本文選題：節(jié)能　切入點：磁盤　出處：《華南理工大學(xué)》2012年博士論文

【摘要】：節(jié)能環(huán)保是當(dāng)今世界各國政府共同面臨的重大問題，，我國政府已將節(jié)能工作放在了國民經(jīng)濟工作的重要位置。近年來，隨著全球數(shù)字信息總量的快速增長，存儲系統(tǒng)的能耗問題日趨嚴(yán)重，如何降低存儲系統(tǒng)能耗的問題已成為當(dāng)前計算機科學(xué)技術(shù)領(lǐng)域中一個急待解決的關(guān)鍵科學(xué)技術(shù)問題和熱點研究課題。因此，展開這方面的研究對推動存儲系統(tǒng)節(jié)能技術(shù)的發(fā)展和應(yīng)用具有重要的學(xué)術(shù)價值和實際意義。 論文從磁盤存儲系統(tǒng)的節(jié)能原理入手，以提高磁盤存儲系統(tǒng)的能效和性能為目標(biāo)，在認(rèn)真分析和研究了磁盤存儲系統(tǒng)節(jié)能的要求和各種技術(shù)因素的基礎(chǔ)上，重點對磁盤存儲系統(tǒng)的磁盤電源管理、緩存磁盤數(shù)據(jù)管理、磁盤負(fù)載平衡和節(jié)能磁盤存儲系統(tǒng)模型等節(jié)能關(guān)鍵技術(shù)問題進行了深入研究和討論。論文的主要研究工作和創(chuàng)新點包括如下幾個方面： （一）提出了一種訪問模式自匹配和性能保證的磁盤電源管理策略(Access ModeMatching and Performance Guarantee, AMM-PG)。針對現(xiàn)有磁盤電源管理策略的不足，AMM-PG采用空閑間隔長度的數(shù)據(jù)直方圖來對應(yīng)磁盤的訪問模式并導(dǎo)出對應(yīng)的超時閾值，實現(xiàn)適應(yīng)磁盤的訪問模式對磁盤的電源進行合理的管理，從而提高磁盤的能效；另外，AMM-PG對部分磁盤訪問模式增加考慮磁盤何時從待機狀態(tài)返回到空閑狀態(tài)，通過性能要求和空閑間隔長度的數(shù)據(jù)直方圖確定一個磁盤保持在待機狀態(tài)的最大時間，從而在滿足性能要求的同時，盡可能提高能量節(jié)省量。理論分析和實驗結(jié)果表明，與固定超時策略和自適應(yīng)管理策略share相比，AMM-PG具有更好的節(jié)能效果。 （二）提出了一種負(fù)載自適應(yīng)的具能量意識緩存磁盤數(shù)據(jù)管理方法（Workload-Adaptive LRU, WA-LRU）。針對現(xiàn)有緩存磁盤數(shù)據(jù)管理方法的不足，WA-LRU根據(jù)熱門數(shù)據(jù)的數(shù)量和數(shù)據(jù)訪問的密集程度區(qū)分?jǐn)?shù)據(jù)磁盤的節(jié)能優(yōu)先級，當(dāng)需要替換緩存磁盤中的數(shù)據(jù)時，總是替換來自節(jié)能優(yōu)先級最低的數(shù)據(jù)磁盤中的數(shù)據(jù)，從而匹配數(shù)據(jù)磁盤訪問模式，提高節(jié)能效果，并使用基于生命周期管理的LRU隊列來識別和消除緩存磁盤的瓶頸效應(yīng)，從而避免出現(xiàn)因多個數(shù)據(jù)磁盤被緩存的數(shù)據(jù)較少而造成的能效較低問題。理論分析和實驗結(jié)果表明，與現(xiàn)有的LRU、PA-LRU和PB-LRU方法相比，WA-LRU方法具有更好的節(jié)能效果。 （三）提出了一種面向節(jié)能磁盤存儲系統(tǒng)的多隊列負(fù)載平衡方法（Multi QueueLoad Balance, MQLB）。MQLB針對基于整體存儲節(jié)點的節(jié)能方案中負(fù)載傾斜的部分磁盤容易過載的問題，根據(jù)數(shù)據(jù)的訪問頻率、訪問時間局部性等因素，按熱門程度將數(shù)據(jù)分成多個隊列，只維護最近被頻繁訪問的數(shù)據(jù)作為遷移對象，并使用熱度衡量熱門數(shù)據(jù)單元HDU給磁盤帶來的負(fù)載壓力，在磁盤出現(xiàn)過載時，根據(jù)過載磁盤的過載程度計算合適的負(fù)載壓力比LPR，然后根據(jù)該LPR選擇合適的HDU進行數(shù)據(jù)遷移。理論分析和實驗結(jié)果表明，與現(xiàn)有的PDC、WAM和EM-FMS方法相比，MQLB方法可以更有效和快速地調(diào)整負(fù)載且引入更小的開銷。 （四）提出了一種新的基于動態(tài)樹的節(jié)能磁盤存儲系統(tǒng)模型（Dynamic Tree-basedMassive Array of Idle Disks, DT-MAID）。DT-MAID針對現(xiàn)有節(jié)能磁盤存儲系統(tǒng)存在的瓶頸和局部性問題，提出了從系統(tǒng)整體角度來研究節(jié)能磁盤存儲系統(tǒng)模型的思想，即：將緩存磁盤組織成兩層動態(tài)樹的結(jié)構(gòu)，在動態(tài)樹結(jié)構(gòu)中的緩存磁盤分成Master和Slave兩種角色；動態(tài)樹管理將負(fù)載導(dǎo)向Master，以有利于Slave進入待機狀態(tài)而提高系統(tǒng)的能效；Slave并不是一直在旋轉(zhuǎn)，當(dāng)其負(fù)載較輕滿足設(shè)定的條件時，將會進入待機狀態(tài)以節(jié)省能量；當(dāng)Master的負(fù)載較大或系統(tǒng)被訪問的數(shù)據(jù)的覆蓋率較大時，動態(tài)樹管理會從處于非待機狀態(tài)的Slave中選擇一個合適的節(jié)點或恢復(fù)一個待機狀態(tài)的Slave接替其工作，從而提高系統(tǒng)的服務(wù)能力和能效。理論分析和實驗結(jié)果表明，DT-MAID是一種較理想的節(jié)能磁盤存儲系統(tǒng)模型，具有較好的節(jié)能效果、服務(wù)質(zhì)量和可靠性。
[Abstract]:In recent years , with the rapid growth of the global digital information , the problem of energy consumption of the storage system has become more and more serious , and how to reduce the energy consumption of the storage system has become a key science and technology problem and hot topic research topic in the field of computer science and technology . Therefore , the research on this aspect has important academic value and practical significance in promoting the development and application of energy saving technology of the storage system .

This paper starts with the energy - saving principle of magnetic disk storage system to improve the energy efficiency and performance of magnetic disk storage system . On the basis of careful analysis and research on energy - saving requirements and technical factors of disk storage system , the key technical problems such as disk power management , cache disk data management , disk load balancing and energy - saving disk storage system model of magnetic disk storage system are studied and discussed .

( 1 ) An Access ModeMatching and Performance Guarantee ( AMM - PG ) is proposed in this paper . aiming at the shortage of the existing disk power management strategy , the AMM - PG adopts the data histogram of the idle interval length to correspond to the access mode of the disk and derives the corresponding time - out threshold value , so that the access mode of the adaptive disk is realized to reasonably manage the power supply of the magnetic disk , thereby improving the energy efficiency of the magnetic disk ;
in addition , that AMM - PG increase the access mode of the partial disk to the idle state when the disk is returned from the stand - by state to the idle state , and the maximum time of one disk in the standby state is determined through the performance requirement and the data histogram of the idle interval length , so that the energy saving amount is improved as much as possible while meeting the performance requirement .

( 2 ) A load - adaptive energy - aware caching disk data management method ( workload - Adaptive lru , wa - lru ) is proposed . According to the shortage of the existing cache disk data management method , the WA - lru distinguishes the energy - saving priority of the data disk according to the quantity of the hot data and the intensity of the data access , and always replaces the data in the data disk with the lowest energy - saving priority when the data in the cache disk needs to be replaced , so that the bottleneck effect of the cache disk is identified and eliminated by using the lru queue based on the life cycle management .

( 3 ) A multi - queue load balancing method ( MQLB ) for energy - saving disk storage system is proposed . This paper divides the data into a plurality of queues according to the factors such as the access frequency of the data , the locality of the access time and the like according to the factors such as the access frequency of the data , the locality of the access time and the like according to the factors such as the access frequency of the data , the locality of the access time and the like .

( 4 ) A new Dynamic Tree - based Array of Idle Disks ( DT - MAID ) is proposed . DT - MAID aims at the bottleneck and local problems existing in the existing energy - saving disk storage system , and puts forward the idea of studying the energy - saving disk storage system model from the whole angle of the system , that is , the cache disk is organized into two layers of dynamic tree structure , and the cache disk in the dynamic tree structure is divided into two roles of Master and Slave ;
the dynamic tree management loads the load to the Master to facilitate Slave to enter a standby state to improve the energy efficiency of the system ;
Slave is not always spinning , when its load is lighter than the set condition , it will enter standby state to save energy ;
The results show that DT - MAID is an ideal energy - saving disk storage system model , which has good energy - saving effect , quality of service and reliability .

【學(xué)位授予單位】：華南理工大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2012
【分類號】：TP333

【參考文獻】

相關(guān)期刊論文 前5條

1 黃河清;宋曉華;曹元大;;異構(gòu)存儲系統(tǒng)中基于能量模型的文件遷移策略[J];北京航空航天大學(xué)學(xué)報;2007年09期

2 陳超;趙躍龍;王文豐;潘敏;;基于反饋的改進動態(tài)負(fù)載均衡策略[J];計算機工程;2010年14期

3 王文豐;趙躍龍;;一種基于系統(tǒng)負(fù)載的輪流服務(wù)模型[J];華南理工大學(xué)學(xué)報(自然科學(xué)版);2009年10期

4 王少峰;周忠;吳威;;一種面向分布式虛擬環(huán)境的分層迭代負(fù)載平衡算法[J];軟件學(xué)報;2008年09期

5 柴云鵬;都志輝;李三立;;保障服務(wù)質(zhì)量的節(jié)能流媒體存儲系統(tǒng)[J];清華大學(xué)學(xué)報(自然科學(xué)版);2010年01期

本文編號：1718213