Towards a Deployable IP Anycast Service

Since it was first described in 1993, IP anycast has been a promising technology for simple, efficient, and robust service discovery, and for connectionless services. Due to scaling issues, the difficulty of deployment, and lack of application-specific features such as load balancing and connection affinity, the use of IP anycast is limited to a small number of critical low-level services such as DNS root server replication. More commonly, application-layer anycast, such as DNS-based redirection, is used. As the number of P2P and overlay services grows, however, the advantages of IP anycast become more appealing. This paper proposes a new proxy overlay deployment model for IP anycast that overcomes most of the limitations of native IP anycast. We believe that this makes IP anycast a viable option for easing deployment and improving the robustness and efficiency of many P2P and overlay technologies. We describe the new deployment model, some of its uses for P2P and overlay networks, its pros and cons relative to application-layer anycast, and discuss research issues.