APHID Parallel Game-Tree Search Library
APHID is an acronym for Asycnhronous Parallel Hierarchical Iterative Deepening.
The APHID game-tree search algorithm attempts to accomplish many goals:
- to test out the viability of asynchronous search methods for parallel
alpha-beta based search.
- to develop an algorithm that works well across all hardware platforms
and applications.
- to implement a tool that can be easily inserted into legacy sequential
search algorithms.
Synchronous game-tree search methods have been employed throughout the
literature. A synchronous algorithm is one where all processors must
synchronize at a node within the tree before proceeding. For example,
Young Brothers Wait synchronizes at every PV node. However, APHID is
an asynchronous algorithm; it does not synchronize at the root of the
game tree.
The library is written in C, using PVM as the underlying communication manager,
and is available for beta-testers here. Included at
the FTP site is a version of Crafty that was used for my thesis
experiments.
Please e-mail if you have any
further questions.
The algorithm has been tested in many applications, including:
Publications
Mark G. Brockington-
Asynchronous Parallel Game-Tree Search
-
Ph.D. Thesis, Department of Computing Science, University of Alberta, November 1997.
-
Mark G. Brockington and Jonathan Schaeffer. -
The APHID Parallel alpha-beta Search Algorithm.
-
Accepted at Eighth IEEE Symposium of Parallel and Distributed Processing (SPDP'96), New Orleans, Oct. 23-26 1996.
-
Expanded version available as
Technical Report 96-07,
Department of Computing Science, University of Alberta, August 1996.
-
Mark G. Brockington and Jonathan Schaeffer. -
APHID Game-Tree Search.
-
Presented at Advances in Computer Chess 8, Maastricht, June 1996.
-
April 18, 1998