BGP Scalability Simulation

Topology and setup of these kinds of tests largely depend on whether you are testing iBGP or eBGP. In my experience, eBGP testing is fairly straight forward as you are almost always testing reconvergence of the BGP next-hop. iBGP testing scenarios on the other hand can be quite a bit more complex as you may also be testing the reconvergence of the underlying IGP if the BGP next-hop remains unchanged. Can you describe your testing goals and environment in a bit more detail?

Stefan Fouant
Principal Network Engineer
NeuStar, Inc. -
GPG Key ID: 0xB5E3803D

Thanks Stefan for your reply.

Basically the goal of this testing is to study the BGP scalability issues in
the internet sometime in future lets say 10 years from now and try to find
out what problems it could face . I am trying to use ns2 as my simulation

Can you suggest how I can set up the envrionment for this kind of study and
what parameters should I try to caputre.


Vince Fuller has done some projections on what the the routing tables will
be like in the near future which would be useful for you, check out

If you are looking at doing simulations of what it could be like, use
similar figures to his for IPv4 & IPv6 routing table size.


Bradley Freeman

We have a similar analysis (which agrees with Vince Fuller's #s in a general sense) in the middle of a recent sigcomm paper:

See the paper "Accountable Internet Protocol (AIP)".

I point it out mostly because the Fuller presentation said "kinda looks exponential"; we found that the scaling was 17% per year, which could be a bit more useful if you need to come up with #s for the years between when Fuller provides projections for.



Do you have something specific in mind you want to measure? e.g. convergence times, table size, update count, etc? the scope of your study seems to broad as you describe it..


Hi Ricardo,

Basically I want to measure the Convergence times and routing table sizes.
But I am not able to find a good topology of internet which I can utilize
for my experimentations. I am looking at GT-ITM, BRITE and IGen but don't
know what kind of abstraction they provide and if these topologies are
feasible to test the above mentioned parameters.

What challenges I can face if I want to measure all those parameters
convergence times ,table sizes , update count, signal sizes etc.


The topos you mentioned are synthetic (e.g. generated based on math), you might want to check these ones instead, based on bgp tables from public sources:

Also, i don't think using a full internet topology is the way to go to do measure convergence time. The reason is that convergence time is highly dependent on ibgp architecture, router timers, etc and modeling things as one router per AS is at most unrealistic for this purpose. I would suggest to look at a small yet realistic topology of a few ISPs, e.g. as given by rocket fuel:

For routing table size, you just need to grab the existing available routing tables, e.g.
and do an extrapolation of the number of prefixes in RIB n years from now


Hi Ricardo,

I checked the topologies from bgp tables at ucla site. These are in the form
of huge tables . Can you please guide me how this can be utilized in the