云环境下面向多租赁的虚拟资源分配关键技术研究

发布时间：2018-05-11 13:22

本文选题：网络虚拟化 + 虚拟网络　；参考：《东北大学》2015年博士论文

【摘要】：云计算最大的特点是使得IT资源可以像水、电、天然气一样按需租赁并计费。租赁的实现离不开虚拟化技术的支持,主要手段是通过虚拟化技术将物理资源集中在一起形成共享虚拟资源池,实现虚拟资源动态分配的多租赁特性。云环境下的市场被分为两部分：基础设施提供商和服务提供商,前者拥有并维护硬件资源,对其进行抽象形成虚拟资源；后者根据实际需求按需租赁前者的虚拟资源来构建定制的虚拟网络以便于向用户提供服务。云环境下,这种典型的云环境下多租赁市场运营机制对于企业节省成本及提高资源利用率具有重要意义。因此,设计合理的虚拟资源租赁机制,提高物理资源利用率,是云计算虚拟资源分配研究的关键问题。现有的虚拟资源分配研究内容主要集中在租赁交易方式和虚拟网络部署两个方面。前者更加宏观,研究云市场竞争的多租赁环境下基础设施提供商和服务提供商之间的资源供求关系,以最大化社会整体利益为目标,并保证公平、高效的竞争环境。后者以硬件资源利用率为目标,研究如何在有限的硬件资源上尽可能满足更多用户的需求,为用户直接分配资源并创建虚拟网络,属于虚拟网络映射问题(Virtual Network Embedding, VNE)。虽然目前己经有多种虚拟资源分配算法被提出,但现有研究成果仍然存在以下几方面的问题：(1)云环境下资源分配定价机制不够灵活；(2)底层网络的资源利用效率差；(3)虚拟网络映射成功率低；(4)物理网络易出现资源占用不平衡问题；(5)仅考虑虚拟网络请求固定不变的情况；(6)虚拟网络服务可靠性差。因此,针对上述问题,本文对虚拟资源分配问题进行了深入研究,并取得如下成果：(1)考虑到云市场多租赁环境下服务提供商之间的竞争关系,他们不可能完全共享信息。因此利用隐马尔可夫理论,根据服务提供商的历史资源需求情况预测其当前出价,进而构建动态博弈定价模型以激励服务提供商选择最优出价策略,从而实现利益最大化。在资源分配阶段,设计了以多种类资源单位价格为基准的资源分配模型,该模型支持多服务提供商、多种资源同时分配,增加了基础设施提供商的收益并能够提高竞争公平性。(2)针对多租赁模式下的虚拟网络映射问题,以降低底层链路负载,加快映射效率,提高物理资源利用率为目标,将离散粒子群算法与虚拟节点映射规则相结合,提出了物理节点可复用、负载可控制的虚拟网络映射算法。算法能够节约物理链路的带宽资源。在保证网络负载的前提下,获得了较好的物理节点利用率和收益成本比。为了进一步提高算法在大规模网络下的求解效率,引入了交叉算子,设计了混合智能群算法,解决了粒子群算法容易陷入局部最优解,无法达到全局最优而出现早熟收敛的问题,能够使得物理网络获得较高的收益成本比。(3)考虑到网络拓扑结构对虚拟网络映射成功率的影响,重新定义了节点的综合能力,提出一种基于拓扑感知的虚拟网络映射算法。在映射过程中加入了拓扑感知度量方法以辅助选择映射方案,同时引入了滑动窗口技术对虚拟网络请求进行预处理,使得算法能够获得较好的接受率和收益成本比。(4)以提高虚拟网络接受率和底层网络利用率,避免物理网络节点和链路出现瓶颈为目标,建立了虚拟网络重配置问题的多目标优化数学模型,并采用元启发算法提出了一种虚拟网络重配置算法。算法可以显著地降低物理节点和链路的最大负载,并能够保证虚拟网络请求获得较高的接受率。(5)针对虚拟网络请求资源动态变化的实际情况,提出了面向动态虚拟网络请求的虚拟网络映射算法。以混合线性规划理论为基础,建立了以最小映射和迁移代价为优化目标的映射模型,该算法采用多队列存储不同类型的虚拟网络请求,优先映射需要释放资源的请求以获得更多的资源,从而降低了链路映射成本和迁移成本。(6)针对物理网络节点和链路失效问题,从容错角度出发,为虚拟网络增加备份的冗余虚拟节点和虚拟链路,以最小化映射成本为目标建立整数线性规划模型,设计了面向物理网络节点与链路失效的可靠虚拟网络映射算法。算法通过评估虚拟网络节点的重要性来定位需要备份的节点和链路,然后建立附加备份资源的虚拟网络增广图并对其进行映射,使得用户的虚拟网络获得了更好的可靠性支持。总之,文本从云环境下虚拟资源分配和虚拟网络映射两个角度出发,着重研究了基于非完全信息博弈的虚拟资源分配算法和网络虚拟化环境中适用于不同场景的虚拟网络映射算法。理论分析和大量的实验结果证明了这些方法的有效性和高效性。我们希望基于这些方法和技术进一步开发云资源管理、调度系统。
[Abstract]:The biggest feature of cloud computing is that IT resources can be leased and charged on demand like water, electricity and natural gas. The implementation of the lease can not be separated from the support of virtualization technology. The main means is to concentrate the physical resources together by virtualization technology to form a shared virtual resource pool and realize the multi lease characteristics of the dynamic allocation of virtual resources. The market is divided into two parts: the infrastructure provider and the service provider, the former owns and maintains the hardware resources, and abstracts it into virtual resources; the latter leases the former virtual resource according to the actual demand to build a customized virtual network to facilitate the service to the user. In the cloud environment, this typical cloud environment The operation mechanism of the next multi lease market is of great significance for the enterprise to save cost and improve the utilization of resources. Therefore, it is the key problem to design a reasonable leasing mechanism of virtual resources and improve the utilization rate of physical resources. There are two aspects of virtual network deployment. The former is more macro. It studies the supply and demand relationship between infrastructure providers and service providers in the multi rental environment of the cloud market competition, aims at maximizing the overall social benefits, and ensures a fair and efficient competitive environment. The latter aims at the use of hardware resources and studies how it is limited. The hardware resources are as much as possible to meet the needs of more users, directly allocate resources for users and create virtual networks, which belong to the Virtual Network Embedding (VNE). Although many kinds of virtual resource allocation algorithms have been proposed at present, the existing research results still have the following problems: (1) cloud ring The resource allocation pricing mechanism under the environment is not flexible enough; (2) the low utilization efficiency of the underlying network is poor; (3) the success rate of the virtual network mapping is low; (4) the physical network is prone to the imbalance of resource occupancy; (5) only the virtual network request is fixed and fixed; (6) the virtual network service is poor in reliability. Therefore, this paper is aimed at the above problem. The problem of virtual resource allocation is studied in depth, and the following results are obtained: (1) considering the competitive relationship between service providers in the multi rental environment of the cloud market, they can not fully share information. Therefore, the hidden Markov theory is used to predict the current bids according to the resource requirements of the service providers and then build up. The dynamic game pricing model is used to motivate service providers to select the best bid strategy and maximize the benefit. In the resource allocation stage, a resource allocation model based on multi resource unit price is designed. The model supports multi service providers, multiple resources are allocated at the same time, and the revenue of infrastructure providers can be increased and can be increased. In order to improve the fairness of competition. (2) aiming at the problem of virtual network mapping under the multi lease mode, in order to reduce the load of the underlying link, accelerate the mapping efficiency and improve the utilization of physical resources, the discrete particle swarm optimization (PSO) and the virtual node mapping rule are combined to propose a virtual network mapping algorithm which can be reused and controlled by the physical node. The algorithm can save the bandwidth resources of physical links. Under the premise of guaranteeing the network load, the better utilization ratio of physical nodes and the ratio of profit and cost are obtained. In order to further improve the efficiency of the algorithm in the large-scale network, the cross operator is introduced, and the hybrid intelligent group algorithm is designed to solve the particle swarm algorithm easily falling into the local area. The optimal solution, which can not reach the global optimal, has the problem of premature convergence, which can make the physical network obtain higher profit and cost ratio. (3) considering the influence of the network topology on the success rate of the virtual network mapping, the comprehensive ability of the node is redefined, and a virtual network mapping algorithm based on topology perception is proposed. In the process, the topology perception measure is added to assist the selection mapping scheme, and the sliding window technology is introduced to preprocess the virtual network request, which makes the algorithm get better acceptance rate and benefit cost ratio. (4) to improve the acceptance rate of virtual network and the utilization rate of the underlying network, avoid the occurrence of physical network nodes and links. The multi-objective optimization mathematical model of virtual network reconfiguration is established, and a virtual network reconfiguration algorithm is proposed by using the meta heuristic algorithm. The algorithm can significantly reduce the maximum load of physical nodes and links, and can guarantee the higher acceptance rate of virtual network requests. (5) for virtual network request capital. A virtual network mapping algorithm for dynamic virtual network requests is proposed. Based on the theory of mixed linear programming, a mapping model with minimum mapping and migration costs is established. The algorithm uses multiple queues to store different types of virtual network requests, and priority mapping needs to release resources. The request is to obtain more resources, thus reducing the link mapping cost and transfer cost. (6) aiming at the physical network node and link failure problem, from the fault tolerance point of view, the redundant virtual nodes and virtual links are added to the virtual network, and the integer linear programming model is established to minimize the mapping cost. A reliable virtual network mapping algorithm for physical network nodes and link failures. By evaluating the importance of virtual network nodes to locate the nodes and links that need to be backed up, a virtual network augmented map with additional backup resources is established and mapped to make the user's virtual network better support for reliability. In a word, From the two angles of the virtual resource allocation and virtual network mapping in the cloud environment, the text focuses on the research of the virtual resource allocation algorithm based on the incomplete information game and the virtual network mapping algorithm for different scenarios in the network virtualization environment. The theoretical analysis and a large number of experimental results prove the effectiveness and high of these methods. Effectiveness. We hope to further develop cloud resource management and scheduling system based on these methods and technologies.

【学位授予单位】：东北大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TP393.07

【相似文献】