Traditionally in the computational world there were two types of compute cluster configurations. One type is called “Shared Memory” and the other is called a “Beowulf”. With the Shared Memory systems it was boxes of computers plugged together with a very high speed interconnect. The connect made it so that all the pieces connected together appears as just one computer. Just like your computer at home. The advantage of a shared memory system is that usually you could have a lot of memory (RAM) installed in the machine. Load these systems up with a large disk for temporary space and it could really crunch some data sets. With the advent of “cores” lately and the ability to buy memory relatively cheaply (compared to just a few short years ago) many home computers now coming with multiple “cores” are essentially a type of “Shared memory” system. Shared memory systems are great for problems that have large GB sized data sets that need to all be loaded into memory at once to analyze it.

The other type of cluster is a Beowulf cluster. This is where you take multiple computers and hook them together so that they can perform parts of the parallel applications. Each computer generally doesn’t have a lot of resources compared to a single shared memory system but collectively a Beowulf can compete with these numbers. Beowulf’s are best suited for problems where they need lots of CPU time but not a lot of memory or disk storage. So in this scenario just having lots of computers available to run some part of a program is the benefit. A good example of a beowulf cluster is the Seti @ Home or Folding @ Home Projects. Here you can download a screensaver that allows them to use your idle computer to run parts of their very large sets of data that they need to crunch.