GLIB: A Global and Local Integrated Load Balancing Scheme for Datacenter Network

Abstract

Load balancing is an important issue for improving performance of datacenter networks with regard to the large traffic volume. The global load balancing schemes master a global view of the network, but it is costly to maintain accurate view of global network status and they are unable to deal with the transient traffic bursts. The local load balancing schemes respond quickly to transient traffic bursts, but they are unaware of the global status of traffic distribution. As it's challenging to design a complete solution for the problem, there are few efforts devoted to integrate them to take both advantages. In this paper, we propose GLIB as a first step trying to integrate global and local load balancing mechanisms into a single framework. It consists of a local load balancing algorithm using backup routes to cope with transient traffic bursts and a centralized heuristic optimization algorithm to balance network load from a global viewpoint. Simulations have been conducted to verify the proposed scheme and the results have shown that GLIB has good performance in in several realistic datacenter traffic traces. Besides, GLIB explore the effect of our local load balancing algorithm and global route optimization algorithm respectively, and we can conclude that GLIB is a complete load balancing solution in datacenter networks.