I have received results from a TAC case about this familiar issue. For
those coming late to the picnic, the issue was that the presence of OSPF
or IS-IS overrides 'no ip classless' in the router configuration and
would force the router to behave classlessly. The TAC engineer
consulted with some Development Engineers and here is what they said:
Yes, we do this if the default (or the supernet) route
is supplied by OSPF or ISIS (I guess EIGRP should be there,
but it is not) and the part of the code that requested
the RT lookup didn't not specify to ignore the default
route if there's no specific subnet (which is the case
for the locally originated and transit packets).
The assumption is that it is safe to use a default/supernet
route installed by a classless protocol.
So what ever you and me have seen in our testing is correct behaviour.
So, our original guesses were correct. The router assumes that if
we're running OSPF or IS-IS then we want classless routing even if we
didn't specify 'ip classless' in the config. However, an important
point is that this applies only if the supernet was installed by the
classless protocol. If OSPF or IS-IS is running on a router but the
supernet was installed by another process, then classful routing would
still apply without the addition of 'ip classless' to the config.
I've also discovered that if you add 'ip clueless' to the config, the
following occurs:
First, ip classless is overridden but only in cases where no ip
classless was manually configured previously and was not the default
setting, unless the router has not had any previous configuration and is
running at least 12.1(5F)T12. This does not apply for any 12.0 images
except 12.0(6)S but does apply to any 11.2 image after 11.2(26c)P;
Second, the gateway of last resort might be chosen by the RT lookup
process if the GOLR was set by a classless routing protocol with an
administrative distance lower than that of any other classless or
classful routing protocol on the router, except in the case of BGP or
EGP in which case the administrative distance must be at least equal to
that of the routing protocol which previously installed the GOLR, if
already present;
Thirdly, if the lowest-weighted routing protocol is OSPF and the GOLR
is advertised to neighbor as an E2 route, then the neighbor router may
choose to use that route unless another neighbor has advertised the same
supernet route as an E1 route. In which case--but especially when
utilizing IP over Avian Carriers (with QoS)--the RT lookup will choose
the Type 1 External route unless EIGRP is running on this router as
well. In that case, the GOLR will be set via EIGRP because Cisco
prefers EIGRP to OSPF and we should all use that anyway because, don't
ya know, OSPF is harder to configure and requires way too much thought
to begin with. IS-IS is just out of the question. However, if a router
learns a supernet route via OSPF and IS-IS *and* EIGRP then you will be
severely punished. Flogging is generally suggested. As an alternative,
only run EIGRP and leave "ip clueless" configured.
Any other configuration will provide ambiguous results.
HTH,
John
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=1758&t=1758
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
