Dynamic load balancing for delay CDF α-percentile optimization with a global view

Abstract

Load balancing and delay optimizing play significant roles in improving network performance. However, in the current networks, only local load balance (LB) and mean delay optimization mechanisms are adopted, which ignore the jitter and the network's status. In this paper, a dynamic LB algorithm is proposed to optimize the network delay with a global view (LBGV). It makes a traffic distribution decision based on the load of different elements and links in the network, which can significantly reduce the CDF (Cumulative Distribution Function) α-percentile of the network delay, as well as improve the overall jitter and the queuing fairness among traffic from different-loaded areas. Then this algorithm is adapted to and testified in some specific scenarios where the delay variance may be considerable and can't be ignored. Simulation results show it outperforms other existing algorithms.