The impact of the Convergence Sublayer Protocol Data Unit (CS_PDU) size on the end-to-end network delay in a B-ISDN environment is investigated in this paper. Cell loss probabilities are assumed to be bursty and when a cell loss occurs a retransmission is assumed to happen at the CS_PDU level. The cell loss probability is assumed to be such that there are two different probabilities for cell loss. Typically the cell loss probability is very small, in a real ATM network 10-9 or better.There are however instances when the cell loss probability will be significantly higher due to network congestion, such as buffer overflow, say a cell loss probability of 10-3. An ATM network, for simplicity consisting of one single link, has been simulated and the simulation results indicate that it is possible to select a CS_PDU size that will minimize the end-to-end delay. The ATM network has also been analyzed by using mathematical modeling technique. Our first mathematical analysis approach assumed naively a Poisson arrival ss. The resulting closed queueing.network was solved approximately and an expression for the end-to-end delay as a function of the CS_PDU size was obtained. It was then shown that the delay is optimized for a particular CS_PDU size. For high speed networks, burstiness is a very important factor for system end-to-end delay and thus a Poisson arrival process is not a reasonable assumption, furthermore batch arrivals of CS_PDU should also be considered. It is found that queueing models are IBP[x]/GEO/1/K and IBP[x]/D/1/K systems. Numerical results for queue length distributions and system end-to-end delays for both queueing systems have been obtained and are used to obtain the optimal CS_PDU size.
展开▼