Submitted by:

Gwendolyn L. Huntoon, Manager of Communications Jamshid Mahdavi, Network Research Coordinator Matt Mathis, Sr. Networking Specialist

Pittsburgh Supercomputing Center
4400 Fifth Ave
Pittsburgh, PA, 15213
(412)268-4960
(412)268-8200 (fax)


Next Generation Routing

Gwendolyn L. Huntoon
Jamshid Mahdavi
Matt Mathis

The Next Generation Internet, while building upon and leveraging existing network technologies, infrastructure and applications, must also foster the development of new paradigms in networking. Many network components in support of high bandwidth networks and high performance applications are well into the research phase, soon to be deployed and tested in real networking environments. However, it is important for the networking community to stop and re-evaluate current networking components in respect to the new demands and requirements assocaited with NGI.

One such component, critical to the core infrastructure of the Internet, is network routing. Since the inception of the Internet, routing protocols have been developed, standarized, and modified to fit the next evolution of the Internet. Today's Internet technology only provides good support for routing based on destination IP address. Similarly, common routing protocols rely on IP address information, such as network prefix, AS number, etc to build routing tables and make routing decisions. In the Next Gernetation Internet, this information alone will no longer suffice. Thus, it is time to ask the question "Is the current Internet routing correct ?". 

Routing protocols, in order to accomodate new types of network architectures, applications and services associated with NGI, will need to be based on more than destination based routing. They will need to be capable of collecting, distributing and utilizing a wider range of information and of making decisions based not just technical information, but also policy based information as set forth by individual users and/or applications as well as policies of specific providers and networks.

Past research in policy based routing has all too often been focused on the general problems of fully flexible policies. This work has been done within the current paradigm for Internet routing, and the results have often been completely unimplementable in the production Internet. New, revolutionary thinking on the subject of policy routing is needed in order to support the demands of the NGI. 

The current Internet makes no distinction between premium, regular, and bulk traffic. People who are willing to pay for premium services often can not because all traffic to a given destination is aggregated (or blended) regardless of the source or character of the traffic or the service needed. This traffic pooling means that all internet users, including R&E and premium commercial services are forced to share the network which is clogged up with millions of low speed users. The advent of QoS capabilities in networks of the future will provide the capability to segregate traffic. Currently, however, no good solutions exist for solving potential problems associated with routing for segregated flows.

We have already seen the limitations associated with existing routing protocols. The list of networks has grown tremendously over the years, increasing the amount of memory (for storing routing tables) and processing speed (for performance) needed to effectively route in a high performance networking environment. We have already seen restrictions by providers on what networks they will carry and announce (aggregation requirements to reduce the cost of routing) along with serious performance degrations (from routers unable to make effecient routing decisions). We have seen that switched networks are not the answer (such as ATM), particularly since they don't scale well. Current routing technologies are strained to the breaking point; simply expanding existing technologies will not be adequate for the complex routing scenarios posed by the NGI. 

The gigopop architecture currently suggested as the core to NGI, suggests a radical departure from the current Internet topology. Aggregation points will have multiple MAN and WAN network connections, often with specific AUPs or requirements associated with each link. Applications are emerging with very specific service requirements. Routing in the NGI must be capable of taking specific service requirements into consideration in order to meet the needs of these applications. Other issues which need to be considered include: link performance, type of application, billing information, and site security information.