Customization of First Class Tuple-Spaces in a Higher-Order Language

A distributed data structure is an object which permits many producers to augment or modify its contents, and many consumers simultaneously to access its component elements. Synchronization is implicit in data structure access: a process that requests an element which has not yet been generated blocks until a producer creates it. In this paper, we describe a parallel programming language (called T S) whose fundamental communication device is a signiicant generalization of the tuple-space distributed data structure found in the Linda coordination languagee6]. Our sequential base language is a dialect of Schemee19]. Beyond the fact that T S is derived by incorporating a tuple-space coordination language into a higher-order computation language (i.e., Scheme), T S diiers from other tuple-space languages in two important ways: Tuple-spaces are rst-class objects. They may be dynamically created , bound to names, passed as arguments to (or returned as results from) functions, and built into other data structures or tuples. The behavior of tuple-spaces may be customized. A tuple-space is manipulated via a policy closure that speciies its operational characteristics. The representation of policy closures take signiicant advantage of Scheme's support for higher-order functions; there is no fundamental extension to Scheme needed in order to support them. We argue that rst-class customizable tuple-spaces provide an expressive and exible medium for building parallel programs in higher-order languages .