Pragmatic modeling of broadband access traffic

Abstract

Good traffic modeling is a basic requirement for accurate capacity planning. The recent discovery of heavy-tails, and long-range dependence (LRD) in traffic has heralded a new, and more elegant way to model data traffic, particularly characteristics such as extreme burstiness across many time scales. However, most of the measurements used to populate such models have been fine grained packet traces. In reality we are far from being able to obtain such traces from more than a small subset of the Internet, and this is likely to remain true at least in the immediate future. The only source of ubiquitous data is Simple Network Management Protocol (SNMP), but SNMP has many limitations which make it difficult to work with for traffic modeling. These limitations make it impossible to use standard LRD models. However, we show here that for broadband access, SNMP is capable of capturing the most important features of the data traffic. We base this analysis on a large volume (more than 2 months) of SNMP data obtained from a large operating broadband access network. The model is approximate, but is nonetheless quite useful for capacity planning. The results validate our intuition about LRD in data traffic, while allowing the key parameters of the model to be computed solely from SNMP traffic utilization data.