Enabling Traffic-Differentiated Load Balancing for Datacenter Networks

In modern datacenter networks (DCNs), load balancing mechanisms are widely deployed to enhance link utilization and alleviate congestion. Recently, a large number of load balancing algorithms have been proposed to spread traffic among the multiple parallel paths. The existing solutions make rerouting decisions for all flows once they experience congestion on a path. They are unable to distinguish between the flows that really need to be rerouted and the flows that potentially have negative effects due to rerouting, resulting in frequently ineffective rerouting and performance degradation. To address the above issues, we present a traffic-differentiated load balancing (TDLB) mechanism, which focuses on distinguishing flows that necessarily to be rerouted and employing corresponding measures to make optimize routing decisions. Specifically, TDLB detects path congestion based on queue length at the switches, and distinguishes the traffic that must be rerouted through the host pair information in the packet header, and selects an optimal path for rerouting. The remaining traffic remains on the original path and relies on congestion control protocols to slow down to alleviate congestion. The NS-2 simulation results show that TDLB effectively reduces tailing latency and average flow completion time (FCT) for short flows by up to 45% and 46%, respectively, compared to the state-of-the-art load balancing schemes.