Anyone notice strange announcements for

David Barak thegameiam at
Tue Jan 13 18:11:54 UTC 2009

--- On Tue, 1/13/09, Patrick W. Gilmore <patrick at> wrote:
> > AS_PATH != identity, and I would not recommend loading
> the latter onto the former.
> We disagree.  When I am researching something, I frequently
> look at ASNs in the path to figure out not just where but
> who controls the path.

Oh, I certainly think that there is a loose coupling there, and the relationship is highly valuable from a troubleshooting point of view.  However, I would counsel anyone investigating AS_Path relationships to remember that do not completely characterize the relationship between any two organizations, let alone the multipolar relationships between all organizations.  

It's a good first-cut, but it doesn't have the level of authority that you're implying.  I'm not aware of any ASNs being trademarked...

> >> 	Personally, I would be upset if someone injected
> a route
> >> with my ASN in the AS_PATH without my permission.
> > 
> > Why?  Is this a theoretical "because it's
> ugly" complaint, or is there a reason why manipulating
> this particular BGP attribute in this particular way is so
> bad?  Organizations do filtering and routing manipulation
> all over the place.  Is there something worse about doing it
> this way than others?
> Filtering and other manipulation happened on your router,
> prepending my ASN is putting that information into every
> router.  That seems to be a serious qualitative difference
> to me.  Do you disagree?

This is qualitatively similar to an ASN announcing de-aggregated routes - it may be nice if they don't, and you don't have to accept them, but are they permitted?

> This thread has been interesting & educational.  So
> many people seem to be happy to explain why they should be
> allowed to use globally unique identifiers they do not own
> in ways which were not intended, then explain to the people
> who do own those identifiers how they should react, which
> alarms should go off, and even which priority the alarms
> should have.
> As I have repeated probably hundreds of times: Your
> network, your decision.  I have yet to hear a credible
> argument against that stance.  What you do _inside_ your
> network is _your_ decision.  When it leaves your network,
> however, things change.

Exactly!  Provider RB announces $WEIRD.  A bunch of providers receive alarms about the existence of $WEIRD, and they treated this as $IMPORTANT.  The bunch of providers who treated this as $IMPORTANT are informing all of us about their monitoring thresholds and their responses to this threshold being met.  Their networks, their decisions.

It should be pointed out that pre-provisioned AS_Path filters and prefix-lists would actually be effective at defeating this and preventing someone who is actually malicious from using this technique.  This is an excellent argument for implementing SIDR...

> Announcing an ASN which is not yours to eBGP peers means it
> is leaving your network, which means it is not just your
> business.  Doing so and then telling the ASN owner that they
> should not worry about it afterwards - and in fact arguing
> when the owner repeatedly tells you this caused them
> problems - does not seem to be the proper course of action.

Understood, but if this is viewed as problematic, there is a very simple solution: don't allow a BGP customer (or peer!) to prepend someone else's ASN.

> I mentioned earlier in the thread if Cogent prepending
> Sprint's ASN to Verio, people would react differently. 
> Randy said tools can be used for good or bad, obviously
> implying he's the good guy.  He is not the good guy.  He
> used someone else's resources without their permission
> and without even notifying them, costing them time &
> effort.  Randy doesn't get to decide if the ASN owner
> should have alerted or investigated  or whatever, and
> neither do any of you unless it is your ASN.
> How can anyone seriously argue the ASN owner is somehow
> wrong and keep a straight face?  How can anyone else who
> actually runs a network not see that as ridiculous?

Are any providers going to implement ^ASN filtering as a result of this experiment?  This could turn out to be a very inexpensive lesson, which is far preferable to more expensive lessons...

David Barak
Need Geek Rock?  Try The Franchise:


More information about the NANOG mailing list