Assessment of Adaptive Polling Contention Resolution for Optical Switching in Edge Data Center Networking

Abstract

Edge Data Center (EDC) provides delay-sensitive services for end-users in edge computing. In Edge Data Center Network (EDCN), fast optical switching is the most promising technology due to low latency, high bandwidth, and transparency of rate/modulation format. However, packets contention of optical switching causes optical packets loss leading to low throughput and high latency. For buffer-less optical switches, it is challenging to solve packets contention because of lack of backup resources for contended packets. In this article, we propose a contention resolution, Adaptive Polling Contention Resolution (APCR), and assess the throughput, latency, and packet loss performance in DCN. An experimental demonstration based on flow control is implemented to test different contention resolution for the admissible load. The APCR algorithm can be adaptive to different load by buffer occupation status. APCR employs hybrid polling scheme to improve throughput and decrease average latency by global desynchronizing in high load. We assess the proposed algorithm performance in intra-cluster and inter-cluster EDCN. The results show that the proposed APCR can increase throughput by more than 15% and reduce the contention count by 43.72% at 0.4 loads.