网络流量特征对排队性能影响的仿真分析与比较

资源的匮乏和流量的不确定性是造成性能问题的根源,为明确流量的方差特征和自相似性两个因素中究竟哪一个因素对排队性能影响更大,设计了两组独立的仿真试验进行分析比较。第一组实验基于一个渐进二阶自相似流体模型,通过大规模迭代计算近似求解,模型的流量分布取自实际网络流量边缘分布。第二组实验在ns2仿真工具中生成分形高斯噪声模拟流量并统计性能参数。两组实验结果均表明,排队性能受流速方差的影响远远超过受Hurst系数的影响。在实验设定的条件下,由方差引起的丢包率变化幅度比由Hurst系数引起的约高出2个数量级,平均排队时延也相差上百毫秒。而且,Hurst系数对排队性能的影响还受到方差的制约。

Simulation & Comparison of Impact of Traffic Characteristics on Queuing Performance

Variance and self-similarity are two important characteristics of network traffic. Both of them have distinct impact on performance. To analyze and compare their impact on queuing performance, two independent groups of simulation experiments were designed. In the first one, an asymptotically second-order fluid traffic model, whose marginal distributions come from practical network traffics, was employed and substantive iterations were used to approach evaluations of performance metrics. In the second one, statistics of performance metrics were obtained by generating simulation traffic conforming FGN model in ns2. Results of both experiments indicate that the impact of traffic variance on performance always exceeds that of the Hurst parameter sharply. Under the experiment settings, the change in loss rate caused by variance is almost two magnitudes more than that caused by Hurst parameter. Difference in average queuing delay exceeds 100ms too. In addition, the Hurst parameter's effect is restricted by the variance.