Abstract: The management of resources and scheduling in grid environment is a complex problem.In data-intensive application the delay of data access is a critical problem.This paper proposes a novel data replication algorithm based on clustering，called clustering based replication algorithm.With CBR strategy，a number of grid nodes are clustered according to the bandwidth between them to construct a logic region.Also this paper introduces a modified LRU algorithm，in which the other data files required by other applications are in consideration in order to avoid deleting the data files that maybe are required in the future.The algorithm is compared with other two algorithms.The experimental results show the effectiveness of the proposed algorithm to the scheduling problem.

Key words: cluster, data intensive, scheduling, Least Recently Used（LRU） algorithm

摘要： 网格环境下资源的管理和调度是一个非常复杂且具有挑战性的问题。在数据密集型应用中，数据文件的读取延迟时间是至关重要的。提出了一种基于聚类预处理的数据文件复制算法（CBR），将传输带宽满足一定条件的网格结点通过聚类方法构成一个“逻辑区域”；并介绍了一种改进的LRU算法，考虑了其他计算任务需要的数据文件请求，避免删除未来将使用的数据文件。通过实验证明，该算法得到的计算任务完成时间优于其他两种算法。

关键词: 聚类, 数据密集型, 调度, 最近最少使用（LRU）算法