No one behind the wheel at WorldCom

Sat Jul 13 21:54:50 UTC 2002

> > 1. Deaggregation to help spread out traffic flow. As someone who used
> >    to send a lot of traffic toward some big providers, it can be hard
> >    to balance traffic efficiently when all you get is one short prefix
> >    at multiple peering points. Having more-specifics, and possibly
> 
> A slight exaggeration, large providers have more than one assignment
> of IPs and 
> according to the RIR info they are used regionally anyway

Yes, but as the prefix length gets shorter, you lose visibility into
how traffic might efficiently be divided up among the points at which
a prefix is offered (whether you listen to MEDs or manipulate metrics
yourself). Treating North America as a region, a provider might
announce a /8 at five different places in that region. For any given
point, you might be trying to reach a more-specific that's in the same
city, or across the continent. To the extent that providers announce
longer prefixes because that's what the RIRs gave them, and practice
allocation policies that concentrate use of that space topologically
within their network, yes, your comment is a sensible refinement of
mine. 

The practice of announcing more-specifics to help peers with traffic
engineering is certainly in use today (just as the practice of not
doing so is in use today); the extent to which that puts AS701 where
it is on Tony's list is something I don't know. I'm assuming, though,
that application of that practice by AS701 would cause them to be
higher on Tony's list than if they did not engage in that
practice. Whether AS701 thinks that way or not is up to AS701 folks to
say (or not).

> > 2. Cut-outs for those pesky dot-coms; you know, the ones with the most
> >    compelling content on the Internet jumping up and down in your face
> >    with a need to multi-home their /24 to satisfy the crushing global
> >    demand for such essentials as "the hamster dance."
> 
> Overlap the more specific with the main block? (I assume) Tony's report shows
> originating AS, in which case the sub-assignments wont show towards
> UUs count.

I was making the assumption that longer prefixes within a shorter one
did contribute to what Tony is counting.

> > Let's consider the converse, though - what if AS701 were to suddenly
> > become a paragon of routing table efficiency, and collapse all their
> > announcements into one (not possible, I know, but indulge me, please)?
> > 
> > First, some decrease in revenue because all the more-specifics for
> > multi-homed customers would be preferred over the one big AS701
> > announcement.
> 
> They will still announce the customer's BGP more specifics tho?

You're applying reality to the example. This is a contrived example to
illustrate the end of the spectrum where AS701 emits one very short
prefix - kind of like some IPv6 people seem to think that
inter-provider routing should work (to use a current analogy).

> > Second, a traffic balancing nightmare as everyone who touches AS701 in
> > multiple places tries to figure out how to deliver traffic to AS701
> > efficiently.
> 
> As above, it is at least as far as I can tell assigned per country.

What do countries have to do with this? AS701 is UUNET's North
American (for the most part) AS number. It is possible to have a
handful of attachments to it within the United States alone. As noted
above, if AS701 were to announce a short prefix with no more-specifics
at several points, your options to efficiently balance traffic among
those points are less than if you were supplied with more-specific
prefixes that give you clues as to how the short prefix is
partitioned internally (say, east-coast US vs. west-coast US).

Stephen