Abstract

When optical switching is used in Data Center Networks (DCNs), either as a standalone switching system, or together with electronic packet switching in a hybrid way, demand matrix decomposition (DMD) based scheduling methods are frequently considered, as an alternative to flow based solutions, to allocate the resource of the optical switching plane. However, what differs the DMD based resource allocation in optical switching from that in electronic switching is that the reconfiguration of the optical switch matrix often incurs much longer overheads, and may have a significant impact on the system performance. In this paper, we study the DMD based resource allocation methods in the optical switch plane of DCNs, aiming to figure out how partial reconfiguration and interruption may contribute to an improved system throughput, in the presence of a non-negligible reconfiguration delay. We model the scheduling problem into a Mixed-Integer Nonlinear Programming problem. Also, an efficient heuristic algorithm is proposed to solve the problem. Results show that by allowing partial reconfiguration and interruption, the performance of the system may be improved by 33%. We show that the proposed algorithm achieves near optimal performance at much lower time complexity. We also show how the traffic characteristics such as network load, skewness of traffic, and reconfiguration delay, may affect the performance gain.