Congestion Management in Interconnection Networks

Abstract

Interconnection networks (INs) are becoming more and more powerful and hence place increasingly higher demands on the networks that interconnect their processors or processing nodes. Many of the applications running on such systems, especially embedded systems applications, with increasing application demands, high-performance networks are the hearts of these systems. These applications play an essential role in such networks. Congestion management remains yet a critical problem for interconnection networks. When traffic load is unfairly distributed across the network, some resources could be idle while others are quite congested. Imbalance of communications can produce network congestion. If congestion is not efficiently controlled, it is possible that these resources reach the saturation. This leads to a rise in message latency. In addition, global system performance is degraded due to congestion and this performance degradation is quickly propagated to the entire network. Therefore, a high speed routing scheme that minimizes congestion and avoids hot-spot areas should be included in the network components. This thesis presents a Load-Balanced Distributed Adaptive Routing (LBDAR) algorithm that is used to manage congestion in interconnection networks. LBDAR algorithm aims to achieve high throughput, while delivering packets with low latency. LBDAR is adaptive routing mechanism that balances the communication traffic over the interconnection network routers and switches. This communication balancing is based on load-control and path expansion in order to maintain low message latency values. Experimentation shows that LBDAR achieves significant performance improvement over other algorithms used to manage congestion in INs.