Author: Robert Hyatt
Date: 07:48:15 01/07/04
Go up one level in this thread
On January 07, 2004 at 03:50:23, Richard Pijl wrote: >On January 06, 2004 at 16:18:58, Robert Hyatt wrote: > >>On January 06, 2004 at 15:32:49, David Rasmussen wrote: >> >>>I just got my wish for a thesis subject granted. I will write about >>>parallelizing a chess program, or more specifically, the alpha beta like >>>algorithm in a chess program. So my hobby (Chezzz) has now become my "job". >>> >>>But the precise details of the project are not ready yet. I will have to decide >>>what exactly to focus on. One idea is to implement several forms of >>>parallelizing and test their relative effeciency: with and without Young >>>Brothers Wait, that scheme where two threads are searching the same, but they >>>benefit because they share hash, AABAB or something? >>> >>>Those of you who have fiddled with parallelizing: Do you have any ideas about >>>what to implement or examing etc. ? >>> >>>/David >> >>I don't recall the acronym either, but I think it might be ABBADA, and I would >>forget about it personally. Depending on serendipitous communication thru the >>transposition table to make the search faster leaves me cold. It would be nice >>if it were that easy, but alas... > >ABDADA is nice if you want to have SMP capabilities fast, without having to >rewrite major parts of your program. Adding SMP capabilities took me just a few >days, then a few weeks more for testing and rooting out problems. With some >adaptations I got a fair speedup, for the amount of work I put in it (about >1.5x). I agree that if given the time a proper splitting method is preferable >though. >Richard. 1.5 is not bad for a dual. But the implementations I have seen get stuck around that speedup even with 4+ processors, which is why I don't particularly like it. It reminds me of the old proverb "if something sounds too good (easy) to be true, it probably is." :) I don't believe there is a "quick and dirty" approach to SMP that will work well. > >>Best approach is a YBW-like approach, ie something like what I do in Crafty. >>If you don't at least to YBW, your search overhead will go up. The DTS >>approach I did in Cray Blitz had a touch of YBW, but it was better overall, >>but also _way_ more complicated.
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.