Analysis of the Two-Step Scheduling Framework with Transfer Speedup under Non-Uniform Traffic

We recently proposed the two-step scheduling framework (TSS) for multi-card optical switches: first, input ports on each card are assigned a different wavelength, then for every wavelength a matching is found among all cards. The proposed framework significantly improves the scheduler scalability and seamlessly adheres to the modular switch architectural constraints. In this paper TSS is evaluated from both the complexity and performance standpoints. The complexity analysis demonstrates the advantages of TSS in terms of required signaling and performed operations compared to a classical single-step scheduler. The performance analysis is conducted under non-uniform traffic, i.e., with a packet distribution highly skewed toward specific source-destination pairs. TSS latency using heuristic scheduling algorithms is shown to increase, compared to the one using computationally intensive optimal algorithms, especially at high loads. To counteract this issue, the transfer speedup technique is exploited, which allows to send more than one packet within a switching interval, while issuing a single scheduling decision. Results on different switch configurations demonstrate that the smallest transfer speedup allows TSS using suboptimal algorithms to achieve performance comparable to using optimal ones.