Impact of ATM cell delay on multimedia applications q H.-K. Tan a , R. Radhakrishna Pillai b, * , C.-K. Tham a , L. Wong a , J. Biswas b a Department of Electrical Engineering, National University of Singapore, Singapore, Singapore 119260 b Kent Ridge Digital Labs, 21 Heng Mui Keng Terrace, Singapore, Singapore 119613 Received 23 December 1998; received in revised form 4 January 2000; accepted 5 January 2000 Abstract The impact of ATM cell delays on multimedia application performance is studied through measurements using a broadband network analyser. A method is developed to measure the ATM cell delays of an application using the analyser. It is found that the cell delay distribution becomes continuous as the traf®c load on a connection becomes lighter. The effect of background traf®c on cell delay for a reference connection is studied. Though the cell delay of the reference connection is unaffected by the presence of background traf®c, the cell-delay auto-correlation is found to decrease. A gamma distribution is used to approximate the cell delay distribution. The measurement results closely match the delay prediction using the approximation. In a network with multiple ATM switches, the gamma distribution only slightly overestimates the delay. The frame level statistics and the cell delay performance of a video conferencing application based on motion JPEG are measured. Using the gamma cell delay distribution, the frame loss of an MPEG-I video source is predicted and compared against the measured values based on MPEG-I traces. q 2000 Elsevier Science B.V. All rights reserved. Keywords: Traf®c modelling; Asynchronous Transfer Mode; QoS; Multimedia applications 1. Introduction Asynchronous Transfer Mode (ATM) is widely used as the transfer mode for Broadband Integrated Services Digital Networks (B-ISDN) networks. Multimedia applications are expected to be the killer applications for ATM networks. The application data is segmented into ®xed size cells of 53 bytes and transferred across the network. It is important to study the cell level performance of these applications and its impact on the application level performance. The cell level performance measures that affect a multi- media application performance include cell errors, cell losses and cell delay. Only cell delays are considered in this paper. Real time interactive multimedia applications like video conferencing are very sensitive to cell delays. End-to-end packet delay consists of segmentation and reas- sembly delay and cell transfer delay. The cell transfer delay includes propagation delay, switching delay and queuing delay within the network. The propagation delay is depen- dent on the distance whereas the switching and queueing delays are dependent on the switch architecture, switch scheduling mechanism, and the network traf®c. To investigate cell transfer delay, referred to simply as cell delay in this paper, a Hewlett Packard Broadband Series Test System (BSTS) [6] was employed. Traf®c was gener- ated by using the tester as well as by using a software traf®c generator and the cell delay is measured using the tester. The background traf®c load, number of background sources and the number of network hops are varied and the effect on cell delays on a reference connection is studied. The cell level parameters considered are the cell inter-arrival time, cell transfer delay, cell delay variation (CDV) and auto- correlation. A gamma distribution is used to approximate the cell delay distribution. The frame and cell level statistics of a video conferencing application based on motion JPEG are measured. The effect of cell delay on MPEG-1 traf®c is analysed and a method is devised to predict frame losses, assuming a gamma distribution for cell delay. Section 2 de®nes the cell level parameters considered for the study. Section 3 describes the cell delay measurement technique and Section 4 shows the cell delay measurement results using both the tester and software as traf®c genera- tors. Section 5 presents the frame and cell level measure- ments of a video conferencing application. Section 6 analyses the effect of cell delay on MPEG-1 video frame loss and conclusions are drawn in Section 7. Computer Communications 23 (2000) 1215±1222 0140-3664/00/$ - see front matter q 2000 Elsevier Science B.V. All rights reserved. PII: S0140-3664(00)00199-7 www.elsevier.com/locate/comcom q A version of this paper has been presented at the Applied Telecommu- nication Symposium (Part of 1999 Advanced Simulation Technologies Conference, ASTC'99), San Diego, 11±15 April 1999. * Corresponding author. E-mail address: pillai@krdl.org.sg (R. Radhakrishna Pillai).