Dynamic control of data center network and computation resources

In this paper, we present a policy for optimally balancing computing resources in a networked computing environment. In particular, we examine how to balance the computation time between tasks in a virtualized pipeline of distinct resources such as network control elements and servers. We consider a paradigm with limited resources to be divided among pipeline stages for maximum efficiency. This strategy has general applications across a range of hierarchical levels within a data center, from the classical processor pipeline to cloud-level computation tasks. Furthermore, the final control strategy can be simply implemented using a simple switching policy, establishing its suitability for high-speed applications. We show a policy iteration technique to determine and implement the this switching policy. Numerical results are presented as a demonstration of the policy iteration convergence, resulting in significant improvements from more basic control methods. Finally, we discuss potential directions for future work in efficient testing, implementation, and near-optimal approximations to our optimal control policy.