A trace study for characteristics of packet arrivals

Packet arriving patterns have great impact on the design and evaluation of network equipments. In this paper, we present an analysis of four traces collected by NLANR, uncovering the characteristics of their packet arrival processes. Our contributions are two-fold. First, the analysis results show that both the sequences of successive small and large packets appear frequently because of the dependence between different packets. On the other hand, the probability for "worst-case" traffic of back-to-back small packets to show up tends to increase when the link utilization becomes higher. Since most of today's links are lightly loaded, "worst-case" traffic does not account for a large number of packets. Second, we investigate the variations of packet arriving bit rate for long-lived streams (flows) and find that high speed streams are more likely to be accumulated and exhibit high level of burstiness. The influence of application and network dynamics can also be observed in the arriving patterns of low speed streams. Statistical analysis shows that over 60% of streams (flows) have an arriving bit rate variation comparable to or larger than average.