CDN-based Multipath Overly Routing for Real-Time Traffic

This work presents a Content Distribution Network-based approach to uncorrelated multipath routing for real-time traffic in large overlay networks. We propose clustering and routing mechanisms that require fewer direct measurement probes and lower communication overhead by leveraging the pre-existing Content Distribution Networks (CDN's) such as Akamai. Specifically, we exploit the dynamic interrelation between nodes and CDN's servers to organize overlay nodes into a hierarchy of clusters of nodes. Each cluster has a representative node. The representative node exchanges its routing state only with a small subset of other nodes in order to determine uncorrelated multiple paths, thus incurring relatively small communication overhead.;Our main contributions are as follows: (1) We propose a new clustering technique that requires no additional infrastructure and uses fewer direct measurement probes. An overlay node simply contacts its local DNS server to obtain a DNS redirection. Overlay nodes, whose DNS redirections are similar, are then grouped into a cluster. (2) We propose a verification mechanism for the clustering process that not only guarantees the network latency among nodes within a cluster but also adapts to network conditions. (3) We propose a scalable distributed multipath source routing algorithm to determine uncorrelated paths between pairs of nodes that satisfy real-time traffic requirements. The representative nodes of all clusters, called supernodes, form a logical full-mesh network. Traditional multipath routing algorithms would be inefficient if applied to full-mesh networks, as they incur O(N2) communication overhead. In contrast, each supernode in our system exchanges routing information with only a small subset of supernodes called a supercluster. We design a self-organizing system to group supemodes into superclusters to obtain near-minimal O(N1.5) communication overhead. (4) We also develop a prototype for a VoIP application and evaluate its performance, under various network conditions, to observe to what extent the quality of voice and video could improve by sending and mixing multiple streams using multiple codecs, instead of sending only one stream with a single codec.