Network Modeling for the Internet
Changepoints
and Extremes
| Leader | Richard Smith (University of North Carolina, Chapel Hill), rls@email.unc.edu |
| Meeting | Thursday 3:00 - 4:00 pm,
room 203 |
| Members | Robert Buche
(North Carolina State University), rtbuche@unity.ncsu.edu Fred Godtliebsen (University of Tromso, Norway), godtlieb@email.unc.edu Krishanu Maulik (EURANDOM, The Netherlands), maulik@eurandom.tue.nl Cheolwoo Park (SAMSI), cwpark@email.unc.edu Juhyun Park (University of North Carolina, Chapel Hill), parkj@email.unc.edu Haipeng Shen ( University of North Carolina, Chapel Hill) , haipeng@email.unc.edu Murad Taqqu (Boston University), murad@bu.edu Haakon Tjelmeland (Norwegian University of Science and Technology, Norway), haakont@samsi.info Zhengyuan Zhu (University of North Carolina, Chapel Hill), zhuz@email.unc.edu |
| Outline Objectives | Pictures of bursty internet
traffic tend to
show a number of
characteristics: (a) periods of bursty behavior interspersed with non-bursty behavior; (b) seemingly sharp transitions from one to the other (though it is not clear just how sharp these transitions are); (c) within a bursty period, a much greater than usual frequency of extreme observations, both above and below the overall mean. The objective of this working group is to explore approaches in which the transitions between bursty and non-bursty behavior are considered changepoints, with models and methods from extreme value theory used to characterize the bursty periods. The changepoints form some version of a point process (the simplest model would assume a homogeneous Poisson process, but others can be considered) while the bursty periods are viewed as independent realizations of some kind of threshold-exceedance process whose parameters are allowed to vary from one bursty period to another in the manner of a hierarchical model. The structure is flexible enough to allow for a variety of alternative specifications, and can be fitted through the Reversible Jump MCMC paradigm. Finally, it is proposed that the fitted model(s) be used to calculate a variety of "indexes of burstiness" (the development of which should be considered part of the research program) and thus used to compare different internet servers and paradigms. |
| Webpage | http://www.stat.unc.edu/faculty/rs/cpev/hpage.html |
© 2002, Statistical and Applied Mathematical Sciences Institute. All rights reserved.