Self-Similar Network Traffic and Performance Evaluation

Rating:
Summary: Book for new researchers
Review: Teletraffic Theory was laid the foundation by A.K.Erlang at the beginning of 20 century (1908-1918). It's a voice traffic theory. How do you guess,since then anything changed? Undoubtedly. In the modern networks prevail packet traffic. It gave rise requirement to research new form of one - self-similar. Corresponding measurements, collecting statistics and following investigations was reflected in the papers, articles, projects. And, finally, book is being published. It summarized done work. But this book isn't only summarizing, but also excellent assistant for young researchers, devoting himself to study and development new derection of Teletraffic Theory.

Rating:
Summary: Good compilation of recent results in the field
Review: This book is a compilation of articles written for the performance analyst and network modeler who is needing an overview of the recent results in this area. The book does include articles on both the theoretical and experimental results in self-similar network traffic and this makes it useful from a model validation viewpoint. Each article also has numerous up-to-date references for the reader who needs more in-depth information. The first article gives a general overview of the subject along with relevant definitions for those who may be approaching it for the first time. Theorems are stated but not proved but references are given for the proofs. Wavelets are discussed in Chapter 2; the most useful part of this chapter is the discussion on how to generate fractional Brownian motion time series using wavelet expansions. Chapter 3 gives a very quick overview of network simulations with heavy-tailed distributions. This chapter is a little disappointing in that the authors do not discuss how to deal with heavy-tailed distributions (with their large moments) computationally. This is important for developing network simulations and models using these distributions.A discussion of approximations of heavy-tailed distributions by other more manageable distributions would have been helpful here. The next chapter talks about fractional Brownian motion and could be viewed as a first glance at the theory of large deviations. The author does a good job of showing that analytical results are available in these kinds of processes. More general results in Gaussian input processes are discussed in Chapter 5, with asymptotics of queue length distributions characterized in some detail. The M/G/1 model is discussed in Chapter 6, where it is assumed that the service time is heavy-tailed. The most interesting part of this chapter is the discussion of the limit theorems. Chapter 7 gives a good overview of performance degradation under heavy-tail sources; the authors show, interestingly, that the expected time for overflow is polynomially fast. The authors in Chapter 8, give asymptotic bounds for the buffer overflow probability given a self-similar source They do take the time to prove many of their results. M/G/Infinity input processes are discussed in Chapter 9 and the authors do a good job of using large deviations to give asymptotic bounds on these kinds of queuing models. In Chapter 10, the author discusses subexponential distributions. These results were very new to me but the author did I think do a good job of explaining them. Queuing with on/off sources is discussed in Chapter 11, and the author derives many of the results using the Mellin transform. Most importantly he does compare his results with simulation experiments. I did not read the next chapter on VBR so I will omit its review. Just perusing though it seems as though the author holds that long-range dependence is irrelevant for VBR. Chapter 13 discusses transient losses and the impact on performance. The authors do a great job of comparing the difference in performance between Markov and long-range dependent processes. Actually the most interesting part of this chapter is the use of integral equations to prove some of the results. The results in Chapter 14 are more practical, where the authors use network simulations to study the problem of file size distributions drawn from heavy-tailed distributions in client-server models. The most useful part of this chapter is the comparison between the UDP and TCP as a moderator of the degradation experienced when self-similarity is increased. The best chapter for me was Chapter 15, which discusses TCP connection arrivals and their self-similar nature. The author does show how to model empirical distributions, particularly the Weibull distribution. I did not read the rest of the chapters so I will omit their review. This book is recommended for those involved in network modeling and performance. It is somewhat expensive, but worth it both as an introduction and as a reference source.

Rating:
Summary: The comprehensive book in the field!
Review: This is one of those books thats very comprehensive and the only one of its kind available in the new area of Self similar traffic modeling in Communication networks.

Self Similar traffic came to view in 1995 after the publishing of the "Self Similar Nature of Ethernet" paper by one of the authors of the book. Since then there have been a number of papers and book articles on self similarity and long range dependence, but not a single book fully devoted to this.

The book is very well organised and is a good handbook for all references to self similar network traffic. The first chapter provides a comprehensive introduction, which also serves as a pointer to the exact material one is looking for his research. The further chapters are well written.

As a graduate student doing research in this field, I find this book very useful.