“While the programming task for genetic algorithms is relatively
simple, a parallel supercomputer is required to make the approach
effective for antenna modeling and design. Lacking a leading-edge
supercomputer, Blasing and his colleagues decided to take an
increasingly popular approach-a “roll-your-own” model made from
low-cost, commercially available parts. “We just went down to a
local computer store and bought the guts of 20 PCs-we didn’t need
the monitors or disk drives-and mounted the boards in a rack,” he
explained. The 20 computers were linked with the Linux operating
system, which is in the public domain. The entire configuration
cost about $25,000, and the team got it up and running in a few
days.“
“Called a “Beowulf-class” computer, the approach has gained
popularity in academia since the cost is low and the computers
provide an opportunity for teaching parallel-programming
techniques. Blasing said that Endgate’s project is one of the few
commercial instances where the approach has been used. “Linux is
also a key component for practical implementation of this kind of
multiprocessor since it has easy-to-implement networking
extensions,” he said.”