Traffic-Aware Configuration of All-Optical Data Center Networks Based on Hyper-FleX-LION

Hao Yang,Zuqing Zhu
DOI: https://doi.org/10.1109/tnet.2024.3363841
2024-01-01
IEEE/ACM Transactions on Networking
Abstract:Due to the advantages of optical circuit switching (OCS), all-optical data center networks (DCNs) have attracted intensive research interests recently. Hyper-FleX-LION is a highly-flexible all-optical DCN architecture that operates with the OCS based on wavelength-division multiplexing (WDM). In this work, we study how to realize traffic-aware configuration of all-optical DCNs in Hyper-FleX-LION. We formulate an integer linear programming (ILP) model for the problem to jointly optimize the configuration of Hyper-FleX-LION and the provisioning schemes of demands in it for minimizing its port usage. To ensure the practicalness of the optimization, we assume that each top-of-rack (ToR) switch can not only receive the traffic targeting to its rack but also forward traffic to other racks as an intermediate node. We also classifier traffic demands as normal and latency-sensitive ones, and set the maximum hop-count for routing latency-sensitive demands. By analyzing the complexity of the problem theoretically, we prove its $\mathcal{APX}$ -hardness, i.e., there does not exist a polynomial-time approximation algorithm for it unless $\mathcal{P}=\mathcal{NP}$ . Then, we propose a polynomial-time heuristic JTRO based on iterative optimization to solve the problem effectively and time-efficiently. Extensive numerical simulations verify the effectiveness of our proposed algorithm. We also build a small-scale but real all-optical DCN testbed in Hyper-FleX-LION to interconnect four racks, and leverage distributed machine learning (DML) as the network services in it to demonstrate the performance of our proposal experimentally.
telecommunications,computer science, theory & methods,engineering, electrical & electronic, hardware & architecture
What problem does this paper attempt to address?