A Scalable Framework for Global Application Anycast

As a one-to-one-of-many communication method, anycast provides a mechanism for clients to select one of the ldquonearestrdquo servers in a group of anycast servers. Global anycast is a building block for many distributed applications. However, it faces several challenges. First, it should be scalable to world wide. Second, it must minimize the anycast query latency. And third, the anycast result must be the ldquonearestrdquo server to the client or at least near the ldquonearestrdquo one. To handle the first challenge, anycast member are not allowed to keep track of global information, but this will result in long query latency and make anycast result far away from the optimal server. To tradeoff scalability and query efficiency, a scalable framework for global application layer anycast (GAA) is proposed in this paper. In GAA, all anycast servers compose an overlay network in the application layer. To reduce the query latency, a client-centralized on-demanding probing anycast scheme is proposed. Each server only needs to keep track of O(logN) neighbors. Analytical and experimental results show that the query latency in GAA is cut down significantly than in Meridian and OASIS, and the related performance of GAA is also much better than Meridian and visual coordinates based framework.