Delay-cost tradeoff for virtual machine migration in cloud data centers

Providing continuous services for applications such as web services, is one of the most critical issues in cloud data centers. However, physical machines (PMs) in data centers may have to periodically undergo maintenance processes, which might disable the services of virtual machines (VMs) running on them. To reduce waiting delay, one may migrate VMs to other active PMs. However, this will incur extra migration cost. By considering both delay and migration cost, we study two problems. Firstly, we formulate a joint optimization problem with both delay and migration cost considered, which is proved to be NP-hard. For a special case with the same amount of resources required, we show that it can be converted to a weighted bipartite matching problem. We also develop an efficient heuristic algorithm for the general problem. Secondly, considering the limited migration cost at each maintained PM, we aim to migrate the VMs such that the total waiting delay is minimized. It is shown that the problem with a single active PM can be transformed to a 0-1 knapsack problem. We then design a heuristic algorithm for multiple PMs, by solving a series of knapsack problems. Simulation results demonstrate the effectiveness of the schemes, and show that the proposed algorithm is able to leverage the spatial variation in the VMs migration for delay and cost optimization.