CMU-CS-02-167
Computer Science Department
School of Computer Science, Carnegie Mellon University



CMU-CS-02-167

FastCARS: Fast, Corrrelation-Aware Sampling
for Network Data Mining

Jia-Yu Pan, Srinivasan Seshan, Christos Faloutsos

December 2002

CMU-CS-02-167.ps
CMU-CS-02-167.pdf


Keywords: Traffic analysis, netowrk data mining, sampling


Measuring traffic on routers is vital for finding patterns, traffic modeling, and anomaly detection. Unfortunately, technology trends are making it more and more difficult to observe and record the large amount of data generated by high speed links. Traffic sampling techniques provide a simple alternative that reduces the volume of data collected. Real world data is seldom temporally independent and data observed at one time is likely to have important correlations with data observed at close-by instants in time. A good sampling method should be able to give measurements that take this correlation into account. Unfortunately, existing sampling techniques largely hide any temporal relationship in the recorded data.

Our proposed method, "FastCARS", naturally captures statistics for packets that are 1, 2 or more steps away. It has the following properties: (a) provides accurate measurements of full trace's statistics, (b) is simple and scalable for implementation, (c) captures correlations between successive packets, as well as packets that are further apart, (d) evenly separate sampling efforts over time, and (e) generalizes previously proposed sampling methods and includes them as special cases.

We also propose several new tools for network data mining and demonstrate the good quality of the information provided by FastCARS. These tools include: (a) The n-step histograms which give correlated statistics at different levels of temporal correlation, (b) the convolution test which could be used to examine the dependence level between packet arrivals. (c) the n-step packet-size/delay graph which provides accurate bandwidth estimation and load monitoring, and (d) the n-step flow graph which effectively visualizes flow patterns hidden in a trace.

The experimental results on multiple, real-world datasets (479Mb in total), show that the proposed FastCARS sampling method and these new data mining tools are effective. With these tools, we show that the independence assumption of packet arrival is not correct, and that packet trains may not be the only cause of dependence among arrivals. The provided tools may be useful in applications such as monitoring link load and traffic flows.

28 pages


Return to: SCS Technical Report Collection
School of Computer Science homepage

This page maintained by reports@cs.cmu.edu