Sean,

| Whether or not they announce non-(Sprint)-customer routes, such as AS1?

But Alex, you know that many providers would be delighted to filter out peers' routes on your behalf, as well as not announce you to their peers. Yet they still charge you money, no?

Yeah, but this is a semantic argument. I used the word 'peering' as meaning announcement of your own customer routes only, and expecting those routes only to be reannounced to BGP speaking customers; this is the normal use in the term 'peering point'. Equally valid is the definition meaning 'route exchange for no money, networks being of equivalent size blah blah blah' (I remember when ANS were charging $100/month for peering, or, urm, may be it wasn't peering after all then). Equally valid is the use 'peering' meaning running a BGP session and exchange routes. There are 3 main differences as far as I can see between traditional peering connections, and traditional BGP speaking customer transit: 1. Historically, peering has been free, transit hasn't 2. Historically, peering has been provided over open-port or nearly open port lines (MAEs etc), transit hasn't in the main. One could argue that peering has also essentially been best effort. 3. Historically, peering has meant you were only announced customer routes, and transit has meant you got all routes. And similarly for announcements. (2) is already to changing (private peering). (1) has changed in the sense that it's not free to everyone any more, i.e. is there is black as well as white, but not much grey (we'll peer if you pay) yet. (3) as you say already comes in many shades of grey with innovative partial transit and route swap arrangements. One of the problems with these arguments is people always seem to mix up (a) monopoly exploitation (b) people making political decisions who don't understand routing etc. etc. on the one hand, with, on the other (c) change resistance, (d) outdated dogma etc. etc., (f) lack of appreciation that the internet is a commercial environment etc. etc., which tends to weaken the arguments of those arguing (a), (b) etc. by conflation with know-not-much'es arguing (c), (d) ... [No, this is not a comment about BBN peering issues, it's a comment about evolving peering models & settlement]

This raises an interesting diagnostic issue which some people run into from time to time when hunting down weird routing problems: how can a third party know what is being announced and thrown away (or just not selected), versus what is not being announced at all?

show ip b neighbor <blah> received-before-filters on your IOS 13.0 router with a looking-glass behind it? :-) -- Alex Bligh GX Networks (formerly Xara Networks)