Parallel and Distributed Computing: A Survey of Models, Paradigms and Approaches

Rating:
Summary: A good start, but don't stop here
Review: This book is subtitled as "An all-inclusive survey of the fundamentals of parallel and distributed computing." It both succeeds and fails on this point. Leopold does indeed cover a wide expanse of technologies and approaches that characterize the space of high performance computing. It is in many ways still an emerging space, so conclusively nailing down every possible thread (no pun intended) in a coherent fashion is eminently difficult. The author's treatment of these different possibilities is uneven, overlooking some important contemporary technologies and implementations. It does cover a wide range of topics within the fields of distributed and parallel computing. Furthermore, within the chapters Leopold treats us to both high-level discussions of approaches and provides a glimpse into some of the implementation challenges involved. On the latter point especially, this book is very useful in that it gives the noninitiate some understanding and appreciation of the peculiarities of parallel programming, without requiring substantial technical background in the technologies. The examples in High Performance C and Parallel Fortran were very enlightening.

Where the book fails is that it is far from "all inclusive". There are a number of prominent and important developments that have not been included. Similarly, there are other interesting newer technologies that have only received cursory treatment. Examples include:

- No mention of SETI@Home. SETI@Home is the poster child of massively distributed computing, and with 15 teraflops of raw computing power, it is more capable than IBM's ASCI White supercomputer.
- No mention of distributed.net, or other notable exercises in public and commercial grid computing.
- Grid computing gets only a glancing reference at the tail end of one chapter. A comparative analysis of this important and still-forming space is glaringly absent from this text.
- JavaSpaces, Sun's answer to tuple-spaces, gets only a few sentences.
- Java RMI similarly gets less than a paragraph.
- Although DCOM is now basically legacy for Microsoft, it represents an important milestone in the evolution of distributed computing. It receives only a paragraph.
- Talk of web services and .Net would have been hitting the airwaves as the writing of this book as progressing, although possibly late in the effort. However, some cursory mention at least should have been made. There is increasing discussion of exposing grid compute services via web services interfaces, and Microsoft has recently announced their intention to port the Globus toolkit to Windows.
- Oh yeah, about Globus. Barely a mention.

It was clear from the text that the author came from a strong UNIX and CORBA background. The text has the feel of a PhD thesis-turned-book, and the areas of concentration are decidedly academic. There are a few other areas of minor complaint. Some of the wording in the text is clumsy, reflecting inadequate editing. Some topics feel like they are introduced in reverse order, assuming the reader already has some context about the given topic.

The author makes a sometimes-clumsy distinction between paradigms and models. The distinction is important in that an understanding of models brings a reader closer to envisioning how they might tackle a given problem themselves. However, reference to various models are sprinkled throughout the book. A comparative analysis, even brief, would have been very useful had it been centralized.

Those complaints may sound harsh, but overall the book is useful. It demystifies the problems of parallel programming, and provides a reasonably concise starting point for researching the distributed computing space. But, consider this book a starting point, and not an ending point.

Rating:
Summary: A good start, but don't stop here
Review: This book is subtitled as "An all-inclusive survey of the fundamentals of parallel and distributed computing." It both succeeds and fails on this point. Leopold does indeed cover a wide expanse of technologies and approaches that characterize the space of high performance computing. It is in many ways still an emerging space, so conclusively nailing down every possible thread (no pun intended) in a coherent fashion is eminently difficult. The author's treatment of these different possibilities is uneven, overlooking some important contemporary technologies and implementations. It does cover a wide range of topics within the fields of distributed and parallel computing. Furthermore, within the chapters Leopold treats us to both high-level discussions of approaches and provides a glimpse into some of the implementation challenges involved. On the latter point especially, this book is very useful in that it gives the noninitiate some understanding and appreciation of the peculiarities of parallel programming, without requiring substantial technical background in the technologies. The examples in High Performance C and Parallel Fortran were very enlightening.

Where the book fails is that it is far from "all inclusive". There are a number of prominent and important developments that have not been included. Similarly, there are other interesting newer technologies that have only received cursory treatment. Examples include:

- No mention of SETI@Home. SETI@Home is the poster child of massively distributed computing, and with 15 teraflops of raw computing power, it is more capable than IBM's ASCI White supercomputer.
- No mention of distributed.net, or other notable exercises in public and commercial grid computing.
- Grid computing gets only a glancing reference at the tail end of one chapter. A comparative analysis of this important and still-forming space is glaringly absent from this text.
- JavaSpaces, Sun's answer to tuple-spaces, gets only a few sentences.
- Java RMI similarly gets less than a paragraph.
- Although DCOM is now basically legacy for Microsoft, it represents an important milestone in the evolution of distributed computing. It receives only a paragraph.
- Talk of web services and .Net would have been hitting the airwaves as the writing of this book as progressing, although possibly late in the effort. However, some cursory mention at least should have been made. There is increasing discussion of exposing grid compute services via web services interfaces, and Microsoft has recently announced their intention to port the Globus toolkit to Windows.
- Oh yeah, about Globus. Barely a mention.

It was clear from the text that the author came from a strong UNIX and CORBA background. The text has the feel of a PhD thesis-turned-book, and the areas of concentration are decidedly academic. There are a few other areas of minor complaint. Some of the wording in the text is clumsy, reflecting inadequate editing. Some topics feel like they are introduced in reverse order, assuming the reader already has some context about the given topic.

The author makes a sometimes-clumsy distinction between paradigms and models. The distinction is important in that an understanding of models brings a reader closer to envisioning how they might tackle a given problem themselves. However, reference to various models are sprinkled throughout the book. A comparative analysis, even brief, would have been very useful had it been centralized.

Those complaints may sound harsh, but overall the book is useful. It demystifies the problems of parallel programming, and provides a reasonably concise starting point for researching the distributed computing space. But, consider this book a starting point, and not an ending point.