基于强化学习的多机群网格资源调度模型

在由多个计算机集群构成的多机群网格环境下，为了解决数据并行型计算（DPC）与计算资源的有效匹配问题，提出了一个基于强化学习机制的网格资源调度模型；给出了由多个计算机机群组成的多机群网格、逻辑计算机机群、数据并行型计算和一系列Agent的定义；利用多Agent的协作做竞争机制、基于强化学习的匹配知识库的修正方法，研究了逻辑计算机机群与DPC资源供需之间的有效匹配问题；描述了网格的资源调度模型。理论分析和实践表明，该模型有效地解决了多机群网格环境之下数据并行型计算所需的资源优化使用问题。该模型适合于基于多机群网格的数据并行型计算。

Multi-cluster Grid Resource Dispatch Model Based on Reinforcement Learning

For resolving the problem to effectively match between Data Parallel Computing (DPC) and computational resources in Multi-cluster Grid that composed of many computer dusters, a grid resource dispatch model based on reinforcement learning is discussed. A series of formal definitions, such as the Multi-cluster Grid (MCG), the Logical Computer Cluster(LCC), the DPC and the Agents, are given. Using the mechanism of cooperation and competition of Multi-Agent, the knowledge-base re- vising techniques based on reinforcement learning, the effective match methods are studied. The resource dispatch model is de- scribed. The analysis and experiment results show that this model effectively resolves the problems of optimized use of re- sources in Multi-cluster Grid. It can be fit for the data paralld computing in Grid.