Load balancing algorithm based on time series prediction of packet loss rate

Abstract

This paper proposes a new load balancing mechanism, called PPLR (prediction packet loss rate), which is based on time series prediction. Load balancing is needed in many fields; however, current approaches have problems such as lacking of preparing for sudden spikes. In this paper, we describe how PPLR dynamically predicts loss levels over all available network paths and adjust the allocation proportion of packets for next prediction period. We present our simulation work using NS-2 and discuss performance comparison between PPLR and ECMP mechanisms. The results show that PPLR not only significantly reduces the packet loss rate, but also has a good effect on the packet delay.Index Terms-load balancing, multipath routing, prediction, proactive probing I.