DEER: Deadline-driven and Contention-free Central Arbitration in AWGR-based Optical Datacenter Networks

All-optical switching has been a promising technology for intra-datacenter networks to increase network bandwidth and decrease transmission delay. The AWGR combined with fast tunable lasers is being envisioned as one of the mainstream choices for intra-datacenter networks. Congestion control with fulfilling the deadline requirements of flows within the datacenter network influences the network performance largely. In this paper, we propose DEER, a central arbitration algorithm that enables flexible optical timeslot assignment for deadline-driven bursting traffic in an AWGR-based optical datacenter network realizing contention-free and fulfilling the deadline requirements as much as possible. The proposed algorithm takes various traffic volumes and deadlines into account and includes a heuristic algorithm making global decisions for all ToR switches. Numerical results indicate that DEER realizes contention-free transmission and decreases the deadline missing rate by 41.48% versus RRB and 26.29% versus LQF.