi
Thu, 30 Sep 2004 15:32:06 +0300


On 30.09 14:06, Bengt Kleberg wrote:
> i could be wrong here. could you explain how an ''internal timer'' based 
> approach would be more easily disturbed by gc, compared to the current 
> shootout way of doing things?

The current method measures "How much time does it take to do N amount
of work." The new methology would be (if I understand correctly) 
"How much work was done in N seconds". 

Given two identical performance languages and different luck we get:

...WWGWWWGWWW... 8 work 
...WGWWWGWWWG... 7 work

W = work being done, G = garbage collection

This is made even further bad by generational garbage collectors with
patterns like:

WWWgWWWgWWWGGWWWgWWWg (where g < G)

- Einar Karttunen