Gabe Weinstock '01 Receives
Provost Summer Science Research Grant
GABRIEL WEINSTOCK '01 has received a grant from the Provost's Office to support a summer research project during the summer of 1999. During the eight-week project, Gabe will be working on distributed computing using the Linux operating system. Professor David Eck will supervise the project.
The Provost's Summer Science Research Program funds several grants every year. These are awarded to student applicants on a competitive basis. A grand includes a $2,000 stipend plus room and board.
Here is a copy of Gabe's project proposal:
For the Provost Summer Science Program I propose to establish a Beowulf cluster in the computer lab of Lansing hall, using the new Pentium III computers installed there. The Beowulf cluster grew out of a NASA project which later spread to other institutions. "Traditionally, the solution of many leading-edge problems in science and engineering has required computers that provided the highest possible performance. Unfortunately, these computers had associated with them the highest cost of any computing platform. However, today it is possible, using commodity components, to create machines that are substantially cheaper than supercomputers but are nearly comparable in performance. Specifically, Beowulf-class computers utilize cost-effective, mass-market, off-the-shelf components to deliver computational cycles at the lowest possible price/performance." (http://www-hpc.jpl.nasa.gov/PEP/dsk/papers/beowulf_seminar.html)
The cluster itself consists of a number of computers networked via Fast Ethernet connections (preferably 100Mbps, although we intend to use 10Mbps connections), thus enabling parallel processing, or distributed computing (distributing instructions over multiple processors). In order to enable parallel processing, the free operating system Linux will be installed on the PC's. Linux allows parallel processing and applications have been written for the system which take advantage of this feature.
The goal of our research is to investigate parallel processing and distributed computing, using Linux and the Beowulf libraries, to test and develop multiprocessing applications, and to document our findings.
Beowulf software is available free as add-ons to the popular Linux operating system (which is also free). The add-ons implement programming libraries, which would enable students to write programs designed for distributed computing environments, as well as allow us to run programs tailored to run on Beowulf clusters. One aspect of our research would be to quantitatively determine the processing speed of a small Beowulf cluster versus a typical PC. Technology is already leaning towards distribution of computing power over a number of smaller systems, as opposed to higher performance but lower volume systems.
Many universities, including Clemson, Drexel and CalTech have Beowulf clusters. Students could share resources with students at these schools if a Beowulf cluster were implemented here. Additionally, the presence of a Beowulf cluster on campus would be attractive to prospective computer science students. The most commonly used programming environment on Beowulf systems is PVM, which has become the de-facto standard for developing applications. One aspect of our research will be to explore programs designed to operate on a Beowulf cluster. One such program is PVM-POV, a free ray-tracing program based on the popular program POV-ray. Ray-tracing involves creation of photo-realistic computer generated images. It is extremely computationally intensive, so much so that there is a computing benchmark called the POVBENCH, which records the time taken to render a complicated scene, designed to test the speed of a processor. It is interesting to note that a Beowulf cluster consisting of 17 IBM Netfinity servers recently matched the performance of a Cray T3E 900-AC64 supercomputer (http://www.linuxworld.com/linuxworld/lw-1999-03/lw-03-cluster.html).
By installing a Beowulf cluster on the campus, we can stay at the forefront of computing technology. I personally hope to better understand the Linux operating system, parallel processing, and networking protocols through this project. Beowulf clusters can approach supercomputers in their processing power. At Supercomputing 1996 NASA and DOE unveiled clusters costing less than fifty-thousand dollars that sustained over one gigaflop/s performance (http://www.beowulf.org/intro.html). While our proposed cluster will be nowhere near as powerful, it still has the potential to teach students the fundamentals of parallel computing and be used as a powerful tool for the science department. Setting up a Beowulf cluster has been called a "hands on" and "do it yourself" experience, and is highly suited for a research project, as it requires a minimum of resources to properly establish and has long term benefits for students, faculty, and the Colleges. Full tutorials on building a Beowulf system are available online. Additionally, many resources are available for someone attempting to create a cluster, including mailing lists, web pages, and workshops. It is not impossible that we could help shape the future of low cost distributed computing through working with a Beowulf cluster.
In addition to investigating processing speed of programs on the proposed Beowulf cluster and learning about parallel architectures and distributed computing as well as networking protocols, we intend to develop programs written specifically for the cluster. This will enable us to better learn about parallel computing as well as the Linux operating system. Some of the programs suited to parallel processing include the simulation of physical systems, the development of high capacity databases (including platforms for high speed web services) and weather forecasting.
Distributed computing in general and Beowulf clusters specifically are quickly becoming the modus operandi for large, computationally intensive projects. There are numerous advantages to setting up a small cluster here at Hobart and William Smith Colleges. These include the low overhead required to build a cluster, the potential for use as a teaching tool for students and as a powerful aid for research, and the ability to put the colleges on the forefront of an emerging technology.
Department of Mathematics and Computer Science
Hobart and William Smith Colleges