云计算多实例市场预测与组合购买决策研究

发布时间：2018-06-20 09:21

本文选题：云计算 + 多实例　；参考：《武汉理工大学》2013年硕士论文

【摘要】：云计算是一种虚拟化的、可伸缩的IT服务,它可以按照用户的需求动态地提供服务。随着各云平台用户需求的不断增加,其资源负荷量也有了较大的变化,如何准确预测客户需求并合理分配云资源成为各平台供应商面临的挑战。因此,各种云资源的市场交易平台应运而生,为用户及时、便捷的获取云服务提供了可靠保障。然而,云计算资源目前出现了多种购买模式,如亚马逊弹性云平台的3种类型的实例计费方案,分别为：按需运行实例、保留定制实例和现货竞价实例。前两类实例的价格一定,但是,关于云供应商如何针对现货竞价实例进行定价及定价趋势的问题,目前的研究结果尚未十分明朗,另一方面,用户如何更好地从云计算环境中获得决策过程所需的决策信息、如何从海量决策信息中处理动态的用户需求信息,都会影响用户最终的购买决策效率。因此研究云计算多实例的市场交易和购买决策十分必要。云计算多实例的市场交易和购买决策面临的主要挑战有：(1)云供应商如何针对现货竞价实例定价及其价格变动的趋势研究只考虑了定性的影响趋势,无法量化；(2)用户对于自身需求信息无法进行全面的处理和分析；(3)在购买决策过程中,只考虑了服务质量这一单一因素,使得服务呈现单一化的趋势,无法满足用户多方面的需求。基于以上问题,本文提出了基于云联盟的云计算市场交易体系,为供需双方的交易提供架构支撑。然后构建云计算市场交易预测模型,包括现货竞价实例价格预测模型和客户需求预测模型。最后研究了基于客户的多实例组合购买决策,以达到满足用户多样化需求的目标。具体的工作包括： (1)提出了基于云计算的市场交易体系； (2)设计了云计算市场现货实例的价格预测模型和算法,充分挖掘云供应商现货实例的定价规律,为云用户的投标决策提供依据； (3)设计了云计算市场客户需求预测模型,并采用灰色BP神经网络进行仿真训练,改进了传统BP神经网络的缺点,使得预测值更为精确,从而充分挖掘云用户自身的需求,使得用户更好的了解自身的需求规律,优化购买决策； (4)构建了基于客户的云计算市场多实例组合购买决策模型,首先从成本优化和服务时间最小入手,结合不同实例类型的特点,构建单目标约束模型,然后综合考虑客户各方面的需求,构建双目标约束模型,同时达到客户成本最小化和服务时间最短的目标。
[Abstract]:Cloud computing is a virtualized, scalable IT service that provides services dynamically according to user needs. With the increasing demand of cloud platform users, the amount of resource load has also changed greatly. How to accurately predict customer demand and allocate cloud resources rationally becomes a challenge to each platform provider. Therefore, various cloud resources market trading platform emerged as the times require, providing a reliable guarantee for users to obtain cloud services in a timely and convenient manner. However, there are many purchase modes for cloud computing resources at present, such as three types of instance billing schemes of Amazon elastic cloud platform, which are: running on demand instance, retaining customization instance and spot bidding instance. The price of the first two types of examples is certain. However, the current research results on how cloud suppliers carry out pricing and pricing trends for spot bidding examples are not very clear. On the other hand, How to obtain the decision information from the cloud computing environment and how to deal with the dynamic user demand information from the massive decision information will affect the efficiency of the final purchase decision. Therefore, it is necessary to study the market transaction and purchase decision of cloud computing multi-instance. The main challenges of cloud computing multi-instance market transaction and purchase decision are: (1) cloud suppliers' research on how to price spot bidding cases and the trend of price change only consider qualitative influence trends, but can not be quantified; 2) in the process of purchasing decision, the single factor of quality of service is considered only, which makes the service present a single trend and can not meet the needs of users in many aspects. 2) the users can not handle and analyze the information of their own needs comprehensively. 3) in the process of purchasing decision, only the single factor of service quality is considered, which makes the service present a single trend. Based on the above problems, this paper proposes a cloud computing market transaction system based on cloud alliance, which provides the framework support for the transaction between supply and demand. Then the transaction forecasting model of cloud computing market is constructed, including spot bidding example price forecasting model and customer demand forecasting model. Finally, the multi-instance purchasing decision based on customers is studied to meet the needs of customers. The specific work includes: 1) putting forward the market transaction system based on cloud computing; (2) designing the price forecasting model and algorithm of cloud computing market spot case, fully mining the pricing law of cloud supplier spot case. This paper provides the basis for cloud users' bidding decision. (3) A cloud computing market customer demand prediction model is designed, and the grey BP neural network is used for simulation training, which improves the shortcomings of the traditional BP neural network. Make the forecast value more accurate, thus fully mining the needs of cloud users, make users better understand their own demand law, optimize the purchase decision; In this paper, we construct a multi-instance combined purchase decision model based on customers in cloud computing market. Firstly, we construct a single-objective constraint model based on cost optimization and minimum service time, combined with the characteristics of different case types. Then, considering the needs of all aspects of customers, a two-objective constraint model is constructed, and the goal of minimizing customer cost and minimum service time is achieved at the same time.
【学位授予单位】：武汉理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F49

【参考文献】