Gamma Programming Paradigm and Heterogeneous Computing

The GAMMA programming paradigm is based on the multiset datastructure. In this model a succession of chemical reactions consume the elements of the multiset or multiset of tuples , and produce new elements according to specific rules. This paper describes how the Gamma paradigm can be integrated with the entity-relationship (ER) model to structure parallel and distributed programs. Such an integration can permit querying as well as application programming in a seamless fashion based on the notion of transactions and provides for heterogeneous processing. We also describe how to design a scheduler for the Gamma paradigm and how this paradigm is useful in heterogeneous programming. The word paradigm has several meanings: a recipe or a pattern or a scheme, or a grammar or a rule, or a theme. By a programming paradigm, we mean, see Floyd [l] a class of programs that solve different problems, yet have the same control structure pattern. That is we can first write a generic program (called a program shell) that implements the common control structure and include a few unspecified datatypes and procedures that can vary from one application to another. Then a parallel application program is obtained by substituting for the types and procedures needed for a specific application. The following features are desirable for a heterogeneous programming paradigm: (i) Foundational: It provides for proving the correctness of programs and its termination (ii) Operational: a.It separates control and compuational rules from parallelism. b.It abstracts away granularity level by associating parallelism to a generic concept of thread. Hence provides for deterministic I nondeterministic I probabilistic parallelism. c.It provides for fault tolerance and recovery. d.It can be abstracted into basic reader, searcher, writer processes that can be scheduled effectively. (iii) Architectural: It can be implemented using: a.Competing servers based on nondeterministic or probabilistic actions b. Synchronous servers pipeline that is deterministic (data parallelism) c. Network of multiprocessors with a sparse message passing network. d. Broadcasting among processors. (iv) Performance: Provides an improved scalable performance over sequential programs run on single processors. According to accepted wisdom, the craft of programming consists of: (i) Selecting a suitable data structure for information-bearing objects in the problem domain. (ii) Designing algorithms to manipulate the data structure to provide a state in which the answer is contained. The GAMMA programming paradigm, proposed by Banatre et al., [2] ,[3],[4] has all the above features. It is a closed and conceptually executable …