With the rapid growth of computational power on compute servers, and the fast maturing of x86 virtualization technologies, Virtual Machines (VM) have become increasingly important to supporting efficient and flexible resource provisioning. Modern virtual machine technologies (e.g. Xen, VMware) allow a single physical server to be carved into multiple virtual resource containers, each delivering a powerful, secure, and isolated execution environment for applications. In addition to providing access to resources, such environments can be customized to encapsulate the entire software and hardware platform needed by the applications and support their seamless deployments.

The management of these VM-based resource containers, e.g. lifecycle management and resource allocation, can be conducted through the interfaces provided by the virtualization platforms. This allows the VMs to be scheduled as processes in typical operating systems, and QoS-aware schedulers, similar to those available in operating systems, can be employed to allow the VMs to time- and space-share resources, and in the meantime provide QoS guarantees for the applications running inside of the VMs.

A hierachical architecture is designed for VM-based resource management. A virtual resource manager provides centralized allocation of distributed virtualized resources, and a per-host VM scheduler manages its local VMs. Intelligent controllers are integrated into this architecture at different levels to provide optimized resource management. Fuzzy-logic based machine learning methods are used to model resource demands for VMs running changing workloads, and resources are allocated dynamically based on a profit-driven model. For applications requiring QoS guarantees, advance resource reservation are also provided for VMs. VM migration can be employed to vacate workloads and dedicate resources for VMs, and a model is developed to estimate the migration overhead for efficient scheduling of reservations.

Fuzzy-logic based Resource Control