A Data Caching Scheme for Hybrid Hard Drives

碩士 === 國立中興大學 === 資訊科學與工程學系所 === 99 === The hybrid hard drive is composed of traditional hard disk and NAND Flash. The hybrid hard drive has a large capacity storage space as traditional hard disk, and is also able to use NAND Flash as a cache to improve I/O performance. Nevertheless, due to the...

Full description

Bibliographic Details
Main Authors: Syuan-You Liao, 廖軒佑
Other Authors: 張軒彬
Format: Others
Language:zh-TW
Published: 2011
Online Access:http://ndltd.ncl.edu.tw/handle/12500647645282911874
Description
Summary:碩士 === 國立中興大學 === 資訊科學與工程學系所 === 99 === The hybrid hard drive is composed of traditional hard disk and NAND Flash. The hybrid hard drive has a large capacity storage space as traditional hard disk, and is also able to use NAND Flash as a cache to improve I/O performance. Nevertheless, due to the cost consideration, NAND Flash usually has a limited capacity. Thus, how to take advantage of both hard drives and NAND Flash and to cache the appropriate data on the NAND Flash is an important issue for the hybrid hard drives. In this paper, we propose the scheme a data caching scheme that analyzes the dada access features, such as read count, write count, random or sequential access and recency, and then giving priority to cache random read data that are recently accessed to NAND Flash. In addition, CFQ (Completely Fair Queuing) is the default I/O Scheduler for the Linux. It may pauses for a short time after a synchronous request to anticipate a close-by read request. We modify CFQ scheduler that, if the data accessed by the most recent synchronous request is on Flash, we do not wait. Otherwise, CFQ is operated normally that pauses for a short period of time. We have implemented our scheme in the Linux kernel. From the experimental results, the proposed data caching scheme not only is low overhead, but also takes the advantages of NAND Flash and the traditional simultaneously to improve the I/O performance.