BGP Confederation deployment


My company has a small network that is now expanding somewhat.  We have
3 data centers in total  2 of which provide eBGP connections to our
customers who have local Ethernet/Metro-E circuit handoffs to each data
center... We currently use eBGP to the client with a public AS.

Between the 2 data centers another engineer decided to run IBGP, which
was then removed and now spanning tree is being used between the 2 data
centers (Over high capacity circuits), Not my idea!

One of my fellow engineers insists running BGP confederation is optimal

My plan is revert to private AS for each data center and run eBGP to the
clients and also between the data centers for faster convergence along
with BFD protocol.. I do not want to use confederation as I dont see it
being needed and i dont have an experience with any routing anomalies
associate with it.

I plan on weighing routes using MED's between the private AS in our networks to control routing.

I have argued that BGP confederation is used to solve the issue of larger
networks iBGP mesh and typically ISP use confederations only..

Is there a need to use confederations ?  Does it make sense to use it
here ??? If we plan on using eBGP between data center I dont see a need
for confederations.


In summary this is how the network will look:


Client ---eBGP---> DC#1  <----eBGP----> DC#2 -----eBGP----> Client


we really dont have any need to run IBGP except maybe between the 2 egress (eBGP) switches that connect to the clients


  • You are right that Idea of running BGP confederation is to minimize
    the IBGP full mesh requirements. But on the other hand same can be done
    by using Route Reflectors design too.

    Now do you need confederation or route-reflectors is little tricky to tell without looking at complete network topology.

    I would recommend you to share the detailed network topology to come up with right design.

    As far confederation goes, that should be configured at client side if you want to do so, only benifit will be that client won't need to own a Public AS number any more.


  • The topology is as follows:

    Data center #1 called Ny4
    Center #2 called Nj2

    eBGP customers connect to both NY4/ NJ2

    have redundant switches and circuits between NY4/NJ2

    we use vlan trunking between the 2 data centers.

    both data
    centers run a public AS and peer to the client
    I want to change each
    data center and convert to private AS and control routing more
    if we do this we will require clients to change their
    BGP peering config... I figure we can use the BGP local-as command to
    help with any client who cannot peer with our private as because of
    duplicate AS numbers or timing.

    In this situation there doesnt
    seem to be a need for confederation.

    don't ISP typically use

Sign In or Register to comment.