Application-aware cross-layer virtual machine resource management.

ABSTRACTExisting resource management solutions in datacenters and cloud systems typically treat VMs as black boxes when making resource allocation decisions. This paper advocates the cooperation between VM host- and guest-layer schedulers for optimizing the resource management and application performance. It presents an approach to such cross-layer optimization upon fuzzy-modeling-based resource management. This approach exploits guest-layer application knowledge to capture workload characteristics and improve VM modeling, and enables the host-layer scheduler to feedback resource allocation decision and adapt guest-layer application configuration. As a case study, this approach is applied to virtualized databases which have challenging dynamic, complex resource usage behaviors. Specifically, it characterizes query workloads based on a database's internal cost estimation and adapts query executions by tuning the cost model parameters according to changing resource availability. A prototype of the proposed approach is implemented on Xen VMs and evaluated using workloads based on TPC-H and RUBiS. The results show that with guest-to-host workload characterization, resources can be efficiently allocated to database VMs serving workloads with changing intensity and composition while meeting Quality-of-Service (QoS) targets. For TPC-H, the prediction error for VM resource demand is less than 3.5%; for RUBiS, the response time target is met for 92% of the time. Both significantly outperform the resource allocation scheme without workload characterization. With host-to-guest database adaptation, the performance of TPC-H-based workloads is also improved by 17% when the VM's available I/O bandwidth is reduced due to contention.