MRI: Development of a High Density, High Performance Beowulf Cluster

Texas A&M Agrilife Research, College Station TX

Investigators

Abstract

Texas A&M University plans to design and build a 256 processor, medium grain parallel machine with as close to maximum floating point performance per dollar as possible. This system, commonly referred to as a Beowulf cluster, will require a modification of the usual paradigm in order to solve certain design problems inherent in these systems. The cluster will contain 256 processors consisting of 128 nodes using dual AMD motherboards, each with 1GB ECC DDR DRAM and will be interconnected with Gigabit Ethernet to form one of the largest and most powerful academic clusters. Components will be chosen to minimize cost while maximizing performance. Novel techniques for mounting and cooling the motherboards will be explored in an effort to maximize the density of the nodes. Multiple operating environments will also be made available, allowing the end user to determine which environment yields the peak performance for their application. This machine will be used to solve a variety of computationally intensive problems in the mathematical and physical sciences and support researchers primarily located in chemistry, mathematics, physics and statistics. One group, with a core in numerical analysis but with emphasis on fluid flowing porous media, will also include researchers with expertise in oil recovery and in groundwater flow (from Petroleum and from Agriculture Engineering) and in parallel architectures (from Computer Science). In addition to problems in large-scale computation, the groups will be solving problems in inverse problems, nonparametric function estimation, statistical methods for computational biology, quantum dynamics, high energy physics, inorganic and organometallic systems, and polyatomic molecules. This effort will also provide an educational benefit, both within the design and the implementation stages. Mechanical Engineering will use the novel power and cooling requirements of the system as source of activity in a senior design class. Once the system is in place, the departments of Mathematics, Statistics and Computer Science will provide courses in efficient utilization and programming to allow maximum benefit to the designated research groups, as well as the rest of the university community.