[routing-wg]Towards a better way to document BGP communities

[Olivier Bonaventure <Olivier.Bonaventure@localhost]

Dear All,

During the last ten years, more and more operators have configured their 
network with various types of BGP communities that are used for traffic 
engineering, blackholing, monitoring and other purposes. A recent survey 
of BGP communities in RIPE and RouteViews routing tables
(  http://inl.info.ucl.ac.be/publications/bgp-communities  )
shows that more than 1000 ASes have defined their own BGP communities 
(mostly for traffic engineering purposes ) and half of the BGP routes 
carry one or more BGP communities.
These BGP communities are important tools for many operators. However, 
they are currently only documented on an adhoc basis, usually either as 
notes on web pages or as comments in RPSL specifications. Some BGP 
communities may interact in strange ways and debugging those 
interactions may be very difficult on the Internet.
We believe that a more precise way of documenting BGP communities would 
be useful for both operators and their clients. This method should be 
vendor neutral but should allow one to express the policies that are 
supported by most router vendors. It should also be possible for a 
client to test how one of his route would be filtered by his provider in 
function of the BGP communities associated to that route. Doing such 
tests in the public Internet is cumbersome and difficult to debug. A 
classical example of this difficulty is that some combinations of BGP 
communities may lead to BGP wedgies as described in RFC4264.
A better approach would be to have a simulation tool that allows 
operators to both define the BGP policies used by their network and also 
provide a model that allows their client to test how their routes would 
be filtered.
The open-source CBGP simulator ( http://cbgp.info.ucl.ac.be/ ) could be 
easily used for this purpose and operators could define a CBGP model of 
their network that specifies clearly how routes containing BGP 
communities are filtered and give this model to their customers to aid 
them to tune their configurations. In the long term, a CBGP model of 
each network could be one of the elements of the RPSL specification of a 
network.
If you have defined BGP communities in your network and are interested 
in developping a BGP model of your network to precisely document the BGP 
communities that you use, let us know. We would like to build a set of 
reference CBGP models that can be used as best current practices by 
other operators.
Best regards,


Olivier Bonaventure and Bruno Quoitin

--
http://inl.info.ucl.ac.be , UCLouvain, Belgium