Adaptive Routing for Datacenter Networks Using Ant Colony Optimization

Modern datacenter networks (DCNs) employ Clos topologies that providing sufficient cross-sectional bandwidth, various load balancing mechanisms are proposed to make full use of multiple parallel paths between end-hosts. Faced with a large number of heterogeneous flows, existing load balancing schemes cannot work well and cause performance degradation, such as latency-sensitive short flows experiencing large tail delay and severe link bandwidth waste due to random rerouting. To solve these issues, we propose an adaptive routing mechanism based on ant colony optimization algorithm (RACO), which adopts different (re)routing strategies for heterogeneous flows. Specifically, RACO uses the improved ant colony optimization algorithm to make optimal rerouting decisions to obtain high throughput and low latency for both elephant flows and mice flows, respectively. The experimental results based on Mininet simulation show that RACO effectively increases the throughput of long flows and reduces the average flow completion time (FCT) of short flows by up to 42% and 61%, respectively, compared with the state-of-the-art load balancing mechanisms.