Placing Virtual Clusters on Cloud Platform Based on Inter-VMs Network Traffic

• Main Authors: Yeh-Lun Du, 杜葉倫
• Other Authors: Wei-Jen Wang
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/6rx6a4

碩士 === 國立中央大學 === 資訊工程學系在職專班 === 106 === The use of virtualization technology allows software-based virtual machines to be able to run on a physical machine, which makes resource allocation on a cloud environment more flexible. With the fast development of the cloud computing technology, clusters of virtual machines become more popular in many applications, such as Hadoop cluster. A virtual cluster is composed by multiple virtual machines, which perform collaborative computing tasks through virtual networks on top of physical networks. Resource allocation for virtual clusters is a key issue to be solved on cloud. It is different from the single VM placement problem. First, all VMs of a cluster share the physical network bandwidth on cloud, and some of them may talk to each other very often. Second, physical network bandwidth consumption among two VMs can be turned into internal memory read/write operations if two VMs are placed at the same physical machine. Most existing scheduling mechanisms only consider/use the single virtual machine placement strategy, instead of the virtual cluster placement strategy. Therefore those VM placement algorithms/mechanisms may lead to poor performance since they do not consider the communication relationship among the VMs. In this paper, we propose a novel scheduling strategy for virtual clusters based on Inter-VMs network traffic. We have compared the simulation performance in terms of physical network usage of the proposed algorithm with existing algorithms such as FCFS, Round-Robin, and the Divide-Aware algorithm. The simulation results indicate that the proposed scheduling algorithm consumes the smallest physical network bandwidth. As a result, the proposed algorithm can reduce the possibility of performance degradation due to physical network bandwidth contention.