EXPERIENCE WITH PARALLEL SYMBOLIC APPLICATIONS IN ORCA

Orca is a simple, portable, imperative parallel programming language, based on a form of distributed shared memory called shared data-object s. This paper discusses the suitability of Orca for parallel symbolic programming. Orca was not designed specifically for this application a rea, and it lacks several features supported by specific languages for symbolic parallel computing, such as futures, automatic load balancin g and automatic garbage collection. On the other hand, Orca does give high-level support for sharing global state. Also, its implementation automatically distributes shared data (stored in shared objects). In a ddition, Orca programs are portable, because the language abstracts fr om the underlying hardware and operating systems. Efficient Orca imple mentations exist on a variety of parallel systems. We first give a com parison between Orca and two other models: imperative message-passing systems and functional languages. We do so by looking at several key i ssues in parallel programming and by studying how each of the three pa radigms deals with these issues. Next, we describe our experiences wit h writing parallel symbolic applications in Orca. We study the perform ance of each application on two platforms: the SP-2 and a collection o f SPARC processors connected by an ethernet. This work indicates that Orca is quite suitable for writing efficient and portable programs for symbolic applications.