On comparing the real and probed packet drop rates of a bottleneck router: The TCP traffic case

In this paper, we address the question of whether the packet drop rate of a bottleneck router measured (experienced) by probe packets can accurately reflect the bottleneck router's real packet drop rate. To answer this question, we built a testbed network and conducted several real experiments on it. In these experiments, using various traffic types and loads, we compared the probed and the real packet drop rates of the bottleneck router. We also collected the internal queue length variation process of the bottleneck router and used it to do several post analyses to answer some 'what-if' questions. This paper presents some interesting findings derived from this study.