OSPF inter-area traffic

[itech]

Hi all,

Been thinking of using OSPF ..but its seems to involve quite some pre-planning in how the protocol will fit into my current network topology....

a) Can i confirm that inter-area traffic will be routed towards area 0 / area0 will be the transit network for all inter-area traffic ?

b) Consider the diagram below

q1) Does R1 send Type3 LSA for Area2 via both Area 1 and Area 0 to R2 ?

Area2 wants to send traffic to Area3

q2) Going by the rule above that all inter-area traffic must transit to Area0, isn't it totally not efficient for traffc from Area 2 <-> Area 3 to pass through Area 0 via a 100Mb link ? - especially when Area2 can reach Area3 through Area1 on 1Gbps link ?

Sorry if i am not getting the full picture..

Regards,

Alan

[Harold Ritter]

a) Yes, all traffic between two areas need to transit via the backbone area.

q1) type 3 LSAs are only exchanged via the backbone area. L3 LSAs received via any other areas will not be propagated beyond this area.

q2) yes, traffic needs to transit via the backbone area. It is not inefficient if you put the link between R1 and R2 in area 0.

Regards,

[Richard Burts]

Many of us are used to the flexibility that we get using EIGRP which would make it easier to accommodate what the original poster is suggesting. But if we adopt OSPF as our routing protocol we much accept that it imposes certain architectural restrictions. 

It is not clear whether there was a particular reason for creating area 1 and trying to use it as a transit area. Perhaps it was intended to keep that as a dedicated link for communication between those two areas? In that case it would not work using OSPF. But perhaps an alternative might be to use OSPF as the routing protocol for the enterprise and to use Policy Based Routing to direct traffic between these two areas over the link (and to not have that link in OSPF at all)?

HTH

Rick

[itech]

Hi Richard and Harold,

Thank you so much for the advices given.

In view of the replies given,  can i confirm the following

a) Base on the original diagram given, traffic from Area2 <-> Area3 will not transit through Area1 (even if Area1 has a lower cost) and will transit through Area0

b) ABR will not propagate Type3 LSA into non-Area0 areas - hence R1 and R2 will not propagate Type3 LSA into Area1.

c) Lastly, how will inter-area traffic from Area1 <-> Area2 flow towards each other ?  Will it transit through Area0  ?

e.g. Area 2's traffic to Area1 will transit via R1 -> (area0) -> R2 -> then (area1)  ?

Regards,

Alan

[Harold Ritter]

Hi Alan,

a) Correct

b) R1 and R2 being ABRs, they will propagate their respective type 3 LSAs into area 1, but these type 3 LSA will not further be propagated (i.e. R1 will propagate type 3 LSA for area 2 into area 1, but R2 will not further propagate them to area 3). 

c) Since R1 is an ABR between area 1 and area 2, traffic will transit directly between the two areas through R1. 

Regards,

[itech]

Hi Harold,

Type 3 LSAs are only exchanged via the backbone area. L3 LSAs received via any other areas will not be propagated beyond this area.

R1 and R2 being ABRs, they will propagate their respective type 3 LSAs into area 1, but these type 3 LSA will not further be propagated (i.e. R1 will propagate type 3 LSA for area 2 into area 1, but R2 will not further propagate them to area 3). 

---

Sorry i am still abit confuse about the 2 statements above... Could you explain further ?

---

c) Since R1 is an ABR between area 1 and area 2, traffic will transit directly between the two areas through R1. 

---

Does that means that the rule in which all inter-area communication must go through area0 does not apply if the souce and destination areas are connected to the same ABR ?

Regards,

Alan

[Harold Ritter]

> Sorry i am still a bit confuse about the 2 statements above... Could you explain further ?

- R1 receives link state information from area 2.

- As it is an ABR, it injects the type 3 LSAs for area 2 in area 0 and area 1. 

- R2 receives the type 3 LSAs from area 0 and propagates in area 3.

- R2 also receives the type 3 LSAs for area 2 via area 1, but does not propagate to area 3. AS an ABR, It can only further propagate type 3 LSAs received via area 0.

> Does that means that the rule in which all inter-area communication must go through area0 does not apply if the source and destination areas are connected to the same ABR ?

The rule still applies, as the ABR is part of area 0.

Regards,

[Joseph W. Doherty]

I believe Harold's response to question "c" is partially incorrect.

"c) Lastly, how will inter-area traffic from Area1 <-> Area2 flow towards each other ? Will it transit through Area0 ?

e.g. Area 2's traffic to Area1 will transit via R1 -> (area0) -> R2 -> then (area1) ?"

That depends.

Traffic from area 2, going to area 1, will hit R1, which as an ABR with a connection to area 1, will pass the traffic directly to area 1.

Traffic from area 1, going to area 2, might go to either R1 or R2. If the traffic from area 1 first hits R1, again since R1 is an ABR, it will be directly passed to area 2. However, if the traffic from area 1 first hits R2, an ABR without a connection to area 1, it will transit area 0 links. Since I'm not sure what the cost or the logical topology of all the area 0 links are, it's possible that traffic might transit R3 and/or R4 to get to R1.

[Harold Ritter]

Hi Joseph,

Thanks for the precision. The point I was trying to convey is that crossing an ABR is considered going through area 0.

Regards,

[Joseph W. Doherty]

"The point I was trying to convey is that crossing an ABR is considered going through area 0."

Agreed! However, it's important what non-zero areas the ABR has a foot in. Although R1 and R2, are both ABRs, and both have a connection to area 1, only R1 has a connection to area 2.

Another example, if you were to ping a host IP in area 2 using R2's interface connection to area 1 source IP, R2's outbound packet would flow across area 0, but the ping response/reply packet would flow across area 1; this asymmetric traffic flow regardless of link costing.

[dbeattie]

I always question the multi-area requirement. Even in large deployments, OSPF is efficient enough to support high numbers of routers in a single area. In this case, with the total number of routers being less than 50, if there is no other driver for the other areas, path selection would be best achieved by the use of a single area.

Hope this helps,

Dave

[Harold Ritter]

Hi Dave,

I tend to agree with your comment. Many networks have an over complicated area design and in many cases could just use a single area.

Regards,

[Joseph W. Doherty]

Whether a multi-area OSPF design is needed, is often a difficult question to answer.

How many routers you can "safely" have within an OSPF area depends on many factors. 50 was/is an old rule of thumb. Modern Cisco OSPF equipment can often very much exceed a count of 50 devices within an OSPF area, but that also has much to due with modern equipment's continued improvement in their CPU performance, area topology, how devices are configured, and how OSPF is actually implemented on the OSPF equipment. (The latter can have a huge impact on OSPF operations. Way, way back, Cisco made some major improvements to their OSPF implementation, and recommended all upgrade, if using OSPF, to an IOS version that had those improvements.) Again, other factors are involved in how many devices should be within the same OSPF area. Just understand, the more L3 devices you have within an OSPF area makes more devices subject to an OSPF area meltdown and perhaps increases the chances of having an issue within an OSPF area.

In other words, often large areas or single area OSPF designs work great, until the day they don't (which, though, may never happen). I.e. sometimes a multi-area OSPF area design has more to due with resiliency than what you need for normal everyday operation.

I do agree with Harold that sometimes OSPF area multi-area designs are overly complicated, or would be fine with a single area, but this also assumes you understand and have put in place other good practices. For example, are you using ISPF? Or have you enabled dampening? Or do you have needless parallel links, etc.?

[dbeattie]

A few other items that I have seen contribute to the stability of OSPF designs are:

• Ensuring that transit links are point-to-point
• Ensuring that access links are passive (or redist from connected)
• Not mixing transit and access links i.e. no hosts on transit links)

All of these will help you scale both single- and multi-area topologies.

Obviously, alternatives (e.g. point-to-multipoint) have their place, but these three guidelines promote stable scalability.

Hope this helps

Dave

[Joseph W. Doherty]

BTW, I believe one issue of doing a redist from connected, it makes all those networks as external LSAs.