78. FIBBING TO ALLEVIATE CONGESTION IN WAN AND DATA CENTER NETWORKS
Name: Ashish Kashinath
Grad Year: 2017
Congestion is an all too common problem in networking. With the recent developments of Software Defined Networking (SDN), it is possible to centralize control over the control and forwarding planes in datacenters and the private Wide Area Network (WAN) setting. However, purely SDN based approaches to congestion relief require network devices to support new generation protocol standards, like OpenFlow, making deployment of purely SDN based congestion relief techniques difficult with networks still running with legacy networking equipment. Our solution to congestion relief in data-centers and the WAN is a hybrid centralized SDN/distributed approach. We leverage a technique, called Fibbing, which injects fake nodes and links into hybrid networks so that both OpenFlow and legacy routers compute forwarding tables based on augmented topologies. ECN and congestion monitored based Fibbing simulations are run on common datacenter and WAN topologies showing the benefits of how Fibbing can be a readily deploy-able and effective solution to congestion relief. Throughput graphs, during times of simulated congestion, show that congestion triggered Fibbing increases throughput of flows when compared to ordinary TCP and ECN enabled TCP. We also provide statistical Fibbed routing table update time metrics for advising when congestion triggered Fibbing is most effective. Our evaluation and results show that the Fibbing technique is a viable and effective solution to congestion relief in datacenters and the WAN that is readily deploy-able in hybrid networks.
Industry Application Area(s)
Internet, Networking, Systems | Software, Analytics