Towards a performance interference-aware virtual machine placement strategy for supporting soft real-time applications in the cloud

It is standard practice for cloud service providers (CSPs) to overbook physical system resources to maximize the resource utilization and make their business model more profitable. Resource overbooking can lead to performance interference, however, among the virtual machines (VMs) hosted on the physical resources causing performance unpredictability for soft real-time applications hosted in the VMs, which is unacceptable to these applications. Balancing these conflicting requirements needs a careful design of the placement strategies for hosting soft real-time applications such that the performance interference effects are minimized while still allowing resource overbooking. These placement decisions cannot be made offline because workloads change at run time. Moreover, satisfying the priorities of collocated VMs may require VM migrations, which require an online solution. This paper presents a machine learning-based, online placement solution to this problem where the system is trained using a publicly available trace of a large data center owned by Google. Our approach first classifies the VMs based on their historic mean CPU and memory usage, and performance features. Subsequently, it learns the best patterns of collocating the classified VMs by employing machine learning techniques. These extracted patterns are those that provide the lowest performance interference level on the specified host machines making them amenable to hosting soft real-time applications while still allowing resource overbooking. Keywords-virtual machine placement, cloud computing, performance interference, resource overbooking, application QoS.