分布式文件系统Cappella日志设计与实现-计算机系统结构专业论文.docxVIP

华中 科 技大 华 中 科 技 大 学 硕 士 学 位 论 文 II II Abstract According to CAP theory, distributed file system cant satisfy simultaneously the availability, consistency and network partition. Under the distributed environment, hardware failure, software failure, network partition and power failure may lead to node failure in distributed storage system, its crucial to guarantee the availability of distributed file system under node failure circumstance. Most of the distributed storage systems adopt cache to provide high performance, but in the condition of node failure, the cached data may be lost, so its vital to assure the consistency of the recovered distributed file system. So we adapt journal to satisfy the requirement of recoverability and consistency. Adopt logical log, file updates firstly operate in memory, meanwhile generate log records which record the change of data block; employ write ahead logging, guarantee that not until the log record is flushed to log file, the data in data cache should be not flushed to their original position; apply group commit strategy, the log record will be write into log buffer at first, and then commit to the log file as a group reclaiming the log buffer; introduce fuzzy checkpoint, flush data corresponding to the already committed log records in data cache to their target location and reclaim log file space. Design and implement recovery protocol of distributed file system, the client and server in distributed file system utilize both the synchronous reply and asynchronous reply, the synchronous reply makes sure the subsequent requests are non-blocked, and the asynchronous reply acknowledges the durability of the requests execution result; client maintains request cache for requests in different state, and adjusts the request cache according to the server side response, the server keep information about clients connection and requests, when server crashes down, the recovery will be progressed based on client side request cache an