本文選題：云計(jì)算 + 對(duì)等網(wǎng)絡(luò)��； 參考：《南京郵電大學(xué)》2014年碩士論文

【摘要】：云計(jì)算的概念自提出以來，即在全世界范圍內(nèi)得到了廣泛應(yīng)用。云計(jì)算憑借其高度并行性、虛擬化等特點(diǎn)大幅簡化了程序員的工作，提高了作業(yè)執(zhí)行速率。MapReduce作為云計(jì)算中的核心編程模型，為程序員自動(dòng)完成了容錯(cuò)、死鎖等至今仍無通用解決方案的問題。如今，越來越多的企業(yè)都已經(jīng)將自己的業(yè)務(wù)部署到了云計(jì)算平臺(tái)中。然而，傳統(tǒng)云計(jì)算平臺(tái)中的MapReduce編程模型是基于集中式的C/S框架，模型中由一個(gè)控制節(jié)點(diǎn)管理多個(gè)工作節(jié)點(diǎn)，且用戶提交的作業(yè)全部都由唯一的控制節(jié)點(diǎn)進(jìn)行分配管理，控制節(jié)點(diǎn)一旦失效，就會(huì)造成無法挽回的后果。不但如此，單一控制節(jié)點(diǎn)負(fù)載有限，隨著集群規(guī)模的不斷增大，其勢必會(huì)成為整個(gè)系統(tǒng)的瓶頸，嚴(yán)重限制了集群的擴(kuò)展。 因此，為了解決傳統(tǒng)云計(jì)算平臺(tái)中單控制節(jié)點(diǎn)失效和集群擴(kuò)展性不足兩大問題，本文提出了基于P2P-云平臺(tái)的作業(yè)管理機(jī)制。該機(jī)制首先對(duì)傳統(tǒng)云計(jì)算平臺(tái)中基于單控制節(jié)點(diǎn)的主從結(jié)構(gòu)進(jìn)行改進(jìn)，提出基于Chord協(xié)議的多控制節(jié)點(diǎn)協(xié)同控制多工作節(jié)點(diǎn)的體系結(jié)構(gòu)ChordMR。其次，針對(duì)控制節(jié)點(diǎn)失效所帶來的作業(yè)丟失問題，提出作業(yè)動(dòng)態(tài)管理機(jī)制，該機(jī)制從作業(yè)分配、作業(yè)備份、作業(yè)恢復(fù)以及控制節(jié)點(diǎn)加入時(shí)的作業(yè)遷移等各方面對(duì)云作業(yè)進(jìn)行有效管理，以加快云作業(yè)執(zhí)行速率、提高云作業(yè)執(zhí)行成功率。通過仿真實(shí)驗(yàn)，驗(yàn)證了該作業(yè)管理機(jī)制在可行性和有效性等方面的優(yōu)勢。最后，本文編程實(shí)現(xiàn)了基于P2P-云平臺(tái)的MapReduce作業(yè)處理框架。理論分析和實(shí)驗(yàn)結(jié)果表明，，傳統(tǒng)云平臺(tái)與P2P技術(shù)的有效結(jié)合不但解決了傳統(tǒng)云平臺(tái)中存在的單點(diǎn)失效問題，而且增強(qiáng)了集群的擴(kuò)展性，有效解決了系統(tǒng)的瓶頸問題。
[Abstract]:The concept of cloud computing has been widely used in the world since it was put forward. Because of its high parallelism, virtualization and other characteristics, cloud computing greatly simplifies the work of programmers, improves the job execution rate. MapReduce as the core programming model in cloud computing, automatically completes fault tolerance for programmers. Deadlock is still not a common solution to the problem. Nowadays, more and more enterprises have deployed their business to cloud computing platform. However, the MapReduce programming model in the traditional cloud computing platform is based on the centralized C / S framework. In the model, multiple work nodes are managed by one control node, and all the jobs submitted by the user are allocated and managed by the unique control node. Once the control node fails, there will be irreparable consequences. Moreover, the load of single control node is limited. With the increasing of cluster scale, it will become the bottleneck of the whole system, which seriously limits the expansion of the cluster. Therefore, in order to solve the problems of failure of single control node and insufficient scalability of cluster in traditional cloud computing platform, a job management mechanism based on P2P- cloud platform is proposed in this paper. The mechanism firstly improves the master-slave structure based on single control node in traditional cloud computing platform, and proposes a chord protocol based architecture for multi-control nodes to control multi-work nodes. Secondly, aiming at the problem of job loss caused by the failure of control nodes, a dynamic job management mechanism is proposed, which includes job assignment, job backup, and so on. In order to speed up the cloud job execution rate and improve the success rate of cloud job execution, the job recovery and job migration when the control nodes join are effectively managed in order to speed up the cloud job execution rate. The feasibility and effectiveness of the job management mechanism are verified by simulation experiments. Finally, a MapReduce job processing framework based on P2P-cloud platform is implemented. The theoretical analysis and experimental results show that the combination of traditional cloud platform and P2P technology not only solves the single point failure problem in traditional cloud platform, but also enhances the expansibility of cluster and effectively solves the bottleneck problem of the system.
【學(xué)位授予單位】：南京郵電大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2014
【分類號(hào)】：TP393.02;TP311.1

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 金晶;王妍;李昕;陳山枝;;MapReduce架構(gòu)的多控制節(jié)點(diǎn)改進(jìn)[J];北京郵電大學(xué)學(xué)報(bào);2012年04期

本文編號(hào)：2033701