Statistical Models of Internet Traffic - Internet Tomography

Bowei Xi

Estimating quality of service parameters, such as link loss rates
and delay distributions, associated with computer
and communication networks are of considerable importance to the
network administrators and service providers. A new
class of probing schemes to collect end-to-end measurements
is proposed and its properties studied. Such measurements are used to
recover the underlying link loss rates by solving a statistical inverse
problem. Two types of estimators are introduced and their large sample
properties studied: (1) Maximum likelihood estimator (MLE), which
turns out to be computationally intensive; (2) Least squares (LS)
based estimators that are fast, easy to implement and provide an
explicit form of the covariance matrix of the parameters of
interest. A distributed version of the LS estimators is also
introduced and discussed. Finally, some design issues and
detection of changes over time in the loss rates will also be
briefly addressed.