基于学习方式对Hadoop作业调度的改进研究

随着并行计算、分布式计算和网格计算技术的发展，云计算作为一种新的模型被提出来，发展极为迅速。Hadoop作为一个开源的云计算系统，得到了广泛的运用。作业调度是Hadoop平台的核心问题之一，通过对Hadoop中已有调度算法的了解和分析后，基于学习的方式，利用过去的节点历史记录和作业属性来不断地改进作业调度;应用了基于特征加权的朴素贝叶斯分类器算法来改进任务的分配调度，并通过实验进行了验证，结果表明它对任务分配调度执行效率有一定的提高。

Research on Improving Hadoop Job Scheduling Based on Learning Approach

With parallel computing, distributed computing and grid computing technology, cloud computing was proposed as a new model and developing fast. Hadoop is an open source cloud computing system that has been widely used.Job scheduling is one of the core problem on Hadoop platform. Through understanding and analyzing current scheduling algorithm that has already existed for Hadoop,based on learning approach, the past history of nodes and job attri-butes were used to improve job scheduling. Feature weighting-based naive bayes classification algorithm was applied to im- prove tasks scheduling, then it was verified through experiments. As a result, it improves the efficiency of scheduling of task allocation for Hadoop.