Summary of Issues and trends in router design by George Cragg

The paper attempts to describe what a router is, the three classes of routers, and the design decisions which go into the three classes. It finishes up with discusions on what the hot issues are, and where router design reserach is heading.

The components of a router include input and output ports, a switch fabric tying them together, and some type of processor to allow for the routing to take place. The three types of routers are backbone routers, which require reliability and speed, enterprise routers, which prefer high port density and inexpensiveness, and finally access routers, which need to be cost competitive yet provide significant functionality without port density. The functionality of all three would preferably include quality of service capabilities, but the access routers likely need to support much more: PPP, PPPoE, VPN (IPSec, PPTP, L2TP, etc.) Current trends include high speed route lookup, which is required since that's what routers do: route on a destination address. Also switching fabrics tend to follow an ATM type design requirement due to ATM's ability to rapidly switch small packets of data. The ports also have queues, both in and out. How the ports handle that queing and how the switch fabric fits into the product is important. Finally scheduling of output packets and reducing port costs are fundamental design issues as well. As far as cost is concerned, the trend of memory and prcoessor costs to drop in half every 18 months is a big help here. Other design issues include centralized management, ways to avoid router lookups altogether at the router and value-added features from router operating systems. Current open issues on the block inlude flow identification which is useful for resource reservation (which allow for QoS guarantees and value pricing) as well as Ease of configuration and stability.

4. Conclusion

This article would make a good first introduction to router design for people with some background in computer science or computer engineering. It gives a pretty good, albeit brief, rundown on what state of the art is in router design and what market segments value which features in a router. For me, understanding what the cost drivers are in what stage of the process was most rewarding: often times (actually nearly every time) design decisions are made with cost as the decision maker; without understanding the expensive parts of the process the design will be a mystery.