發(fā)布時間：2018-03-28 04:12

  本文選題：云平臺　切入點：資源分配　出處：《哈爾濱工程大學(xué)》2014年碩士論文

【摘要】：隨著云計算的發(fā)展,按需付費的使用模式逐漸成為趨勢。在按需付費的模式下,云計算平臺為了提高經(jīng)濟效益就必須具備按用戶需求自適應(yīng)資源分配的能力。基于預(yù)測的云平臺自適應(yīng)資源分配技術(shù)能夠根據(jù)云平臺中應(yīng)用的歷史運行數(shù)據(jù)對其未來的資源需求做出預(yù)測,從而準(zhǔn)確的對該應(yīng)用的資源做出動態(tài)調(diào)整。目前,對云平臺應(yīng)用資源需求預(yù)測的研究主要集中在單一預(yù)測模型或者方法上,缺乏對預(yù)測樣本的分類,導(dǎo)致預(yù)測結(jié)果不夠精確;對自適應(yīng)資源分配的研究著重于虛擬機自適應(yīng)資源配置,沒有將虛擬機資源調(diào)整與虛擬機重置結(jié)合起來。據(jù)此本文提出一種基于混合預(yù)測的云平臺自適應(yīng)資源分配方法,通過混合預(yù)測模型和多粒度自適應(yīng)資源分配,獲得更高的云平臺資源利用效率并保障用戶的 SLA(Service-Level Agreement)。本文首先對當(dāng)前云平臺的資源分配現(xiàn)狀進行研究與分析,在其基礎(chǔ)上研究了虛擬機放置問題的建模方法和虛擬機重置問題成本分析方法,并且通過對現(xiàn)有預(yù)測模型的分析研究,選擇了更適合當(dāng)前云平臺應(yīng)用特點的Markov Chain預(yù)測方法和FFT預(yù)測方法作為混合預(yù)測模型的基礎(chǔ),討論了這兩種算法相結(jié)合進行混合資源需求預(yù)測的方法。所提出的基于混合預(yù)測的云平臺自適應(yīng)資源分配方法,按照應(yīng)用資源需求變化的周期性特點進行分類,對周期性或非周期性應(yīng)用采用不同的預(yù)測模型;以預(yù)測結(jié)果為基礎(chǔ),分別采用基于混合預(yù)測的虛擬機資源動態(tài)分配策略,基于混合預(yù)測的虛擬機在線遷移策略,基于混合預(yù)測的虛擬機動態(tài)重置策略三種策略進行多粒度的云平臺自適應(yīng)資源分配,以有效適應(yīng)應(yīng)用需求變化,減少虛擬機遷移的數(shù)量,降低違反SLA概率,減少虛擬機占用的物理機數(shù)量,最終達到提高云平臺系統(tǒng)資源利用效率的目的。最后通過實驗證明,基于混合預(yù)測的云平臺自適應(yīng)資源分配方法可以有效的進行應(yīng)用資源需求的預(yù)測并自適應(yīng)資源分配,在提高虛擬機資源利用效率、減少物理機占用和降低SLA方面達到了預(yù)期效果。
[Abstract]:With the development of cloud computing, pay-on-demand is becoming a trend. In order to improve economic benefits, cloud computing platform must have the ability of adaptive resource allocation according to user's demand. The technology of adaptive resource allocation in cloud platform based on prediction can be based on the historical running data of cloud platform application. Future resource needs are predicted, At present, the research on cloud platform application resource demand forecasting is mainly focused on a single prediction model or method, the lack of classification of forecasting samples, resulting in the prediction results are not accurate; The research of adaptive resource allocation focuses on the adaptive resource allocation of virtual machine, and does not combine the adjustment of virtual machine resources with the reset of virtual machine. Based on this, an adaptive resource allocation method for cloud platform based on mixed prediction is proposed. Through mixed prediction model and multi-granularity adaptive resource allocation, we can obtain higher resource utilization efficiency of cloud platform and ensure user's SLA(Service-Level agreement. Firstly, this paper studies and analyzes the current situation of resource allocation in cloud platform. On the basis of it, the modeling method of virtual machine placement problem and the cost analysis method of virtual machine reset problem are studied, and the existing prediction models are analyzed and studied. The Markov Chain forecasting method and the FFT forecasting method, which are more suitable for the current cloud platform application characteristics, are selected as the basis of the hybrid prediction model. This paper discusses the method of combining these two algorithms to forecast the demand of mixed resources. The proposed adaptive resource allocation method based on hybrid prediction is classified according to the periodicity of the change of resource demand. Different prediction models are used for periodic or aperiodic applications, based on prediction results, virtual machine resource dynamic allocation strategy based on hybrid prediction and virtual machine online migration strategy based on hybrid prediction are adopted, respectively. The dynamic reset strategy of virtual machine based on mixed prediction is applied to multi-granularity adaptive resource allocation of cloud platform to adapt to the change of application requirements, reduce the number of virtual machine migration, and reduce the probability of violating SLA. Reduce the number of physical machines consumed by virtual machines, and finally achieve the purpose of improving the efficiency of resource utilization of cloud platform system. Finally, through experiments, it is proved that, The adaptive resource allocation method of cloud platform based on hybrid prediction can effectively predict the application resource requirements and adaptively allocate resources, which can improve the efficiency of virtual machine resource utilization. The expected results are achieved in terms of reducing the footprint of physical computers and reducing SLA.
【學(xué)位授予單位】：哈爾濱工程大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：TP393.09

【參考文獻】

相關(guān)期刊論文 前2條

1 師雪霖清華大學(xué)計算機科學(xué)與技術(shù)系 清華信息科學(xué)與技術(shù)國家實驗室(籌);徐恪;;云虛擬機資源分配的效用最大化模型[J];計算機學(xué)報;2013年02期

2 華夏渝;鄭駿;胡文心;;基于云計算環(huán)境的蟻群優(yōu)化計算資源分配算法[J];華東師范大學(xué)學(xué)報(自然科學(xué)版);2010年01期

，

本文編號：1674688