Optical load-balancing tradeoffs in wavelength-routing cloud data centers

Wavelength routing based on arrayed waveguide grating (AWG) devices has been proposed as a disruptive technology for capacity, footprint, and flexibility requirements of cloud data centers. The limited buffering capacity of AWG-based routers, however, affects network throughput and makes them highly susceptible to data center traffic imbalance. A load-balancing stage, capable of evenly distributing the incoming traffic in both space and time, can improve network robustness and remedy congestion penalties. Robustness and scalability due to load balancing come at the cost of physical-layer complexities in terms of added cost, power consumption, and impairments. In this paper, we conduct a novel scalable analysis to study load-balancing tradeoffs between network-layer gains and physical-layer penalties in a multistage routing scenario. Our cross-layer simulations examine the requirements of a wavelength-routing load balancer based on AWG and tunable wavelength converters (TWCs) in order to achieve performance interesting for a data center. Our results point to the critical contribution of Q-factor degradation to load-balancer performance and encourage efforts in the design of novel router architectures, aiming to consolidate routing with scheduling-based load balancing.