Um arcabouço para roteamento em redes MPLS

Abstract

This work addresses the problem of physical route selection for Label Switched Paths (LSPs) in Multi-Protocol Label Switching (MPLS) networks. It consists of defining routes for LSPs trying to minimize the total number of hops and the total number of rejections. The LSPs path selection problem is addressed through two different approaches: on-line and offline. We propose an ILP (Integer Linear Programming) model to solve the offline problem in an exact manner. This model tries to balance the network load while minimizing the network rejection rate. The goal is to optimize the overall network performance by routing requests through under-utilized links. Issues concerning the execution time necessary to solve the model are noticed and a genetic algorithm (GA) is developed to address theses issues. The GA is based on the combination of routing policies and adaptive route movements. In order to analyze the genetic algorithm's performance, various simulations are conducted. Results show that the GA executes much faster than the commercial tool used to solve the model in an exact manner. Results also show that the use of the GA considerably decreases the amount of rejected requests in the network. Experiments are conducted and satisfactory results are found showing that the GA is also suitable for on-line routing. This work also addresses the problem of adjusting the network routingin other cases besides the arrival of new requests. The process ofadjusting the network routing is called Grooming and it occurs when a new network element is configured, LSPs are torn down or a network element recovers from failure. It is necessary when routing decisions taken in the past are no longer efficient. A traffic/resource change detection mechanism is proposed to trigger the Grooming process. This mechanism is based on control charts and is used in the context of defining the moment the Grooming should take place. Simulation results indicate that better performance is achieved using the proposed mechanisms.