Gre tunnel Design Branches and dynamic Routing

[parhamkiani]

Hi All

As can be seen in the digram below , we have got a route which is connected to over 500 branches. All baranches have got point-to-point tunnel to this central router.We aim to implement dynamic routing between central router and branches instead of static route from there, all routes should be redistributed to ospf1.So, we have got some qouestions to ask. first of all, which routing protocol do you recommend in regard to this topology? should it be find that we replace point-to-point gre with MGRE or We just keep all point to point tunnels and add only dynamic routing ? MGRE would be fine with this scalabe design?what design would be best practice for this kinds of topologies?

Many thanks

[Marwan ALshawi]

hi there

the best solution is to use mGRE to reduce the amount of point to point tunnels and management overhead and the many routing peers

EIGRP can be used and configure the branches as stub

or OSPF where each branch has to be in its own area configured stub, totaly stubby or NSSA depends if you have any external redistributed route in the branches or local Internet for example

hope this help

[parhamkiani]

Hi

I appreciate your help. I find your information very useful.

Best regards,

[Richard Burts]

In determining which design would work better it would help if we knew more about the network environment. And especially to know whether most traffic from a branch is to the head end or is most traffic from a branch to another branch. If most of the traffic is branch to branch then it is a good argument for MGRE. With MGRE you need to learn the individual routes from all the branches and so a dynamic routing protocol would be a big advantage. And since you need to see all the routes from all the branches there is no benefit in trying to filter out any of the advertisements. And so a routing protocol like OSPF might be good.

But if most of the traffic from a branch is to the head end then I would suggest that MGRE has more overhead than it is worth since a branch does not really need to see all the details from all the branches. In this case I would keep it as point to point tunnels. I have implemented a network very much like what you describe. This network has more than 400 remote locations. Each location has a point to point tunnel to the head end. (actually in this network there are two head end routers for redundancy and each branch has a tunnel to both head end routers). We use EIGRP as the routing protocol and believe that it gives several advantages that we would not get with OSPF. We configure each branch as EIGRP stub. From the head end router we advertise only a default route to the branch. These features keep the routing table at the branch very small and routing is quite efficient. The head end knows where every subnet is but the branch needs to know only how to route to the head end.

I believe an implementation like this could work well for you if most of your traffic from a branch is to the HQ and not so much to other branches.

HTH

Rick