1. BMC Beowulf Supercomputer
Our goal is to create a low-cost version of a Beowulf cluster (see below for description). In the first year, we hope to solicit funds from interested groups such that we can immediately assemble a working cluster. After gaining additional experience, we hope to pursue outside funding for related computational projects, enlarge the number of computers in the cluster, and develop custom software for experiments at BMC.
1.1. What is a Beowulf cluster?
A Beowulf Computer Cluster is a set of low-cost, off-the-shelf computers that are connected via a standard network and special software. These connected computers can then perform as a single programmable "supercomputer." Typically, the individual computers are inexpensive PCs running Linux connected together through regular networking. Beowulf-class computer clusters can span the range from a few, older PCs to an entire building of fast, ultra-modern workstations.
A Beowulf-class supercomputer could be utilized by anyone performing computations. In its simplest usage, the cluster can be used as a convenient method to run multiple copies of the same program. For example, say you had a Java program that crunches some numbers. You could simply run many copies of that program at once, all in parallel. More sophisticated parallelization would require more complicated programming. However, there are intermediate tools available that allow one to reap the benefits of parallel processing without having to learn esoteric programming skills.
With access to a Beowulf-class cluster, students and faculty could attempt to tackle problems that have been too computationally expensive. Our hope is that this would lead to expanded opportunities in research and the classroom. Typical uses could include modeling, simulation, analysis, 3D computer animation rendering, and optimization.
1.2. What can be done on such a cluster?
Most anything that you compute on a computer could be sped up by running the code on a cluster.
Projects include:
-
rendering visualizations
-
simulations (from economics, social studies, physics, chemistry, biology, and others)
-
computational and optimization problems (searches, sorts, proofs)
-
artificial intelligence problems (artificial neural networks, genetic algorithms)
-
other distributed problems (SETI, finding large prime numbers, etc)
What have we already done? Tom Carroll (Physics) has run some number crunching programs in Java, and reports:
| Without cluster | With Cluster |
| 369.06 hrs (15.38 days) | 10.34 hrs (0.431 days) |
1.3. Principal Investigators
Michael Noel and Douglas Blank conceived and began the design of the project in the Fall of 2003, and began collecting equipment and funds in January 2004.
This project has been funded in part by the following BMC groups and grants:
-
Information Services
-
Computer Science Department
-
Physics Department
-
Geology Department
-
Chemistry Department
-
Howard Hughes Medical Institute Grant
Food and beverages at the build party supplied by the
Center for Science in Society.
This is an on-going project and is continually collecting funds for future planned software and hardware enhancement and upgrades.
1.4. What's New?
To keep abreast of current status, see BMCBeowulfSuperComputer:WhatsNew
1.5. Logs
1.6. HOWTOs
Hardware
Instructions on how to assemble a node can be found at BMCBeowulfSuperComputer:HOWTO. A poster for the party in
OpenOffice format, and in
HTML format. Feel free to use for your own use.
Software and Programming
Here is a quick demonstration of the type of programming one can do on a Beowulf-class cluster: BMCBeowulfSuperComputer:ProgrammingDemo
1.7. Links
-
History of the idea, and lots of information - http://beowulf.org
