登錄結構化發散式檔案系統之設計

Abstract

本篇論文提出一種〝登錄結構化發散式檔案系統(LDFS)〞之設計方法。其 基本的概念是將作業系統中的檔案處理子系統﹐從一個邏輯上的處理行程 轉變為一個實體上的處理元件。LDFS 的主要功能有三：(1)藉著將檔案資 料交錯分佈在由多個磁碟處理機(Disk Processor)所各自管理的磁碟上﹐ 使得檔案傳輸速率(File Transfer Rate)不再受限於單一磁碟機的效能 。(2)藉由將檔案的處理工作分工在兩個層次(邏輯層與實體層)的處理機 上﹐並配合檔案快取(File Caching)技術的使用﹐可大量減少磁碟機輸出 入動作對檔案存取效率的影響。(3)藉由登錄結構化的儲存管理方式(Log- Structured Storage Management)﹐以及存在檔案處理機(File Processor)與各個磁碟處理機間之內部處理協定 ( Internal Transaction protocol ) 的使用﹐即使系統發生意外當機情形﹐仍能在 系統重新啟動時恢復各磁碟中檔案資料的一致性。目前我們已建立一個登 錄結構化發散式檔案系統的雛型﹐並在此雛型上進行系統效能的量測。其 結果顯示 LDFS 的雛型的確有良好的系統效能﹐亦具有更進一步發展的潛 力。 This thesis presents the design of Log-Structured Diverging File System (LDFS). The basic concept of LDFS is to change the role of file system from a logical process in operating system to a physical processing component of operating system. There are three principle functions provided by LDFS : (1) By strip- ping file data across multiple disks, which are management by Disk Processors, the file transfer rate is not limited by the performance of a single disk. (2) By dividing file processing into two levels(logical level and physical level) of processors and using file caching technique in each processor, it greatly reduces the impact of disk I/O on file access. (3) By the log- structured storage management and the internal transaction pro- tocol between FP and each DP,the consistency of file data stored in all disks can be restored when system crash occurs. We have built a prototype of LDFS and performed some evalu- ations on it. The results show this LDFS prototype is performing well and has the capability to performs still better.