RE: A proposal for the improvement of the delay pools

From: Chemolli Francesco (USI) <>
Date: Mon, 15 Apr 2002 10:49:41 +0200

> On Monday 15 April 2002 09:33, Chemolli Francesco (USI) wrote:
> > It depends on the approach again. With "yours" it mostly relies on
> > configuration parameters, with "mine" the overhead is all in
> > keyed (or "class" as I called them) pools, and it means basically
> > an hashtable or a splaytree or whatever accessor is faster for
> > each defined pool.
> I would say the two approaches are identical there. The best indexing
> method depends on the "class attribute" or "key" used.

My point exactly.

> The difference between the two is mainly tree or matrix hierarchy.
> With tree hierarchy and "first match", then the most common kinds of
> configurations becomes quite easy. The matrix approach allows for
> some increased flexibility, but puts more demands on the acl
> configurations.

Yes. Not that much in terms of definition IMO, but in terms of
performance there is a definite O(n/2)*acl_eval_cost hit
for each request, where N is the number of pool associations
defined in the configuration, plus extra costs in pool useage:
two passes in a linked list versus single recursive
leaf-to-root tree walk are the requirements for each write

Received on Mon Apr 15 2002 - 08:36:44 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:15:10 MST