基于分布式包簇映射的云资源分配策略

Cloud Resource Allocation Based on Distribute Package Cluster Mapping

传统以虚拟机为中心的管理方式在可扩展性和灵活性方面存在严重问题,其资源分配往往被视为约束优化问题和简化成多维装箱问题,随着云计算市场规模越来越大,很难在规定的时间内求得最优解,针对该问题提出了一种基于分布式包簇映射的资源管理框架。本文以系统负载均衡为目标,将服务器划分为若干个服务器集群,减小问题规模,然后在每个服务器集群完成包簇映射进一步减小问题规模,并利用模拟退火算法实现全局最优解。将提出的基于分布式包簇映射框架与基于包簇框架的遗传算法进行比较,试验结果表明,随着簇个数的增加,提出的基于分布式包簇映射调度方法在大规模云资源分配负载均衡上有明显的性能优势,有效的减少了簇的个数。

The traditional virtual machine-centric management has serious problems in scalability and flexibility, resource allocation is often regarded as constraint optimization problem and reduced to multidimensional packing problem. As the cloud market becomes large and large, it’s difficult to find optimal solution within the specific time, a resource allocation management framework based on distribute a distributed package-cluster mapping was proposed.This paper takes load balancing as the target, divides the servers into several server clusters, reduces the problem scale, and then completes the packet cluster mapping in each cluster to further reduce the problem scale. Moreover, an Simulated Annealing Algorithm was used in this framework. Experimental results showed that, compared with the tradition virtual machine centered framework based genetic algorithm, The distributed packet cluster mapping scheduling method has obvious performance advantages on the load balancing of large-scale cloud resource allocation, effectively reducing the number of clusters.