Team : Phare
Departure date : 07/11/2009
Supervision : Guy PUJOLLE Co-supervision : PENNA Manöel
Fuzzy Logic Based Method for Modeling Queue Behavior of Network Nodes
Capacity planning of IP-based networks is a difficult task. Ideally, in order to estimate the maximum amount of traffic that can be carried by the network, without violating QoS requirements such as end-to-end delay and packet loss, it is necessary to determine the queue length distribution of the network nodes under different traffic conditions. This relationship is strongly dependent on the incoming traffic profile and the queuing discipline adopted by the network node. Analytical models for queue length distribution are available only for relatively simple traffic patterns. The characterization of a model for the queue length distribution is a vastly studied subject, but the associated mathematical apparatus becomes more and more complicated when we have to deal with multiple priority queues, non-exponential service times and long term correlated traffic. Also, when per-flow guarantees are required, it is necessary to determine the impact of the queue behavior on the performance of individual flows.
This thesis addresses the design of performance models of network nodes. We propose a generic method for modeling queue behavior aiming to provide a methodology for building the corresponding performance models. We take into account scenarios in which it is very hard to develop an analytical model. The proposed method combines non-linear programming and simulation to build a fuzzy model capable of determining the performance of a network node. Using such strategy, which is based on a general off-line method to produce the equations capable of representing outputs for the whole space of the specified input traffic parameters, it is possible to find out the optimal values that can be used for the configuration of network nodes, with important applications on traffic engineering and capacity planning. This approach does not require the derivation of an analytical model and can be applied to any type of traffic. Also, this methodology includes a training method that permits the application of any type of performance metric.
Defence : 07/10/2009 - 15h00 Jury members : M. TOHMÉ Samir, Professeur à l'UVSQ
M.ROBERTS James, Directeur unité R&D, France Telecom
M. MINOUX Michel, Professeur à l'Université PMC
M. PERROS Harry, Professeur à NCSU
M. PENNA Manoel Camillo, Professeur à PUCPR
M. FONSECA Mauro, Professeur à PUCPR
Mme. MUNARETTO Anelise, Professeur à l'UTFPR
M.PUJOLLE Guy, Professeur à l'UPMC