Check out the new USENIX Web site. next up previous
Next: External routing Up: IP space Planning Previous: IP space Planning

Internal routing

Having the IP space mapped to the physical map, a number of areas are created that contain different subnetworks. Between these areas there are different network links. This approach creates multiple routes from a source to a destination. It is no longer possible to manually manage the routing tables in all nodes that have more than one interconnect. Using the Open Shortest Path First (OSPF) routing protocol minimizes the routing configuration of a node. As soon as the appropriate interlinks are made (or broken) on the IP level, the OSPF processes start to exchange information about their knowledge of the network topology. This also allows us to add or remove nodes from the network without any routing reconfiguration.

Using the hop count and cost factors on various routes, the system will select the route with the lowest total cost. If any link fails, it is detected by the OSPF daemons and all routing tables in the network will be updated to reflect the new situation. Currently, routing is done solely on hop count. All cost factors are the same.

By using OSPF routing and different interconnects on one node, the reliability of the network on the IP level is greatly enhanced at a negligible cost of processor overhead and network bandwidth.

For the implementation of the OSPF protocol the Zebra [Zebra,Routing] package is used. Zebra is an Open Source package which provides a number of different routing protocols by various daemons, all controlled by a master daemon. Currently OSPFv2 routing is used throughout the backbone network because the protocol is fit to the topology and does not impose a large overhead.

next up previous
Next: External routing Up: IP space Planning Previous: IP space Planning
Rudi van Drunen 2003-04-08