Impact of Aggregation on Scaling Behavior of Internet Backbone Traffic", Sprint ATL

We study the impact of aggregation on the scaling behavior of Internet backbone traffic, based on traces collected from OC3 and OC12 links in a tier-1 ISP. We make two striking observa-tions regarding the sub-second small time scaling behaviors of Internet backbone traffic: 1) for a majority of these traces, the Hurst parameters at small time scales (1ms- 100ms) are fairly close to 0.5. Hence the traffic at these time scales are nearly uncorrelated or independent; 2) the scaling behaviors at small time scales are link-dependent, and stay fairly invariant over changing utilization and time. To understand the scaling behavior of network traffic, we develop analytical models and employ them to demonstrate how traffic composition – aggregation of traffic with differ-ent characteristics – affects the small-time scalings of network traffic. The degree of aggregation and burst correlation struc-ture are two major factors in traffic composition. Our trace-based data analysis confirms this. Furthermore, we discover that traffic composition on a backbone link stays fairly consis-tent over time and changing utilization, which we believe is the cause for the invariant small-time scalings we observe in the traces