Scyld ClusterWare System Overview

System Components and Layout

Scyld ClusterWare streamlines the processes of configuring, running, and maintaining a Linux cluster using a group of commodity off-the-shelf (COTS) computers connected through a private network.

The front-end "master node" in the cluster is configured with a full Linux installation, distributing computing tasks to the other "compute nodes" in the cluster. Nodes communicate across a private network and share a common process execution space with common, cluster-wide process ID values.

A compute node is commonly diskless, as its kernel image is downloaded from the master node at node startup time using the Preboot eXecution Environment (PXE), and libraries and executable binaries are transparently transfered from the master node as needed. A compute node may access data files on locally attached storage or across NFS from an NFS server managed by the master node or some other accessible server.

In order for the master node to communicate with an outside network, it needs two network interface controllers (NICs): one for the private internal cluster network, and the other for the outside network. It is suggested that the master node be connected to an outside network so multiple users can access the cluster from remote locations.

Figure 1. Cluster Configuration