Reseaerch On Traffic-aware Virtual Machine Migration In Virtual Computing Environment

Virtualization, which acts as the underlining technology support for cloudcomputing, enables huge amounts of third-party application being packed intovirtual machines. Virtual machine (VM) migration enables the servers in datacenters to be re-consolidated or reshuffled to reduce the operational cost of datacenters. However, few researches have been conducted on network cost-awarevirtual machine migration. It has been shown in recent research that traffic andbandwidth demand between VMs in a data center account for a lot of the totaltraffic. Besides, virtual machine migration brings additional data transferringoverhead, which would also enhance the data center network cost. This paperconsiders the Overloaded VM migration problem from the perspective ofminimizing the network cost, in which we considers the inherent dependenciesbetween VMs comprising a multi-tier application, and the underlining topology ofphysical machines. We divide our VM migration problem into static consolidationproblem and dynamic consolidation, analyze the difference between the two andprovide solutions to make a trade-off between network traffic cost and virtualmachine migration cost.The VM migration in data center consists of static consolidation and dynamicconsolidation. As for the former, the placement of VM lasts for a long time inservers, maybe several months or years. In fact, webmasters do not prefer migratingvirtual machines automatically. They prefer the static way: evaluate the bestplacement scheme manually before the consolidation takes place. Therefore, themechanism that the swarm intelligence algorithm tries to find an approximateoptimal solution through repeated iterations makes it a good solution for staticconsolidation. In this paper, genetic algorithm (GA) and artificial bee colony (ABC)is adapted and changed to suit the VM migration problem so that the network costis minimized. Simulation experiments show that when the number of VMs is small,GA has better network cost. But when problem size increases, it is evident ABC isadvantageous to GA. Besides ABC’s running time is about only a half of GA, whichis another advantage. To the best of our knowledge, we are the first to use ABC tosolve virtual machine migration problem. Dynamic consolidation of virtual machines lasts for a much shorter timeinstead. Webmasters can migrate one or several virtual machines based on thevariation of workloads in data centers. As for dynamic consolidation, we proposedthree heuristic algorithms step by step: LM, nCaM and nCaM2. LM relies on theheuristic information to minimize the network cost brought by each VM migration.In mCaM, each VM is allocated to the destination server to achieve the optimaltrade-off between the profit of network communication cost and migration cost.mCaM2is a two-step optimization algorithm. For each VM, it chooses thedestination host based on the minimal profit of network communication cost; as forselecting which VM to migrate in a server, it considers the impact of VM migrationcost. Simulation experiments show that our proposed algorithm is muchadvantageous to these old VM consolidation algorithms, and decrease a lot inrunning time compared with AppAware.