Alternative ATM LAN architectures for connectionless traffic and their performance analysis

• Main Author: Asgari, A.
• Published: Swansea University 1997
• Subjects: 621.382
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.635800

ATM has been chosen as the switching and multiplexing technique for B-ISDN. ATM supports a wide range of services, including voice, data, and video, and provides a flexible means to multiplex these traffic types on the same physical network. While ATM is intended for public area networks, there has been growing interest in supporting ATM in the local area environment. ATM LANs are expected to offer a more attractive backbone network than other solutions and a gradual migration to desk top, as well as the gateway to B-ISDN when it becomes available. Integrating connection-oriented ATM into a local area environment poses a challenge with respect to interconnecting connectionless LAN traffic. The main objective of the research reported in this thesis is therefore to investigate alternative ATM-based LAN architectures for handling connectionless traffic. Three novel ATM LAN architectures based on Frame Sequencing at the source switch, Local Broadcasting at the destination switch, and a centralised Connectionless Server (CLS), are designed, which collectively contribute to existing research in this area. An alternative architecture proposed by ATM Forum is known as LAN Emulation. The ATM Adaptation Layer type 5 (AAL5) is used in these architectures as the preferred choice for carrying data traffic. ATM network models are developed in this research to study the performance of the centralised CLS architecture and also LAN Emulation architecture. In the centralised CLS architecture, the CLS itself and in the LAN Emulation architecture, the Broadcast/Unknown Server (BUS), play a key role in handling connectionless traffic and are likely to become bottlenecks. A set of metrics defined as quality of service parameters (cell/frame loss rate, cell/frame delay, jitter) is used to evaluate the performance of these two architectures and their associated servers. It is shown that these metrics are not only affected by design parameters such as queuing discipline, buffer capacity, link bit rate, the processing speed of servers, and number of sources but also by input traffic characteristics such as traffic load, traffic burstiness, average frame sizes, maximum frame sizes, and incoming and outgoing connection rates.