Virtual Machine Dynamic Redeployment Based On Multi-objective Optimization

With the emergence of cloud computing,server providers can implement infras-tructure as a service(IaaS)more easily.Customers use resources on demand and there is no need for them to buy hardware facilities,which greatly reduces cost and the time of application deployment.In cloud data center,resource has been provided in the form of virtual machine(VM),customers use VMs like real machines.The usage of VMs improves resource utilization of servers and makes data center maintenance more easy.With the increase of user services,data center continues to scale up,consequently current data centers face many challenges,like imbalanced server loads and bottleneck switch links caused by large internal data center communication traffic.These factors lead to increase of user request delay and violation of service level agreement(SLA).VM live migration provides an effective way of adjusting resource usage pattern in cloud data centers,like migrating VMs from high load server to low load server to realize load balance.The problem of optimizing resource usage by VM live migra-tion can be called VM redeployment or VM consolidation,general VM redeployment works usually focus on one challenge in data center and pay little attention to others or just ignore them,while effective VM redeployment should make tradeoffs between these challenges,and more importantly,should not let other challenges become worse.On the other hand,since VM live migration leads to performance degradation of appli-cations,consolidation work should control migration cost.In this paper,we proposed a VM dynamic redeployment algorithm based on multi-objective optimization.Firstly,we define five optimization objectives during VM rede-ployment according to the status of data center,which include the number of used PMs,load balance degree,internal communication traffic,number of VM migration and ac-cumulative VM migration time.Then we formalize these optimization objectives into multi-objective optimization problem,rather than optimizing all these five objectives,we allow data center administrators to optimize part of these objectives and let others become the constraints of multi-objective optimization problem,by this way adminis-trators can focus attention on some objectives and restrict others from becoming worse.Simulation experiments based on real world workload trace and real world experiment show that compared with single objective optimization approaches our method effec-tively make tradeoffs between optimized objectives and has better overall performance,which is more practical in real data centers.On the other hand,a migration plan is needed to execute VM migration for con-verting data center from initial mapping to target mapping.Since VM migrations have been executed in the internal of data center,there may be dependency relationship be-tween VM migrations,like the destination PM may have no spare resource to receive new VMs until it migrates some VMs out.Current works focusing on VM migra-tion plan usually consider how to allocated bandwidth for migrated VMs to implement quickly VM migration,they cannot resolve the dependency relationship between VM migrations,and even migration dead lock caused by PM resource constraint.In this paper,we also proposed a VM parallel migration algorithm for implementing mapping conversion,our algorithm not only resolve migration dependency,but also implement parallel VM migration,reduce total migration time.