G. Pujolle et al. (Eds.): NETWORKING 2000, LNCS 1815, pp. 714-726, 2000 © Springer-Verlag Berlin Heidelberg 2000 Performance Evaluation of Resource Division Policies for PGPS Scheduling in ATM Networks Amr S. Ayad, Khaled M. F. Elsayed, and Mahmoud T. El-Hadidi Department of Electronics and Communications Engineering Faculty of Engineering, Cairo University, Giza, Egypt 12613 asaad@ie-eg.com, khaled@ieee.org, hadidi@frcu.eun.eg Abstract. The paper addresses the issue of reserving resources at ATM switches along the path of calls requiring a deterministic bound on end-to-end delay. The switches are assumed to schedule outgoing packets using the Packet-by-Packet Generalized Processor Sharing (PGPS) scheduling discipline. We propose an algorithm for call admission control (CAC), and present the simulation results used to evaluate the performance of the resource division policies used for mapping the end-to-end delay requirement into service rates to be reserved at each switch. The simulation results also show the performance gain when a simple resource-based routing algorithm is used. 1 Introduction One of the main promises of ATM networks is to provide users with Quality-of- Service (QoS) guarantees, such as Cell Transfer Delay (CTD) and Cell Loss Ratio (CLR). Handling the variety in QoS requirements of different applications requires the network to use a mechanism for serving packets from different applications according to their contracted QoS level. Many packet-scheduling disciplines have been proposed in the literature to implement such mechanisms (see [4], [8], [9]). Each scheduling discipline requires algorithms for performing call admission control (CAC) and resource reservation. In this paper, we propose such algorithms for the case of PGPS service discipline and calls requiring a hard (deterministic) bound on end-to-end delay. The paper addresses the following problems: 1. How to map the end-to-end delay requirement of a call into a local resource requirement to be reserved at each switch along the call’s path? 2. How to divide the resource requirement among the schedulers on the call’s path? That is, a simple even division policy would be to reserve the same amount of resources at all schedulers. However, it may be more efficient to use a policy that takes schedulers capacities and/or loading into account. 3. How much gain (if any) would be obtained from applying non-even resource division policies? What are the factors controlling the gain value? The following terms will be used throughout the paper: