Consumer oriented state aggregation using reinforcement learning approach

Abstract

Hierarchical routing strategy as a scalable solution to the QoS routing problem defines a particular kind of network state information; each domain advertises only its aggregated state to the outside. Unfortunately, topology aggregation (TA) which is the process of summarizing the topological information of domains introduces some inaccuracy due to the compaction of exact network state. To reduce inaccuracy, both the TA method and the routing algorithm must be carefully selected. In this paper, we propose a novel aggregation model which intelligently adapts the aggregated states of domains according to the needs and topological locations of consumer nodes. Based on this model, we modify one of the current TA methods to enhance its performance. Through extensive simulations, we show that appropriate adjustment of compact state based on its relevance to sources, results in a lower miss-admission ratio and longer update intervals for the hierarchical QoS routing protocols.