SCALABLE SUPERCOMPUTING: INTEGRATING DISTRIBUTED PROCESSINGIN A GENERAL PURPOSE PARALLEL PROGRAMMING LANGUAGE

THE OBJECTIVES OF THIS PROPOSED RESEARCH ARE TO DESIGN AND IMPLEMENT A PRELIMINARY VERSION OF A PARALLEL PROGRAMMING LANGUAGE THAT WILL: 1. INCLUDE EFFICIENT IMPLEMENTATIONS OF GENERAL PURPOSE CONSTRUCTS FOR BOTH SHARED AND DISTRIBUTED MEMORY PARALLEL PROGRAMMING, 2. BE CAPABLE OF INCREASING THEEXECUTION SPEED OF APPLICATIONS BY AT LEAST TWO ORDERS-OF-MAGNITUDE, 3. RUN ON A COMPREHENSIVE SET OF PROCESSORS, 4. ADHERE TO EXISTING STANDARDS AS CLOSELY AS POSSIBLE, 5. BE EASY TO USE AND UNDERSTAND, AND 6. REQUIRE MINIMAL RECODING OF EXISTING ALGORITHMS AND LIBRARIES. THE PROPOSED STRATEGY FOR THIS PROJECT IS BASED ON EXTENDINGAN EXISTING GENERAL PURPOSE PARALLEL PROGRAMMING LANGUAGE WHICH ALREADY SUPPORTS SHARED MEMORY PARALLELISM. THE LANGUAGE, TOP LEVEL COMMON LISP (TOPCL), WILL BE EXTENDED BY ADDING CONSTRUCTS SUPPORTING DISTRIBUTED MEMORY PARALLELISM. BY EXTENDING TOPCL'S EXISTING SUPPORT OF PARALLEL PROGRAMMING TO INCLUDE DISTRIBUTED MEMORY PARALLELISM (THE "NODE" LEVEL), A SINGLE APPLICATION WILL BE CAPABLE OF EXPLOITING THE COMPUTATIONAL RESOURCES OF ENTIRE NETWORKS OF CONVENTIONAL COMPUTERS, INCLUDING SHARED MEMORY MULTIPROCESSORS. AS A RESULT, SOFTWARE DEVELOPERS WILL BE ABLE TO ACHIEVE SUPERCOMPUTER PERFORMANCE WITH NETWORKS OF LOW-COST HARDWARE. FURTHERMORE, TOPCL CURRENTLY SATISFIES MANY OF THE STATED OBJECTIVES. LEVERAGING UPON THE EXISTINGSYSTEM WILL MINIMIZE THE COSTS AND RISKS OF THIS PROPOSED PROJECT AND AT THE SAME TIME MAXIMIZE THE POTENTIAL FOR SUCCESS.