Parameter estimation and performance analysis of several network applications

Abstract : In this thesis, we have looked at several modeling problems to estimate some parameters or to evaluate the performance of some mechanisms. The first subject studied is about unicast applications performing rate control to adapt to network conditions. To that end, we have proposed to estimate the cross traffic intensity and the buffer size at the bottleneck of a connection. We have developed two inference models based on the $M/M/1/K$ and $M/D/1/K$ queues, and derived eleven schemes for estimating the above-mentioned characteristics. The performance of these schemes have been evaluated and compared using ns-2 simulations, and the best scheme has been identified. The second subject investigated is about the on-line estimation of the membership of dynamic multicast groups. We first model the multicast group by an $M/M/\infty$ queue and show that under heavy traffic the estimation problem can be solved using a Kalman filter. We next use the same model, but with a general traffic regime, and derive the estimator using Wiener filter theory. We last design an efficient estimator using the $M/H_2/\infty$ queue model. The performance of these estimators have been evaluated over simulations driven by both synthetic and real session traces. The third subject studied is about location mechanisms in a mobile agent environment. There are two widely used mechanisms to ensure communications between components of applications. We have proposed Markovian models to evaluate the cost of these mechanisms in terms of response times. We have validated our models both via simulations and experiments over a LAN and a MAN. We were then able to use the models to formally compare the performance of both mechanisms.
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-01115023
Contributor : Sara Alouf <>
Submitted on : Tuesday, February 10, 2015 - 1:58:45 PM
Last modification on : Thursday, January 11, 2018 - 4:55:56 PM
Long-term archiving on : Saturday, September 12, 2015 - 10:41:07 AM

Identifiers

  • HAL Id : tel-01115023, version 1

Collections

Relations

Citation

Sara Alouf. Parameter estimation and performance analysis of several network applications. Networking and Internet Architecture [cs.NI]. Univeristé Nice Sophia Antipolis, 2002. English. ⟨tel-01115023⟩

Share

Metrics

Record views

269

Files downloads

241