Hi, we are Joy Sarkar, and Dr. S. Balasubramanian , (bala to all in JNC) and this is an account of our developing the first Beowulf Cluster in JNCASR. For the uninitiated, Beowulf Cluster is the generic name given to a low cost and viable alternative to a supercomputer or a parallel processing computer. It consists of an array of many computers that have been configured to compute the same problem parallely, or in other words, in tandem. The catch is that for a Beowulf cluster, all commodities are mass-market (COTS:-Commodities Off The Shelf, to the Beowulf community) so that the building cost of the cluster comes out to be very low compared to a commercial parallel supercomputer like the Cray.
                                In this web document we wish to tell you about our adventure of bringing kamadhenu to life! This document is not meant to be a HOWTO or a manual, only an informal  retrospective. However, I(JS) have tried to document all the challenges that we faced and how we overcame them, with the hope that perhaps it will help a couple of fellow researchers like us to beat the same odds with a little less effort !
 

First, the specifications:

Design:

The cluster consists of a fully-configured PC called the master, and 7 other headless (read diskless) nodes that provide the other processors for parallel computing.
                    All the nodes consist of processors sitting on motherboards and connected to each other (and to the master) with network cards. All the nodes are diskless, i.e. they do not have hard-disks or floppy disks. Other than the processors and motherboard, the only important component that the nodes have is the all important Network Interface Card (NIC). The master additionally has 2 hard-disk-drives, 1 floppy disk-drive, 1 CDROM drive and an extra network card for connectivity to the LAN.

 Components: