Dyncast: Use Dynamic Anycast to Facilitate Service Semantics Embedded in IP address

Abstract

Edge computing services are deployed at edge sites that are closer to users compare to the central cloud. There are quite a number of edge sites in a city hosting those service instances. The user’s request is normally served by the geographically closest one in order to get faster response. However the shortest distance does not necessarily mean the lowest latency from the user’s experience. With increasing number of edge sites, computing capacity and load, and network path status rather than geographical location, are playing key roles in determining the best edge site or service instance to handle the user’s request to achieve the optimal overall load balance and user experience. In this paper, we propose a dynamic anycast (Dyncast) networking architecture to optimally route the computing request to the most appropriate service instance by considering the real time computing loads and the network status simultaneously. The field testbed experiments demonstrate the effectiveness of the proposed dyncast architecture. Dyncast shows 9.5% to 159.9% improvement in terms of job completion time (JCT) over traditional scheduling strategy under different computing load and network status scenarios.