From: ssaponas (ssaponas@u.washington.edu)
Date: Wed Oct 27 2004 - 07:52:25 PDT
Review by T. Scott Saponas
"Supporting Real-Time Applications in an Integrated Services Packet Network: Architecture and Mechanisms" presents an Integrated Services Packet Network (ISPN) that supports real-time applications through guaranteed service and predicted service. It is further shown how such a system could be built using a new packet scheduling mechanism that supports these real-time services and datagram traffic. Their system exploits the property of real-time applications that while an absolute statistical bound is needed is needed on delay, as long as data arrives before the play-back point all is well. What this translates to in queuing is that isolating real-times streams from one another like in token bucket approaches might not be good because it won't always allow burst through for one flow at the expense of others. The reason this is desired is that the property above allows us to make a temporary small delay in all other flows because they will still get to their endpoints in time and then the flow with the burst also gets to the end in time. So, this suggests some more FIFO looking approaches could actually satisfy the needs of all flows instead of all but one flow as long as there aren't malicious users. (So a little something extra can be put in to make sure someone isn't continually bursty)
The ISDN idea is clearly a goal worth shooting for and this paper presents much of what is needed for a good ISDN. However, as mention by the authors, there are political and administrative challenges we must overcome before any ISDN design is really useful. At the end of the paper the authors discuss briefly the need for pricing, but I believe pricing and identifying exactly how market competition can play a role in the ISDN design and in getting an ISDN design implemented and running is important. I would like to see the authors further show how such a system could become implemented and used into today's Internet - particularly in the form of what will incentives ISPs to provide the service since there currently isn't market competition for your second or third hop ISP.
The problem of how to provide good service for real-time applications in the same networks that support other datagram activity such as the web is becoming even more relevant as more multimedia content is available for consumers over the Internet. As the paper points out, real-time application demand is also being by what appears to be an eventual merger of much of the telephone networks and our data networks. Today's Internet suffers from the problem that there are no facilities for pre-scheduling resources (or they are not used for various reasons).
This archive was generated by hypermail 2.1.6 : Wed Oct 27 2004 - 07:52:37 PDT