Re: Doc 1: Symmetric Routing

[Sean Doran <[e-mail]>]

|       o   Compatibility with other BGP implementation needs to be
|           tested.

Bwa ha ha ha ha!

There's one other problem, and that has to do with the
fact that one has to subnet short prefixes in order
to make this work in the current general case of provider-based
delegations.   That's bad juju.

Of course, the other problem is that there are practical limits
on a/ how much you can affect how other people route towards
you, and this is proportional to the number of defaultless networks
and the number of peering points among them, and b/ the amount
of policy that can be handled by routers and their maintainers everywhere.

| 3.2 Splitting AS
| 
|    This approach requires an AS to be split into multiple ASs and run
|    external BGP between these ASs (possibly with the MED configured for
|    load balancing among multiple links). 

BGP confederations are great for this; forget MED and so forth,
use one IGP and (in a Cisco implementation) say "router bgp 1
no sync
bgp confederation peers 2"

Of course, what I would like to see is:

[say, on big provider A]
ip route 1.1.0.0 255.255.128.0 Serial 0/0 withdraw-when-down
ip route 1.1.128.0 255.255.128.0 Serial 0/0 withdraw-when-down as-origin 16385 
[say, on big provider B]
ip route 1.1.128.0 255.255.255.0 Serial 0/0 withdraw-when-down
ip route 1.1.0.0 255.255.128.0 Serial 0/0 withdraw-when-down as-origin 16385
 
for this kind of thing.  Screw using BGP unless you *REALLY* need to.
(Mind you, you still want to give the multihomed net all routes
so that they can do outbound load-balancing).

The multihomed network could do this too with a route-map
that does as-path prepends, as you suggested. 

	Sean.

P.S.:
|       o   An AS number has been traditionally tied to an organization.
|           Splitting AS means loss of coherence for some customers.

So what?  Besides, IDRP will solve everything.

|	o   AS number exhaustion.

In my example, one could quite easily reuse 16385 in similar
circumstances, and call it "PROVIDER-CRUFT".  

This assumes -- reasonably, I think -- that usually the idea is to 
facilitate load-balancing of traffic, rather than to try to 
get nth party people to use a multihomed network's AS as a policy tool.

|    As has been illustrate in Section 3, it is not easy to implement
|    routing symmetry in the multi-provider Internet with the current ver-
|    sion of BGP. 

Dennis had a great message at one point to a multihomed network;
his step-by-step plan was an infinite loop, which is about right.
My thought is that things that actually care about symmetry
at allought to cope OKishly with RTT symmetry in the absence of 
physical path symmetry, which simply cannot be taken for granted
not only with IP routing, but the routing of lower-level
protocols in wide use.

In some cases, RTT symmetry won't happen (networks with radically
different designs and bandwidths, for example, where both
use closest-exit routing).  IMHO, things that can't cope
with asymmetry ought to be fixed.

Yes, this includes people trying to do diagnostics, which is
probably the hardest problem.

P.P.S.:
|    We also acknowledge Sean Doran of
|    Sprint as the first person (to the authors knowledge) to make use of
|    the AS path manipulation technique.

Oh, sure, blame me for all these kinky things.  (Shut up Paul :-) )

	Sean.