The efficient management of shared resources in virtualized environments has become an important issue with the advent of cloud computing. This is a challenging management task because the resources of a single physical server may have to be shared between multiple virtual machines (VMs) running applications with different performance objectives, under unpredictable and erratic workloads. A number of existing works have developed performance differentiation and resource management techniques for shared resource environments by using linear feedback control approaches. However, the dominant nonlinearities of performance differentiation schemes and virtualized environments mean that linear control techniques do not provide effective control under a wide range of operating conditions. Instead of using linear control techniques, this paper presents a new nonlinear control approach that enables achieving differentiated performance requirements effectively in virtualized environments through the automated provisioning of resources. By using a nonlinear block control structure called the Hammerstein and Wiener model, a nonlinear feedback control system is integrated to the physical server (hypervisor) to efficiently achieve the performance differentiation objectives. The novelty of this approach is the inclusion of a compensation framework, which reduces the impact of nonlinearities on the management system. The experiments conducted in a virtual machine environment have shown significant improvements in performance differentiation and system stability of the proposed nonlinear control approach compared to a linear control system. In addition, the simulation results demonstrate the scalability of this nonlinear approach, providing stable performance differentiation between 10 applications/VMs.
↧