Author: Robert Hyatt
Date: 13:33:56 12/09/97
Go up one level in this thread
On December 09, 1997 at 16:27:42, Stuart Cracraft wrote: >So I've read that the recommended strategy is >a two-tiered table based on brand new entries >and replacement based on the largest subtree >replacing smaller subtrees, in preference >to deeper subtrees. > >What other strategies are there that are worth >looking at? > >Are we talking only a few percentage improvement >or is the extra effort worth anything above that? > >--Stuart The main other approach is to use "buckets"... Ie something like set-associative cache, where you hash to a set of entries, and replace the set entry that is least useful. The drawback to this, is that the PC has a limited memory bandwidth, and trying to fetch several hash entries at one time overloads this and causes a bottleneck. This approach is faster when the table is heavily over-subscribed, but with today's large memory machines, this really is not much of a problem any more... And the two-tiered model works well since only one entry or two entries are fetched for each lookup and store...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.