【摘要】：作為虛擬化的個重要特性，檢查點可以對虛擬機(jī)進(jìn)行系統(tǒng)級的狀態(tài)保存，并確保完成后虛擬機(jī)可以繼續(xù)運行，目前已經(jīng)廣泛應(yīng)用于集群和數(shù)據(jù)中心的系統(tǒng)維護(hù)、容錯災(zāi)備及測試調(diào)試上。盡管全系統(tǒng)級檢查點是種廣泛使用且有效的狀態(tài)保存策略，但由于它需要向磁盤寫入大量的數(shù)據(jù)而導(dǎo)致檢查點時間過長，進(jìn)而影響虛擬機(jī)的服務(wù)質(zhì)量及系統(tǒng)的可擴(kuò)展性，而且隨著虛擬機(jī)內(nèi)存配置的增加，檢查點時間隨之線性增加。 虛擬機(jī)檢查點優(yōu)化機(jī)制有效地解決了上述問題。該機(jī)制針對虛擬機(jī)在不同負(fù)載下的內(nèi)存變化特征，首先分析臟頁集隨時間的變化情況，然后以字節(jié)為單位細(xì)粒度分析臟頁面內(nèi)臟數(shù)據(jù)的變化情況，，從而代替?zhèn)鹘y(tǒng)方法中的頁面粒度，并分析了臟字節(jié)在臟頁面內(nèi)的分布情況。根據(jù)上述分析結(jié)果，該機(jī)制采用臟頁記錄機(jī)制記錄虛擬機(jī)運行過程中產(chǎn)生的臟頁，通過臟頁位圖進(jìn)行標(biāo)記，采用Delta內(nèi)存壓縮方案無失真地提取出臟頁面中的臟數(shù)據(jù)，減少了虛擬機(jī)狀態(tài)中的冗余數(shù)據(jù)，并使用多線程技術(shù)加快了內(nèi)存壓縮。另外，該機(jī)制通過分析虛擬機(jī)的臟頁分布特征，提出了種快速掃描虛擬機(jī)臟頁位圖的方法找到變臟頁面，有效減少了虛擬機(jī)的檢查點時間。 實驗表明，與Xen中原始的全系統(tǒng)檢查點方法相比，在內(nèi)存配置1GB時，該機(jī)制平均可減少70.54%的檢查點時間，對于大內(nèi)存的虛擬機(jī)還可以取得更好的性能；與Remus以頁面為粒度的增量式檢查點方法相比，該機(jī)制平均減少52.88%的檢查點文件大�。慌cXen中傳統(tǒng)的臟頁位圖掃描方法相比，該機(jī)制的平均掃描時間減少了91.13%。
[Abstract]:As an important feature of virtualization, checkpoint can save the state of virtual machine at system level and ensure that the virtual machine can continue to run after completion, which has been widely used in system maintenance of cluster and data center. Fault-tolerant disaster preparedness and test debugging. Although the whole system checkpoint is a widely used and effective state preservation strategy, because it needs to write a large amount of data to the disk, the checkpoint time is too long, which affects the quality of service of virtual machine and the expansibility of the system. And with the increase of virtual machine memory configuration, checkpoint time increases linearly. The checkpoint optimization mechanism of virtual machine effectively solves the above problems. According to the memory characteristics of virtual machine under different loads, the mechanism firstly analyzes the change of dirty page set with time, and then analyzes the change of visceral data of dirty page with fine granularity in bytes. In order to replace the traditional method of page granularity, and analyze the distribution of dirty sections in dirty pages. According to the above analysis results, this mechanism uses the dirty page recording mechanism to record the dirty pages generated in the running process of the virtual machine, marks the dirty pages by the dirty page bitmap, and extracts the dirty data from the dirty pages without distortion by using the Delta memory compression scheme. The redundant data in virtual machine state is reduced and memory compression is accelerated by multi-thread technology. In addition, by analyzing the distribution characteristics of the dirty pages of the virtual machine, this mechanism proposes a method of quickly scanning the dirty page bitmap of the virtual machine to find the dirty pages, which effectively reduces the checkpoint time of the virtual machine. The experimental results show that compared with the original full-system checkpoint method in Xen, the mechanism can reduce the checkpointing time by 70.54% on average when the memory is configured with 1GB, and it can also achieve better performance for the large memory virtual machine. Compared with the incremental checkpoint method with page granularity in Remus, this mechanism reduces the average checkpoint file size by 52.88%, and reduces the average scanning time by 91.13% compared with the traditional dirty page bitmap scanning method in Xen.
【學(xué)位授予單位】：華中科技大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2013
【分類號】：TP302

【參考文獻(xiàn)】

相關(guān)期刊論文 前2條

1 劉鵬程;陳榕;;面向云計算的虛擬機(jī)動態(tài)遷移框架[J];計算機(jī)工程;2010年05期

2 溫抿雄;李沁;;虛擬機(jī)遷移系統(tǒng)網(wǎng)絡(luò)連接重定向技術(shù)[J];計算機(jī)應(yīng)用研究;2009年05期

本文編號：2386430