Summary: | Communication through shared objects is a simple model for parallel programming of distributedmemory machines. We describe our experiences with this model for implementing a parallel algorithm for the Arc Consistency Problem. We provide performance measurements of the program on two distributed systems, which use the same network (Ethernet) and operating system (Amoeba), but differ in processor speed and size of local memory. Keywords: Shared objects, Orca, Amoeba, distributed systems, broadcasting 1 Introduction Collections of workstations connected by a local area network are attractive for running parallel programs. There are, however, two problems in exploiting their power: the lack of easy-to-use parallel programming systems and the communication overhead. In our research we advocate the use of a simple model for parallel programming, called shared dataobjects [1]. Its greatest advantage is that it abstracts from the underlying distributed system, and offers the programmer a .
|