Load balancing in grid networks

A load balancing algorithm functioning as an extension to the topological routing scheme is proposed. The algorithm is designed to evenly distribute the traffic load in a 4-regular grid network. This is done by letting each node calculate how loaded it is with regular intervals and feed this information to its neighboring nodes, which then use it to derive link probabilities applied in probabilistic routing. The traffic load is defined as the average link utilization of each link and the average expedition delay for each output queue, i.e. the waiting time of packets in queues. The algorithm has been tested on the i3 Demonstrator, a 6×6 grid network testbed, using different traffic patterns. Results show that better performance is achieved when a high level of traffic including hotspot traffic is applied. Furthermore, packet drops and lengthy packet delays are avoided. With low traffic levels or with a uniform-like P2P traffic pattern, the performance is neither improved nor degraded noteworthy.