Summary: | 博士 === 國立清華大學 === 資訊工程學系 === 88 === In this dissertation, the scheduling problems in ATM-PON (ATM-Passive Optical Network) system are studied. ATM-PON system has been developed by ITU-T and NTT as an access network system for the supporting of full service, in which several ATM (Asynchronous Transfer Mode) network systems are interconnected by a PSC (Passive Star Coupler) based WDM (Wavelength Division Multiplexing) network system. The design of scheduling disciplines for WDM and ATM network system has different criteria. As the backbone network system, WDM network with large bandwidth and high transmission rate needs a stable, adaptive and efficient scheduling discipline for accommodating the traffic changes. On the other hand, the scheduling discipline for ATM switching network system has to provide various QoS (Quality of Service) -guaranteed services. Based on these studies, we propose two scheduling disciplines for the WDM and ATM network system. These two scheduling disciplines can be implemented easily in the ATM-PON system.
In the WDM network system, we propose a traffic prediction scheme that considers all the bottlenecks and critical factors caused in the WDM network system. The simulation results show that the performance of proposed prediction scheme is much better than the traditional TDM and reservation scheme. Besides, the traffic prediction scheme is very suitable to be applied in backbone networks, in which the transmission performance is much more important than the services that the network can provide. Moreover, the traffic prediction scheme also can be applied in the smaller local area network system.
As to the ATM network, although Earliest-Deadline-First (EDF) is recognized as the optimal scheduling discipline to support the bounded-delay service, it lacks of the support of a prioritizing scheme for cell loss control. Consequently, we modify the simple FIFO to implement a near-EDF scheduling discipline with loss control. The proposed scheme can guarantee the stringent delay bound and minimize the cell loss rate. The adaptive queue lengths in the proposed buffer structure can be dynamically determined according to the traffic load and cell delay constraints to minimize the cell loss rates. With the different deterministic adaptive queue lengths, the proposed scheme is equivalent to the other previous schemes.
The simulation results show that the traffic prediction scheme has much shorter message delay time than the typical schemes. The accessible offered load of the Near-EDF with loss control scheme in ATM network system is higher than the other schemes. Further, the mathematical analysis also shows that the proposed scheme''s schedulability condition is approximate to that of EDF. We also derive the formulas of the cell loss probability in Near-EDF with loss control scheme to evaluate the cell loss rate of each priority class.
|