|
ALS 2000 Abstract
Maximizing Beowulf Performance
Robert G. Brown, Duke University
Abstract
At this point in time the beowulf (and other related compute cluster) architectures has come of age in Linux. Few indeed
are those in any realm of technical computing that are unaware of the fact that one can assemble a collection of
commodity off the shelf (COTS) computers and networking hardware into a high performance supercomputing
environment. However, a detailed knowledge or appreciation for the bottlenecks and special problems associated with
beowulf design is not so common. A review of the important bottlenecks and design features of a beowulf is given
along with associated benchmarking and measurement tools to illustrate how to bridge the gap between the simple
``recipe'' of a beowulf as a pile of compute nodes, interconnected with a fast network and running linux and the
realities of engineering a parallel code and beowulf-style cluster to achieve satisfactory performance.
- View the full text of this paper in
HTML form, and
PDF form.
- If you need the latest Adobe Acrobat Reader, you can download it from Adobe's site.
- To become a USENIX Member, please see our Membership Information.
|
Need help? Use our Contacts page.
