DESIGN: Cost increments from Intra-device edges

[Raajay Viswanathan]

Currently, to factor administrative distance (local preference) we order the intra device edges and assign them cost in increments of the network diameter. However, that might not be right, because there might be more than one intra-device path in a end-to-end path.

What should be the increments when there are multiple such edges in the path?

[Raajay Viswanathan]

An example highlighting this would be nice.

[Aaron Gember-Jacobson]

If the instance graph is a DAG, we scale down costs such that the cost of the downstream path will not impact the selection of a path within the current routing instance.