Constraint-aware VM placement in heterogeneous computing clusters

Virtualized systems consist of a large number of machines that are configured with different hardware and software, and execute a large number of virtual machines (VMs) for diverse applications. There can be various constraint conditions of placing VMs in such systems due to the concerns on security, availability, performance, etc. However, VM placement constraints can limit the choice of hosts for VMs, affecting the performance of the systems negatively. In this paper, we study constraint-aware VM placement in heterogeneous computing clusters. We first present a model of VM placement constraints that supports all types of constraints between VMs, and between VMs and hosts. Second, we discuss six constraint-aware VM placement algorithms which optimize the performance for either energy saving or load balancing. Third, using simulations, we analyze the effects of different types of VM placement constraints on VM placement, and evaluate the performance of the algorithms over various settings. We also run experiments of the algorithms in a small cluster. Our extensive simulation results demonstrate that the effects of VM placement constraints vary, depending on the optimization goal, the types of the constraints, and the system configurations. For the constraint-aware algorithms, we show that an energy saving algorithm which attempts to place a new VM on one of active hosts by utilizing VM migrations, and a load balancing algorithm which attempts to migrate some VMs from a selected host for a new VM, i.e. a potential hotspot, to another host provide good performance