Synchronizing operations on multiple objects

Abstract

Parallel programming on distributed memory systems is one of the most challenging research areas in parallel computing today. Objectbased parallel programming languages are an important class of languages for such systems. Shared objects allow the programmer to deal with data partitioning, communication, and synchronization in a highlevel manner. Synchronizing operations on a single object is well. understood. Dealing with synchronization on multiple objects distributed over the available processors, however, is still an open issue. In this paper, we will present an abstraction, cMled weavers, that is used to synchronize operations on multiple objects, and show how weavers are applied in a runtime support system for atomic functions on multiple objects.