[Help] Last update at http://inet.nttam.com : Mon Aug 7 21:39:57 1995

Abstract -- Multimedia Experiments at the University of Pisa: From Videoconference to Random Fractals Network and Application Engineering Track
N5: High Speed Networking

[Previous] [Table [Next]
[Paper [Paper


Multimedia Experiments at the University of Pisa: From Videoconference to Random Fractals

Giordano, Stefano ( giordano@iet.unipi.it)
Russo, Franco ( russo@iet.unipi.it)
Pierazzini, Giuseppe ( peppe@pisa.infn.it)

Abstract

Real time services in support of multimedia applications are emerging over networks originally designed for asynchronous services. The traffic generated by these applications represents a very critical component of the whole amount of traffic to be supported by future broadband telecommunication networks. Broadband experiments were carried out in Tuscany over a network infrastructure represented by a 140 Mbit/s DQDB MAN. The architecture of the MAN is organized into three switching levels: (level 1) the Customer Access Network (CAN), (level 2) the MAN Switching System MSS and (level 3) the I-MSS (Inter MAN Switching System). A level zero distributed switching environment is represented by the LANs connected by multiprotocol routers to the MAN. The main application areas, based on the connection less data service provided by the MAN, are the following: The paper deals with the experiments carried out in the field of desktop conferencing in a regional environment using the MBONE's tools . Particular interest is directed to the multicast multimedia IP traffic offered to the MAN. To support these time constrained applications, new QoS (Quality of Service) requirements have to be consider and their provisioning is critical due to the bursty nature of the sources. High variability of the traffic is at the same time the pros and the cons of modern traffic control strategies. Statistical multiplexing plays a central role in trying to obtain a better utilization of resources but, on the contrary, its control is extremely critical and could drive to conservative solutions. These problems are complicated by the need to maintain a complete integration of new services and trasport technologies with the oldest ones. To provide those services on a "best effort" access class it is necessary to make performance provisioning based on the statistical behaviours of traffic sources.

The paper presents measurements of Ethernet traffic offered to the MAN collected during its regular use. The traffic is represented by two main components: traditional data traffic (i.e. file transfer, remote login, access to WWW servers, etc.), and multimedia real time traffic generated by voice, video, shared whiteboard.

The two main traffic components revealed to be highly variable and bursty. Variability in packet arrivals has direct consequences in queueing delays: greater variability correspond to longer delay. The interest in modeling bursty traffic offered to a broadband network is central in problems related to statistical multiplexing. Our measurements show that aggregated traffic offered to the broadband network is well-described by self-similar processes instead of traditional Renewal or Markov Modulated Poisson Process (short range dependence processes).

These results, obtained over a broadband network, confirmed previous analysis carried out at Bellcore. Increasing the number of sources multiplexed over an Ethernet LAN (thus increasing the average offered load), we obtained a self-similar process with an increasing Hurst parameter, which characterizes the correlation function of the self-similar arrival process. The interest of this analysis is directed to the performance evaluation of a broadband network which provides a best effort, asynchronous interconnection of several remote LANs. Different effects of this arrival processes on network performances are presented considering a model of the DQDB network. Our work is directed to point out the need for more realistic traffic scenarios in the analysis of broadband telecommunication networks. We considered the analysis of the MAC protocol standardized by the IEEE 802.6 supporting multimedia voice and video applications. The rational behind this choice is that real traffic sources correspond to the multiplexing of several services over the network and not to a single isolated video or voice source. This will be particularly true for LAN interconnection that will represent the first service to be provided by broadband networks in a business and educational environment. The implications of the self similar or "fractal" behaviour of real traffics must be taken into account in dimensioning future telecommunication networks.