Message Dissemination under the Multicasting Communication Mode

We discuss algorithms, complexity issues, and applications for message dissemination problems under the multicasting communication mode. These problems arise while executing in a parallel or distributed computing environment iterative methods for solving scientific computation applications, dynamic programming procedures, sparse matrix multiplication, etc. Our message communication problems also arise when disseminating information over ad-hoc wireless networks. Given a communication environment and a set of messages that need to be exchanged, the message dissemination problem is to find a schedule to transmit all the messages in the least total number of communication rounds. Generating an optimal communication schedule (with the least total number of communication rounds) for message dissemination problems over a wide range of communication environments is an NP-hard problem. To cope with intractability efficient message dissemination approximation algorithms have been developed for different types of communication environments and message communication patterns (the communications that must take place). The communication environment consists of the communication network (the direct communications allowed for each processor), primitive operations (the basic communication operations allowed by the system), and the communication model (possible operations during each communication round or step). Our goal is to present scattered research results developed during the last decade to establish that the multicasting (one-to-many) communication environment is a powerful communication primitive that allows for solutions that are considerable better than those achievable under the telephone (or one-to-one) communication environment. The multicasting communication environment has been available for quite some time in parallel computing systems. We also establish that, within the multicasting communication mode, forwarding plays an important role by allowing solutions that are considerable better than when restricting to direct communications, even when the communication load is balanced and the network is complete (all possible bidirectional links are present). We show that off-line communication scheduling allows for considerably better solutions over on-line scheduling. However, on-line scheduling provides added flexibility and it is applicable to a larger set of scenarios.