Title: An Independent-Connection Model for Traffic Matrices

Authors: Vijay Erramilli, Mark Crovella Dept. of Computer Science, Boston Univ. , Nina Taft, Intel Research, Berkeley

Date: 09/06/2006

Abstract:

The `gravity' model has been used both for traffic matrix (TM)
estimation and for generating synthetic TMs.  It is based on the
assumption that a packet's network egress is independent of its ingress.
We argue that in real IP networks, this assumption should not and does
not hold.  The fact that most traffic consists of two-way exchanges of
packets means that traffic streams flowing in opposite directions at any
point in the network are {\em not\/} independent.  In this paper we
propose a model for traffic matrices based on independence of {\em
connections\/} rather than packets.  We argue that the
independent-connection (IC) model is simpler, more intuitive, and has a
more direct connection to underlying network phenomena than the gravity
model.  Using publicly available TMs, we show that the IC model fits
real data better than the gravity model.  We then
characterize the parameters involved in the IC model based on our
datasets; these results can be used to construct synthetic TMs.
Finally, we turn to the well-studied problem of choosing a prior for TM
estimation.  Assuming that certain parameters of model can be measured
in advance and remain constant in time, we show that the IC model yields
a better prior for TM estimation than the gravity model.