云计算环境下基于时间的公平带宽动态分配算法

发布时间：2018-04-26 12:21

本文选题：云计算 + 数据中心网络　；参考：《浙江工商大学》2014年硕士论文

【摘要】：云计算技术在近几年越来越成熟,越来越多的数据密集型信息中心部署了自己的云。为了满足现在社会大众对于计算能力、存储空间以及相关计算服务的复杂多变的需求,云计算技术很好的提供了相应的解决方案,并从经济学角度上诠释了它固有的优势。利用云计算,企业、科研机构以及政府纷纷建立了自己的数据中心来满足自身日益增多的业务及计算需求。云计算使用虚拟化技术,将物理服务器中的CPU资源、内存资源以及存储资源虚拟化处理后,根据不同需求划分给不同租户。同时这些划分的计算资源的使用都是严格隔离的,例如,划分给每一个租户的资源的所有计算资源保证全部都归这个租户使用,我们称划分出去的计算资源为虚拟机(VM)。众所周知,云计算中的网络表现对于数据中心网络中的虚拟机而言,是非常重要的。数据中心网络中,提供数据密集型应用服务的虚拟机,就是非常占用带宽的典型代表。例如多媒体在线服务、MapReduce类应用服务,在服务的过程都会产生大量的数据通信,导致在整个数据中心网络中不管是核心层、汇聚层还是接入层,都需要大量的带宽需求。那么,对于云数据中心网络的网络管理就是非常中的工作,且毫无疑问的是,接入层物理服务器(或Hypervisor层)的带宽分配策略是整个数据中心网络资源分配的基础解决方案,因为物理服务器(或Hypervisor)能够直接的监控以及管理其内部虚拟机的带宽竞争。解决这个层面上的网络带宽分配问题,就能使得整个数据中心的整体网络性能提升到相当可观的地步。但令人遗憾的是,处于这个物理主机(或Hypervisor)层面上的数据中心网络资源分配,并没有引起人们的足够重视。大部分数据中心网络带宽的分配政策都是针对于全局层面,可是具体的策略实施,却都是需要在最终接入物理服务器的这个层面上开始执行,而并不是将大量资金花在交换机硬件设备升级。当前,还没有一个确实可行的数据中心网络带宽分配策略,能够同时保证虚拟机带宽动态分配、公平、延时最优化的分配策略。因此,物理服务器(或Hypervisor)层面上网络带宽分配的研究是非常有必要的。本文就针对于数据中心网络接入层的网络带宽进行研究,在云计算环境下基于时间的数据中心网络公平动态分配,利用租户应用历史流量吞吐率,公平调整数据中心网络带宽的分配政策,实现网络资源的充分利用。主要贡献如下： 1)我们通过具体实际问题阐述了云计算环境下数据中心网络内的虚拟机带宽冲突问题,引出对于带宽资源公平分配的研究。 2)我们提出了一个与租户花费相匹配,并且能够实现高网络利用率的公平分配的要求。并针对这个要求给出了一个基于时间的公平带宽动态分配策略。 3)构建云计算平台,针对本文公平分配带宽的需要,对于虚拟网络进行控制。通过与TCP/IP内部的拥塞控制策略以及静态公平带宽分配策略的对比实验,验证了,本文基于时间的动态带宽分配策略在保证对租户公平分享带宽的同时,实现最小化延时。
[Abstract]:Cloud computing technology has become more and more mature in recent years , and more and more data - intensive information centers have deployed its own cloud . In order to meet the needs of today ' s social public for computing power , storage space and related computing services , cloud computing technology provides the corresponding solutions . Cloud computing , enterprises , scientific research institutions and governments have set up their own data centers to meet their growing business and computing needs . The use of cloud computing , enterprises , scientific research institutions and government has set up their own data centers to meet their growing business and computing needs . The use of these divided computing resources is strictly isolated , for example , all computing resources allocated to each tenant are guaranteed to be used by this tenant , and we call the divided computing resources as virtual machines ( VMs ) .

As is well known , the network performance in cloud computing is very important for virtual machines in the data center network . Virtual machines providing data - intensive application services , such as multimedia online services , MapReduce class application services , generate large amounts of data communication throughout the data center network , resulting in large bandwidth requirements across the entire data center network , whether core , aggregation or access layers .

So , the network management of the cloud data center network is very important , and there is no doubt that the bandwidth allocation strategy of the access layer physical server ( or hypervisor layer ) is the foundation solution for the whole data center network resource allocation , because the physical server ( or hypervisor ) can directly monitor and manage the bandwidth contention of the internal virtual machine .

In this paper , the network bandwidth of data center network access layer is studied . In the cloud computing environment , based on the time - based data center network fair dynamic allocation , the distribution policy of data center network bandwidth is adjusted fairly by using the application history traffic throughput rate of the tenant , and the utilization of network resources is realized .

1 ) We expound the virtual machine bandwidth conflict problem in the data center network in the cloud computing environment through specific practical problems , and lead out the research on the fair distribution of bandwidth resources .

2 ) We put forward a requirement to match the cost of the tenant , and can realize the fair distribution of the high network utilization rate . And a dynamic allocation strategy based on time is given for this requirement .

3 ) Constructing the cloud computing platform , aiming at the need of the fair distribution bandwidth in this paper , and controlling the virtual network . Through the comparison experiment with the congestion control strategy and the static fair bandwidth allocation strategy in TCP / IP , this paper proves that the dynamic bandwidth allocation strategy based on time guarantees the minimum delay time while ensuring the fair share bandwidth of the tenants .

【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP308

【参考文献】