DESIGN AND DEPLOYMENT OF A COST EFFECTIVE GRID COMPUTING MODEL FOR SCALABLE AND HIGH PERFORMANCE COMPUTING REQUIREMENT BY CHARACTERIZING HOST AVAILABILITY: THE CASE OF ARBA MINCH UNIVERSITY

Abstract

Any major Ethiopian institution has a substantial number of computer systems on its campus, often in the scale of thousands. Mutka [16] measured the overall available time of a group of workstations at the University of Wisconsin, and found that a workstation is idle and available for computation 70% of the time. This study can be taken as indication that most of the computational resources available in Ethiopia higher education are underutilized. Given that a large amount of scientific computing isappropriate for execution in an opportunistic environment, a campus grid is an inexpensive way to build a powerful computational resource. Desktop grids, which use the idle cycles of many desktop PC’s, are currently the largest distributed systems in the world. While the motivation for using desktop resources is clear, namely the opportunity to perform large computations at low-cost, the main challenge is that the resources are volatile. Despite Desktop grid systems have been successfully used for many high throughput applications, there has been little insight into the detailed temporal structure of CPU availability of desktop grid resources and the performance of the underlining networks that hinder its usability. Host characterization is essential for accurate performance measuring and modeling of such platforms. In Ethiopia such platform is in the infant state and no institution has been building such infrastructure. This structure is critical to characterize the performance of desktop grid platform. In this study attempt has been done to address the following questions: (i) what are the temporal characteristics of desktop CPU availability in public higher institutions setting? (ii) How do these characteristics affect the performance and availability of desktop grids? (iii) Based on these characteristics, can we construct a model with quantifiable performance that satisfy higher institution needs and requirement with cost effective way? To realize the fact we developed and implement a model of desktop grids with a research settings in Arba Minch University, and present host availability and volatility measurements for 500 utility workstations and measure CPU Execution availability within desktop grids which constitute 30 hosts by running HTcondor open source desktop grid software. We utilize these measurements to characterize CPU and Execution availability and develop a performance model for desktop grid in case of Arba Minch University.