In a world where peer-to-peer networks flourish, and where Internet path congestion and oscillations are daily events, the need for new efficient routing mechanisms is more and more pressing. Andersen, Balakrishnan, Kaashoek, and Morris present resilient overlay networks (RONs) as groups of nodes, distributed over large areas, whose users agree to engage in cooperative networking, and whose paths are formed over actual Internet routing paths.
The main characteristics of a RON are that the number of participating nodes is small (up to 50), and that communication between sites follows paths that circumvent temporary failures of the actual Internet paths. This is achieved by continuous probing of the direct links between sites, and by employing a new link-state routing protocol (different than open shortest path first (OSPF) and border gateway protocol (BGP)). Simply put, when the underlying segments of a direct path between two RON nodes fail (BGP failures are often cited), the overlay network redirects the entire path toward an intermediary RON node, apparently lengthening the entire path, but still offering connectivity. RON nodes have addresses different than Internet protocol version 4 (IPv4) or Internet protocol version 6 (IPv6). Actual experiments, performed by the authors, included a 16-node deployment in the USA and Europe.
As expected, another distinguishing trait of RON networking is the ability of applications at the uppermost layer to make routing decisions (traditionally, routing and application layers are separated, with the inconvenience of application interruption when routing fails).
The authors pay detailed attention to motivating the overlaying routing approach. Not only do they describe an actual implementation, including simulation, test deployment, and performance measurements, but they also address, in a separate discussion section, tough questions on potential violation of the deployed Internet policy routing (presumably due to tunneling), limited RON size and scalability, and network address translation (NAT) traversal.
Finally, one aspect whose treatment seems to be overlooked is one that lies at the very heart of a routing protocol: loop avoidance. While a description of path lookup and building by using link-state exchanges is given, proofs (at least conceptual) of loop avoidance are not mentioned at all.
The paper provides a comprehensive bibliographical list. Many of the references are correctly used as explanations of the current BGP routing instabilities, as well as of how these influence transmission control protocol (TCP) applications; thus, they offer perfect motivations for the need to overlay network routing.
