CAREER: Effective Grid Programming with EveryWare and G-commerce

University Of Tennessee Knoxville, Knoxville TN

Investigators

Abstract

As network performance has increased, the high-performance computing community has begun to turn to distributed systems to meet their performance needs. Research that combines the concurrency management techniques from parallel computing with scalable and adaptive distributed systems design has resulted in a new software architecture known as the "Computational Grid". Its main tenet is that user programs should be able to draw computational "power" from a distributed resource pool the way electrical appliances draw electricity from a power utility: ubiquitously, seamlessly and reliably. This project will investigate a novel approach to building software systems that implement the" Computational Grid" model of computing and how to foster new curricular directions based on this investigation. As vehicles for this proposed work, two research thrusts will be combined in the form of EveryWare and G-commerce to define a new programming methodology for the Grid. EveryWare is a software toolkit for building robust, high-performance distributed applications that can use existing Computational Grid middleware as building blocks. An EveryWare program is structured as a set of application customizable services that use the EveryWare toolkit as an adaptive server framewirk while leveraging the functionality provided by existing Grid systems. G-commerce is a computational economic system that allows applications to purchase resources from computational suppliers. A key feature of G-commerce is that it is a macro-economic system which provably achieves economic equilibrium and market stability. By combining these two we will define a new and powerful approach to programming Grid applications. To do so we will: .extend and implement the EveryWare software toolkit, .study and develop economic formulations that will form the basis of G-commerce based resource allocation strategies, .develop a computational exchange for implementing resource economies an infrastructure for empirical G-commerce resource scheduling studies and, .verify our results using non-trivial Grid applications in "live" Grid settings. The research approach of this project is both analytical and experimental. It will approach the problem of ensuring resource allocation stability and equilibrium theoretically and using simulation. Then implement promising economic formulations using the computational exchange, and test their validity with EveryWare implimentations of Grid applications. At the same time, EveryWare will be enhanced so that it enables the aplication programmers and schedulers to consider computational price when making resource allocation decisions. EveryWare and G-commerce will also serve as the basis for new directions in curriculum development. A course sequence will be designed that combines distributed computing techniques, statistical analysis of load conditions, and economic modeling methods from the literature with practical implementation using Computational Grids. The resulting multi-disciplinary course will allow students to compare and combine different strategies easily for the purpose of making quantitative evaluations, which will be a primary focus of each sequence. Practica will also be organized in which students from different Universities will be invited to participate. Each practicum will focus on the implementation of a chosen application using different Computational Grid infrastructures. Students will compare different techniques and use the EveryWare toolkit and G-commerce to combine those that are the most effective.