Comparison of Fair Bandwidth Allocation Techniques in Multi-rate Loss Networks

摘要

While allocating network resources to competing flows in a fair manner has gained much attention in recent publications, it is still an open question whether there is any clear economical or technical reason for preferring one bandwidth allocation strategy over another. Therefore, in this paper we consider the problem of bandwidth allocation to contending rigid and elastic flows in multi-rate loss networks under various fairness eriteria. We formulate the bandwidth allocation problem as an optimization task, and extend previous works by allowing a minimum and a maximum bandwidth demand to be associated with each flow. By doing so we model both elastic and rigid flows in a multi-rate loss model framework and by solving the problem numerically in a simple case we gain insight into network performance measures including utilization, flow throughput, flow delay and blocking. Since we also model the impact of window based flow control (assuming fix window size), we believe that this work helps understand traffic engineering trade-offs in the Internet. We conclude that in terms of the most important performance measures the difference between the max-min, the proportional rate and the minimum-delay fairness allocations strongly depends on the network utilization and therefore the application of the various fairness criteria (and corresponding implementational complexity) will ultimately depend on the network operator's business model.