MoN16: Sixteenth Mathematics of Networks meeting
Sotiris Skaperas (University of Macedonia, Greece) – Change point detection analysis for load balancing based on content popularity
In this talk, we suggest that early detection of traffic changes can be achieved through studying user behaviour in terms of surfing habits. Along these lines, we investigate the web content popularity dynamics with a change-point detection analysis. We propose such methodology as a promising way to detect traffic load peaks and drive load balancing mechanisms. We use a database of real youtube video visits to produce the input time-series. We apply off-line and on-line Cumulative Sum Control Chart (CUSUM) change point tests to detect the existence, the number, the magnitude and the direction of changes in the mean and the variance of the studied sample. For example, traditional relevant detection methods do not consider the direction of changes. To tackle this issue, we propose a variation of the Moving Average Convergence Divergence (MACD) indicator to detect the direction of estimated change points. Our off-line analysis provides useful insights for the problem requirements and reference results for the on-line methods evaluation. Our results from our on-line approach highlight its usefulness to detect early changes in traffic behaviour. We conclude our presentation with the open-issues and our next steps.
Contact:
Keith Briggs
(
)
or
Richard G. Clegg (richard@richardclegg.org)