Configurable Cost-Quality Optimization of Cloud-based VoIP

In this paper, we formulate configurable cloud-based VoIP call allocation problem as a special case of
dynamic multi-objective bin-packing. We consider voice quality influenced by CPU stress, cost
contributed by the number of billing hours for Virtual Machines (VMs) provisioning, and calls placed
on hold due to under-provisioning resources. We distinguish call allocation strategies by the type and
amount of information used for allocation: knowledge-free, utilization-aware, rental-aware, and loadaware. We propose and study a variety of strategies with static and dynamic policies of VM provisioning.
To study realistic scenarios, we consider startup delays for VM provisioning, and three configurable
parameters: utilization threshold, rental threshold, and prediction interval. They can be configured and
dynamically adapted to cope with different objective preferences, workloads, and cloud properties. We
conduct comprehensive simulation on the real workload of the MIXvoip company and show that the
proposed strategies outperform ones currently in-use.