Contrrle Des Ressources De Rrseaux Sur Des Chelles Temporelles Multiples

des erreurs de frannais (ce qui est fort improbable), priire de prendre directement contact avec lui. Avertissement: ce paragraphe ne peut tre entiirement apprrcii que par J-C. Bolot et l'auteur :-) 6 Abstract Real-time multimedia applications require resource allocation because of their quality of service (QoS) requirements. On the other hand, network eeciency depends crucially on the degree of resource sharing inside the network. A key problem in concurrently achieving both goals is caused by the uctuation over multiple timescales of the traac bandwidth emitted by these applications, because it makes it hard to predict resource requirements with suucient accuracy. This in turn requires a careful design of control mechanisms so that they cover all timescales. In this thesis, we examine resource control over three natural timescales. On the packet timescale , we evaluate the performance of traac smoothing as a mechanism to accommodate bandwidth uctuation. We show that there exists a correlation horizon that separates relevant from irrelevant uctuation timescales for the purpose of performance prediction. This illustrates the general principle that the traac, system, and performance metric timescales together determine the set of candidate traac models. On the burst timescale , we advocate renegotiation as an eecient mechanism to accommodate uctuations over timescales beyond the correlation horizon. A new network service model called RCBR (Renegotiated Constant Bit Rate) combines network simplicity with desirable quality of service guarantees, while achieving much of the potential statistical multiplexing gain of bursty traac. On the ow timescale , we discuss measurement-based admission control as a means of relieving the application of the burden of a-priori traac speciication. The impact of measurement uncertainty and the interplay of burst and ow timescale dynamics are captured in a single analytical model. This model provides useful guidelines for robust admission control in the presence of unknown traac statistics. A key result shows that robustness can be achieved by matching the traac measurement timescale to the critical timescale of the system, which depends on the average ow holding time and the system size. We then address two issues in multicast transmission. First, we extend the RCBR service model to multicast ows to illustrate the tradeoo between network and application complexity. In particular, we compare the complexity of applications in a best eeort network and in a network ooering controlled sharing through renegotiation. For hierarchically encoded ows, we examine the issue of making the network aware of the …