This paper presents the results of an investigation of collective communication operations for distributed computing across asynchronous transfer mode (ATM) networks. Several collective operations have been implemented and studied on a three-switch ATM network testbed at Michigan State University. The methods use virtual topologies constructed from ATM virtual channels. A particular type of virtual topology is described that efficiently implements several collective operations through the use of hardware-supported ATM multicast channels. Performance measurements are presented that illustrate how a thread-based software design can take advantage of such underlying hardware features.

This paper defines and describes the properties of a multicast virtual topology, the M-array and a resource-efficient variation, the REM-array. It is shown how several collective operations can be implemented efficiently using these virtual topologies, while maintaining low complexity. Because the methods are applicable to any parallel computing environment that supports multicast communication in hardware, they provide a framework for collective communication libraries that are portable and yet take advantage of such low-level hardware functionality. In particular, the paper describes the practical issues of using these methods in wormhole-routed massively parallel computers (MPCs) and in workstation clusters connected by Asynchronous Transfer Mode (ATM) networks. Performance results are given for both environments.

The paper presents the results of an investigation into the efficient implementation of reduction operations for cluster-based parallel computing across asynchronous transfer mode (ATM) local area networks. The study combines graph theoretical analysis with experimentation on an ATM network. Different reduction algorithms are analyzed in terms of the amount of message traffic produced and the number of message-passing steps required. This analysis, which indicates how to take advantage of switch-based interconnects and hardware multicast communication, is applied to the development of both N/1 and N/K reduction protocols. Performance measurements from implementations on a three-switch ATM testbed are presented to support the analytical results.>

This paper presents the results of an investigation into the efficient implementation of multicast operations for cluster-based parallel computing on asynchronous transfer mode (ATM) networks. Both software- and hardware-based multicast operations have been implemented and studied on a three-switch ATM network testbed. Performance measurements are presented that illustrate how software approaches can best take advantage of switch-based network architectures, and what additional advantage can be gained from using underlying hardware support.>