CISE Instrumentation: Acquisition of High-Performance SMPs and a High-Speed Network

点击翻译按钮获取中文摘要

9729828 Chaudhary, Vipin Wayne State University CISE Research Instrumentation: High Performance Computing on an ATM-Connected Cluster of Symmetric Multiprocessors This research instrumentation grant serves to support the purchase of one Sun Enterprise 4000 (with 6 processors), two Sun Enterprises 3000 (4 processors each), and a 12-port Switch with 8 ATM interfaces which will enable the following projects:- Development of an Automatic Parallelization Environment (APE) for Scientific Engineering Programs in a Cluster of SMPs,- Real Time Optimization of #D Conformal Radiation Therapy Treatment Planning, - Parallel Techniques for 3D Cone Beam Tomography, and- Measurement of Resource Requirements for ATM ABR Connections. The first project integrates compile-time and run-time parallelization techniques focusing on loops with non-linear subscripts on a cluster of SMPs. An implementation of APE will reduce the execution time for various scientific and engineering applications. The second project serves radiation therapy where current computer assisted optimization techniques are too slow or too expensive to be adopted in clinical practice. Genetic algorithms will be used to optimize the dosage. A real-time cost effective implementation on a cluster of multiprocessors will make the planning process applicable in clinical practice. The third project encounters the bottleneck in parallel implementation of 3D Cone Beam Tomography technique on ethernet in the low bandwidth of the network. Using the virtual connectivity provided by ATM, the multiphase broadcast can be efficiently implemented. Results from this implementation will apply to other problems like volumetric imaging applications and finite element analysis of 3D structures. The last project utilizes profile data from various applications to enable sources to request just the necessary bandwidth and facilitate fair allocation of bandwidth. It will permit a better utilization of the network resources and improve the application perf ormance.