2.2 A Taxonomy of Parallel Computing

2.3 Benefits of Beowulf
In addition to the significant price/performance advantage of Beowulf-class systems, Beowulfs provide several other important advantages, ranging from rapid inclusion of COTS technology advances to widely-used APIs and programming languages.
Ironically, in today's dynamic technology market it is the low end systems that first incorporate the latest and highest performance devices. This is because they represent the largest market share and will provide the largest early return on investment. Since Beowulfs require no development lead time, as soon as the latest advances in chip technology are incorporated into the mass-market PCs they can be immediately included in Beowulf configurations. It takes much longer for the same devices to find their way into commercial MPPs. Thus, an advantage of the Beowulf model is rapid response to technology trends.
One of the greatest obstacles hindering the acceptance and application of high performance parallel computing systems is that every manufacturer produced its own particular architecture. Not only were these architectures not compatible between manufacturers, but almost routinely, different generations from the same vendor would be largely incompatible with one another, despite claims to the contrary, e.g., the switch in architecture from the Thinking Machines CM-2 to the CM-5. The problem of disparate architectures is aggravated by the high mortality rate of companies producing MPPs, resulting in no follow-on machines and lost application software investment. Examples include Thinking Machine Corp., Kendall Square Research, Alliant and others. Beowulfs, however, are not susceptible to the vagaries of commercial offerings and corporate mortality. They are assembled from subsystems that can be acquired from multiple vendors who make mass-market systems. Thus an advantage of Beowulfs is that they are not subject to the whims of a single source.
There is no specific, fixed Beowulf system architecture. A broad range of processor nodes and system topologies are possible and can be optimized on-site and at the last minute. They can even be changed later in response to new needs or new technology opportunities. System form is limited not by vendor specification, but only by the imagination of its users. Therefore, another benefit of the Beowulf approach is its property of what may be called just-in-place configuration.
Beowulf systems are scalable. A wide range of system sizes is possible from a small number of nodes connected by a single low cost hub to a system incorporating complex topologies of many hundreds of processors. And, these systems can be expanded over time as additional resources become available or extended requirements drive system size upward.

 



How to Build a Beowulf
How to Build a Beowulf: A Guide to the Implementation and Application of PC Clusters (Scientific and Engineering Computation)
ISBN: 026269218X
EAN: 2147483647
Year: 1999
Pages: 134

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net