In Search Of Clusters

In Search of Clusters (The Ongoing Battle in Lowly Parallel Computing (2nd Edition) by Gregory Pfister, Prentice Hall, 1997 ISBN 0138997098 or try ISBN 0134376250

This is an easy-on-the-brain introduction to the notion of clustering computers.

Traditionally, parallel processing (or even distributed processing, or even dealing with concurrency issues) is an arcane subject, full of tricky side-passages and the occasional assertion that, in order to correctly design code, you must use the correct modal logic and thrice ask for the god's blessing (If memory serves, the appropriate logic is S5).

Clusters are a recent notion in parallelism. The basic notion is that there are two dominant trends in computing: the ever-faster single processor, and the ever-increasing need for reliability.

Faster processors means that we don't need 65,536 processors wired together in a liquid-nitrogen cooled box to solve difficult problems.

And the need for reliability means that we do need at least one more processor, preferably in a different box. And that this better be easy to set up / use (because the need is widespread and there ain't but a few people who can hang with the earlier notions of parallelism).

Which is, sorta, what clusters are all about.

The book is interesting, funny, and manages to cover an awful lot of information without either becoming overly technical *or* glossing over difficult issues.

I'm not saying it'll be the last book you'll ever buy on clustering or parallel programming. But it probably ought to be the first.

WilliamGrosso

If I'm understanding what "clusters" means correctly, for a real-world example of clusters at work, take a look at http://www.distributed.net/ which uses idle time on Internet-connected computers to solve DES/RC5 keys. --KatyMulvey

Oh, there are billions of examples. Many in the real world and many announced. That's part of the point of the book (which, alas, I neglected to mentioned). While academics and other researchers mostly focused on the "crunch a hard problem" notion of parallelism (things of the form "simulate the world's weather in real time"), out in the world of everyday computing different notions of parallelism arose. Clusters being one of them.

The book starts with a genial introduction and then, in chapter 2, has an extensive list and discussion of examples. The idea being that a long list of specific instances helps to ground the rest of the discussion.

WilliamGrosso

Pfisters book is very enjoyable if you are into this stuff. He is obviously a man of great experience, which he seems to enjoy passing on to others.

He explains well how cluster systems are both similar to, and different from, both parallel systems and distributed systems.

RichardCollins

CategoryBook