Experimental Implementation of Parallel TRAM on Massively Parallel Computer

Kazuhiro Ogata and Hiromichi Hirata anShigenori Ioroi and KokicFutatsugihi d

Abstract
Parallel TRAM is an abstract machine for term rewriting systems thatcontrols parallel rewriting. One may design and implement an efficientprocessor for algebraic specification languages based on Parallel TRAM.Parallel TRAM was originally designed to be executed on multiprocessors.We have improved it with the message passing model as its parallelcomputational model so that it can be efficiently implemented on massively parallel computers and workstation clusters whenever MPI is available. The improved Parallel TRAM is called PTRAM/MPI. We have also implemented PTRAM/MPI on Cray T3E carrying 128 processing elements. The preliminary experiment using a set of rewrite rules to compute Fibonacci numbers showed that PTRAM/MPI was 57 times as fast as TRAM.
Contact
Kazuhiro Ogata
Graduate School of Information Science,Japan Advanced Institute of Science and Technology,Tatsunokuchi, Ishikawa 923-1292, JAPAN,
ogata@jaist.ac.jp