SIFS: A Scalable and ISP-Friendly Server Selection Scheme in Cloud Services

A large number of servers (e.g., CDN servers, datacenter servers) are placed in multiple geographical locations within different network domains. In order to minimize user perceived latency while ensuring high data availability, the server selection problem raises great challenges in modern Internet architecture. This paper presents SIFS, a novel server selection scheme for modern network paradigms exemplified by cloud computing. SIFS has good scalability and ISP-friendliness. SIFS proposes a configurable global performance function that allows ISPs and service providers to leverage the cost (i.e., inter-domain traffic) and the quality of service (i.e., network latency). Results from theoretical analysis and simulations show that, compared with existing server selection approaches, SIFS, with the overall burden increasing linearly with the number of clients, reduces each end user’s measurement cost from O(N) to O(1) for clouds containing N server clusters and decreases about 50% inter-domain traffic.