Routing Issues in the Next Generation Internet by Barbara Denny Consulting Engineer denny@nsd.3com.com work: (408) 326-1883 fax: (408) 764-5002 and Cyndi Jung Protocol Architect cmj@nsd.3com.com work: (408) 764-5173 fax: (408) 764-5002 3Com Corporation 5400 Bayfront Plaza P.O. Box 58145 Santa Clara, CA. 95052-8145 The Next Generation Internet (NGI) will be a much larger, more diverse infrastructure that will interconnect a wide variety of devices and provide a wide variety of services. The services and types of devices will be driven by new applications and the network itself will foster new applications. The "interconnectedness" of network, machine, and user will continue to foster innovation and evolution of the network. These improvements must be designed and incorporated carefully so that the Internet remains a unified, consistent whole. The NGI should be: composed of networks whose speeds are orders of magnitude greater than today. Many of the new applications that this network will need to support require gigabit throughput. Distance-learning, health-care, environmental monitoring, and scientific research all require massive amounts of bandwidth to provide the greatest benefit. Optical networks are already reaching transmission speeds in the gigabit range and this will continue to increase. composed of wireless and wired networks seemlessly integrated. This will include the use of satellites for point and multipoint distribution as well as terrestrial wireless networks. The NGI will therefore have an even greater speed mismatch than has been experienced before. enabled for mobility. Mobility is in its infancy and should become commonplace as the technology and cost improves. A mobile, wireless device should have the same capabilities as a wired device in terms of services and applications it can support. pervasive. Global connectivity will continue to improve and the number, type, and location of devices that will communicate via the Internet will continue to grow. able to satisfy different quality-of-service parameters. These parameters will reflect reliability, timeliness, and accuracy. They will be selected not only on the basis of the application requirements and networking capabilities but on the ability of the end system to meet the demands. secure. The network should be able to support privacy, authentication, and confidentiality. This applies to the applications and the networking protocols to ensure their correct behavior and inspire a high-degree of confidence in the "trustworthiness" of the net. The current protocols and algorithms in the Internet today are insufficient for the NGI. In particular, research in routing will be needed to meet the demands the above network will create. It is routing that binds the Internet together and new techniques must be developed. Some of the routing issues which need to be addressed to achieve the promise of the NGI are discussed briefly below. The speed required for high-performance computing applications and the desire to provide fault-tolerance in terms of recovery time for link failure requires faster dissemination of update information and faster convergence of the route computation process when a link state changes. Furthermore, faults resulting in network instabilities are magnified at these speeds, raising the importance of containment techniques beyond those in the current state-of-the-art routing algorithms. The incorporation of satellite networks introduces a unidirectional link. This link could be used in the core of the network or at the edge. Routing protocols and algorithms, including multicasting, assume a bi-directional link. To effectively take advantage of this technology requires changes in the generation and distribution of update packets and in the route computation process. Mobility is treated as a special case within the current Internet. Tunnels are dynamically created to enable the routing of packets to the mobile devices. As the number of mobile devices increase, the use of tunnels becomes impractical. Furthermore, multicasting to a mobile host will not work in the current Internet because the join mechanism assumes that the receiving device is on a directly attached network. The preservation of the original IP address breaks this assumption as it is viewed by the router. Mobile devices need to follow the same processes as other devices in the Internet. This includes the ability to route to them without the use of tunnels. The sheer size of the Internet in terms of the number of networks and interconnections requires a new look at localization. This will require a stronger hierarchy which reduces the volume of information at and between levels without forcing traffic onto non-optimal routes. Current approaches have not achieved sufficient information hiding to allow the Internet to scale to a single network of immense proportions. Satisfying different quality-of-service requirements will increase the complexity of the algorithms and protocols. The current scalar metric routing algorithms are not sufficient for supporting such diverse traffic types. Also, routing all traffic over the same path may result in a link becoming effectively unuseable. Previous attempts to route around congested links have met with catastrophic failure, network oscillation, and eventual meltdown. The right of passage on these paths must also be controlled by policy, and in order to have the routing policy be consistent throughout the network, the policy must be distributed as is the reachability - it does no good to compute a path through a router that will not admit the traffic. No routing protocol today has this capability, and the centralized mechanisms used today are difficult to administer and enforce. Multicast can provide efficient parallel distribution of data. The use of this technology simplifies and enables the development of many key applications in the NGI. We have previously mentioned some issues that need to be resolved for unidirectional links and mobile hosts. Another issue is scalability. Far too much state is required by all of the existing protocols. They do not even scale to the size of the current Internet. The development of the NGI will enable the realization of many national goals and missions. However, in order to achieve all the benefits that this kind of Internet can provide, much research still needs to be done. Routing is only one area that we have addressed.