Draft of RIPE81++

 cengiz at ISI.EDU (Cengiz Alaettinoglu) writes:
  * 
Okay, firstly I would like to say that clearly the RS is a special
case and may in fact not fit into the generalised routing registry
description and this not a basis for the current discussions. 
Although I am not convinced just yet. Let's look at what
you say.

  * I agree with what you say in the above paragraph: The border routers
  * which peer with an RS know the address of the RS. I just want to
  * clarify a point though. 
  * 
  * In specifying policies, RS will be transparent.  That is if AS100
  * and

This is very much how you wish to interperate and implement a route
server. In protocol terms the RS in certainly not transparent. In
terms of policy yes but you still should express your policy with
respect to the RS in terms of the policy it can provide. 

  * AS200 peer with an RS, administrators of AS100 specify policies for
  * AS200, not for RS. If multi-exit discrimination is allowed in this
  * case, there are two options for administrators of AS100:
  * 	1) only specify local border router, i.e. 
  * 		interas-in: from AS200 local-address - ...
  * 	2) use RS's address, i.e.
  * 		interas-in: from AS200 local-address rs-address ...
  * 
Right but is more like

interas-in: from AS-RS local-address rs-addres <policy with respect to AS200>

or even as-in for that matter.

Now this may look odd semantically but is correct. The bottom line is
this where the routing information is derived from. The RS has a
policy to both AS200 and AS100. Of course if it is multiple image this is 
hard to express in ripe-81++ (although can be kludged). However, 
remember ripe-81++ is modelled  on the premise of
destination based policy decisons and not making use of source for
policy which is what the multi-image RS is doing.

  * There are 4 problems with the second option:
  * 1) RS is not completely transparent. (though RS' AS is).
                    
No.....Wrong way round I think. RS AS will be there or are you
gonna break the BGP protocol ?

  * 2) It is confusing since RS is not in AS200.  

No but if expressed in terms of RS-AS it is not confusing.
 
  * 3) If the address of the RS changes, many changes by many
  *    administrators will be necessary in the database even though
  *    policies stay the same.

And what is the problem ??? This is no differnet at any DMZ and a router
changes its address. Why the "pilot RS" we have at Mae-east has 12
peers, if I change the address I inconveience 12 providers but thats
life - the Internet is dynamic and volatile and we shouldn't pander to
making things like this especially easy. The RR (yes RR - not RS) 
should be an integral part of all Internet service and to be quite
honest if you want the RS to fly using a general RR you want to make 
it important that providers respond to changes and realise how import
the RR is.

I think we should be careful when looking at the RS in general.
Clearly much more understanding of what is needed in terms of the
policies wanted from the RS is in order. However, the basic argument
changes very little in terms of whether this is optional or not.

		--Tony



-------- Logged at  Thu Sep 8 19:17:37 MET DST 1994  ---------