Network Tomography for estimation of temporal link characteristics

Vijay Arya

NICTA, University of Melbourne, Australia


Résumé:

Similar in spirit to medical tomography which is used to diagnose patients without operating on them, network tomography is an approach to diagnose IP networks without deploying additional monitoring equipment inside them. Typically the approach involves collecting 'slices' of network measurements ('tomos') and correlating them to statistically estimate the underlying characteristics of networks ('graphias'). MINC (Multicast Inference of network characteristics) is a tomography approach which can infer characteristics of internal network paths from end-to-end measurements of multicast probes. This helps network service providers to monitor network paths and identify bottlenecks in their network. Previous work in this area has focussed on estimating average loss rates and delay distributions of network paths. However, since Internet traffic is bursty, loss rates and delay distributions do not provide fine-grained view of link behavior, in particular temporal characteristics such as typical durations of loss bursts and bursts of long delays. Knowledge of temporal characteristics has applications for services such as VoIP which are sensitive to loss and delay bursts, as well as for bottleneck link detection. In this talk, I will present our work on inferring temporal characteristics of internal network paths, from end-to-end measurements of multicast probes, by focussing on the case of loss. (This is joint work with Nick Duffield (AT&T Labs-Research) and Darryl Veitch (CUBIN, Univ. of Melbourne) )


[Vijay Arya]
NICTA, University of Melbourne, Australia