OSPF ABR cross connect design

[Tyche]

Hello,

Is it a good design practice to have ABRs connected to each other using a physical link in area 0 ?

Without this link can we expect some surprises when doing summarization etc ... ?

Kind Regards

Tyche

[rais]

There are two ABRs, so the connectivity is already redundant without the red cable.

If it's easy to install the red cable, it doesn't hurt.

[balaji.bandi]

You already have mesh network other side, why do you want to add another link ?

You already have dual router connected to uplink even one fails other route path will be in place.

[Tyche]

A few years back I was involved in troubleshooting an OSPF problem. The issue was due to the absence of interconnection between the ABRs. I can't remember if it was the interco in Area 256 or in area 0, in any case there was an issue because one link was missing in the design.  I am trying to remember and understand what the rule was so I can avoid it in a future design ...

RFC 2328 Section 3.7 speaks about partitioned areas (backbone and non-backbone). Problem occur when there is aggregation on both ABRs and partitioning occurs ....

I will have to lab it and see how I can recreate the issue.

Thanks for your input.

[Joseph W. Doherty]

In your diagram, depending on how your area 256 is configured (e.g. stub), you might want your red link, as your diagram only shows 1 link connecting to area zero.

Regarding when using summarization, for area 256, if you only had one area 256 link, yes you also want a redundant path into the area (much like above mention for area zero).  The red link would not help with that.

BTW, years ago, we had two "WAN" ABRs that only had single links to different WAN side OSPF areas and also only had one LAN side area link (also doing area summarization on them).  As these two routers were side by side, addressed both area zero and non-zero areas by having one link between those ABRs, setup as a trunk, with subinterfaces for all the areas on both routers.

[Tyche]

Thank you for the the tip to create added redundancy using subinterfaces ...

Tyche

[David Ruess]

Hello,

Adding the red link in Area 0 will prevent partitioned Area for Area 0.

I agree in that it looks like adding that will add redundancy but you already have that. As long as you understand all the LSAs involved and how the network is converged with and without the link then you’ll be fine.

That being said adding that link may provide further optimal routing. 

let us know when you lab it up.

-David

[Tyche · Accepted solution]

Hello,

Here are the results of my lab.

Let's say I want all traffic from area 256 to reach destination D in area 0 via ABR1.

When an access router in area 256 looses its link to ABR1 (the red cross in the diagram),  without the red link in area 0, traffic is not re-routed from ABR2 to ABR1 because intra area routes are preferred over an inter area route regardless of the cost.

We can conclude that a link between ABRs is needed in area 0 to have a deterministic exit out of a non-zero area (in addition, in this setup, to providing missing redundancy in area 0 to the ABRs).

Thanks all for your input 

Tyche

[Joseph W. Doherty]

"We can conclude that a link between ABRs is needed in area 0 to have a deterministic exit out of a non-zero area (in addition, in this setup, to providing missing redundancy in area 0 to the ABRs)."

Actually, with the information provided, we cannot conclude that.

For ABR2 to go "sideways" to ABR1, that additional hop, and the path from ABR1 to D, has to have an overall lessor path cost than going directly from ABR2 to D.

With possibly needed manipulated link costing, you can achieve ABR2 as always seeing transiting ABR1 as the better path.

Generally, though, I would not consider using the RED link just for that purpose.  Again, as I and others have described, depending on usage of certain OSPF features, without a redundant path, an ABR can easily black hole network traffic, even though another possible path exists.

Further, the cross link between ABRs, might have better bandwidth and/or less latency vs. other paths bouncing off OSPF routers deeper in the area.  For example, if you don't have the area 256 ABRs cross link, and the link fails (as you crossed out on the diagram), traffic arriving on ABR1 would need to bounce of the drawing's left bottom router, back to ABR2 then to the drawing's left top router.

[MHM Cisco World]

As I know this suboptimal path is happened if you not interconnect the both edge router.

[Richard Burts]

In OSPF area partition is an interesting problem. For non backbone areas partition is generally considered not so serious. But if summarization is being done there could be issues. Partition of the backbone area is a bigger problem. The proposed red link would be a good way to prevent partition of the backbone.

[Tyche]

I agree:  adding the red link prevents partitioning of area 0, if the link between the core routers in area 0 fails. 
Conversely, if the core routers had proper link redundancy between each other, the red link would not be needed to prevent partitioning of area 0, however it would be needed so that the ABR still has a link to area 0.

Appreciate your feedback

Tyche

[MHM Cisco World]

but your original post talk about summary and this red link, so I answer according to that.
and for partition of area0, @Richard Burts  totally right.

[Tyche]

@MHM Cisco World ,  I appreciate your answer.
Thank you for taking the time to reply.