EECS 290Q: Project Abstracts

Evaluating the Feasibility of Estimating QoS Parameters and Bypassing Modelling for Real ATM Traffic
Chen-Nee Chuah

One challenge of ATM network is to support traffic streams with different characteristics and service requirements on the same network. Therefore, these networks must exercise admission control as well as reserving bandwidth and buffer capacity for each connection. Statistical multiplexing offers large potential gains for real time traffic, because it allows efficient sharing of network resources. However, for statistical multiplexing to work in practice, we need good descriptors of the traffic to allow rational decisions to be made with regards to resource allocations. Existing schemes involve traffic modelling and thus are heavily dependent on the knowledge of traffic statistics. For example, simple call admission policy has been proposed based on the concept of effective bandwiidth, which depends on cell loss probability and statistics of the stream. To remove this dependence, Duffield et al has proposed to estimate the rate function of the arrival stream itself at the input switch and bypassing the modelling process completely([1]). In [2]-[3], experimental results are shown using real ATM traffic to test the feasibility of method proposed by Duffield et al.

In this talk, I will give an overview of the main ideas discussed in the referenced papers ([1]-[3]). The advantages as well as problems of the proposed schemes will be pointed out.

UCB Networking / Matt Siler siler@eecs.berkeley.edu/ April 19, 1996