Optimal placement of multi-wavelength integrated tunable transmitters in reconfigurable optical data center interconnection networks with multi-hop routing

Abstract

Circuit-switched technologies have created the enticing possibility of guaranteeing the performance of data center networks under bursty and skewed traffic. In our prior work [J. Opt. Commun. Netw.16, 990 (2024)], we proposed to use multi-wavelength integrated tunable transmitters (MWITTs) to provide dynamic capacity sharing in modular arrayed waveguide grating-based interconnection networks. However, with one-hop routing, the scope of wavelength sharing is limited, and the required number of MWITTs to maintain good performance will increase drastically when hotspot top-of-racks are distributed sparsely. In this paper, we propose expanding the scope of dynamic capacity sharing of MWITTs by incorporating a multi-hop routing mechanism, such that fewer MWITTs are needed to achieve comparable performance levels. We model the problem of the placement of MWITTs with multi-hop routing into a binary quadratic-constrained linear programming problem. Two feasible placements, namely, star-topology placement and interconnected-group placement, are proposed and proved optimal in specific scenarios. Furthermore, an efficient heuristic algorithm combined with the above two feasible placements, the simulated annealing-based MWITT placing algorithm (SAMPA), is proposed to solve this placement problem. Results show the high accuracy and low computational complexity of SAMPA. We compare the minimum number of required MWITTs and the blocking probability under one-hop and multi-hop routing through simulation to prove that multi-hop routing can achieve better performance with fewer MWITTs. The number of MWITTs is 128 and 30, to achieve the blocking probability of 0.0169 and $1.4 \times {10^{- 3}}$, respectively, for one-hop routing and multi-hop routing, when the traffic burstiness is 0.5, traffic skewness is 0.6, and number of hotspot racks is 16 in a 128-rack interconnection network.