Abstract:  In Hadoop Distributed File System（HDFS）, the imbalance of load distribution has become the bottleneck for the speed of distributed access. Based on the status quo of load in the current hospital information system, the multi-attribute dual-threshold dynamic load balancing algorithm is proposed to improve the defect of only using the initial and default load balancing strategy. Under the different environmental circumstances, the same size of ophthalmic medical image file is tested in the HDFS cluster, leading to the conclusion that the load balance of multi-attribute and dual-threshold is more conducive to make the average allocation of the load of large amount of image data to the each node in the cluster and maintain the dynamic balance in the process of file access, therefore improves the efficiency and distributed processing performance of the cluster.

Key words: Hadoop Distributed File System（HDFS）, distributed cluster, load balancing, ophthalmology medical, image file access

摘要： 在HDFS系统中，集群负载分配不均衡成为影响分布式文件存取速度的瓶颈。针对医院信息系统的负载现状，提出多属性双阈值决策的动态负载均衡算法，对分布式集群中使用HDFS默认的单属性评价、单阈值的负载均衡策略存在的缺陷加以改进。通过负载性能测试，对比证明运用多属性双阈值的负载均衡算法相比较HDFS默认的负载均衡更有利于将大量的影像负载数据相对均衡地分配到服务器集群中的各个节点上，大幅度地缩短了增加数据规模时数据服务器节点的平均响应时间，有利于提高HDFS集群整体的工作效率。

关键词: Hadoop分布式文件系统（HDFS）, 分布式集群, 负载均衡, 眼科医疗, 影像文件存取