Cluster Ohio Sites
Cluster Ohio Testbed
Waleed Smari
Dept. of Electrical and Computer Engineering
University of Dayton
Summary
In an effort to understand installation procedures, time commitments, and requirements for Cluster Ohio awardees, SGI dontated a number of Pentium processors as a testbed for Cluster Ohio. These processors were delivered to Bowling Green State University, University of Dayton, and Wilberforce University prior to Cluster Ohio awards.
The future of high-performance computing relies on the efficient use of clusters with symmetric multiprocessor nodes and scalable interconnection networks. This new approach represents an appealing alternative for cost effective parallel computing by configuring a platform that is suitable for a large class of applications and workloads. We propose to use a cluster of high performance PCs to study, develop, test, and evaluate several aspects of cluster computing, ranging from applications to system issues, algorithms, and performance. The effort will identify potential solutions to existing challenges and verify their effectiveness. A set of experiments will be devised for computationally intensive applications that will be run on the cluster under different setups. Lessons learned will be analyzed and new approaches will be attempted to achieve high performance, scalability, and high throughput of these distributed parallel platforms. Some of the specific tasks to carry out are:
- Architectural issues that have major impact on the performance of a cluster, including cluster’s topology, reconfigurability, and scalability.
- Communication and networking issues which include overhead and latency in the system.
- Resource management issues, which cover many aspects of application-domain related to partitioning, mapping, scheduling and load balancing.
- Performance monitoring and modeling issues, which will also include
visualization and steering tools that are cluster based.
We intend to investigate these and other issues.
Installation Date
14 June 2001
Configuration
6 single processor nodes |
