North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: AS 701 local-pref answer.

  • From: Alan Hannan
  • Date: Mon Dec 17 03:12:39 2001

> Do they expect consistent route annoucements from their peers?
> Many networks out there insist upon this as a requirement when peering.

  While many networks insist on this as a requirement when peering,
  few folks audit it, and fewer still take action as a result of
  noticing inconsistent announcements.

  "it's only illegal if you get caught".

  I think [email protected]/genu used to actively audit and unicastedly communicate
  this stuff and catch errors folks made.  There was always discrepancy 
  between what the views should look like, but it was usually noise.
  I think the filter was something like 1% or such, scaled to the
  number of routes announced.  This was a good first pass filter to
  see if someone was actively screwing you.

  I recall that a network at which I worked ran a similar consistency
  checking script, and we'd find and watch what folks were doing.
  Normally it was 'in the noise' but if it was significant we'd ask
  them to fix it.  If they didn't fix it, well the implied threat was
  that we'd 'depeer' but it never got to that.

  The real reason for this 'consistent announcement' thing was (at
  least to me, being one who wrangled the lawyers that wrote the
  peering policy before jsb took that stick) to prevent folks from
  forcing us to carry more of a peer's traffic, and to prevent a
  customer from doing closest exit on us, w/out letting us do closest
  exit to them.

  jmalcolm + sherk even had a scheme by which we'd set the BGP NH of
  all learned routes to a particular IP address, say "" and
  then static route that destination out all applicable
  peering interfaces.  In this manner we'd just have a given /32
  destination for each peer, and assign their routes to that.  We never 
  put it into practice because a/ the problem wasn't great, and b/ there 
  were some "quote" technical "unquote" issues with this solution.

  Bottom line, this inconsistency issue is not significant.

> wouldn't this mean that they wouldn't have consistent route announcments
> in various parts of their network?

  My recollection of the issues being discussed are that they are
  inconsequential in practice, and resolution lays in the hands of the