Considering the fact that overlay networks have experienced overwhelming growth over the past two decades, Kurian and Sarac’s survey of related technologies will be useful to newcomers. Since overlay networks make up a rather large field, this survey focuses on application-specific overlay networks. (It notably distances itself from peer-to-peer (P2P) networks, the largest overlay network application area in terms of research.)
Overlay networks are networks that operate on virtual connectivity. The virtualization of computer networks plays an important role in overcoming shortcomings at a logical level. Moreover, they can also be used as testbeds for new services prior to full-blown installations.
The authors begin with a definition for an overlay network: “a virtual network that is built on top of another.” So, the definition remains open for multi-layered overlays, though the composition of such networks limits them to native and overlay layers. A multi-layered or virtual network on top of another virtual network might be useful for merging different overlay services. Thus, the broader definition is more appropriate.
Kurian and Sarac classify the numerous overlay network applications into five categories: content delivery networks (CDNs), overlay multicast, quality-of-service (QoS) guarantees, resilient routing, and security and denial-of-service (DoS) defense. The most successful application to date is the Akamai CDN, which they discuss first, followed by other commercially less-successful applications.
Kurian and Sarac also survey overlay enhancements. They propose three generalized overlay models: a routing overlay model, a service overlay model, and a security overlay model. This modeling seems a bit inappropriate since they refer to an application-specific solution instead of architectural separation. Hence, rather than introducing models, they could have used service names in sections 3 and 4. They also discuss topology, routing performance, and overhead/scalability issues, and the crosscutting approach for overlay and underlay traffic.
Finally, the authors discuss the pluralist versus purist argument. The pluralist view sees a future in having a heterogeneous network with many overlay services, whereas the purist view reserves the overlay network as a testbed for future networks.
In summary, this is a quite extensive survey, as evidenced by its five pages of references.