Beowulf clusters

The definition of Beowulf clusters might be strict but I see Beowulf clusters as a set of machine used for scientific and/or technical computations. A typical Beowulf cluster consists of

- a frontend (two Ethernet adapters) - a set of computational nodes - high speed interconnection (network switches and cabling)

Often the frontend acts as file server (NFS) and manages the users and their rights.

Building and installing a Beowulf cluster is not rocket science but if you are going to install a 20 nodes cluster you do not wish to do it by hand. In the recenct years a number of specialized Linux cluster distributions have been developed. In recent time I have tried Rocks Cluster, Debian Cluster Components, HP XC, and IBM CSM.

Rocks Cluster is a highly specialized Linux distribution for Beowulf clusters. It is based on the CentOS distribution which is repackaging of Red Hat Enterprise Linux (with only freely distributable software is included). Moreover, Rocks Cluster has a concept called rolls. A roll is a metapackage i.e. a collection of packages which provides a certain functionality. Another useful concept is appliances. Each server - or node - in the cluster has to be at least one type of appliance. A cluster has compute nodes and a file server - both types are appliances.

Hewlett-Packard XC is based on Red Hat Enterprise Linux i.e. all the open source components. You seems to be locked to HP hardware (servers and switches). The installation is done by Kickstart, and you only have to answer a few questions depending on your hardware. It is strange but you have to specify the kickstart file, and if you don't, a plain RHEL like distribution is installed. Of course you should remaster the installation DVD (modifying the kickstart file ks.cfg) so it meets your hardware. The frontend (or head node as HP calls it) will be installed with complete GUI.

The Debian Cluster Components is a project which builds a set of packages for Beowulf clusters.