We introduce a mathematical model formulation for scheduling virtual machine (VM) migrations in data centers. The model is aimed at minimizing the number of active physical host servers over time. Our goal is not only to avoid overload situations resulting from aggressive consolidation mechanisms but also overload situations caused by overhead-intenisve VM migrations. Although various VM scheduling approaches have been proposed in the literature, so far predictable resource demands caused by VM migrations are not directly considered in mathematical scheduling models, which can easily entail unplanned overload situations and resulting performance degradation. We propose a new model formulation, which explicitly takes the migration overheads into account while recalculating and executing VM allocations over time. Experimental results based on VM resource demand time series from a data center show that the model allows for significant server savings compared to a static assignment of VMs to physical hosts, while avoiding overload situations.
展开▼
