(An) efficient priority-scheduling algorithm forintergrated services packet networks

Abstract

This dissertation work is devoted to the study of handling heterogeneous traffic mix efficiently in an integrated services packet network. First, for an efficient scheduling strategy we propose and analyze a combined preemptive/nonpreemptive priority discipline. As the discretion rule for preemption, we suggest three schemes based on the parameters of the low-priority job;elapsed service time, elapsed to original service time ratio, and remaining service time. In this work, by using the busy-period analysis technique we analyze an M/G/1 queueing system with multiple priority classes of jobs. Considered preemptive rules are the preemptive-resume, preemptive-repeat-identical, and preemptive-repeat-different policies. These three preemptive rules will be analyzed in parallel. The key idea of analysis is based on the consideration of a busy period as a composite of delay cycles. As results, we present the exact Laplace transforms associated with waiting time, response time, and the number of jobs in the system in addition to their expectations as well. We also show some numerical examples to investigate the system performances with different discretion schemes. In addition, practical applications of the proposed model are presented. Second, for improving the performance deterioration due to the retransmissions of preempted packets, we newly propose and analyze a packet-segmentation technique in an integrated services packet network. In the proposed system, it is assumed that messages arrive according to a Poisson process with an arbitrary length. After entering the switch, the messages are divided into packets with a constant size, and then the header is added. As the scheduling strategy the combined preemptive/nonpreemptive priority discipline with the third discretion scheme is considered. The proposed system is advantageous in that if a low-priority packet with a long size is transmitted through a network under a light traffic load due to higher-priority classe...