Skip to Main content Skip to Navigation

Transport de Flux Temps Réels dans un Réseau IP

Abstract : In this thesis, we study several implementation methods of the QoS (Quality of
Service) in an IP network. Before starting our research, we will first reveal the various
mechanisms of QoS which we will study in the thesis. MPLS (Multi Protocol Switching
Label), DiffServ (Differentiated Services) and scheduling algorithms will form the basic
mechanisms of our backbone network which we will study. In accordance with several other
works, we define three classes of services to differentiate in the network. The first class
includes voice flows. It requires a small delay and a reduced jitter. Flows of "critical data",
requiring a low loss probability and a limited delay, form the second class. The third class,
gathering the applications such as file transfer or email exchange, do not need any condition
from the network.
As a first step, we model outgoing/entering flows of a mobile wireless network. We
model the inter arrival law of the aggregation of the flows at the MAC (Medium Access
Control) layer, entering the UTRAN (UMTS Terrestrial Radio Access Network) network, the
UMTS (Mobile Universal Telecommunication Service) access network. The CDMA (Code
Division Multiple Access) protocol used in this network, proposes different access probability
according to the QoS required. We determine also the characterizing law of the inter arrival
aggregated traffic leaving a WiFi (Wireless Fidelity) network, using the basic MAC 802.11
layer. For these two networks, we propose various models of aggregation of voice, Web, file
transfer flow and a multiplexing of these various classes. We measure the precision of two
distribution models to those created by the traces. The first distribution used is the MMPP
(Markov Modulated Poisson Process) process which represents a Markovian model. We
experimented two values of the states number: 2 and 4. The second law that we consider is the
Gaussian law. Our results show that both the type of aggregate flows and the network used
influence the resulting model.
In a second step, we develop the equations determining the stationary probabilities of
a queue implementing a GPS (Generalised Processor Sharing) scheduler with three classes of
services. By using the DiffServ mechanism to differentiate flows, we measure the QoS at the
output of a single queue using WRR (Weighted Round Robin), one of the algorithms
approximating GPS. We then plot the various curves of delays and loss probabilities observed
at the output of this queue according to weight and the load created by each class. We apply
the various conclusions of the choice of the parameters which we draw from one server to a
whole network. Moreover, we add the traffic engineering capability of MPLS to quantify the
gain added by each policy. From this work, we can generalize our observations which become
valid as well on a queue as in a network.
In a third step, we develop a method of dynamic adaptation of the routing. We propose
it in order to solve the problem with the variations of the delay distribution on the links
forming the end to end path. This mechanism is based on network tomography techniques in
order to estimate the delay distribution on the various sections of the paths observed. If the
average delay on the used path remains greater of the ε threshold for a time τ than the average
delay of another path, then, the mechanism initiates the procedure of path modification. The
use of the MPLS protocol allows this mechanism to be flexible and fast.
Complete list of metadata

Cited literature [50 references]  Display  Hide  Download
Contributor : Hazar Aouad Connect in order to contact the contributor
Submitted on : Wednesday, October 4, 2006 - 4:58:18 PM
Last modification on : Friday, July 31, 2020 - 10:44:02 AM
Long-term archiving on: : Tuesday, April 6, 2010 - 6:13:20 PM


  • HAL Id : tel-00103576, version 1


Hazar Aouad. Transport de Flux Temps Réels dans un Réseau IP
Mobile. Réseaux et télécommunications [cs.NI]. Télécom ParisTech, 2005. Français. ⟨tel-00103576⟩



Record views


Files downloads