Computer Chess Club Archives


Search

Terms

Messages

Subject: tip for "simulating" an MP computer & performance of ABDADA

Author: Tom Kerrigan

Date: 17:43:18 05/28/00


I found out that running multiple threads on one processor is not necessarily a
good way to simulate an MP computer. You have to make sure each thread only
searches one node per timeslice. If you're running Windows, you can do this by
calling Sleep(0) once per node.

I implemented Jean-Christophe Weill's ABDADA parallel search algorithm and I'm
very happy with it. In terms of node counts, I get the following speedups when
running BK:

Threads     Speedup
1           1
2           1.81
4           3.32
8           5.66

My implementation is pretty crude, so hopefully these numbers will improve over
time. I'm not sure how this algorithm compares to others, but I like it for a
few reasons:
1) it's very easy to understand and implement
2) it doesn't require any communication between threads, except for a fancy
shared hash table
3) if some threads barf, the search can still finish (this hasn't happened to me
yet, but it's cool for peace of mind)

-Tom



This page took 0.04 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.