Abstract: Distributed file systems need a mechanism to control the concurrent access from various clients and maintain the consistence of file data.Locking is one of most popular mechanisms to implement concurrency control.This paper researches the distributed I/O lock mode of Lustre file system and optimizes its various lock applications.It introduces the basic concept of the distributed lock，and analyzes the algorithms of client data writeback cache and dynamic file size acquirement based on the distributed lock model.To improve the performance and scalability of Lustre’s lock service，it proposes strategies of adaptive I/O locking and lock conflict detection based on interval tree and client-cached lock discard.

Key words: concurrency control, distributed lock, Lustre file

摘要： 分布式文件系统需要有一种机制对来自各个客户端的并发访问进行控制，维护文件数据的一致性。锁是实现并发控制最流行的机制。研究了Lustre文件系统的分布式I/O范围锁的模型，并对它的各种应用进行了优化。介绍了Lustre分布式锁的基本概念，对基于锁实现数据客户端写回缓冲以及多写者文件大小的动态获取的算法进行了分析；提出自适应I/O锁策略，基于区间树的范围锁冲突检测优化策略以及客户端锁淘汰策略来增强Lustre锁服务的性能和扩展性。

关键词: 并发控制, 分布式锁, Lustre文件系统