使用具適應性的檔案重整技術來提升Linux檔案系統效能的研究

Improving Performance of Linux File System with Adaptive File Rearrangement

鄭凱宇(Kai-Yu Cheng)

指導教授 ： 姜美玲

國立暨南國際大學/管理學院/資訊管理學系/碩士(2008年)

https://doi.org/10.6837/NCNU.2008.00318

摘要

隨著科技的發達，電腦許多週邊裝置也不斷創新，例如：CPU速度加快、RAM的存取速度不斷提昇、儲存裝置的容量加大…等。其中硬碟一直是影響電腦系統效能的一大瓶頸。硬碟的資料存取時間包含搜尋時間（seek time）、旋轉延遲（rotational latency）和傳輸時間（transfer time），而其中以搜尋時間所耗費最多的時間。當資料是不連續地存放於硬碟上，為了存取特定資料，硬碟讀寫頭（disk head）需大幅地移動，增加移動距離會增加搜尋時間，導致硬碟傳輸速率的降低。本研究透過將熱門資料集中儲存於硬碟中央所切割出的保留區（Reserved Area）和動態的判斷是否將存取頻繁的資料搬移至保留區，以達到改善磁碟資料的佈局（disk layout）和減少資料在存取時硬碟的搜尋時間為主要目標。
本研究透過修改Linux Ext3檔案系統，將熱門檔案集中存放於硬碟中央所規劃出的保留區，由於熱門檔案有極高的存取頻率，因此，當系統存取熱門檔案時，硬碟讀寫臂將來回於硬碟中央的保留區進行存取。此目的是藉由縮短硬碟讀寫臂的移動距離來降低資料在存取時硬碟的搜尋時間，並進一步提升檔案系統的存取效能。此外，本研究也提出一個判斷是否將檔案搬移至保留區的方法。當檔案經過不斷的存取，系統判斷該檔案已變成熱門時，此時進一步地判斷該檔案存放於硬碟上的磁柱（cylinder）離硬碟中央保留區的磁柱之距離是否接近，來決定是否將該檔案搬移至保留區內。如果檔案原先的儲存位置已距離保留區很近，則此檔案並不會被搬移至保留區內，以減少檔案搬移的系統負擔（overhead）。
本研究實作於Linux 2.6核心內，我們修改Ext3檔案系統，並透過自行開發的測效程式來模擬BBS系統的存取特性，實驗證明，修改過的Ext3檔案系統可以有效的縮短資料存取時硬碟11.39-30.14%的存取時間。此外，當冷門檔案欲搬移至保留區時，判斷檔案於硬碟中所在位置是否已接近保留區，而決定是否搬移，此方法可以有效改善3.29-18.86%的系統效能。

關鍵字

檔案系統 ； 搜尋時間 ； 磁柱

並列摘要

Many computer devices are innovated continuously with the evolution of the technology. For example, the speed of CPU is accelerated, the access speed of RAM is increased, and the capacity of storage is getting larger and larger. However, hard disk is still the bottleneck of the system performance. The data access time of the hard disk includes seek time, rotational latency, and transfer time. Among them, seek time still wastes most of the total access time. When the data is stored discontinuously in the disk, the disk head needs to move more distances in order to access the data. The increased moving distances of the disk head not only increases the disk seek time but also decreases disk transfer rate. In this thesis, we will focus on how to store hot data in the central cylinders of the disk and dynamically determine whether the cold data should be moved into the central cylinders of the disk, when it has been accessed frequently. This can improve the disk layout and decrease the seek time of accessing in the hard disk.
In our research, we store hot files (i.e frequently accessed files) in the Reserved Area located at the center of the hard disk by modifying Linux Ext3 file system. Because the access frequencies of hot files are higher, the disk head will access files inside the Reserved Area when the system accesses hot files. The purpose is to decrease seek time for accessing hot file in the hard disk by shortening the disk head movement distance to improve the file system performance. Besides, we also propose a method to determine whether the cold file when it becomes hot should be moved to the Reserved Area. When the cold file is accessed frequently, the system will determine the file has become hot. The system will also determine whether the location of file in the hard disk is close to the Reserved Area. If the location of the data is closed to the Reserved Area, the data will not be moved, because the benefit is less than the system overhead of file moving.
We have implemented this system by modifying the Ext3 file system of Linux kernel 2.6 and created a performance benchmark to simulate the operation behaviors of the BBS system. The result of experiments shows that the modified Ext3 file system can effectively shorten 11.39-30.14% of the disk access time. Besides, when the cold file becomes hot, the file will be moved to the Reserved Area. If the location of this file is close to the Reserved Area, the system will determine whether the file needs to be moved into the Reserved Area. This method can improve 3.29-18.86% of the system performance.

並列關鍵字

File System ； Seek Time ； Cylinder

參考文獻

[1] S. Akyurek and K. Salem, “Adaptive Block Rearrangement,” ACM Transactions on Computer Systems, 13, (2), 89-121, 1995.

[2] S. Akyurek and K. Salem, “Adaptive Block Rearrangement Under UNIX,” Software-Practice and Experience, 27, (1), pp.1-23, January 1997.

[3] M. McDonald and R. Bunt, “Improving File System Performance by Dynamically Restructuring Disk Space,” Phoenix Conference on Computers and Communication, pp. 264-269, Mar. 1989.

[4] Wenguang Wang, Yanping Zhao, and Rich Bunt, “HyLog：A High Performance Approach to Managing Disk Layout”, In Proceedings of the 3rd USENIX Conference on File and Storage Technologies, pages 144--158, March 2004.

[5] C. Staelin and H. Garcia-Molina, “Clustering Active Disk Data to Improve Disk Performance,” Technical Report CS-TR-283-90, Department of Computer Science, Princeton University, (September 1990).