Experiments with QoS Driven Learning Packet Networks

With the growth of the Internet, in numbers of users, IP (Internet Protocol) addresses, and routers, and the increasing diversity of the types of links that are being used (wired, optical or wireless), the IP based network architecture needs to evolve. Reliability, security, scalability and QoS (Quality-of-Service) have become key issues. This paper discusses a research project which revisits IP routing and suggests a novel packet network architecture called a “Cognitive Packet Network (CPN)” , in which intelligent control for routing and flow control are exercised under greater control of the user connections with the help of “smart packets”, rather than being completely left to nodes and routing tables. CPN provides a reliable packet network infrastructure which incorporates packet loss and delay directly into user QoS criteria, and uses these and other criteria to conduct routing. We present the QoS based routing algorithm that we have designed and implemented, and report extensive measurements related to network performance, reliability and QoS. Our experiments also show how CPN can address the QoS needs of packetized voice. Other experiments show how CPN can be used to route IP packets which travel to and from a web server and tunnel through a CPN sub-network so as to obtain improved QoS.