Short version: I disagree with Toni's suggestion that the interdomain
routing system should operate according to ASes, and
not be concerned with routing packets addressed to
particular prefixes to particular border routers.
This concept of "AS-based routing" or "route on ASN"
has been discussed before. I provide 3 strong
objections to it, any one of which would be sufficient
to rule the idea out of serious consideration.
There are no-doubt other objections.
Hi Toni,
In your message (2010-09-29 msg07400), you wrote about my lack of
response (and lack of responses by anyone else) to your message:
Consensus on node reference (was: Consensus on identity/location
separation) (2010-09-24)
http://www.ietf.org/mail-archive/web/rrg/current/msg07379.html
However, I have now replied to this:
Re: [rrg]Consensus on node reference
http://www.ietf.org/mail-archive/web/rrg/current/msg07406.html
You quoted your msg7379 in full and then wrote:
> OK. Half of us (who expressed opinion) are for, the other half
> are against location/identity separation.
You and I expressed an opinion in that thread - you support it and I
am against it. (BTW, each of your paragraphs is one long line in the
archives.)
> So, the easy part passed. Now the big mess and its even bigger
> clearance. Reiterating.
I don't understand this.
Then you raised a question for this new thread:
> Inter/Intra Distinction/Separation
> Inter-domain path selection and intra-domain next-hop resolution
> are unnecessarily entangled with the use of intra-domain prefixes
> in inter-domain path selection.
Yes, in the current arrangements, each interface has an IP address and
that IP address (in the destination field) is used for deciding how to
forward the packet at each router, no matter whether the router is in
the local network or outside it (i.e. in the Interdomain routing system).
I support the continuation of these arrangements.
CES architectures such as LISP and Ivip (I don't know all the details
of the latest version of IRON) are additions to the routing system in
general (primarily or entirely the interdomain routing system), which
don't affect node (host) behavior at all. These architectures
involve tunneling packets from ITRs to ETRs. LISP and Ivip do this by
encapsulating the packet in an outer header, with the ETR's IP address
in the outer header's destination field. So the ordinary local and
interdomain routers still work the same as they do today, but the
inner packet is actually being sent towards an ETR which will
decapsulate it, according to the outer destination address, without
reference to the destination address of the packet which is being
tunneled.
(Ivip in the long-term would transition to a "Modified Header
Forwarding" arrangement in which there is no encapsulation, but
the header would contain sufficient bits for routers to forward
the packet to, or most of the way to, the correct ETR.)
> So the inter-domain routing system has been filled up with
> routing domains' inner intelligence that is spilling all over.
Yes, the interdomain routing system (the BGP routers in the DFZ) need
to concern themselves with each prefix which is advertised by one or
more border routers of any AS.
CES architectures aim to provide very large numbers of multihomable,
portable, prefixes for end-user networks, which can support inbound TE
and ideally mobility, without adding much to the number of prefixes
advertised in the interdomain routing system (AKA "advertised in the
DFZ").
> But there are AS paths. They are the necessary and sufficient
> pointers for conducting inter-domain path selection; provided
> that routing domains start to convey inter-domain traffic as
> whole entities, acting similarly to nodes in intra-domain
> hop-by-hop forwarding.
>
> Thus, inter-domain path selection will be based on AS numbers
> (only) and intra-domain next-hop resolution – on locators
> (currently IP addresses).
This change to the behavior of routers would enable considerable
simplification of the interdomain routing system.
I understand your suggestion would involve something like:
Every router would somehow obtain a list of all the prefixes
currently handled by the system. Each prefix would have a
single ASN, indicating which ASN is advertising this prefix.
Then, the routers would somehow (perhaps with a BGP-like protocol)
decide how to forward packets based entirely on which ASN
advertised the prefix the destination field best matches.
Since the number of ASNs is a lot smaller than the number of
prefixes, this would significantly reduce the workload of the
route processors of all the interdomain routers.
This way, it would be relatively simple to add very large numbers of
prefixes, without the scaling difficulties which would occur if we
kept the current system and expected the BGP-based interdomain routing
system to cope with all those prefixes.
However, it is not necessarily our task to make life easier for the
routing system - especially if it involves placing extra processing,
packet number, packet size and/or delay burdens on some or all hosts.
I argue against CES (Locator/Identifier Separation) because it would
do all these things.
Your idea of routing on ASN alone has been discussed on the list
before. As far as I know, the major objections to it include:
1 - This would only be acceptable if, for all ASes, the AS would
be happy to accept packets for any of its prefixes on any of
its routers.
This is most definitely not the case for most ASes, AFAIK.
ASes want and need to be able to advertise particular prefixes
on particular border routers. This is how they share the load
and achieve various other objectives.
2 - Your proposal would only work if it could be assumed that each
AS was actually a cohesive network with all its border routers
(BRs) able to easily send packets anywhere inside the network.
This is not a reasonable assumption. I could be an AS and have
routers at a dozen sites all around the world, with one or more
BRs at each site - but no links between these sites. I might
want to advertise some of my prefixes at one site, or one BR at
one site, and other prefixes at another.
Your plan would compel me to accept incoming packets for any
of my prefixes on any of my BRs at any of my sites. But this
would be completely at odds with what I want and need to do.
3 - Your plan would mean that an AS would have no control
whatsoever over where its traffic enters its network. Let's
say the AS has only 2 BRs. There's nothing to prevent the rest
of the routers deciding to send all the packets to BR-1 and
none to BR-2.
Likewise, an AS with 100 BRs couldn't prevent the rest of the
routing system deciding to send all packets, addressed to any
of its prefixes, to a single BR.
There are no-doubt other objections, but I think any one of these is
sufficient to rule out the possibility that "routing on ASN" is either
a good thing in principle, or has a hope of being accepted by networks.
> Do you agree that inter-domain and intra-domain routing have to be
> separated?
They already are separated. Between ASes BGP is used. Inside ASes,
another routing protocol is used.
- Robin
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
