From: Howard C. Berkowitz (hcb@gettcomm.com)
Date: Fri Jun 13 2003 - 10:27:55 GMT-3
At 8:36 AM +0100 6/12/03, McCallum, Robert wrote:
>Howard you could have a scenario like this quite often in fact when
>you are connecting customer routers via a vrf within an MPLS
>environment. I agree that it is a design issue but hey you know
>what customers are like ;-> I have seen this on a few occasions
>when I place a customer into a vpn with
>OSPF between the PE and the CE. The customer seems to forget that
>the way they used to connect their routers together has now
>completely changed. But it still works. The other beauty when
>doing ospf is that the customer then decides at a later date to
>interconnect two of his sites together. Then
>moans the face of us when the traffic still goes via the MPLS
>core.....enter SHAM link.
Thanks for the thoughtful response. To me, a scenario such as you
describe is a much better learning experience than simply asking the
question, in isolation, "how do you connect two discontiguous areas
of the same number?"
In practice, I'll use the teaching scenario where an area is laid out
with half the address space on the left and half on the right, with a
single link between the ABRs, and let students fall into the trap of
sending only one summary route (covering the area) from both ABRs.
Of course, when the area partitions, you have the chance of each of
the ABRs blackholing traffic for half of the addresses sent to it.
The lesson to be learned is not to summarize quite as far as is
possible, but to do it consistently with the addressing plan.
As an aside, Cisco and Bay handle the situation of ABRs advertising
quite differently, and both methods are valuable for enforcing
certain policies. I wish Cisco would support both. Cisco, of course,
will always advertise the summary. Bay/Nortel RS will, if some of
the more-specifics of a summary become unreachable, will stop
announcing the summary and only advertise the more-specifics.
I will admit that people may not have enough routers to set up a
realistic situation with discontiguous areas and alternate paths. But
my advice is to put the problem in the slightly broader context of
where it would solve a problem, rather than just asking the minimal
configuration problem. People, I believe, will learn better when
they see a reason for doing it.
When posters discuss having multiple areas with the same number, and
don't bring up partitioning as the problem they are trying to solve,
I shake my head in wonder. Yes, of course -- there are reasons that
partitions will take place, and to have recovery mechanisms for them.
There are all sorts of techniques. Let's say you have a destination
address in your area that is normally reached intra-area, and the
partition causes the intra-area link to go down. If you also had a
static route to that destination, run through the backbone,
redistributed into OSPF, it normally would not be used due to the
rule of preferring intra-area over inter-area over external 1 over
external 2. But the external (type irrelevant here) would activate
if the areas became partitioned.
>
>> -----Original Message-----
>> From: Howard C. Berkowitz [mailto:hcb@gettcomm.com]
>> Sent: 12 June 2003 03:59
>> To: ccielab@groupstudy.com
>> Subject: Re: OSPF partitioning issue
>>
>>
>> At 8:42 AM +0800 6/12/03, huang gang wrote:
>> >hi,
>> > ospf don't require that all areas except area 0 should
>> be continuous.
>> >huangg
>>
>> Only in a very, very limited sense, that there's nothing violated by
>> having more than one area with the same ID. Another limited sense is
>> that a nonzero area can be discontiguous with area 0.0.0.0 if it uses
>> a virtual link.
>>
>> But WHY would anyone deliberately want to have multiple area x's
>> (other than as a failure mode as a result of a partition)? There
>> have been a lot of questions about doing this, and I am utterly
>> mystified why anyone would WANT to do this deliberately. If nothing
>> else, it's a nightmare for documentation. If you moved a router from
>> one area 0.0.0.2 to another 0.0.0.2, do you really want the same
> > network statements?
>>
>> Even if one is exploring the effect of nonzero area partitioning, set
>> it up as an single area with two ABRs, and perhaps one critical
>> internal link. By taking that link up and down, you'll get better
>> understanding of what happens when an area is partitioned.
>>
>> The idea of having multiple areas with the same number comes up often
>> enough on the list that I wonder if some practice lab somewhere is
>> using it as a horrible example. I'd never permit it in any
>> operational network for which I had responsibility.
>>
>> I've heard of many strange configurations rumored to be on the CCIE
>> lab, but this would be beyond my worst imagination if Cisco actually
>> asked you to do it.
>>
>> >
>> >> Hi all,
>> >>
>> >> In the following scenario-
>> >>
>> >>
>> >> Area0---------Area2
>> >> \ |
>> >> \ |
>> >> \ |
>> >> \ |
>> >> Area2
>> >>
>> >> If the link between Area2 fails, it becomes discontiguos,
>> this makes the
>> >> intra area routes to be shown as inter area routes in
>> both the area 2
>> >> routers, I tested this in the lab and found no visible
>> reachability issues.
>> >> Does this disconinuity create any hidden problems?
>> >>
>> >> The other question is, that if we need to repair this,
>> can i use a virtual
>> >> link between Aree2 to Area 0 and then to Area 2
>> >> or should I use a tunnel interface on each router putting
>> them into area 2??
>> >> I actually tried it using tunnel but was not able to make
>> it work , as I was
>> >> still seeing some of the Area 2 routes as O IA routes.
>> >>
>> >> Thanks as always for your inputs.
>> >>
>> >> Smiles,
>> >>
>> >> Mohit.
This archive was generated by hypermail 2.1.4 : Fri Jul 04 2003 - 11:10:58 GMT-3