OSPF Design Discussion

OSPF Design – In the below picture, where should you place an OSPF ABR (Area Border Router) to scale OSPF design ? Why ?

Please share your thoughts in the comment box below.

First 5 correct answers will get my CCDE Preparation Workbook for free. Please subscribe to email list so I can see your email address for communication.

OSPF Design

99 Replies to “OSPF Design Discussion”

  1. Obviously on G or H. There is only one way to reach another side of network, so it’s not necessary to compute LS metric to other side routers and networks.

  2. G & H
    Make G & H as a backbone area (area 0).
    Rest of the routers are directly connected to either G or H.
    This may be the best in this scenario.

    1. You are the first correct answer for the G or H but I didn’t understand your answer for the ” WHY ” part Raheel, what do you mean rest of the routers are directly connected to either G or H ?

  3. ABR should be placed on G and H in this scenario.
    Why ? because all the routers in the area are linked to these and to scale use summarization and type 3 filtering feature

    1. G AND H should be chosen as ABR and both of links between G and H should be placed in area 0 to limit area 0 DB and prevent it from being affected by any LSA flooding that might happen inside each area (supposing we have area 1 up and area 2 down). this can reduce spf full calculations in area 0.

  4. A. Just because I like to be different . there is no reason all routerds can’t be in the same area.
    There is redundency in the network with two links and dependant on the requirement you might need all routerds to be in one area TE? . also eurgan said AN meaning singular ?

    1. ABR should be placed on G and H only and have to make G&H backbone area 0 which allow to do summarization on ABR it will help to scale routing table.

      1. G or H is the correct answer because, they are a point of contact between the two domains( so it respects the formal definition of ABR) and doesn’t impact scalability (iodeally new routers will be added in one of the two domains or in a third , 4th or nth domain.

  5. G and H are the connection point between the 2 mesh, I think it will be the best place for an area border. It will also make summarization and filtering easier

  6. G and H should be the ABR so that it can divide the complexity of the network and at the same time provide the ability for each Area to converge faster and making the IP addressing sample to be able to summarize between them.

  7. G and H.

    The job of an ABR is to receive lsa 1 from routers in same area with it and convert it to lsa 3 to flood into other areas connected to it. For this to effectively happen the ABR should have at least a leg in the area it receives lsas 1 from and the mandatory leg in area 0 for star like look prevention mechanism.

    G and H would do the job of connecting to non area 0 routers, receiving lsas from them, sending to each other as lsa 3 and flooding to the other non area 0 routers with minimal flooding scope based on the link connectivity.

  8. @G & H . they would make up are Zero , for route filtering , route Summarization; retuction of the hardware overhead from large routing tables ; reduction of the fault domain & to speed up convergance time

    1. G and H. Put the links in between G and H in area 0. Make the above and below areas totally stubby areas. Send only default route from Both ABRs to the corresponding Totally stubby areas. This will reduce the topology DB in the same area to only internal routes as there are quite few links and routers. External routes and destinations are not allowed into a totally stubby area. I think by doing this, we can achieve simplicity, scalability and stability in the design 🙂

  9. asnwer : G, H


    1 .because once determined where are the points of complexity in the newtwork, and this is the optimal place to
    divide the flooding domains

    2.because we can do optimal aggregation and summarization of topology information if G and H are the ABR

  10. Not here for the book but for the discussion.

    We have two complex areas of the network which are fully meshed. One of the main principles of design as mentioned in “Optimal Routing Design” by Russ White is to separate complexity from complexity. This means that we want to divide the two fully meshed areas of the network from the others.

    Based on this the most logical place to place the ABR is at G or H. Many people have already mentioned this. What interests me is if placing it at G vs H will make an impact on the network. Here is what I’m thinking. If we place it at G then we can summarize towards Router H and make the area totally stubby or totally NSSA or whatever is desirable. This will lead to a smaller LSDB and routing table at H and all the routers below.

    If we make H the ABR, then it will be part of area 0 so it will have to react to whatever happens in area 0 and it can’t have a summarized routing table and also have to keep two LSDBs. Router H can still summarize toward all routers below and make that area totally stubby or totallly NSSA.

    It would seem that it is slightly more optimal to place the summary point at G because more routers could use the summary then vs placing it at H.

    1. You got the book already as my gift 🙂 So thanks for the comment. For the first part making G or H an ABR is for sure.

      But selecting among G or H cannot be. Topology is symmetrical so , without assuming anything else , in this topology any of them would be equally optimal.

  11. since you ask for scalable ospf design shouldn’t we worry on the single point of failures due to G & H single routers though they have dual links between them ?

  12. I guess the point is about the full mesh networks and not about summarization/filering/hiding topology (which is only a bonus here).
    Using ABRs allow to use mesh groups and avoid that each router sends the same information again. For instance a flap on A is forwarded only once by H.

  13. G and H are the only logical boundaries. Aside from lsa reduction, summarization, etc… Making any other routers ABRs has the potential to create parralel links to a border area. Obviously since traffic entering an area can not traverse another area and return, in the event of a link failure you may end up with an unsupported topology and drop traffic as a result.

  14. At the first glance it would come to mind that G and H must be the ABR but if we really want to optimize the LSDB so no need for G and H to know the type 1 and 2 of both areas if they are not included in their intra-area transit path .
    C , D , E and F is the correct answer also will provide ABR redundancy .

  15. I would go with router G and H as ABR. We can configure G and H in backbone area 0. As G and H are ABR, and by definition ABR is a member of two or more areas which is true in our case, and as ABR advertises the prefixes it learns in one area to its other areas by flooding Type 3 LSAs into the areas, which is basically saying that “I know you can reach these prefixes through me” or in other words “I know how to reach these destinations.” So as we have parallel links with router G and H and if one of the link fails between G and H then we still have reachability to other areas through other link.

  16. G or H depend how you designed the areas you are using , which two links between G&H are ospf enabled and which network will going to be extended in the Future , the upper one or lower one ……keep in mind max 50 router per area is the cisco recommendation

  17. As we are looking in a scalable design we can choose H&G ,In this topology its a full mesh.Any how if we make the G&H as the ABR we can make summarize the area routs and reuce the lsa flood. also these routers having a redundancy.

  18. Cisco states that the factors that influence OSPF scalability include the number of adjacent neighbors for any one router, the number of adjacent routers in an area, the number of areas supported by any one router, and the designated router (DR) selection.

    Since it is not clear from your question (or the responses I’ve read) whether I am adding a new device or indicating which of the eight labeled devices are becoming the ABR, I would reach back to you for clarification.

    A current design concern I have is that there is a possibility of an outage between the two meshed networks if either router G or H where to fail.

    Based on Cisco’s guidance, with the single design goal of scaling OSPF, the ABR should be added (if that is the method) between G and H.

  19. I think its a Backbone splitted in two separated locations. Since, was not mentioned BGP and we have few routers on topology, in this case, i will place the ABRs at the edge of these locations (A,B,C,D,E,F) and so on, for scale. Only G and H will connected these two parts, participating in area 0.

  20. Hello,

    I would pick C,D, E,F. Area 0 with CDEFHG + ABRs CDEF.

    Why ?
    1. TOP in AREA1
    2. Bottom in AREA2
    – less LSA1 flooding in AREA0 (not that many routers) – less SPF changes
    – allows you to split in more areas yet again (C in area 11, D in 12, E in AREA 21, F in area AREA22)
    – redudancy of ABRs

  21. My choice of placing ABR would be between C & E and D & F, though it will increase the DB size but will provide more resiliency and also can achieve load sharing.

  22. Hello Orhan, nice scenario!

    I would choose G and H.

    – The upper and lower networks are full mesh, so you have to separate the complexity of one full mesh from the other.

    – With this design route aggregation would not cause suboptimal routing, because there would be only one exit/entry point for each area.

    – Also you can easily turn any of the full mesh areas to stub areas. Because, again, only one exit/entry point from the area.

    I’m assuming both links between G and H are reliable. I would also recommend to load balance between them.

    1. The prior answer is in case general ABR placement is the discussion.

      In case only one ABR should be placed it would still be G or H. The reasons are the same that were detailed in my prior anwer. The pending question would be, which of the two should be ABR.

      If both full mesh networks are identical, this meaning: same number of routes (internal, external), same reliability of the links (outage frequency), ip addressing is nicely distributed in both, etc. Then it should not matter which one is ABR, as it would be the same.

      In case both full mesh are not equal. Care should be taken in seeing which of the two would get the most load from the following: LSDB size, LSA flooding, LSA type-7 to type-5 translation, etc. This would have to be taken into account for choosing G or H as ABR, the end result should be the most fair load balance of cpu and memory possible between both.

  23. ABR is either G or H. Because with that topology, it is impossible for other router to be Area border router.

    Example: router A must have same area with router C, D and G. A cannot has same area with C while in same time has different area with D and G. Since D and G are connected to C also.

    1. With that topology, for scaling, number of area should be two areas, area 0 and non zero. So, there is only one abr. Either G or H should be same.

  24. if the purpose is to scale and expand beyond the diagram, it makes sense to create another area to lessen traffic within an OSPF area. Although it may look like G or H may be the best place for an ABR, you can technically have ABR anywhere A through H. As you scale up, you just have to ensure that all areas are attached (physically or virtually) to Area 0. In fact, I’d prefer to make A-F an ABR than G & H.

  25. I think, depending on type of routers either one of G/H can be ABR. In case it’s a cisco shop, we’ll need to put one of the loopback (pref.) interface of one of these routers in area 0 and one interface each in either side of the network areas. In case it’s Juniper, just adding one interface each to the either side of the network areas should make it ABR.

    1. Faisal, Thanks for the comment.Vendor implementation ( How you do the router an ABR based on their RFC implementation ) is not important at all. In this topology there is obvious place which you should make an ABR and it doesn’t change based on vendor.

      1. True, I’m thinking on the lines of convergence of the protocol, as there are only two links between G & H, hence convergence takes place much quickly then rest of the full mesh networks on either side. Also in case of one link failure alternate route calculation is going to take place much faster. Otherwise, within full mesh too much wait is observed for the DR/BDR process to complete and as the network grows it’ll become more difficult.

        1. DR/BDR election always add some tax to convergence , thus we recommend to use OSPF Point-to-point network type for some topologies for fast convergence and reducing the LSDB size. But here is the question is not stating anything about convergence and topology type. Anyway what you are saying is correct.

  26. Flooding routing information through a full mesh topology is the MAIN concern,so I would choose H or/and G to be ABR in order to eliminate the scope of LSA flooding in each full mesh part .

  27. for scalability reasons I will divide the network to areas and make each border acts as ABR
    The arias should be divided as below:
    Area 0 Includes G-H and A-B for redundancy and to connect with other new areas
    Area 1 Includes C-G-D
    Area 2 Includes E-H-F
    based-on that the better place to set the ABRs is routers G and H

  28. I will not prefer G and H as ABR because if either of the node goes down it will isolate the either network. I would propose ABRs from one of the following pair either AB or CE or DF.

      1. Hi Orhan, As you mentioned scaling the network in your question. So I considered whole network as backbone ( area0 ) network. If I would need to add more areas so I would choose either AB,CE or DF ( or any combination from these where ABRs should be in same region) as ABR to connect with other non-backbone areas. If any of the ABR goes down then at least I would have connectivity to speak to other areas. If I go with GH and if any one of them goes down then the half of the network will be isolated.

  29. My choice is G or H, having it as ABR will have a lesser max paths.

    If to choose CD and EF it will introduce sub optmal path – because intra-area is preferred than inter-area thus, it has to be carefully manage.

  30. ABR should be configured as Router G and H as they have the link to all other routers in a segment and configuring G,H as ABR has lesser impact on over all LSA transmitted with in the network.

  31. This scenario does not have complete information, may be author need to provide some additional details about current area details and Layer 2 topology.

  32. Normally, it should be the G and H routers are ABR. The reason is: – if G and H are ABR, it will have less LSA flooding (type 1 and or type 2)

  33. I realize it may be too late but I’d still give it a go.
    If the network will stay like it is and not going to change I’d place ABR at either G or H. Answer for that has already been provided above.
    However if the network is going to evolve and more links are going to be added (e.g. C-E or D-F), in that case we’d have to follow the same rule and allocate either one of the pair to be ABR again. I think here we need to ask the question if it’s worth doing multi-area OSPF at all.
    I believe it’s a widely accepted fact that OSPF can handle up to several hundreds routers in a single area. The times when 10, 20 or 50 routers per-area were the upper limit are long gone. The only thing that splitting OSPF in multiple area gives us is limit the scope of SPF calculations. With the help of incremental SPF any routing changes in different areas will only trigger routing metric update (and not a full SPF recalculation). However, if all the links are stable (e.g. no aerial links) topology changes are very rare even in a big network and even with a few (un)scheduled link failures modern routers will be able to do best path computation in the blink of an eye.
    So my rule of thumb would be to put all of them in one area and only segregate links that are constantly flapping/unstable into their own OSPF area.

    1. Hi Michael , Thanks for the comment. Problem here is the topology type. In this network you need to do multiarea OSPF since it is a fullmesh and not only calculating SPF is an issue but mostly the flooding will be overwhelming thus you should put a boundary.

  34. good and healthy networks 😉 get it. These are the ones connecting both ospf areas. Where one would imagine them being the good and healthy.

    If one considers extending links beyond the current topology view, the ABRs can also reside on C(orrect)&E(normous) or D(efault)&F(earless) but if this isn’t happening go for g(ood) and h(ealthy).

    My $0.02

    1. Ric , It was hard for me to read more than 60 answers , answering most of them and choosing the first 5 correct ones 🙂 You are the 4th one. Congrats ! State your email address please , I will send the book

  35. The OSPF ABR would be placed at either point G or point H. This is to ensure that the LSDBs of the OSPF routers are reduced in size due to limiting of LSA type 1 and 2 flooding. The SPF run would not have to be triggered for the whole area since routers E,H,F, B as well as any other with in that part of the OSPF design will be in an independent area from the Routers A,D,C which will help to demarcate any network failures in those particular locations in the OSPF design rather than have the failure propagate the entire network domain. This will also lead to partial updates rather than full updates between the different OSPF intra routers and the inter-area routers.

  36. The OSPF ABR would be placed at either point G or point H. This is to ensure that the LSDBs of the OSPF routers are reduced in size due to limiting of LSA type 1 and 2 flooding. Another fact would be that the SPF run would not have to be triggered for the whole area since routers E,H,F, B as well as any other with in that part of the OSPF design will be in an independent area from the Routers A,D,C which will help to demarcate any network failures in those particular locations in the OSPF design rather than have the failure propagate the entire network domain incase of any type 1 or 2 LSA update. This would also lead to partial updates between the intra-area routers and he inter-area routers rather than full updates.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.