Full length article: On coding for reliable communication over packet networks
Physical Communication
| Hi-index | 0.00 |
Modern data networks are heterogeneous in that they often employ a variety of link technologies, such as wireline, optical, satellite and wireless links. As a result, internet protocols, such as Transmission Control Protocol (TCP), that were designed for wireline networks, perform poorly when used over heterogeneous networks. This is particularly the case for satellite and wireless networks which are often characterized by high bandwidth-delay product and high link loss probability. This thesis examines the performance of TCP in the context of heterogeneous networks, particularly focusing on interactions between protocols across different layers of the protocol stack. First, we provide an analytical framework to study the interaction between TCP and link layer retransmission protocols (ARQ). The system is modelled as a Markov chain with reward functions, and detailed queueing models are developed for the link layer ARQ. The analysis shows that in most cases implementing ARQ can achieve significant improvement in system throughput. Moreover, by proper choice of protocols parameters, such as the packet size and the number of transmission attempts per packet, significant performance improvement can be obtained. We then investigate the interaction between TCP at the transport layer and ALOHA at the MAC layer. Two equations are derived to express the system performance in terms of various system and protocol parameters, which show that the maximum possible system throughput is 1/e. A sufficient and necessary condition to achieve this throughput is also presented, and the optimal MAC layer transmission probability at which the system achieves its highest throughput is given. Furthermore, the impact of other system and protocol parameters, such as TCP timeout backoff and MAC layer retransmissions, are studied in detail. The results show that the system performance is a balance of idle slots and collisions at the MAC layer, and a tradeoff between packet loss probability and round trip time at the transport layer. Finally, we consider the optimal scheduling problem with window service constraints. Optimal policies that minimize the average response time of jobs are derived and the results show that both the job lengths and the window sizes are essential to the optimal policy. (Copies available exclusively from MIT Libraries, Rm. 14-0551, Cambridge, MA 02139-4307. Ph. 617-253-5668; Fax 617-253-1690.)