CSR-AES: Adaptive Real-Time Scheduling for Grid Computing

University Of Nebraska-Lincoln, Lincoln NE

Investigators

Abstract

Adaptive Real-Time Scheduling for Grid Computing Automated performance control and resource management is crucial for the successful continued evolution of grid computing. An important first step is new scheduling algorithms and analysis techniques that provide high performance with assured quality-of-service (QoS), while also simplifying the management of grid resources. The long-term goal of this research is to investigate and develop adaptive real-time scheduling theory and technology for grid computing. Specifically, this project is developing adaptive real-time scheduling theory and technology for divisible grid applications executing in heterogeneous clusters. This project has three aims: (1) To establish real-time theoretical foundations to guide the design of efficient divisible load scheduling algorithms. The application of divisible load theory, design of scheduling algorithms and schedulability analysis techniques are investigated. 2) To develop scheduling theory and tools in heterogeneous clusters that support resource reservations for real-time grid computing. and 3) To design real-time cluster-based scheduling algorithms that are adaptive and robust to system uncertainties. The integration of adaptive feedback control theory and real-time divisible load scheduling are investigated. Research results will be disseminated to the research community via published papers, public domain tools, and participation in conferences. Once the tools are mature and robust, they will be made available to the Open Science Grid community.