Simulate slow BGP convergence...

[martyvandeveerdonk]

For a client I would like to simulate BGP failover to investigate the impact of slow convergence in an IPT network.

The problem (challenge) is that we would like to do this on 1 physical link/interface. Maybe via subinterfaces?

Can somebody help/advise?

regards, Marty

[tdrais]

Sounds like you are going to test on a live connection which is not a good idea.

I assume you mean failover of direcly connected line. These are the timeouts that normally cause slow change over. If you get a withdraw message from a upstream router it takes effect very quickly.

BGP does not really know anything about lines since it runs at layer 3 you can setup sessions between loopback addresses rather than the physical interfaces. You should be able to simulate a failure of connection with a access list.

[martyvandeveerdonk]

The plan is to simulate this situation in an development lab, not in a live production environment.

Therefor I'm looking for a way to simulate slow BGP converengence which occurs in large data networks.

Any idea?

[tdrais]

If you mean the time it takes to send bgp updates from router to router this is the old how do you simulate latency problem.

To a point you can rate limit interfaces to very low rates but this does not help if all you are sending small amounts of traffic. I suppose you could generate lots of traffic and shape the traffic but it gets tricky since the router always likes to send its traffic first.

There are linux based machines that will delay traffic for specified amounts of time but the software I used is no longer available. I assume other people write this.

My normal method to do this is to build multiple tunnels to a remote location and force the traffic using policy routing to use the tunnels. This method does not work if you do not have access to a remote site or if the lab network is isolated from the production network.