Clusters are surprisingly be powerful, firstly they offer the kind of speed that is available with a supercomputer, the University of Manheim in Germany complies a semi-annual listing of the top 500 supercomputers and in recent years clustered Linux systems have take more and more of a place in this ranking, competing with manufacturers such as Cray, Silicon Graphics and IBM.
Supercomputers are an expensive game; often the places that need supercomputers do not have the money to fund these, and even if they could afford to buy one may not be able to afford to upgrade it at a later date. With clusters this is slightly different because to upgrade a cluster all you need to do is to simply add another node to the cluster, and it doesn’t have to be a new computer system, some of the first clusters have been built out of very old computers but because of the multiples of machines they can be extremely powerful.
An interesting story is of the “stone supercomputer” that was built by a group at Oak Ridge National Labs. Oak Ridge built a Beowulf cluster, and rather than spend any money on the hardware at all, they solicited donations from other groups in their organization. So, for no investment other that their time, they built a several-dozen node system out of discarded 486 machines, slow Pentiums and other systems. It may not have been pretty, or have had the latest greatest processor, but it got the job done.
Clusters also fairly simple because they are in fact just a set of computer systems tied together with a network all working on a large problem that has been broken down into smaller pieces. Of course there is a variety of ways you can link these up and there a lots of different software packages that can be used to make these work.