MULTI-CLASS RESOURCE SHARING WITH BATCH ARRIVALS

Abstract

A cloud provider hosts virtual machines (VMs) of different types, with different resource requirements. There are bounds on the total amounts of each kind of resource that are available. Requests arrive in batches of different sizes. Under the ‘complete blocking’ policy, a request is accepted only if all the VMs in its batch can be accommodated. The ‘partial blocking’ policy would accept a request if there is room for at least one of the VMs in the batch. Blocked requests are lost, with an associated loss of revenue. The trade-offs between costs and benefits are evaluated by means of appropriate models, for which novel solutions based on fixed-point iterations are proposed. The applicability of those solutions is extended, by means of simplifications, to very large-scale systems. Numerical examples and comparisons with simulations are presented.