操作系统_清华大学、参考lecture5-filesystem.pptxVIP

文件系统File System;2;OSes today face new requirements and challenges ;Speed mismatch is one of the key roots of poor efficiency ;Data deluge presents new and more challenging problems;全球数据量高速增长;Data type distribution;Some figures about data deluge ;New development: Data-Intensive Computing as the 4th Paradigm;其他一些说法;11;File System;Disk Organization;File Directory;File Attributes/Metadata;Basic File Operations;Procedure to Open a Directory;File System Implementation;Internal Implementation;Open File Table;File System Requirements;File System Requirements (cont’d);Key Issues;How to set block size?;25;FAT;FAT-16;FAT-32;NTFS: Basics;NTFS: Basics;NTFS: Other Features;NTFS: Architecture;MFT File;NTFS Metadata;File Records;36;EXT;ext2;ext2;ext3;ext4: on-disk layout;ext4: extent tree layout ;ext4: feature summary;44;Distributed File System Overview;Distributed File System;分布式文件系统关键技术;分布式文件系统关键技术;49;NFS;NFS Architecture;NFS;NFSv3;NFSv4;Google File System;GFS Architecture;GFS: Master Design;GFS: Chunkserver Design;GFS: Client Design;GFS: Data Integrity;Carrier: Architecture;Carrier: metadata server ;Carrier: R/W Lock;Carrier: 去冗余的ACL;Carrier: data server;Carrier: Supervisor;DFS设计经验;Open Questions;69;Parallel File System Overview;Parallel File System Overview;PVFS Overview;PVFS2 vs. PVFS1 ;PVFS2 Architecture;PVFS2 Functional Modules;多种使用方式;元数据管理;数据放置策略;通信： Buffered Messaging Interface;无状态的客户端;文件操作语义;容错方案;性能评测;连续读写测试;Blue Gene/P 系统架构;聚合带宽测试：IOR;元数据操作性能测试: Metarates;IBM General Parallel File System;GPFS;GPFS: Node Roles;GPFS: Software Structure;GPFS: Files;GPFS: Parallelism and Consistency;GPFS: Allocation Map;GPFS: Allocation Manager;GPFS: High AvailabilityLogging;GPFS: Node Failure Recovery;GPFS: Handling Disk Failures;GPFS: Concurrency Control;GPFS: Concurrency Control;101;Data Deduplication;103;重复数据的发现 基于Fingerprint的比对 适合使用密码学哈希函数(极低的碰撞概率) MD5、SHA-1最为常用 切块方法 定长：使用固定的块大小 实现简单，无明显的IO性能开销 对数据插入没有很好的稳定性 变长：依据上下文内容确定块大小 通常利用Rabin Fingerprint寻找切块点 对数据插入有很好的稳定性 实现复杂，计算开销大;Summar