Optimal Hiring of Cloud Servers

A host uses servers hired from a Cloud in order to offer certain services to paying customers. It must decide dynamically when and how many servers to hire, and when to release them, so as to minimize both the job holding costs and the server costs. Under certain assumptions, the problem can be formulated in terms of a semi-Markov decision process and the optimal hiring policy can be computed. Two situations are considered: (a) jobs are submitted in random batches and servers can be hired for arbitrary periods of time; (b) jobs arrive singly and servers must be hired for fixed periods of time. In both cases, the optimal policies are compared with some simple and easily implementable heuristics.