RAID5讀寫性能優(yōu)化研究
本文選題:存儲(chǔ)子系統(tǒng) + 零拷貝策略 ; 參考:《華中科技大學(xué)》2012年碩士論文
【摘要】:隨著計(jì)算機(jī)技術(shù)的迅猛發(fā)展,磁盤存儲(chǔ)設(shè)備的存取速度已經(jīng)遠(yuǎn)遠(yuǎn)不能匹配CPU的處理速度。為了縮短速度上的差異,目前許多大型服務(wù)器的磁盤大多采用RAID技術(shù)。RAID5磁盤陣列在一定程度上縮短了I/O請(qǐng)求的響應(yīng)時(shí)間,提高存儲(chǔ)系統(tǒng)的I/O性能。如何合理地優(yōu)化改進(jìn)RAID5系統(tǒng)讀寫性能,使其發(fā)揮最大的效益,始終是存儲(chǔ)子系統(tǒng)關(guān)注的焦點(diǎn)。 當(dāng)前IO存儲(chǔ)子系統(tǒng)在實(shí)現(xiàn)時(shí)都使用各自的緩存系統(tǒng),這存在著重復(fù)的內(nèi)存拷貝現(xiàn)象,會(huì)導(dǎo)致CPU開銷的增加,從而降低磁盤陣列I/O性能。為了減少不必要的內(nèi)存拷貝,對(duì)于RAID5上面的讀性能優(yōu)化,采用零拷貝策略加以改進(jìn)。該方法的核心是在處理IO讀請(qǐng)求時(shí),繞開內(nèi)核系統(tǒng)在驅(qū)動(dòng)層提供的緩存,直接進(jìn)行數(shù)據(jù)交互,從而提高系統(tǒng)的訪問性能。測(cè)試表明,改進(jìn)后的RAID5系統(tǒng),,同一條帶內(nèi)的讀請(qǐng)求提高更為顯著。 在多用戶小數(shù)據(jù)訪問RAID5系統(tǒng)時(shí),存在著小數(shù)據(jù)寫性能的缺陷。深入Linux內(nèi)核代碼分析研究,發(fā)現(xiàn)內(nèi)核處理寫I/O請(qǐng)求會(huì)在非滿條帶寫處理時(shí)出現(xiàn)磁盤IO等待現(xiàn)象,采用添加輔助緩存策略優(yōu)化改進(jìn)RAID5寫性能。該策略的核心是內(nèi)核在處理寫請(qǐng)求時(shí),將要計(jì)算的新校驗(yàn)信息存放在添加的輔助緩存中,實(shí)現(xiàn)了校驗(yàn)信息的計(jì)算和數(shù)據(jù)的讀寫分離進(jìn)行,消除了磁盤IO等待時(shí)間,從而提升了系統(tǒng)的小寫性能。測(cè)試表明,改進(jìn)后的RAID5系統(tǒng),在非滿條帶寫的情況下性能優(yōu)化良好,較好的解決了小寫性能差的問題。
[Abstract]:With the rapid development of computer technology, the access speed of disk storage devices is far from matching the processing speed of CPU. In order to shorten the speed difference many large servers mostly use RAID technology. RAID5 disk array to some extent shorten the response time of I / O request and improve the I / O performance of storage system. How to optimize and improve the read and write performance of RAID5 system and maximize its efficiency is always the focus of the storage subsystem. The current IO storage subsystem uses its own cache system, which has repeated memory copy phenomenon, which will lead to the increase of CPU overhead, thus reducing the I / O performance of the disk array. In order to reduce unnecessary memory copy, zero copy strategy is adopted to optimize the read performance of RAID5. The core of this method is to bypass the cache provided by the kernel system in the driver layer and directly interact with each other in order to improve the access performance of the system. The test results show that the improved RAID5 system can improve the read request in the same band more significantly. There is a flaw in the performance of small data writing in multi-user small data access RAID5 system. Further research on Linux kernel code shows that the kernel processing of writing I / O requests will lead to disk IO waiting phenomenon in non-full stripe write processing. Adding auxiliary cache strategy is used to optimize and improve the write performance of RAID5. The core of this strategy is that when the kernel processes the write request, the new check information will be stored in the added auxiliary cache, which realizes the separation of the calculation of the check information from the reading and writing of the data, and eliminates the disk IO waiting time. Thus, the lowercase performance of the system is improved. The test results show that the improved RAID5 system has a good performance optimization under the condition of non-full stripe writing, and solves the problem of poor lowercase performance.
【學(xué)位授予單位】:華中科技大學(xué)
【學(xué)位級(jí)別】:碩士
【學(xué)位授予年份】:2012
【分類號(hào)】:TP333.35
【參考文獻(xiàn)】
中國期刊全文數(shù)據(jù)庫 前10條
1 陳華英;;磁盤陣列RAID可靠性分析[J];電子科技大學(xué)學(xué)報(bào);2006年03期
2 章宏?duì)N;薛巍;;集群RAID5存儲(chǔ)系統(tǒng)可靠性分析[J];計(jì)算機(jī)研究與發(fā)展;2010年04期
3 王湘娜,蔣本珊,徐漸;基于RAID5的磁盤陣列Cache的研究與實(shí)現(xiàn)[J];計(jì)算機(jī)工程;2003年03期
4 李潔瓊,馮丹,張熳;提高磁盤陣列小寫性能的方法研究[J];計(jì)算機(jī)工程;2003年18期
5 譚毓安;王婉星;于強(qiáng);朱立谷;張雪蘭;;基于磁盤異或引擎的RAID-5小寫性能優(yōu)化[J];計(jì)算機(jī)工程;2006年16期
6 趙亮,劉光明;利用DCMT來解決RAID5的小寫問題[J];計(jì)算機(jī)工程與科學(xué);2002年05期
7 謝長生,姜國松,劉瑞芳;RAID控制器中多級(jí)Cache的研究[J];計(jì)算機(jī)工程與科學(xué);2004年12期
8 陳康林;謝長生;;RAID5中的效率優(yōu)先倒盤算法[J];計(jì)算機(jī)工程與科學(xué);2006年02期
9 熊建剛;馮丹;;高可用的磁盤陣列Cache的設(shè)計(jì)和實(shí)現(xiàn)[J];計(jì)算機(jī)工程與科學(xué);2006年08期
10 劉冬;王麗芳;蔣澤軍;劉志強(qiáng);;RAID小數(shù)據(jù)隨機(jī)訪問性能分析與優(yōu)化[J];計(jì)算機(jī)應(yīng)用;2009年02期
中國博士學(xué)位論文全文數(shù)據(jù)庫 前2條
1 曾碧卿;分布式計(jì)算中并行I/O調(diào)度策略研究[D];中南大學(xué);2005年
2 吳素貞;磁盤陣列高可用技術(shù)研究[D];華中科技大學(xué);2010年
本文編號(hào):1788677
本文鏈接:http://www.sikaile.net/kejilunwen/jisuanjikexuelunwen/1788677.html