Load Balanced Mapping of Distributed Objects to Minimize Network Communication

The use of distributed software and hardware architectures is widespread and accelerating. In order to utilize a distributed architecture fully, software engineers need to develop cooperating software modules running on different machines. One well-known approach to constructing distributed applications is through the use of remote procedure calls (RPC’s). The RPC model is based on the socalled client–server paradigm, and is similar to local procedure calling. A major difference from the local call is that an RPC allows a program to call procedures on a different computer or in a different address space on the same computer. The call is sent in the form of a request message to a remote server process, which sends back a reply message to the client. In contrast to local procedures, remote procedures cannot share the client program’s address space, and as a result, input and output parameters have to be copied between processes. The RPC construct is simple and powerful because it provides a transparent remote calling mechanism which is semantically similar or identical to the traditional local call mechanism. A great many distributed computer applications have substantial computation and network communication requirements. For instance, a number of hardware and software vendors have recently announced strategic commitments to multimedia computing—a distributed application area well known for its use of CPU-intensive algorithms and bulky data for representation, manipulation, and transmission of images, voice, and video [1]. Distributed multimedia systems can be considered a natural extension of today’s distributed systems. Typical examples are computer-supported education, desktop multimedia processing, office and hospital information systems, information kiosks, point of repair, browse, and sell, and entertainJOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING 34, 117–136 (1996) ARTICLE NO. 0050