Improving Load Balancing with Multipath Routing

Internet service providers have to provision network resources to optimize bandwidth utilization. Dynamic routing protocols take traffic variations into account to control the load distribution. Multipath routing protocols attempt to take advantage of the path diversity to bring network robustness and reliability. Indeed, with a specific traffic engineering policy, they enable load balancing across several paths. Our aim is to compute a set of loopfree paths in order to allow routers to share the load on several next hops depending on current load measurement. In this paper, we first describe our original Incoming Interface Multipath Routing technique, DT(p), then we present a scheme for load balancing, DT(p)-TE, based on link monitoring. We evaluate and compare our technique with several existing approaches by a set of simulations, using different scenarios and topologies. The simulations results suggest that the path diversity achieved with our proposition can significantly improve the network response time.

I. Introduction

Traffic engineering (TE) is a critical issue in large IP backbones. Dynamic routing is able to circumvent congested links in order to improve the quality of applications such as HD video and VoIP services. However, overhead imposed by the frequency of link-state updates, new path activation, and signaling hampers its deployment. Multipath routing can balance network load to improve the streaming quality of long lived flows. Short flows are not specifically concerned with dynamic routing because their duration can be smaller than the link state updating period (the duration between two consecutive messages of traffic measurement). Despite the potential benefits in terms of resource control, most backbone networks still use static routing (OSPF [13] or IS-IS [15] consider the topology changes but not residual bandwidth fluctuations) because dynamic routing can lead to route flapping, strong traffic oscillations and excessive signaling messages overhead. Packets routed on outdated information can lead to serious load oscillations if the system does not react quickly enough. TE objectives can be achieved by routing traffic demands on multiple paths. Our approach to reach an efficient multipath routing is divided into four tasks: 1-

Compute and position loop free paths.

2-

Analyze local traffic activities (and advertise the availability of local resources in the network).

3-

Define load balancing policy depending on the computed (and received) information.

4-

Split the traffic among routes.