On Wednesday, August 27, 2003, at 10:40 PM, matthew green wrote:
>    On Wed, Aug 27, 2003 at 11:47:19PM +0100, Richard Earnshaw wrote:
>> GCC 3 makes fairly heavy use of MAP_ANON.  Can you try a build with 
>> that?
>
>    Well, I did that. No differences again:
>
>    non-rbtree  real 8m57.498s  user 7m49.490s  sys 0m48.121s
>        rbtree  real 8m57.650s  user 7m50.497s  sys 0m48.430s
>
>
> hmmm.. this is the second real world figure, and while the difference
> is absolutely tiny, the rbtree version is once again a slightly slower.
> .3 seconds isn't much over nearly 8 minutes, but it is something.
>
> i think the rbtree code isn't worth while.  i'm certainly not 
> convinced.

I think the difference for us is Andrew Brown's improved vm map entry
merging code so that NetBSD tends to have less entries to search such
that the rbtree code is really never used.  This is why I asked for
benchmarks.
-- 
Matt Thomas                     email: matt@3am-software.com
3am Software Foundry              www: 
http://3am-software.com/bio/matt/ Cupertino, CA              
disclaimer: I avow all knowledge of this message.