EXPERIENCES WITH NONNUMERIC APPLICATIONS ON MULTITHREADED ARCHITECTURES

Distributed-memory machines have proved successful for many challengin g numerical programs that can be split into largely independent comput ation-intensive subtasks requiring little data exchange (although the amount of exchanged data may be large). However, many irregular applic ations - e.g. in the AI field - have a fairly tight data coupling that often results from the use of shared data structures, making them in many cases not amenable to parallelization on distributed-memory machi nes. EARTH is an efficient multithreaded architecture that supports in particular large numbers of small data exchanges by means of low star t-up times and the ability of tolerance of even small latencies. In th is paper, we show the benefits provided by EARTH for applications of t his sort by presenting experimental results from several AI applicatio ns run on the MANNA machine, which is a distributed-memory machine wit h a very high-performance communication network. EARTH-MANNA is shown to extend the range of programs that can be parallelized and run effec tively on distributed-memory machines.