원문정보
초록
영어
As computers are pervading in all aspects of life with advancement of technology, a growing demand is being felt for low-cost, huge computational need. Grid computing paradigm provides an attractive alternative where enormous computational and data processing and management power can be generated at a much cheaper budget by means of large scale sharing of resources. However, the efficacy of such systems largely depends on the efficacy of scheduling policies employed. In the past few years, myriad, novel scheduling algorithms have been proposed. Besides, for long running and similar types of applications the performance of grid systems can be improved further if the scheduled jobs can be tuned at run time. It is a very important aspect for enhancing performance, considering the drastic fluctuation of resources availability, variations in communications bandwidth, fluctuations in job submission frequency etc that characterize a grid scenario. In this paper, the design and implementation of a Comprehensive Performance Tuning Framework (CPTF) is reported that initially schedules jobs to resources, mines all such job-to-resource mapping information and thereafter tunes certain parameters for subsequent scheduling of submitted jobs. CPTF aims to minimize the overall throughput of the system instead of minimizing single job execution time. Experiments result shows the efficacy of the proposed framework under varying load conditions.
목차
1. Introduction
2. Related Work
3. Desktop Grid Scheduling: A Brief Review
4. The Comprehensive Performance Tuning Frame Work (CPTF): A Brief Overview
4.1. System Framework
4.2. Job Execution Life Cycle
5. The Adaptive Scheduling Model
5.1. Job Analyzer and Job Decomposition:
5.2. Computational Parameter Selection:
5.3. Heuristic Scheduling
6. Performance Tuning Model
6.1. Execution Analyzer
6.2. Re-scheduler:
7. Experiments and Results
7.1. Environment
7.2. Experiments Conducted and the CPTF/ PES
7.3. Analysis of Results:
8. Conclusion
Acknowledgements