Balancing HTTP traffic using dynamically updated weights, an implementation approach

Abstract

In this paper we present a load balancing application for HTTP traffic that uses dynamic weights. We introduce a load balancing policy based on two criteria: "process time" and "network delay". The former describes Web servers ability to process a forthcoming request, while the latter tries to estimate network conditions. Calculation of the two criteria is periodically updated. A Weighted Round Robin algorithm was implemented using the two aforementioned metrics in order to dynamically estimate the balancing weights. We confirm that the combination of the two criteria increases sensitivity and responsiveness of the application towards network conditions and therefore the performance of the whole load balancing system. Balancing decisions should not be only "load" or "connection" dependent, but also contention dependent. © Springer-Verlag Berlin Heidelberg 2005.