An event-driven architecture for master-worker network computing is proposed to meet the needs in large-scale network computing systems such as high-concurrency
dynamic resource control and stability. The architecture combines the advantages of multi-threaded mode and event-driven mode
and a new strategy of thread management that is suited for master-worker network computing is derived based on queuing theory. The introduction of network queue extends the application domain of staged event-driven architecture from single host to wide area network environment
and the use of delay queue improves system response and reliability. The priority queue is designed for effectively supporting a variety of job scheduling mechanisms. Moreover
systems have the characteristics of modular construction and rapid development. Experiment results and the application of drug discovery grid with systems based on the proposed architecture show that the average job submission time is reduced from 1 850 s to 1 350 s and the average job processing time is reduced from 1 910 s to 1 420 s for 1 000 jobs
and that system resources have more rational use.
关键词
Keywords
references
DUBREUIL M, GAGNE C, PARIZEAU M. Analysis of a master-slave architecture for distributed evolutionary computations [J].IEEE Transactions on Systems, Man and Cybernetics, part B: Cybernetics, 2006, 36(1): 229-235.
HEYMANN E, SENAR M A, LUQUE E, et al. Efficient resource management applied to master-worker applications [J]. Journal of Parallel and Distributed Computing, 2004, 64(6): 767-773.
CESAR E, MORENO A, SORRIBES J, et al. Modeling master/worker applications for automatic performance tuning [J]. Parallel Computing, 2006, 32(7/8): 568-589.
MORAJKO A, MARGALEF T, LUQUE E. Design and implementation of a dynamic tuning environment [J]. Journal of Parallel and Distributed Computing, 2007, 67(4): 474-490.
WELSH M, CULLER D, BREWER E. SEDA: architecture for well-connected scalable internet services [J]. Eighteenth Symposium on Operating Systems Principles, 2001, 35(5):230-243.
LIU Yang, GUI Xiaolin, XU Yuwen. Study on scheduling methods based on priorities in grid workflow [J].Journal of Xi'an Jiaotong University, 2006, 40(4): 411-414.
WANG Yongjian, LUAN Zhongzhi, QIAN Depei, et al. DDGrid: a grid computing environment with massive concurrency and fault-tolerance support [C]∥Proceedings of 7th International Conference on Grid and Cooperative Computing. Piscataway, NJ, USA:IEEE Computer Society, 2008: 5-14.