“What is beowulf? in this context, it’s not the epic poem
describing the adventures of a great Scandinavian warrior in the
sixth century. Instead, Beowulf is a small set of open-source
technologies that tie multiple Linux or FreeBSD systems into a
shared resource cluster. How those resources are shared and the
specific setup and software required depends largely upon what you
want to do. Beowulf clusters are typically used for parallel
processing of complex problems and have gained the title of “Poor
Man’s Supercomputer.”
“Clusters of computers are commonly used these days to build
reliable Web and file servers, but these are not Beowulf clusters.
There’s a subtle, but important distinction here. In clusters that
support reliability and scaling for a server farm, the applications
are replicated on each node, and a redundant control system (such
as a load-balancing switch) routes messages to the appropriate node
based on one or several load-sharing algorithms, such as
round-robin or least-used nodes. Beowulf clusters have applications
that are divided across the cluster nodes and run in parallel,
exchanging messages between nodes to keep the overall computation
in synchronization….”
“But building a Beowulf cluster does not automatically enable
faster processing. It is usually necessary to modify the source and
recompile it with PVM or MPI calls to exchange data between the
separate systems. The old standby Fortran is better for this than C
or C++, because C uses pointers that cannot be automatically
parallelized.”