分布式文件系统自适应负载调度分析-计算机系统结构专业论文.docx

华 华 中 科 技 大 学 硕 士 学 位 论 文 II II Abstract In the long-running process of the distributed file system, old nodes failure eliminated and new nodes are added dynamically as well as the dynamic creation and deletion of stored data. There is a certain gap between the old and new storage nodes in the storage capacity and performance. It will result in the load unbalancing problem. The distributed file system can’t maximize the performance, and even affect the availability. It is necessary to maintain the system load equilibrium by load scheduling mechanism. The load scheduling mechanism for distributed file system is divided into two stages: replica placement scheduling and load rebalancing scheduling. For replica placement scheduling stage, existing programs are generally using the static algorithm or dynamic algorithm with combination of various load information collected by storage nodes. For load rebalancing stage, existing programs are generally using adjustment means such as increasing replicas and replicas migration to correct the unbalanced state, but the process itself will bring some overhead, affecting the system services. For the distributed file system features, this paper presents an adaptive load scheduling mechanism called SALS (Simple Adaptive Load Scheduling), including a replica placement scheduling method and a load rebalancing scheduling method. The replica placement method use reading and writing request queue length in storage node as judging basis, computing heterogeneous node’s currently the longest response time to characterize the load state. Then it divided storage nodes into three states: light load, middle load and over load. The load rebalancing scheduling method use replica adjustment as basis, while ensuring the Client according to the load state of the storage node adaptively adjust its request transmission rate, reduce the too frequent retransmission timeout, to ensure that the storage node overload less impact on system performance. By