Research And Implementation On Network-Aware Virtual Ma- Chine Placement And Consolidation

The emerging and development of "Cloud Computing" is changing the way in which people use computing resources. With the help of "Cloud", people no longer need to buy and maintain their own physical devices but lease computing resources from Cloud providers to serve their needs. Infrastructure as a Service (IAAS) is one major form of "Cloud Computing" which offers virtual machines to serve the users. Virtual machine is independent in logic and people can install Operating System and any software on it as they did on their own computers. More and more people choose to put their job into "cloud" which leads to the expansion of datacenter’scale. Mean-while, the amount of virtual machines in datacenter rises quickly. The management of datacenter needs a good strategy which is called virtual machine placement to map these virtual machines to physical servers. Different virtual machine placement strate-gies have different goals such as to improve the utilization of physical servers or to lower the energy cost. Current research focus on the requirement of CPU and Memory, which try to lower the computing cost or storage cost by virtual machine placement. However, the prevalence of distributed computing architecture such as Map-Reduce has made the ratio of distributed tasks risen. Deploying a distributed task means the virtual machines which hold this task have large network traffic among themselves. Therefore, the communication ability between these virtual machines would affect the performance, which requires us to consider the network requirement while we doing virtual machine placement.This paper thoroughly studied on network-aware virtual machine placement prob-lem and proposed a traffic-aware min-height-tree virtual machine allocation algorithm. Our algorithm aims to improve the communication ability of virtual machines which have large traffic among themselves and keep the traffic of switches balanced. For those tasks which didn’t get best network status in previous algorithm and already know their traffic matrix, we proposed a traffic-aware virtual machine clustering algorithm which conducts more accurate virtual machine placement.Online virtual machine allocation algorithm suffers performance decline due to resource fragments. When a job was finished, related virtual machines would be shut down and leave the system, which leads to resource fragments. Resource fragments lead virtual machines been allocated dispersed which had poor network bandwidth. Therefore, This paper further proposed a a network-aware virtual machine consolida-tion algorithm to keep the whole system running in an optimal status. Our algorithm had two stages. In the first stage, we checked whether consolidation was necessary and in the second stage we performed a heuristic consolidation which focused on lowering communication cost among virtual machines through conditional and automatic virtual machine live migrations.We built a experimental cloud platform in a mock datacenter which is consist-ed of 20 servers and applied our network-aware virtual machine placement algorithm into the platform. We used two benchmarks in our experimental cloud platform to ex-amine network influence on different tasks. And by using real workload traces to do experiments both in our experimental cloud computing testbed and in simulations, we compared the performance of different virtual machine allocation algorithms with or without our consolidation algorithm. Results in our testbed showed that our algorithm greatly shortened the average finish time of map-reduce tasks and reduced time delay of web applications. Simulation results showed that our algorithm considerably reduced the amount of high-delay jobs, lowered the average traffic passed through high-level switches and improved the communication ability among virtual machines.