Abstract

Although efficient support for data-parallel applications is relatively well established, it remains open how well to support irregular and dynamic problems where there are no regular data structures and communication patterns. Tree search is central to solving a variety of problems in artificial intelligence and an important subset of the irregular applications where tasks are frequently created and terminated. In this paper, we introduce the design of a multithreaded distributed runtime system. Efficiency and ease of parallel programming are the two primary goals. In our system, multithreading is used to specify the asynchronous behavior in parallel game tree search, and dynamic load balancing is employed for efficient performance.

Citations