發散式檔案系統之可靠性研究

Abstract

本篇論文旨在於發散式檔案系統之可靠性研究，並根據此研究提出具有可 靠性的發散式檔案系統(RDFS)的設計。本系統的主要目的為增進檔案系統 的可靠性，可用性及具備高效率等性質。本系統採用不同檔案存取性質的 磁碟機組來達成高效率的目標。在本系統中，每一個磁碟機組皆具備讀取 或寫入較優的性質。根據磁碟機組的特性，系統會自動依檔案的需求，將 此檔案儲存於較合適的磁碟機組。本系統亦會依據檔案存取型式的變更， 自動將檔案遷移到符合其存取特性的磁碟機組去，以求得更佳的系統效能 。另外，為達成檔案系統可靠性的要求，本系統會自動將檔案複製一份， 儲存到另一個磁碟機組去，以防止單一磁碟機組的損毀而造成檔案無法存 取的情況。本系統會完全處理雙份檔案間一致性的問題，不需要使用者的 介入。另本系統也處理磁碟機組毀壞時的復原管理。本系統的另外一個目 標為檔案系統的可用性，其方法為採用數個磁碟機組，搭配可靠性的檔案 存取服務，因此只要有一個磁碟機組存在，則本系統可持續的提供檔案服 務。 This thesis presents the studies of reliability on Diverging File System and proposes the design of Reliable Diverging File System (RDFS). The goals of this file system are to enhance file system reliability, availability, and performance. The high performance goal is based on the multiple disk storage structures each implemented on an independent personal computer. Each storage structure is designed to optimize a different set of file system access characteristics such as file attribute requests or large sequential reads/writes. The system also implements dynamic file migration automatically to achieve higher performance by detecting the change of file system access patterns. The system combines the log and file replication method to increase the file system fault tolerance. The fault tolerance is completely transparent to users. The system maintains automatically file replicas, handles file inconsistency problem during file access, and perform recovery for disk storage crash. The goal of availability is through the management of multiple disk storage. The system will serve file requests continuously providing that one disk storage is operational.