semantics ambiguity with interas-in/out: clarification sought

Tony Bate (Tony.Bates at ripe.net) on August 26:
> Descriptions of interas policies do  not  replace  the  global  pol-
> icy  described   in  as-in, as-out and other policy attributes which
> always describes the  global  policy  between  the  two  ASes.   The
> interas-in/out  attributes only specify local variations to the glo-
> bal policy described in the other attributes. If the  global  policy
> mentions  more  routes  than the local policy then local preferences
> for these routes are assumed to be equal for all links.

as-in: AS1 pref1 polexpr1
interas-in: AS1 lid rid pref2 polexpr2

with the above description, the routes imported through (rid, lid)
connection is:
	polexpr1 or polexpr2
with the following preferences:
	pref2 polexpr2
	pref1 (polexpr1 setminus polexpr2)

Is this right? (If it is, including above more formal formulation
might help more implementors).

Eg.
as-in: AS1 100 AS2 AS3 AS4
interas-in: AS1 lid rid 90 AS2 AS4 AS5

that is import routes in
	AS2 AS3 AS4 AS5
with the following preferences
	90 AS2 AS4 AS5
	100 AS3

Is this right?

More complicated E.g.
as-in: AS1 100 AS2 AS3 AS4
interas-in: AS1 lid rid 90 (AS2 AS4 AS5) AND NOT AS3

Note that "(AS2 AS4 AS5) AND NOT AS3" is equivalent to "AS2 AS4 AS5".
Hence this example is identical to previous one and there is no way to
avoid importing routes in AS3.

Actually with the current definitions there is no way that the local
policies can be a restricted version (subset) of global
policies. (Unless the global policy is "NOT ANY" and everything is
done through local policies. Of course, this is not intended.)

Is this right?

Hence, Merit's example:
epg at merit.edu (epg at merit.edu) on September  6:
> Guess, our thoughts on this are colored by working with folks like
> BARRNET which have always had at least two gateways that peer with
> our one gateway.  BARRNET has then preferred to accept one set of
> nets at one router and the rest of the nets at another router.

can not be implemented using interas-in/out unless the global policies
of BARNet is accept NOT ANY and the local policy over one connection
is to accept the first set of networks, and over the second connection
is to accept the second set of networks.


The simple fix to this problem is that the local policies replace the
global policies. That is the routes imported through (rid, lid)
connection is:
	polexpr2
as opposed to 
	polexpr1 or polexpr2.
This is a VERY SIMPLE solution. No confusion for implementors or
administrators. 

This fix makes the interas-in/out policies more global which some of
you do not like. To me, interas-in/out policies are as global as
as-in/out.

Any thoughts.

Cengiz

-- 
Cengiz Alaettinoglu       Information Sciences Institute
cengiz at isi.edu            University of Southern California




-------- Logged at  Wed Sep 14 19:23:44 MET DST 1994  ---------