Abstract: Hadoop Distributed File System（HDFS） is a low-cost, highly fault-tolerant distributed file system that suitable for running on commodity hardware, and offers high-throughput data access for applications on large datasets. However, there are some performance optimization problems in HDFS, such as under-load balancing. Although Hadoop system comes with a load balancer to achieve balanced adjustment, but users need to give a static threshold in advance. In order to solve the fixed threshold and subjectivity, through the analysis, evaluation and optimization of disk space utilization, CPU utilization, memory utilization, the disk I/O occupancy rate, the network bandwidth occupancy rate and other parameters, this paper forms a calculating expression for a threshold, and through the theoretical analysis and simulation experiments, this paper verifies the threshold calculation and load balancing. The experimental results show that this method achieves a better balance effect and improves the utilization of computing resources compared with the Hadoop static input threshold algorithm.

Key words: cloud environment, Hadoop Distributed File System（HDFS）, load balancing, dynamic threshold

摘要： Hadoop分布式文件系统（Hadoop Distributed File System，HDFS）是一种适合在通用硬件上运行的低成本、高度容错性的分布式文件系统，能提供高吞吐量的数据访问，适合针对大规模数据集上的应用。然而，HDFS中还面临一些性能优化问题，如负载均衡不足。虽然Hadoop系统自带的负载均衡器可以实现均衡调整，但需要用户预先给出静态的阈值。为了解决阈值的固定性和主观性，通过对磁盘空间使用率、CPU利用率、内存利用率、磁盘I/O占用率、网络带宽占用率等参数的分析评估优化，形成对阈值的计算表达式，并通过理论分析和仿真实验对阈值的计算和负载均衡进行验证。实验结果表明，相比较Hadoop静态的输入阈值的算法，该方法达到了更好的平衡效果，提高了计算资源的利用率。

关键词: 云环境, Hadoop分布式文件系统（HDFS）, 负载均衡, 动态阈值