[c-nsp] ospf cost

[Jon Lewis]

I've run into an odd situation with a few routers that both talk to
multiple ethernet networks, all doing OSPF.  The routers have GigE to one
network, FE to another, and then a PTP GigE connecting them.  For talking
to each other, these two routers seem to be treating all 3 paths as equal
even though they're obviously not.  The whole point behind the PTP GigE
connection was to give each a more efficient/less congested path to the
other.

I think what I've run into is a limitation in the default ospf cost
calculation which causes any interface with BW >= 100mbit to end
up with a cost of 1.  I see I can raise that with something like:

router ospf 1
 auto-cost reference-bandwidth 10000

but that causes a warning telling me I better make sure all my routers
have the same reference-bandwidth setting.  I assume that's so other
routers don't end up with whacked out SP trees based on wildly different
interface costs across the network.

If anyone's already changed this, was it the sort of thing you rolled out
across the whole network as quickly as possible during a late night
maintenance window?

Is it better to just manually alter a few carefully chosen interface
costs?

----------------------------------------------------------------------
 Jon Lewis                   |  I route
 Senior Network Engineer     |  therefore you are
 Atlantic Net                |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________