High-performance message-based supercomputers have only recently emerged from the research laboratory. The commercial manufacturing of such products as the Intel iPSC, the Ametek s/14, the NCUBE/ten, and the FPS T Series - all based on multicomputer network technology - has sparked lively interest in high-performance computation, and particularly in the message-passing paradigm for parallel computation.
This book makes readily available information on many aspects of the design and use of multicomputer networks, including machine organization, system software, and application programs. It provides an introduction to the field for students and researchers and a survey of important recent results for practicing engineers. The emphasis throughout is on design principles and techniques; however, there are also descriptions and comparison of research and commercial machines and case studies of specific applications.
Multicomputer Networks covers such major design areas as communication hardware, operating systems, fault tolerance, algorithms, and the selection of network topologies. The authors present results in each of these areas, emphasizing analytic models of interconnection networks, VLSI constraints and communication, communication paradigms and hardware support, multicomputer operating systems, and applications for distributed simulation and for partial differential equations. They survey the hardware designs and the available software and present a comparative performance study of existing machines.
About the Author
Daniel A. Reed holds the Edward William and Jane Marr Gutgsell Professorship at the University of Illinois at Urbana-Champaign. He is also Director of the National Center for Supercomputing Applications (NCSA), Director of the National Computational Science Alliance, and Chief Architect, NSF TeraGrid.