Packet Loss Minimization in Load-Balancing Switch

Due to the overall growing demand on the network resources and tight restrictions on the power consumption, the requirements to the long-term scalability, cost and performance capabilities appear together with the deployment of novel switching architectures. The load-balancing switch proposed in [1,2] satisfies to the above requirements due to a simple distributed control and good performance characteristics. However, as it was proven in [3, 4] a set of specific assumptions applied to the load-balancing switch restrains the above advantages. In particular, due to the limited information availability, cental stage buffers can overflow and, correspondingly, a packet loss can occur. In this paper we present a novel load-balancing service protocol which uses the congestion allocation technique to allow the drop of the arriving packets also at the input stage. If congestion indication is detected, the input stage will drop the whole packet upon its arrival while reducing the probability of congestion at the output reassembly unit. In the following the mathematical model for joint input/central stage packet loss evaluation will be presented. Based on the presented analysis we account the ways to minimize the mentioned packet loss.