Dissecting Distributed Computations

Abstract

Beginning with the earliest assembly languages, programming even simple centralised and sequential applications was very cumbersome. Major advances in computing came with the adoption of programming languages that promoted the use of machine independent data abstractions like sets, arrays, and records, as well as control abstractions like while, ifand repeat. With the arrival of parallel machines and multi-threading, the need for additional abstractions that factor out the difficulty of concurrent computing has driven several research activities. These activities resulted in new abstractions like semaphores, monitors and rendez-vous. Not very long ago, these abstractions used to sit beside programming languages in the form of libraries, e.g., concurrency libraries in C++. They have now been integrated as first class citizens into modern programming languages, e.g., through synchronised methods in Java. Despite the many decades of research in distributed computing, abstractions for programming distributed applications are still lacking. Currently, the only abstraction that is of a relatively wide use in distributed programming is the RPC (Remote Procedure Call) [13.1]. Surprisingly enough, RPC actually aims at hiding distribution: the goal of RPC is to make an invocation of a remote object looks like the invocation of a local object. This illusion is however broken when the remote object is located on a machine that crashes or that becomes inaccessible because of a link failure [13.2].