Many books devoted to ATM or to ATM and broadband ISDN are now available. Some, but not all, of these books devote a chapter or two to ATM traffic management and congestion control algorithms. Even then, the treatment is almost exclusively a description of the algorithms, with little coverage of quantitative design and performance issues. With this book, the need for such coverage is now admirably met. Focusing on modeling and performance analysis in high-speed ATM networks, the book has two goals: first, to provide an up-to-date survey of ATM performance issues, and second, to enable the reader to understand the large and growing body of literature on this subject.
The book begins with a description of the types of services provided by ATM networks and the types of traffic patterns to be expected for those services. The next chapter examines traffic characteristics in more depth. It uses Poisson arrival models to approximate real-world traffic. As the author points out, the book does not cover self-similar traffic models, although there is substantial evidence that much ATM traffic exhibits self-similar characteristics. Unfortunately, this omission means that the results presented throughout the rest of the book do not reflect self-similar traffic patterns. On the other hand, this is an evolving area, and there is no consensus on how to model and analyze such traffic.
The next chapter deals with two important aspects of ATM traffic management: admission control and access control. Admission control is essentially a decision function: Given a request for a new connection, stated in terms of a desired quality of service (data rate, burst rate, delay, delay variation, loss probability, and so on), determine whether the capacity to support this new connection exists. Admission control is complex because of the desire to use statistical multiplexing for efficiency. That is, the admission control algorithm must attempt to accommodate as many connections as possible by assuming that not all connections will be transmitting at their peak rate simultaneously and that, therefore, it is possible to share capacity by allocating less than the peak rate to each connection. This problem is stated in all the books on ATM, but Schwartz provides a detailed analysis of the problem and shows how to determine a reasonable capacity to be assigned to each new connection. Next, access control (traffic enforcement) is discussed. Schwartz focuses on the several versions of the leaky bucket algorithm and evaluates their performance with various types of traffic.
The next chapter covers ATM switch design and buffering policies. A number of ATM switch architectures are analyzed, and the performance implications of input versus output queueing are modeled.
The remaining two chapters examine end-to-end performance issues. Chapter 6 deals with bounds on performance and the effective capacity of a network. The issue addressed is how to generalize from performance results at a single node to end-to-end performance for a set of competing virtual connections.
Chapter 7 is devoted to feedback control of congestion. The most important example of this technique is the rate-based control algorithm used in the available bit rate (ABR) service. This is a comparatively new scheme, and little quantitative material is available. Schwartz clarifies the issues involved and gives performance results. The rate-based technique is also compared to a window-based technique, which was an alternative considered for ABR but subsequently rejected.
Finally, to make the book self-contained, a tutorial on queueing analysis is included as an appendix.
Schwartz has packed an impressive amount of material into this medium-sized book. The book is by no means an easy read and will require quite a bit of effort to master, but the reader who persists will gain a good grasp of this field and an ability to understand its ever-growing literature.
