Client-side mechanism for improving busy Web server's performance

The HTTP/1.1 standard reduces latencies and overhead from closing and re-establishing connections by supporting persistent connections as a default, which encourages multiple transfers of objects over a single connection. HTTP/1.1, however, does not explicitly define the connection-closing time but specifies a certain fixed holding-time model. This model may induce wasting of a server's resources when the server maintains a connection with an idle-state client that requests no data for a certain time. This paper proposes a mechanism for heuristic connection management supported on the client-side under persistent HTTP, in addition to the fixed holding-time model on the server-side. The client exploits the tag information within transferred HTML documents so that it decides the connection-closing time. We compare this to the fixed holding-time model used in the current implementation of the Apache Web server. An experimental evaluation of connection management policies, conducted using Web server logs, shows that our policy achieves a 15-20% reduction on a busy Web server in terms of cost with respect to the fixed holding-time policy. As a result, the mechanism allows busy Web servers to use server resources more efficiently without additional server load, and it distributes services to clients more fairly.