Re: [apops] Fwd: Korea Telecom leaking >1000 prefixes to Internet

  • To: Geoff Huston <gih at telstra dot net>
  • Subject: Re: [apops] Fwd: Korea Telecom leaking >1000 prefixes to Internet
  • From: Philip Smith <pfs at cisco dot com>
  • Date: Wed, 13 Dec 2000 10:35:04 +1000
  • Cc: apops at lists dot apnic dot net
  • In-reply-to: <4.3.2.7.2.20001213084926.00b27420@localhost>
  • References: <5.0.2.1.2.20001213050250.071ecec0@localhost><4.3.2.7.2.20001213052553.00c24100@localhost><5.0.2.1.2.20001213040422.07691610@localhost><20001212081205.B23618@goose.automagic.org><5.0.2.1.2.20001212141332.07842ec0@localhost><5.0.2.1.2.20001212042122.00ab3a40@localhost><5.0.2.1.2.20001212042122.00ab3a40@localhost><20001211135958.Q22842@goose.automagic.org><5.0.2.1.2.20001212141332.07842ec0@localhost>
  • Sender: owner-apops@lists.apnic.net
    • At 08:56 13/12/00 +1100, Geoff Huston wrote:
      
      >So I send a bill to a Slovakian ISP who has entreis in my routing table?
      >
      >hmm - probability of payment: 0
      
      You never know...? ;-)
      
      >                                       +--+
      >     Rational Optimally routed network |  |
      >                                       +--+
      >
      >I'd like to know how you get to the latter outcome from this mode of behavior.
      
      Well, what other big sticks have been used at the time of CIDR deployment 
      in 1994? I remember the "flag" day for us in the UK when UUNET converted 
      frmo BGP3 to BGP4. We kinda had to support BGP4, or stick in a static 
      default to see the bits of the Internet which disappeared. And after then, 
      there seemed to be a lot of peer pressure to get most of the Internet over 
      to BGP4. And in the subsequent years (the linear growth in the BGP table), 
      I was certainly very conscious of the "do the right thing" attitude for 
      prudent announcements to the Internet. In the last 3 or 4 years, this seems 
      to have been diminishing...
      
      Following from that we had Tony's CIDR report attempting to use peer 
      pressure on providers so that they aggregate prefixes, rather than 
      announcing specifics (I accept this is a slightly different problem). Could 
      peer pressure of sorts be made to work? Or do we even want it to work? It 
      comes back to the question of does any of this really matter, or do we 
      decide that some of it matters when we are 30seconds from the brink?
      
      >But my point is that NO export is broken - what if I want to bias the 
      >route selection of the 2 AS away provider - no export is useless and the 
      >next step after no export is global.
      
      Yup, for more than one AS hop, there are no tools apart from cooperation 
      and e-mail.
      
      >Now is it KT's fault that the routing technology tools are simply not good 
      >enough to accurately represent policy promulgation?
      
      It's not clear to me that a lack of tools caused this particular problem. 
      I'm told it was a missing prefix filter on one of the border routers. But 
      granted, in the big picture, there is no tool to aid more detailed 
      multihoming or route preferences between the detailed community policy 
      possible for you and your immediately neighbouring ASes, and dealing with 
      detailed policy for you and the rest of the world.
      
      >It could, but if you want fine grained control of multiple links then the 
      >BGP table IS the only traffic engineering vehicle we have, and adjusting 
      >and announcing arbitrary prefixes does not work on a day by day basis - 
      >the /24's allow fine grained rapid (minutes) response multi-provider 
      >traffic engineering.
      >
      >Lousy outcomes are often the outcome of not having the right tool.
      
      What would be the best way forward? Some better defined communities, say in 
      the vein of RFC1998, which can have meaning more AS hops away. I can't see 
      this scaling too well though (I seem to remember this being talked about 
      recently somewhere)...
      
      Or maybe we should say "BGP was nice, but we need a new routing protocol 
      for the Internet today"?
      
      philip
      --
      
      *             APOPS: Asia Pacific Operations Forum              *
      * To unsubscribe: send "unsubscribe" to apops-request at apnic dot net *