From: Howard C. Berkowitz (hcb@gettcomm.com)
Date: Sat Apr 15 2006 - 05:28:58 GMT-3
May I understand a little better why there is the original
requirement to break OSPF rules? I understand that it may be in a
practice lab -- I wish I knew if it was a real question, because I'd
be inclined to find whoever wrote it and beat them with a clue stick.
If anyone is stating this as a requirement to do _in OSPF_, even if
some implementations let you do it, it is a Really Bad Idea. It
breaks all sorts of internal architectural assumptions of OSPF.
Now, I have, on many occasions in the real world, make a certain
route more preferable than an OSPF-learned one. The most common case
is when traffic engineering has me run a direct medium, or at least
tunnel, between two destinations in different nonzero areas. I do
that with static route, NOT redistributed into OSPF, on each router
on the path. I set their administrative distance to a value below
that of OSPF. Do note that OSPF, except for the cross-area link, can
provide the next hops to the static routes.
Yes, there are some experimental proposals in the IETF to allow the
core to be bypassed if the two nonzero areas are connected to the
same ABR. Note, however, that is a special case, as controlled by
configuration as my static route example.
I can still allow an OSPF-learned backup path to form, but it will go
through area 0.0.0.0. This (interarea) path will be taken only if
the traffic engineered path is down.
For that matter, I can provide a second set of ip route statements
defining a third-preference backup to OSPF. This is effectively your
everyday floating static route.
If, in the real world, you have a situation where an inter-area route
is more desirable to you, as opposed to preferable to OSPF, than an
intra-area route, I suggest that your area topologies/definitions
need to be rethought.
Does this example of needlessly avoiding static routes because the
lab doesn't like them give an example of why large network operators,
especially ISPs, often feel as if they have to retrain CCIEs in the
right way to do things?
This archive was generated by hypermail 2.1.4 : Mon May 01 2006 - 11:41:57 GMT-3