The elastic scaling of services permits us (1) to meet service provisioning requirements (i.e., not to violate its SLAs), and (2) to efficiently use available resources (i.e., to minimize the cost of running the application). In this paper, we describe a novel online controller design built on top of the Xen hyper visor. The controller facilitates the elastic provisioning of applications using a combination of adapting the resource constraints of VMs and dynamically plugging in new virtual CPUs. We have implemented our design and evaluated our approach using real world workload traces from the RUBiS online auction benchmark. The results show that our controller allows to reduce total CPU time in comparison to statically allocated CPU with of minimum SLA violation rate and provides stable response time for high priority VM, when we apply it to provision prioritized VMs.