Supporting Real-Time Applications in an Integrated Services Packet Network: Architecture and Mechanism

This paper considers real-time applications in an Integrated Services Packet Network (ISPN) and proposes a modified ISPN architecture. The authors observe that real-time applications may not always require a pre-computed worst-case delay bound provided by a guaranteed service. Instead, a predicted service that tracks network performance and uses it as a guide in computing dynamic delay bounds. The introduction of this predicted service would provide not only superior application performance due to the use of adaptive playback, but also higher link utilizations and thereby greater apportioning of service costs. The authors provide a unifying mechanism for real-time as well as datagram traffic - the concepts of isolation and sharing are central to this mechanism. Finally, there is an emphasis on the importance of pricing to encourage use of suboptimal service. This will avoid having to enforce sharing by refusing a large fraction of requests for real-time traffic. Price discrimination is also used to explain the potential value of the predicted service.

Given that I worked with real-time streaming and multimedia over the summer, I definitely found myself quite engaged in this paper. The innovative concept of predicted service also intrigued me after having worked with fixed-delay packet-dropping mechanisms in video streaming. This paper definitely builds upon the previous one by giving an explicit case of providing various classes of service to suit different application requirements. The discussion of price discrimination felt a little incomplete as a few assumptions were made about lower prices leading to requests for lower-quality service.