The virtualization spectrum from hyperthreads to GRIDs

Modern virtual machines (VMs) can be opaque to conventional performance management tools because VM technology has surpassed standard measurement paradigms. We attempt to ameliorate that problem by (1) observing that disparate types of VMs lie on a discrete spectrum bounded by hyperthreading at one extreme and GRID-like services at the other, and (2) recognizing that poll-based scheduling is the common architectural element in many VM implementations. The associated polling frequency (from GHz to μHz) positions each VM in its respective region of the VM-spectrum. Several case studies are analyzed to illustrate how this insight can make VMs more visible to performance management techniques.