• hyperv
• clips
• cluster
• virtual machine
• hyper-v
• virtualization

Virtual Machines have become a standard unit of resource allocation for cloud environment, and virtualization has been used for implementing cloud infrastructures because of its natural ability to decouple the physical hardware from logical servers. The increasing density of computational power allows packing a significant number of virtual machines on a single node, leading to an exponential growth of the number of logical servers to be manages by the cloud infrastructure. To investigate smart policies to administer such a large number of virtual machines we have developed Octopus, a lightweight system for scheduling virtual machines on a cluster of hypervisors.

Octopus has been implemented using Microsoft Hyper-V in order to exploit the WMI interface to control the hypervisor programmatically with the F# programming language. The original goal was to design a system capable of moving virtual machines across different computing nodes in order to optimize the workload and pack computations to save energy by turning off nodes. Moreover, the VM creation is under control of the final user through a web page where it is possible to ask for specific requirements about computing cores, memory and the OS image to be provisioned.

More recently, our investigation has focused on the possibility of using expert systems to express complex policies and to govern this ever-increasing set of virtual machines. For this reason, we have embedded the CLIPS expert system inside Octopus in order to rely on a full rule-based expert system engine to define the resource management policies: the Octopus code asserts facts about VMs in the CLIPS systems and rules access system primitives exposed as functions invokes by triggered rules. The well-known RETE algorithm ensures a fast execution of policy while ensuring the ability to define policies that may even contain conflicting rules.