Realtime Burstiness Measurement

Bursty traffic causes excessive queuing delay and packet losses. Many researchers have proposed the modeling of burstiness and burstiness mitigation schemes. However, there is no consensus on a quantitative definition of burstiness. In this paper, burstiness is defined as the queue size of a virtual bottleneck buffer. This definition is based on the fact that packets are queued when there is a bottleneck, and then the difference between target and service rates is reflected in queue length, i.e. the amount of data stored at the buffer before the bottleneck. We design and implement a realtime burstiness measurement method on a fully programmable network testbed, GtrcNET-1. Furthermore, we measure actual traffic on two networks, and we present the relationship between burstiness and the packet loss rate, and the behavior of flow aggregation. The results show the effectiveness of both the definition and our measurement method.