Computer Architecture

ELTN 493

Lab 1

4-16-2001

Beowulf Clusters

At the time, the market for computers was very competitive and in high demand causing quick development. This rapid development drove prices down and reliability up. Due to this development the cost of the commodity computer was much more affordable. Also the free operating system Linux was developing, along with other publicly available software, which provided hardware independent software. With all the standards, programmers can now write programs that will run on all future systems. This development is what made technology of Beowulf clusters available.

Beowulf clusters are similar to supercomputers. They were built for high performance computing just as supercomputers. Many people though could not afford supercomputers and began to find an alternative route. It was found that it was possible to build a massive parallel computer using commodity components. The first people to try this were Thomas Sterling and Don Becker under the sponsorship of the ESS project. One of the goals of the ESS project (Earth and space sciences project) was to determine the applicability of massively parallel computers to the problems faced by the Earth and space sciences community (http://dune.mcs.kent.edu/~farrell/equip/beowolf/). At the time, only a supercomputer could address problems associated with such large data sets. Tom and Don built a cluster computer consisting of 16 processors, which they called Beowulf, to address these computations. Their machine was a success and the development effort grew quickly into what is now called the Beowulf Project.

There are many important characteristics of beowulf clusters. Due to changes in processor type and speed, network technology, and relative costs of components do not change the programming model. Another key component is that the system software can be used on other and future systems now that there is a “standardization” of message passing via PVM and MPI (http://dune.mcs.kent.edu/~farrell/equip/beowolf/). Supercomputers have one programming model and its very expensive to develop one-of-a-kind software.

The project’s scope and the number of Beowulf installations have grown over the years. It spread to NASA sites, other R & D labs, and universities around the world (http://dune.mcs.kent.edu/~farrell/equip/beowolf/). Many people were in need for high performace computing and developed their own aspect of the Beowulf system. Beowulf class cluster computers range from several node clusters to several hundred node clusters all over the world.