A MAC/flow level modeling of data and voice integration in WLANs
Mariana Dirani, Tijani Chahed · 2006
We focus in this paper on modeling the integration between streaming and elastic flows in IEEE802.11-based sys- tems. In such a CSMA/CA shared context, both types of flows do not behave in the same way. Streaming flows, characterized by a constant bit rate, are not able to get more than their need; they however suffer a degraded performance when their share is lower than their constant rate. We then model them as non- saturated sources. Data sources are on the contrary saturated ones and share the bandwidth in a fair manner, as given by a Processor Sharing model. In the presence of the two types of flows, a joint model is hence considered at the MAC layer; the results of which, in terms of QoS for streaming flows and service rates for data ones are then fed into a Markovian flow level analysis. I. INTRODUCTION IEEE 802.11-based networks implement CSMA/CA at the MAC layer which shares the radio resources among con- tending flows. The latter are typically data ones which have an elastic property, ie, they are able to share the resource evenly among themselves, as given by a Processor Sharing (PS) model. Introducing streaming flows in such a shared medium brings into the picture their non-saturated nature which makes them different from data ones in sharing the resources at the MAC layer. They in effect may or may not have a frame to send depending on their (frame) arrival rate and cannot thus take advantage from, say, a higher than needed share of resources. If on the contrary, their share of resources is below this nominal rate, they suffer from a degraded performance. At the flow level too, streaming flows are different from elastic ones in that they precisely do not exhibit an elastic property. That is, their service rate is independent of the share of resources they get from the network. In this work, our aim is to model the presence of both types of traffic simultaneously in the network. We first consider the MAC layer and introduce a variation of the Bianchi model (1) for streaming traffic. The idea is to introduce a new state denoting the non presence of a frame to transmit. We are then able to determine the mean rate obtained by a streaming flow in a presence of a given (constant) number of streaming and elastic flows as well as the obtained QoS as given by the inter- frame delay. These results are then used at the flow level to determine the service rate of data flows as well as QoS of streaming ones. A quasi-stationary analysis makes it possible to derive closed form expressions for mean number of voice and data flows as well as performance metrics such as QoS and mean transfer times. The remainder of this paper is organized as follows. Section II is devoted to the MAC layer where we review Bianchi's model for elastic traffic and introduce our new one for streaming traffic. Section III contains our exact Markovian modeling at the flow level as well as a quasi-stationary analysis. Section IV shows our numerical results. Eventually, Section V concludes the paper.