[c-nsp] BGP peer/customer routes

[Pete Templin]

On 5/31/2011 1:31 PM, vince anton wrote:
> thanks for feedback. seems like different people are going around this in
> different ways, some allow transit through peering links, and some outright
> block this from day0
>
> it surprises me that some people seem to be ok with passing transit traffic
> over a peering link. I dont understand why you would want to do this, as to
> me this seems abuse or misconfiguration (possibly not intentional), and
> potentially very expensive, or loss of revenue.
The problem is that transit traffic vs. peering traffic is dependent 
upon source address, yet networks route by destination address.  
Obviously, there's no easy way to route traffic FROM your network to 
your customer via your peer while routing traffic THROUGH your network 
to your customer via your transit link. However, if your customer has 
two providers and one of them is your peer, shouldn't your customer be 
eligible for redundant services through the peer link if your 
customer-aggregation router should be down?

Additionally, where do you point the finger?  Where do you implement 
policy?  Perhaps you should filter your peer such that the longer 
prefixes aren't accepted, and exact-match prefixes are only accepted if 
no prepends are added.

Have you covered MED in any of your configuration or policy?  You might 
want to forcibly set MED=1 on all peering connections, so you have 
predictable behavior later if you ever multi-path to a peer.

pt