Email peering

Steven M. Bellovin smb at cs.columbia.edu
Fri Jun 17 17:16:22 UTC 2005


In message <OF5546988E.7DE79026-ON80257023.0034BB4D-80257023.00357C69 at radianz.c
om>, Michael.Dillon at btradianz.com writes:
>
>> Similar concept, same scaling problems; it just hides the explicit 
>routing
>> from the user (as would any modern "peering" system, presumably).
>
>Then you are presuming wrongly. Nowhere in what I wrote have
>I suggested any changes in the existing email technology.

Quite apart from anything else, this requires an email routing protocol.
Getting the policy statements right in such a protocol is a non-trivial 
task; it will make BGP look simple, because of the implied liability a 
mail sender would incur under this scheme.

Let me be very specific.  I own a 1U server in a rack somewhere.  (The 
concept has been discussed here many times, of course; thanks to some 
friends, I can actually do it.)  How do I send email?

Well, maybe the colo operator has a mail server.  Maybe the rack 
operator has mail-forwarding contracts with his upstream IP (i.e., BGP) 
peering providers.  To whom can they send mail?  More precisely, with 
whom have they signed contracts that will let them deliver mail, and 
under what conditions?  Maybe one of the upstreams has better contracts 
in some countries than the other does, or maybe the other will charge 
me less for delivering my email, but with a hefty chargeback to me if 
it's found to be spam.  Or maybe one of them won't talk to (or listen 
to) mailers in certain parts of the world -- we've seen that alleged 
and/or bragged about on this list -- because of perceptions of where 
spam comes from.  But the better mail server I'd prefer to use is down 
today, because of a fiber cut/DDoS attack/spam overload.  How do I 
know, and how do I fall back to an alternate?

We can all invent more scenarios.  I think we can all see the analogies 
to BGP, too -- and we all know how hard it is to get that to do what we 
want.

The scheme you're suggesting might work without new protocols in a 
purely hierarchical world.  It might even work with a fully-connected 
cluster of Tier 1s, each of which is a tree root.  But the Internet 
doesn't work that way, or we'd all be using EGP for routing.

Besides, as I mentioned the other day, there are policy side-effects.  
See http://news.com.com/Your+ISP+as+Net+watchdog/2100-1028_3-5748649.html?tag=nefd.top
for an example of the kind of thing I'm worried about.  

		--Steven M. Bellovin, http://www.cs.columbia.edu/~smb





More information about the NANOG mailing list