|
Activity Number:
|
16
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, August 2, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #304553 |
|
Title:
|
Probabilistic Modeling and Statistical Inference for Computer Network Traffic
|
|
Author(s):
|
Stilian Stoev and George Michailidis*+ and Joel Vaughan
|
|
Companies:
|
University of Michigan and University of Michigan and University of Michigan
|
|
Address:
|
439 W. Hall, Ann Arbor, MI, 48109,
|
|
Keywords:
|
kriging ; fractional Brownian motion ; wavelet spectrum ; spatio-temporal ; long-range dependence ; network traffic
|
|
Abstract:
|
Accurate and physically interpretable models of computer network traffic are essential for the management, provisioning and design of networks. We propose a new spatio-temporal statistical model for network-wide traffic, which is based on the physical behavior of the individual users. The temporal long-range dependence and the routing structure of the network play key roles. In the proposed framework, we solve the kriging-type problem of predicting the traffic fluctuations on a un-observed set of links from time series measurements of an observed set of links. The temporal long-range dependence and the self-similarity present in the model make wavelets a perfect tool for the estimation of the involved parameters. Kriging in the wavelet domain, on the other hand, allows the practitioner to detect synchronized attacks and anomalies on multiple time-scales.
|
