檔案系統I/O效能與區塊大小影響之分析

Abstract

在Linux環境下，當Process要將目標檔案寫入至硬碟，或者從硬碟讀取資料至記憶體時，Process皆會透過檔案系統對硬碟送出讀寫目標檔案的請求。由於在讀寫的過程中皆須使用到檔案系統，因此根據檔案系統參數的不同設定，即可能影響到整體的I/O效能。 本論文主要針對檔案系統下與區塊大小之相關設定，觀察這些不同的設定在不同的使用環境下，對於I/O效能會造成甚麼程度的影響，並且分析造成這些影響的原因。 首先，我們會從熟悉檔案系統的基礎架構與資料結構開始，這對於之後我們的分析會有所幫助；接著藉由Linux Kernel Source Code來瞭解整個檔案開啟與讀寫操作的流程；最後則是透過實際的I/O效能測試與Linux Kernel Debugger的協助，來分析我們所做的設定對於I/O效能的影響原因。

In the Linux, when a process wants to either write the file back to HDD or read the file from HDD to Memory, the process need File System help to send the write/read request to HDD. Since the write/read procedure need the File System help, the I/O performance of the whole system may be influenced by the different setting of File System parameter.

This thesis is mainly concerned about the setting related to block size. We will observe the influence on I/O performance caused by different setting under different condition, such as operating large or small files, and then analyse the benchmark results.

First, we will get familiar with the basic structure of File System; then realize the write/read procedure by study the Linux Kernel Source Code; finally, by the I/O test and asistance of Linux Kernel Debugger, we will analyse the cause of the influence of I/O performance.