Clustering is the name of a technique in which multiple computers are linked together through networking technology and configured to work together to perform a specific function. There are three main configurations for clustering: parallel processing, fault tolerance, and load balancing.

Parallel Processing
Parallel processing consists of many small devices that are working together to complete a task; it is a powerful design but can be bulky and costly to implement. Super computers such as the Cray are parallel processing devices that are some of the fastest computers in the world, but they can be the size of a minivan and extremely expensive, making their practical use limited. Clustering using parallel processing is a less expensive alternative.

The idea of using several smaller computers working together became a reality with the invention of the personal computer. With so many older units sitting around unused, some researchers found they could use their older computers, open source operating systems like Linux, and networking equipment to connect the PCs in a parallel fashion to harness an impressive amount of computing power for almost no cost.

Because the system was a group of computers clustered together, the name cluster server was coined. Each computer in the cluster is called a node and the nodes communicate with each other across a network. The cluster server has a master node and several slave nodes. The master node is the central brain of the operation and the main interface for the cluster, while the slave nodes are the workers. The master node divides large tasks into several small pieces and distributes them to the slave nodes. Each slave node works on the task it is given until it is finished and then asks for more. It is the job of the master node to collect the finished portions of the job and recombine them into the final output.

Fault Tolerance and Load Balancing
Instead of using a cluster of computers to create a single server, fault tolerance employs several identical high-powered servers operating in the same way as a server cluster to provide a service (e.g., Web server, CRM, Siebel, etc.). Because each machine performs the same function, they are exact duplicates of each other. In a clustered configuration, if one server node fails, the remaining nodes take over preventing an interruption of service.

Load balancing is similar to fault tolerance in design and configuration, but load balancing distributes the workload across several identical computers such that no one computer gets overloaded.

It is possible to create a fault tolerant system or a load balancing system individually, but because they are so similar and their functions are desirable, it is common to encounter a server cluster with both functions.

Machines that make up the cluster can all be in one room in a centralized configuration, or they can be scattered around the world in a distributed configuration - the only requirement is that they are in the same network environment.