Real-time application performance in differentiated services network

Abstract

The Internet traditionally provides service that is commonly characterized as a best-effort service. Many applications run very well using this service model but some new interactive applications such as telephony or video conferencing impose stringent demands to the network. We discuss a proposal for changing the architecture of IP networks in order to provide some level of service discrimination. Architecture for differentiated services (DS) and proposed mechanisms have been analyzed and discussed to a great extent in the past years. The two most known approaches are schemes for throughput sensitive traffic and schemes for delay sensitive traffic. Differentiated services are attractive because they maintain connectionless data transfer in IP networks without the need for per-flow state information in network elements and because they use simple signaling. We first outline both principles for service discrimination and then we use simulations to evaluate them. A scheme for the delay sensitive traffic is realized through strict priority queuing in network elements, while a scheme for throughput sensitive traffic is realized through pre-emptive buffer processing policy. With simulations we evaluate behavior of a differentiated network architecture for two widely used (WWW, FTP) and two emerging Internet applications (video conferencing and IP telephony) and compare them with the best-effort network. The results show that proposed schemes have great positive impact on real-time applications that are assigned "high priority". Queuing delays are reduced to great extent and packet delivery without losses in queues is possible. On the other side, the performance of elastic applications is not highly degraded.