综合负载均衡度最小优先：一种实现云数据中心负载均衡的新方法

发布时间：2018-06-27 01:09

本文选题：云计算 + 数据中心　；参考：《电子科技大学》2012年硕士论文

【摘要】：随着数据中心规模的不断扩大，数据中心服务器的性能越来越受人们的关注，性能低在很大程度上是由于服务器负载过高而效率低下。同时，能源消耗成为日益严重和备受关注的问题，负载均衡使得数据中心不会因为某些机器负载太低而浪费不必要的资源。本文首先介绍了云计算的背景和负载均衡的研究意义。然后重点设计了最小综合负载优先：一种高可变环境下的云计算数据中心动态调度算法。云计算数据中心的一个挑战性的调度问题是分配和迁移虚拟机，以及物理主机的集成功能。动态负载均衡调度算法是一个NP-hard问题。传统的负载均衡调度算法普遍只考虑一个因素，比如物理服务器的CPU。综合负载均衡度最小优先算法(lowest integrated load first，以下简称LILF)同时考虑了多个维度的资源，，包括：CPU、内存和网络带宽，将他们整合进了物理服务器和虚拟服务器进行调度，用综合负载值用来作为分配虚拟机的依据。本文还研究了云计算数据中心整体的多维度不均衡度的衡量、物理服务器的平均不均衡度的衡量，以及CPU、内存和网络带宽的平均利用率。这些指标用来比较算法之间的优劣。最后的模拟结果显示了LILF在数据中心总体不均衡度、物理服务器不均衡度，以及平均利用率三个指标上均有非常好的性能。作为应对预订任务的业务需求的扩展，本论文还详细的设计和实现了静态离线负载均衡调度算法。静态离线负载均衡调度算法的目标是使得未来一段时间内的物理服务器平均负载均衡。调度系统知晓一段时间内前来的所有任务以及其生命周期，从而可以计算出一段时间内平均每一台物理机应该承担的负载，然后以平均负载作为标准分配，使得所有物理机的负载在一段时间内接近该平均负载。
[Abstract]:With the expansion of data center scale, people pay more and more attention to the performance of data center server. At the same time, energy consumption is becoming more and more serious and concerned. Load balancing makes data center not waste unnecessary resources because of low load of some machines. This paper first introduces the background of cloud computing and the significance of load balancing. Then we design a dynamic scheduling algorithm for cloud computing data center in high variable environment. A challenging scheduling problem for cloud computing data centers is the allocation and migration of virtual machines and the integration of physical hosts. Dynamic load balancing scheduling algorithm is a NP-hard problem. Traditional load balancing scheduling algorithms generally only consider one factor, such as physical server CPU. Integrated load balancing minimum priority algorithm (lowest integrated load first,) takes into account multiple dimensions of resources, including: CPU, memory and network bandwidth, and integrates them into physical servers and virtual servers for scheduling. The synthetic load value is used as the basis for the allocation of virtual machines. This paper also studies the measurement of multi-dimensional imbalance of cloud computing data center, the measurement of average imbalance of physical server, and the average utilization of CPU, memory and network bandwidth. These indexes are used to compare the advantages and disadvantages of the algorithms. Finally, the simulation results show that LILF has very good performance in the data center overall imbalance, physical server imbalance, and the average utilization rate. As an extension to meet the business requirements of reservation tasks, this paper also designs and implements a static off-line load balancing scheduling algorithm in detail. The aim of static off-line load balancing scheduling algorithm is to balance the load of physical servers in the future. The scheduling system knows all the tasks that come over a period of time and its life cycle, so that it can calculate the average load per physical machine over a period of time, and then distribute the average load as a standard. The load of all physical machines is close to the average load for a period of time.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP308

【参考文献】