As the series editor states, Reed and Fujimoto have written the first comprehensive treatment of the architecture and performance modeling of those non–shared-memory MIMD computing systems in which a large number of processors (with large private memories) are connected so that they can exchange messages efficiently. This book is a very good survey. Although it is not a textbook with examples and exercises in the usual sense, it could certainly be used in a suitable course. The writing is clear and straightforward; both students and researchers would benefit from a book like this. The references seem to be quite adequate, and I did not find any typos. A chapter-by-chapter summary follows:
(1) “Introduction.” The authors define a multicomputer network and compare it with other parallel architectures. A history of multicomputers is given and multicomputer building blocks are described.
(2) “Analytic Models of Interconnection Networks.” Basic definitions and formulas are given. The authors describe single-stage interconnection networks and examine the effects of certain design decisions on network performance and reliability.
(3) “VLSI Constraints and the Optimal Number of Ports.” Physical design constraints and performance-related issues are covered. The central result of this chapter is that a small number of high-bandwidth ports is preferable to a large number of low-bandwidth ports.
(4) “Communication Paradigms and Hardware Support.” This chapter covers design alternatives for important issues (transport mechanisms, routing mechanisms, buffer management, and flow control). Simulation analyses of important communication component design parameters are given.
(5) “Multicomputer Network Operating Systems.” After an overview, the authors discuss the scheduling of static and dynamic tasks. Simulation studies are included.
(6) “Applications: Distributed Simulation.” This chapter discusses the simulation of discrete systems and gives distributed simulation strategies.
(7) “Applications: Partial Differential Equations.” The authors give a formal method for analyzing the performance of a stencil-and-grid partition when the partitions are mapped onto the nodes of a multicomputer network.
(8) “Commercial Hypercubes: A Performance Analysis.” The hardware and software organizations of one research and several commercial multicomputers, each configured in a hypercube topology, are considered.