發(fā)布時(shí)間：2018-06-05 08:27

  本文選題：容錯(cuò)能力評(píng)測(cè) + 云計(jì)算��； 參考：《哈爾濱工業(yè)大學(xué)》2013年碩士論文

【摘要】：隨著信息系統(tǒng)中數(shù)據(jù)量的迅速增長(zhǎng)，傳統(tǒng)的計(jì)算和存儲(chǔ)模式已經(jīng)不能滿足日益增長(zhǎng)的數(shù)據(jù)處理和存儲(chǔ)需求。在早期分布式處理、并行處理以及網(wǎng)格計(jì)算技術(shù)的基礎(chǔ)之上發(fā)展而來的云計(jì)算技術(shù)成為目前炙手可熱的海量數(shù)據(jù)解決手段。但是，隨著云計(jì)算平臺(tái)的普及與推廣，可靠性成為它面臨的一項(xiàng)重大困難與挑戰(zhàn)。容錯(cuò)能力能夠從一個(gè)層面上反映出系統(tǒng)可靠性的高低。因此，評(píng)測(cè)云計(jì)算平臺(tái)的容錯(cuò)能力對(duì)其可靠性研究具有重要意義。 由于云計(jì)算平臺(tái)的復(fù)雜程度高、軟件規(guī)模大，使得云計(jì)算平臺(tái)的容錯(cuò)能力評(píng)測(cè)成為一項(xiàng)艱巨的任務(wù)。目前針對(duì)云計(jì)算平臺(tái)測(cè)試已有的研究成果中，涉及容錯(cuò)能力評(píng)測(cè)的內(nèi)容較少并且測(cè)試方法較為單一，需要更加深入的研究與完善。 評(píng)測(cè)容錯(cuò)能力的最有效手段是基于故障注入思想的測(cè)試方法。本文以開源云計(jì)算平臺(tái)—Hadoop作為研究目標(biāo)，深入研究Hadoop中核心組件的容錯(cuò)機(jī)制�；诂F(xiàn)實(shí)應(yīng)用中可能出現(xiàn)的故障類型，結(jié)合Hadoop平臺(tái)的部署結(jié)構(gòu)以及運(yùn)行機(jī)制的特點(diǎn)，，提出了一種多層次的Hadoop容錯(cuò)能力測(cè)試框架。從軟件健壯性測(cè)試、MapReduce故障注入測(cè)試、網(wǎng)絡(luò)故障注入測(cè)試和HDFS故障注入測(cè)試四個(gè)層次出發(fā)，分別模擬了云計(jì)算平臺(tái)在應(yīng)用中可能出現(xiàn)的多種軟硬件異常。 依據(jù)Hadoop多層次容錯(cuò)能力測(cè)試框架，設(shè)計(jì)了針對(duì)Hadoop的容錯(cuò)能力評(píng)測(cè)平臺(tái)，實(shí)現(xiàn)了多種故障注入工具，形成了對(duì)Hadoop的軟件健壯性測(cè)試以及可能發(fā)生節(jié)點(diǎn)故障、網(wǎng)絡(luò)故障、硬盤故障等故障類型的容錯(cuò)能力測(cè)試的覆蓋。在故障注入過程中，監(jiān)控和回收云計(jì)算平臺(tái)對(duì)于故障的反饋信息，進(jìn)行結(jié)果分析從而向研究人員提供真實(shí)可靠的評(píng)測(cè)結(jié)果，最終為云計(jì)算平臺(tái)的容錯(cuò)能力評(píng)測(cè)提供有力的數(shù)據(jù)支撐。 為了驗(yàn)證Hadoop容錯(cuò)能力評(píng)測(cè)方法的可行性，本文使用Hadoop搭建了一個(gè)小型的云計(jì)算平臺(tái)環(huán)境進(jìn)行實(shí)驗(yàn)。軟件健壯性測(cè)試發(fā)現(xiàn)了Hadoop接口和實(shí)現(xiàn)中存在的不足并進(jìn)行了缺陷定位。在測(cè)試環(huán)境中進(jìn)行的節(jié)點(diǎn)或進(jìn)程級(jí)失效故障、數(shù)據(jù)操作失效故障、數(shù)據(jù)校驗(yàn)故障、資源過載故障和網(wǎng)絡(luò)故障的注入測(cè)試有效地證明了各個(gè)故障注入工具的有效性。通過結(jié)合Hadoop基準(zhǔn)性能測(cè)試程序進(jìn)行故障注入前后性能變化情況的對(duì)比，能夠?qū)adoop平臺(tái)的容錯(cuò)能力進(jìn)行定性的評(píng)測(cè)。
[Abstract]:With the rapid growth of data in information systems, the traditional computing and storage mode can no longer meet the increasing demand for data processing and storage. Cloud computing technology developed on the basis of early distributed processing, parallel processing and grid computing technology has become a hot solution to mass data. However, with the popularization and popularization of cloud computing platform, reliability becomes a major difficulty and challenge. Fault-tolerant ability can reflect the reliability of the system from one level. Therefore, it is important to evaluate the fault tolerance of cloud computing platform. Due to the complexity of cloud computing platform and the large scale of software, it becomes a difficult task to evaluate the fault tolerance of cloud computing platform. At present, among the existing research results of cloud computing platform testing, the content of fault tolerance capability evaluation is less and the test method is relatively single, so it needs more in-depth research and improvement. The test method based on fault injection is the most effective method to evaluate fault tolerance. In this paper, the open source cloud computing platform-Hadoop is taken as the research goal, and the fault-tolerant mechanism of core components in Hadoop is deeply studied. Based on the possible fault types in practical applications and the characteristics of deployment structure and running mechanism of Hadoop platform, a multi-level Hadoop fault-tolerant capability testing framework is proposed. From the four levels of software robustness test MapReduce fault injection test network fault injection test and HDFS fault injection test several software and hardware anomalies in cloud computing platform are simulated respectively. According to the testing framework of Hadoop multi-level fault-tolerant capability, a fault tolerance evaluation platform for Hadoop is designed, and many kinds of fault injection tools are realized. The software robustness test of Hadoop and the possible node faults and network failures are formed. Hard disk failure and other types of fault tolerance test coverage. In the process of fault injection, the feedback information of cloud computing platform is monitored and recycled, and the result analysis is carried out in order to provide the researchers with real and reliable evaluation results. Finally, it provides powerful data support for fault tolerance evaluation of cloud computing platform. In order to verify the feasibility of Hadoop fault-tolerant capability evaluation method, this paper uses Hadoop to build a small cloud computing platform environment for experiments. Software robustness test found the shortcomings of Hadoop interface and implementation, and carried out defect location. The injection tests of node or process level failure, data operation failure, data check fault, resource overload fault and network fault in the test environment effectively prove the effectiveness of each fault injection tool. The fault tolerance of Hadoop platform can be evaluated qualitatively by comparing the performance changes before and after fault injection with Hadoop benchmark performance test program.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2013
【分類號(hào)】：TP302.8

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 朱鵬,張平;基于單片機(jī)的故障注入系統(tǒng)[J];計(jì)算機(jī)測(cè)量與控制;2004年10期

2 王建瑩,孫峻朝,李運(yùn)策,楊孝宗;FTT-1:一個(gè)基于硬件的故障注入器的設(shè)計(jì)與實(shí)現(xiàn)[J];計(jì)算機(jī)工程與設(shè)計(jì);1998年04期

3 王建瑩,楊孝宗,徐海智;用軟件實(shí)現(xiàn)的故障注入工具評(píng)估錯(cuò)誤檢測(cè)機(jī)制[J];小型微型計(jì)算機(jī)系統(tǒng);2000年05期

4 賀R

本文編號(hào)：1981333