Server Consolidation Algorithms with Bounded Migration Cost and Performance Guarantees in Cloud Computing

Consolidation of virtual machines is essential to achieve energy optimization in cloud computing environments. As virtual machines dynamically enter and leave a cloud system, it becomes necessary to relocate virtual machines among servers. However, relocation of virtual machines introduces run-time overheads and consumes extra energy, thus an careful planning for relocation is necessary. We model the relocation problem as a modified bin packing problem and propose a new server consolidation algorithm that guarantees server consolidation with bounded relocation costs. We also conduct a detailed analysis on the complexity of the server consolidation problem, and give a upper bound on the cost of relocation. Finally, we conduct simulations and compare our server consolidation algorithm with other relocation methods, like First Fit and Best Fit method. The experiment results suggest an interesting trade-off between server consolidation quality and relocation cost. Our algorithm is able to trade about 1% in server consolidation quality for a reduction about 50% in relocation cost, when compared with other well known bin packing algorithms. We also note that the relocation cost incurred in our method is much less than the theoretical bound we provided. The reason is that we overestimate the amount of relocation from theoretical analysis, and the actual amount of relocation found from experiments is much less than the worst-case bound from theoretical analysis.