【摘要】：隨著信息技術(shù)的不斷發(fā)展,全球信息存儲量正在以57%的復(fù)合年增長率膨脹。大量數(shù)據(jù)密集型應(yīng)用對現(xiàn)有的網(wǎng)絡(luò)存儲架構(gòu)提出了更高的要求。對象存儲技術(shù)同時具備傳統(tǒng)SAN快速訪問的特點和NAS跨平臺數(shù)據(jù)共享和強安全性的優(yōu)勢。同時對象存儲設(shè)備本身具備的智能管理功能。因此,對象存儲系統(tǒng)具有高性能、可擴展、高可用性和強安全性等特點,是未來網(wǎng)絡(luò)存儲的發(fā)展方向。元數(shù)據(jù)服務(wù)器(MetaData Server, MDS)是對象存儲系統(tǒng)的重要組成部分,在對象存儲系統(tǒng)中的主要功能包括：管理文件系統(tǒng)名字空間、文件／目錄和對象物理存儲位置之間的映射、用戶訪問控制等。因此,MDS集群的處理能力是整個對象存儲系統(tǒng)高效運作的重要一環(huán)。 本文分析了當前主流網(wǎng)絡(luò)存儲架構(gòu)中元數(shù)據(jù)服務(wù)器集群的負載均衡方案。針對元數(shù)據(jù)服務(wù)器性能異構(gòu)造成的集群負載不均衡問題,本文提出了一種基于權(quán)重的元數(shù)據(jù)服務(wù)器集群負載均衡策略(Weight-based Metadata Server Cluster Load Balancing Strategy, WLBS),并對其在Ceph框架中進行實現(xiàn)。本文的主要研究和開發(fā)內(nèi)容包括以下四個方面： (1)在分析對象存儲與傳統(tǒng)網(wǎng)絡(luò)存儲技術(shù)的基礎(chǔ)上,總結(jié)了MDS負載均衡技術(shù)所涉及的包括元數(shù)據(jù)分區(qū)和服務(wù)器熱點評價兩個關(guān)鍵問題； (2)提出了元數(shù)據(jù)服務(wù)器熱點評測模型,主要解決準確獲取服務(wù)負載狀況的問題。該模型在傳統(tǒng)計數(shù)法的基礎(chǔ)上增加了對MDS性能值的考慮,提出了基于操作時延的MDS性能模型。該性能模型避免了傳統(tǒng)“白盒”性能模型量化資源影響力的難題,準確評估服務(wù)器的當前性能值； (3)提出了一種基于權(quán)重負載均衡策略,主要解決MDS性能異構(gòu)情況下的負載均衡問題。該策略根據(jù)(2)中熱點評測模型獲取節(jié)點當前的負載狀況,并計算出集群的平均負載水平。為了實現(xiàn)負載與性能相對應(yīng)的目標,高于平均負載水平的MDS會主動以目錄子樹遷移方式進行負載遷移,將一部分目錄子樹遷移到低于平均負載水平的MDS,最終實現(xiàn)異構(gòu)MDS集群的負載均衡； (4)搭建實驗平臺,并對策略進行實驗評測。對WLBS是否具備對異構(gòu)MDS集群負載進行調(diào)整而使負載均衡化和保持MDS集群處理能力進行驗證。 大量實驗證明本文提出的負載均衡策略可以在異構(gòu)環(huán)境下使元數(shù)據(jù)服務(wù)器集群保持負載均衡,提高了系統(tǒng)的元數(shù)據(jù)處理能力。
[Abstract]:With the development of information technology, global information storage is expanding at a compound annual growth rate of 57%. A large number of data-intensive applications put forward higher requirements for the existing network storage architecture. Object storage technology has the characteristics of traditional SAN fast access and NAS cross-platform data sharing and strong security advantages. At the same time, the object storage device itself has the intelligent management function. Therefore, object storage system has the characteristics of high performance, scalability, high availability and strong security, which is the development direction of network storage in the future. Metadata server (MetaData Server, MDS) is an important part of object storage system. Its main functions in object storage system include managing file system namespace, mapping file / directory and object physical storage location. User access control, etc. Therefore, the processing capability of MDS cluster is an important part of the efficient operation of the whole object storage system. This paper analyzes the load balancing scheme of metadata server cluster in current mainstream network storage architecture. Aiming at the problem of cluster load imbalance caused by heterogeneous metadata server performance, this paper proposes a load balancing strategy (Weight-based Metadata Server Cluster Load Balancing Strategy, WLBS),) for metadata server cluster based on weight and implements it in the framework of Ceph. The main research and development contents of this paper include the following four aspects: (1) based on the analysis of object storage and traditional network storage technology, This paper summarizes two key issues involved in MDS load balancing technology, including metadata partition and server hot spot evaluation. (2) A metadata server hot spot evaluation model is proposed, which mainly solves the problem of accurately obtaining service load status. Based on the traditional counting method, the MDS performance value is considered in this model, and the MDS performance model based on operation delay is proposed. The performance model avoids the problem of quantifying the resource influence of the traditional "white box" performance model and accurately evaluates the current performance value of the server. (3) A load balancing strategy based on weight is proposed. It mainly solves the load balancing problem in the case of heterogeneous MDS performance. According to the hot spot evaluation model in (2), the current load status of the node is obtained, and the average load level of the cluster is calculated. In order to achieve the goal of corresponding load and performance, MDS with higher than average load level will actively migrate the load in directory subtree mode. Part of the directory subtrees are migrated to MDSs which are below average load level, finally the load balance of heterogeneous MDS cluster is realized. (4) the experimental platform is built, and the policy is evaluated experimentally. This paper verifies whether WLBS has the ability to adjust the load of heterogeneous MDS cluster to balance the load and maintain the processing ability of MDS cluster. A large number of experiments show that the proposed load balancing strategy can keep the load balance of metadata server cluster in heterogeneous environment and improve the metadata processing ability of the system.
【學(xué)位授予單位】：杭州電子科技大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2012
【分類號】：TP368.5

【參考文獻】

相關(guān)期刊論文 前9條

1 郭全生,舒繼武,毛希平,溫冬蟬,鄭緯民;基于LVS系統(tǒng)的負載動態(tài)平衡設(shè)計與實現(xiàn)[J];計算機研究與發(fā)展;2004年06期

2 楊德志,黃華,張建剛,許魯;大容量、高性能、高擴展能力的藍鯨分布式文件系統(tǒng)[J];計算機研究與發(fā)展;2005年06期

3 喻莉;石冰心;;基于Web服務(wù)器的性能模型與參數(shù)分析[J];計算機科學(xué);1999年07期

4 唐丹,金海,張永坤;集群動態(tài)負載平衡系統(tǒng)的性能評價[J];計算機學(xué)報;2004年06期

5 郭成城,晏蒲柳;一種異構(gòu)Web服務(wù)器集群動態(tài)負載均衡算法[J];計算機學(xué)報;2005年02期

6 曾國蓀,263.net,陸鑫達;異構(gòu)計算中的負載共享[J];軟件學(xué)報;2000年04期

7 單志廣,戴瓊海,林闖,楊揚;Web請求分配和選擇的綜合方案與性能分析[J];軟件學(xué)報;2001年03期

8 于磊,林宗楷,郭玉釵,林守勛;多服務(wù)器系統(tǒng)中的負載平衡與容錯[J];系統(tǒng)仿真學(xué)報;2001年03期

9 王娟;馮丹;王芳;廖振松;;一種元數(shù)據(jù)服務(wù)器集群的負載均衡算法[J];小型微型計算機系統(tǒng);2009年04期

，

本文編號：2176871