How many Tunnels [Re: Re: pTLA request]

[Bob Fink]

Simon,

At 05:47 PM 8/25/98 +0200, Simon Leinen wrote:
>Yiorgos Adamopoulos <Y.Adamopoulos@noc.ntua.gr> writes:
>> I think they should be assigned the pTLA.  But, just like in our
>> case (GRNET, 3FFE:2D00::), they should open more tunnels the just
>> with SWITCH and SURFNET.
>
>Yiorgos raised an interesting point here: How many tunnels should
>pTLA/transit sites have to other sites.
>
>There seems to be a tendency to build a full mesh between backbone
>sites.  I think this is very undesirable, because it makes the 6bone
>less realistic as a prototype for routing on a global IPv6 network:
>
>* The diameter of the network is kept artificially small.
>* Abundance of tunnels introduces an unrealistic level of redundancy,
>  which reduces the pressure on sites to fix routing problems to a
>  given neighbor.
>* Some "single hops" over the 6bone represent many many hops over the
>  IPv4 Internet, sometimes including congested subpaths.  Those
>  "shortcuts" attract transit traffic that could use a different path
>  consisting of several hops between 6bone neighbors that are actually
>  close to each other in the IPv4 Internet topology.  This has a bad
>  impact on global 6bone connectivity.
>
>Personally I would prefer the 6bone topology to resemble the IPv4
>topology closer than is the case today.  SWITCH's policy as a pTLA has
>been to set up tunnels to other networks that we either peer with or
>that we have good connectivity with.  In particular, we try to avoid
>the "shortcuts" I mentioned above, as well as links that regularly
>experience congestion.

Actually there is not anything near a full mesh between pTLA sites.  The average seems to be in the 3 to 4 range.  Only a few establish many peerings, like UUNET-UK with 15 or so.  In fact I believe that many/most 6bone pTLAs try to peer realistically based on service quality.  Though I should defer to others to respond to this whom are more active in these options and choices for their own 6bone backbone peering.


>Is there any opinion on mentioning these problems in the "6bone
>Routing Practice" I-D, not as a recommendation but as a problem
>statement?

Not the place for a problem statement, though it might ellaborate a bit in the section 4 comments on at least 3 peerings.  Maybe saying that a full mesh is NOT the goal (Bertrand?)


>Personally, I find BME-FSZ's connectivity quite sufficient - SURFNET
>and SWITCH are both connected to Hungary over few hops the TEN-34
>research backbone.

Thanks for the comments.


Bob