Traffic-control methods for performance enhancement of computer networks
V. Shurbanov, D.R. Avresky · 2000
Since the main objective of this research is enhanced network performance, a significant part of it focuses on the means for performance evaluation. We introduce a novel closed-queueing model for input-queueing space-division packet switches, which explicitly reflects their structure and the traffic distribution, as parameters that can be varied. This distinguishes it from current models, the majority of which either are based on specific switches and traffic distributions or are forced to assume such to ensure tractability. Our model is widely applicable due to its unrestricted tractability for arbitrary switch structures and traffic distributions. To enhance network performance we first focus on the utilization of alternative communication paths. Much work has been devoted to employing such paths for fault tolerance. However, the problems of utilizing them for performance enhancement have been neglected, especially for static- routing networks. This work formally proves that the throughput of multiple paths is maximal for uniformly balanced traffic. Based on this, we introduce a method for traffic balancing in static-routing networks. The method is applicable to arbitrary multi-path topologies and traffic patterns that lend themselves to balancing. The second focus is on the arbitration of contending router ports. In practice the arbitration mechanism usually provides equivalent output bandwidth to all inputs. However, this strategy contradicts the widely accepted principle of fair arbitration. We propose a method for enforcing fairness by biasing the arbitration based on the network topology and traffic pattern. The method results in enhanced throughput, scalability, and latencies, in comparison with current arbitration strategies. The third focus is on end-to-end flow control. Two new mechanisms are evaluated and compared with the widely employed static window flow control. The first mechanism is a practically oriented modification of the packet-pair flow control. The second one is designed based on the analysis of dynamic performance data. It is demonstrated that it provides similar throughput but lower latencies than the other approaches. The purpose of the proposed methods is to provide a foundation for traffic control that provides enhanced performance in computer networks. The methods are general and potentially applicable to a variety of network architectures.