A new flow-level load balance in MPLS network based on distributable traffic

Abstract

Load distribution across multiple parallel paths is an important consideration. In many practical contexts, the aggregate traffic from source to sink may be such that no single link can carry the load. In an MPLS domain, this problem can be addressed by instantiating multiple paths. The main objective of this paper balances traffic at the flow level among the parallel Label Switched Paths (LSPs) in MPLS networks. Different from other proposals, our new framework is based on the distributable traffic (DT), where cross-traffic in real networks is considered, and each LSP is modeled as an M/G/1 processor-sharing queue. We define a flow to be a sequence of packet having the same identifier, and dispatch packet belonging to one flow to the same path, so the packet disorder problem is avoided effectively. This mechanism only needs to be implemented in the ingress LSRs and the egress LSRs. A new defined cost function is being used to distribute traffic to path. We computer the cost function based on the delay and packet loss of each LSPs, and minimize the cost function. The minimized cost function is inverse ratio to DT. If the cost function of a certain LSP is smaller, it means that more traffic can be distributed on this LSP. Extensive simulations using NS2 are performed with MPLS modules. Simulation results show that our approach so effective that the throughput is increased significantly and reduces the end-to-end delay and the packet drop rate, and it can distribute the traffic onto parallel LSPs more evenly and fairly.