Client/server architectures over wide area networks

Designing cost-effective architectures over wide area networks is becoming increasingly important with the rapid growth of client/server applications (e.g., WWW) over private and public wide-area networks. This paper presents a heuristic algorithm that can be used to (1) design cost-effective client/server architectures in an wide-area network environment; (2) analyze the tradeoffs between network transport cost and information storage cost; and (3) dynamically determine which server should be the default for a user. The algorithm works by determining server locations and default server assignments so that total network cost can be minimized. The algorithm can handle many realistic network parameters that cannot be taken into account easily by other approaches in the literature. These parameters include arbitrary physical and logical network topologies; physical and logical user groups; arbitrary number, capacity, and locations of servers; and arbitrary geographical distributions of user demands. Furthermore, the algorithm has a low time complexity of O(N/sup 3/). Quantitative analysis using the algorithm provides new insight into the design of client/server systems over wide-area networks. For example, a small number of servers could provide cost savings close to that achievable with large number of servers. Furthermore, cost savings using multiple servers can be made virtually insensitive to the ratio of transport cost to storage cost.