Dynamic dispersion framework for router load balancing

Load balancing is a key technology to achieve high performance as well as high reliability. Most of traditional approaches are based on a critical assumption that the service qualities of channels are already known. With such information, they attempted to preventively disperse work-loads to attain fairness and therefore high efficiency. However, in practice, such information is difficult to obtain, especially when dispersion is performed at high-level layers. In this paper, we propose a novel scheme Dynamic dispersion frAMework (DAM), to adoptively disperse the traffic. Based on collected statistics, DAM dynamically adjusts the dispersion strategy to achieve fairness in terms of real capabilities of channels. Our experimental results show that DAM provides an accurate dispersion strategy compared with an offline optimal solution. The other advantage of DAM is that it is transparent to both the user and the server.