U.S.-Egypt Cooperative Research: Efficient Scheduling Strategies for Real-time Applications in Cluster Environments with Possible Processors' Failure

University Of Connecticut, Storrs CT

Investigators

Abstract

0411653 Ammar Description: This award is to support a cooperative research between Dr. Reda Ammar, Department of Computer Science & Engineering, University of Connecticut, Storrs, Connecticut and Dr. Ayman El-Dessouki, President, Electronics Research Institute, Cairo, Egypt. They plan to investigate, develop, and evaluate scheduling strategies of satisfying the performance and reliability requirements of real-time applications. They expect to significantly improve scheduling strategies for computationally intensive real-time applications in cluster computing environments with possible processors' failure. The scheduling methods involve three related models: the Cluster Reference Model, the Application Model, and scheduling algorithms. First, the PIs will define the specific cluster structure and relevant expected behavior in order to establish a firm basis for the study. Then the application performance model is presented. This model contains information about the expected behavior of the computation from both a structural and performance perspective. Third, using the above two models as basis, they plan a cluster scheduling scenario which can integrate the use of the application model to better achieve high performance and an acceptable reliability level in case of failure. Within this scenario, they will investigate, specify, and evaluate appropriate scheduling algorithms. They will apply these results to case studies of selected applications to be executed within the clusters currently available at both of participating institutes. These algorithms should also provide higher reliability, reduce the communication cost among sequential tasks, minimize fragmentation introduced by scheduling several tasks into processors, and maximize the overall admission rate. Broad Impact: The project addresses scheduling real-time task graphs while optimizing the available processing power utilization in the cluster environment, achieving a required level of reliability, reducing communication overhead, and minimizing processing power fragmentation. Dr Ammar is a leading expert in the field and has published extensively in this field. Dr El-Dessouki is leading the research activities at the ERI in Egypt, and they have been cooperating for several years including building a computing grid between the two institutes. Using and developing expertise in the Grid technology will provide the Egyptian scientific community with a local high performance computing power at low cost. The grid -along with the proposed techniques will provide a new powerful facility that can be used to solve problems with extensive computation needs. The PIs plan to build a simulator for comparing their approach with the existing approaches, and this could be used for educational purposes. The project involves training of graduate students. This project is being supported under the US-Egypt Joint Fund Program, which provides grants to scientists and engineers in both countries to carry out these cooperative activities.