A Dictionary Architecture for Optimized Intra-Domain Routing

Abstract

The unpredictability of outages and traffic dynamics require network engineers to keep careful watch on their infrastructures.When significant changes are identified, engineers often tweak link weights (i.e., traffic engineering), which can have unwanted or ineffective outcomes. In this paper, we describe a new intra-domain routing architecture that seeks simultaneously to enable complex route optimization and eliminate route convergence delays to best meet operational objectives in dynamic network environments. Our approach uses centralized, multiobjective optimization to generate a set of forwarding tables that we refer to as a dictionary for each node in a network. When deployed, nodes select from among tables in their dictionary in an autonomous fashion based on the measured state of the network. The set of tables only needs to be recomputed when there are significant changes in network structure or operational objectives. We show that this approach can be implemented with relatively modest resources. We demonstrate the feasibility and capabilities of our method through a series of simulations run on synthetic and real-world network topologies. Our results show that dictionary-based routing is able to effectively maintain performance objectives by automatically adapting to a variety of structural and dynamic changes. We also show that in the common case, dictionary-based routing outperforms link-state algorithms in terms of speed of convergence after a failure and in terms of scalability of failure recovery.