Product: 0131448536

Cluster supercomputers became originated when NASA had a need for another supercomputer but no budge to buy one. The particular tasks that they needed to perform tended to have a lot of parallelism. That is, they could benefit by having a bunch of systems work on small tasks and then bring the results together. An example of an ideal parallel application is provessing seismic data. Basically you have the signals to process from a bunch of sensors and each data stream is independent of the other until the end of the process. An example of a non-parallel system is inventory. When you are increasing or decreasing inventory, you only have one bin of this part, and two CPU’s better not sell the same part twice.

Anyway Don Becker and some others proposed using a cluster of PC’s to handle the tasks. They wrote a software package called Beowulf to run under Linux. Linus had two real advantages: stability and cost. It’s much more stable than Microsoft systems and it’s free. This, as is often the case in the computer biz, started a landslide.

This book is the most complete I’ve read on how to build a cluster. It goes extensively into every aspect. It talks about the architecture of the various CPU’s in use (Itanium, Xeron, Opteron), high speed interconnection systems (Myricom, Infiniband, Dolphin, Quadrics), available software (Suse, Red Hat, management systems) – in short everything you need to get the system up and running.

If you’re new to clustering, here is an integrated, step by step approach to getting the system up. Of course you’ll need more books later (FORTRAN, Message Passing Interface, etc.) but this is the place to start.
Rating: 5
0131448536
Count: 5