Collaborative informed gateway selection in large-scale and heterogeneous

Abstract

In wireless community access networks, clients tend to reach the Internet through multiple gateway nodes instead of a single default gateway. The mapping of gateways to clients should take into account the perception of network performance from each client node. Network conditions and traffic load can fluctuate and make repeated client-gateway measurements necessary. However, frequent measurements would result in a high communication overhead as well as high processing overhead in gateways and clients. We propose a lightweight client-side gateway selection algorithm by crowd-sourcing monitoring information from neighbor clients, without requiring explicit topology information or a detailed view of the network, while providing an accurate selection as compared to an ideal omniscient approach. Our collaborative gateway selection algorithm achieves good end-to-end performance, such as low latency perceived at client nodes, and fair distribution of the measurements over the gateway nodes. The number of performance measurements triggered by clients are reduced drastically, from n down to 2 measurements per node in each period. An experimental evaluation of our approach shows more than 80% similarity estimation of the gateway performance in the majority of the considered cases. We propose two variants of the gateway selection algorithm, collaborative-best and collaborative-fair, which yield near optimal gateway selection while utilizing partial information.