PERFORMANCE CONTROL OF HIGH-CAPACITY IP NETWORKS FOR COLLABORATIVE VIRTUAL ENVIRONMENTS L. A. Rønningen, A. Lie Norwegian University of Science and Technology (NTNU), Norway ABSTRACT The support for low latency high throughput IP networks for multimedia streaming is today very limited. This paper shows that in high-capacity Gbps IP networks, RTP/UDP/IP packets of 1500 bytes/packet will achieve extremely low queuing latency (<10ms) and still utilise the network capacity almost 100%. A router must implement a low-complexity algorithm that monitors the queue length and the rate of change of queue length. By sending this information to neighbouring routers and hosts, the traffic can be dynamically controlled to achieve the desired balanced throughput and latency. The only traffic needing QoS controlled channels are these low-rate control messages. The video streams are assumed to include error resilient coding in order to cope with packet losses up to 15% of the stream, alternatively to scale down the output data rate with the same factor. The low latency of this network make it a very promising network candidate for Collaborative Virtual Environments such as Distributed Multimedia Plays, where distributed musicians and actors can practice and perform live concerts and theatre, provided total latency do not exceed 10ms. INTRODUCTION Basic IP packet switching networks support the “best effort” philosophy, and do not give any Quality of Service (QoS) guarantee. However, extended with schemes like IntServ or DiffServ (see below) the QoS can be controlled. A network providing QoS guarantee is ATM. ATM uses cell (packet) switching, and virtual circuits. MPLS in turn, uses the label switching principle, guarantees QoS by offering traffic classes and resource reservation (RSVP-TE or LDP), and is more flexible than ATM when interconnecting other networks through the multi- protocol encapsulation principle. Quality Of Service Control In IP Networks Distributed Multimedia Plays (DMP) by Rønningen (1), i.e. Collaborative Virtual Environments for musicians, in some cases require delays as low as 10 ms for audio and 20 ms for video. During a collaborative session (see Figure 1), the requirements may vary, and an adaptive QoS scheme is needed. We will in this paper put focus on controlling time delays, and at the same time obtain high resource utilisation. Coding schemes shall be adaptive and designed so that artefacts from packet drop/loss are minimised. The spatial and temporal resolution as well as the service profile shall be scalable dynamically, based on traffic measurements in the network.