Last Modified: 2002 October 21
Charlie Torres / charlie@phys.utb.edu
Him þa ellenrof andswarode,
wlanc Wedera leod, word æfter spræc,
heard under helme: "We synt Higelaces
beodgeneatas; Beowulf is min nama."[The honorable one answered him then, pride of the Weathergeats, hard under helm: "We are Hygelac's table-companions; Beowulf is my name."]-- Beowulf, ll.340-343
The page you're reading now contains general information about Beowulf clusters and our motivation to build one. The main source of information about the UTB Beowulf (called Lobizón) is the UTB Beowulf Homepage.
In the summer of 1994 Thomas Sterling and Don Becker, working at the Center of Excellence in Space Data and Information Sciences (CESDIS) located at the Goddard Space Flight Center in Greenbelt Maryland, under the sponsorship of the Earth and space sciences (ESS) project, built a cluster computer consisting of 16 DX4 processors connected by channel bonded Ethernet. They called their machine Beowulf. This "home made" computer was an instant success and their idea of constructing powerful computers utilizing off the shelves equipment quickly spread through NASA and into the academic and research communities. Roughly speaking, then a Beowulf machine is simply a cluster of microprocessors that communicate together via a network. Although this method of forming a parallel-processor is fairly simple, there are several options for the architecture of these computers. Beowulf clusters are now recognized as a genre within the High Performance Computing community. More information on Beowulf clusters in general can be found on the Beowulf Project Homepage.
The computational resources needed to search for gravitational waves in the output of a detector depend on the type of astrophysical source one is trying to detect. For example, the computational requirements for stochastic background searches are minimal (i.e., a typical workstation can do the job ), while that required for an all-sky search for periodic sources is demanding for even the most powerful computers (i.e., on the order of flops -see i.e. P. R. Brady and T. Creighton, "Searching for periodic sources with LIGO. II: Hierarchical searches", gr-qc/9812014.) Somewhat less intensive, but still computationally-demanding, are searches for the inspiral of two compact objects---e.g., neutron stars and/or black holes (see B. J. Owen and B. S .Sathyaprakash, "Matched filtering of gravitational waves from inspiraling compact binaries: Computational cost and template placement", gr-qc/9808076 .) Thus, if one wants to be able to analyze LIGO data for general astrophysical sources, one must find a way to meet these computational challenges. Benchmarking studies carried out by researchers at Caltech and members of the LIGO Scientific collaboration at the University of Wisconsin at Milwaukee (UWM) have indicated that a Beowulf machine is the most cost-effective way to meet LIGO's data analysis needs (see, e.g., UWM LSC group Beowulf system).
Lobizón means "werewolf" in Spanish. This is the name chosen for our Beowulf which is currently under construction. Prof Fitratullah Khan from the Computer Science department is directing this project with the assistance of several computer science students. For a more complete description of the technical specifications of our Lobizón Beowulf, including benchmarks, components description and listing of the team members you can visit the UTB Beowulf homepage.