Hi,

I'm new to the this list and I joined it last week so please be gentle
though I'm might ask or state something stupid.
The reason why I joined is the research work that the LISP team have
carried out - the work have inspired me and I have a proposal that
like to share with you.

In Dino's last draft the LISP v2 is put on hold. My idea is build
around DNS and I would prefer to take it further - open the door for
hosts as well so that greater scalability could be achieved. You could
call this proposal version 4 of LISP - I rather call it a hierarchical
IPv4 framework since it will require changes to the Internet topology,
changes to some ISP routers and extensions to the DNS. And it opens up
the possibility that the IPv4 stack at the hosts could be enhanced as
well.

Other influence sources are MPLS' shim header and forwarding mechanism
- but LISP starts to look more and more as a MPLS solution. Also good
old PSTN, why does it not have scalability issues though there are
more globally unique numbers than in the Internet....hierarchical
numbering architecture is the answer and thus subscribers do not have
to migrate to hexadecimal keypads. Recently I have been working with
enterprises, in order to receive a budget and a project from upper
management you have to provide decent answers to "what is the return
of investment?", "how does this improve my business?" for them. And I
have no answers to push IPv6 into the enterprises, IMHO it is a
forklift upgrade that enterprises will not take - they will use NAT
instead as long as possible.

All that has been in my "mixer" for a while, I'm trying to write a
draft about the proposal but it is time consuming and I prefer to
check at this moment if there is any major showstopper. The proposal
consist of some basic rules:

- Allocate a globally unique IPv4 block for RLOC allocations;
hereafter called the Global LISP IPv4 Block, GLIB
- Divide the current flat Internet topology into five areas; we have
today five Regional Internet Registries and it is logical to do the
split after the structure of the RIRs. Later the RIR domain can be
further divided into sub-domains.
- A subnet from the GLIB will be allocated to each RIR domain;
hereafter called RIRID
- Divide the service providers in Tier-1 and Tier-2 categories; a
Tier-1 service provider have connections to another RIR domain and a
Tier-2 provider have only connections internally in the RIR domain
- The RIR admin will allocate a host address (RLOC) for each Tier-1
provider at the local RIR domain
- Only GLIB prefixes are exchanged between the five RIR domains, i.e.
each RIR domain will have its own IPv4 realm (almost full block,
except the GLIB) when hierarchical IPv4 (hIPv4) framework
implementation is completed. When sub-domains are generated each
sub-domain can have their own IPv4 realm.
- When an IP connection is established inside the RIR-domain
(intra-RIR traffic) IPv4 solutions are used. But when the IP
connection is established to another RIR-domain (inter-RIR traffic)
the hIPv4 framework will be used
- DNS; add RLOC and weight extensions to the system
- Do not exclude extensions to the end systems (hosts), better
scalability is achieved if extensions are allowed at the end systems
- Current IGPs and BGP are still valid routing protocols; perhaps
introduce new capabilities to BGP called hierarchical BGP (which will
only allow GLIB prefixes between inter-RIR BGP peers).
-Use a shim header that contains RLOC and EID information. The shim
header is called a LISP header.

The framework will require most likely new hardware on the connections
between the RIR domains. A hIPV4 header consists of an IPv4 header and
a LISP header (that is inserted somewhere between the IPv4 header and
the payload). The LISP header has a RLOC (=IP address) and an EID (=IP
address) field. How difficult is it implement the tasks below in
hardware on per packet basis on a high speed connection between e.g.
the U.S and Europe?
a.      replace the source address in the IPv4 header with the RLOC address
of the LISP header.
b.      replace the destination address in the IPv4 header with the EID
address of the LISP header.
c.      replace the RLOC address in the LISP header with the destination
address of the IPv4 header.
d.      replace the EID address in the LISP header with the source address
of the IPv4 header.
e.      when the hIPv4 header swap is completed forward the packet upon the
destination address of the IPv4 header.

And here are the carrots, long term
Enterprises
- No need to learn a totally new protocol
- No porting of applications to new protocol
- Get Provider Independent addresses without multihoming requirement,
i.e. achieve site mobility
- Choose preferred Tier-1 provider for ingress and egress traffic with
DNS weight
- When end systems are upgraded to support the hIPv4 framework and
using PI addresses , the NAT solutions can be removed
Internet Service Providers
- No need to learn a totally new protocol
- No need to learn new routing protocols
- Remove address constraints
- Hierarchical BGP, smaller BGP table for each RIR domain
- Internal link flaps are not seen in other RIR domains, only hBGP
link flaps are reflected globally

Not much information and not sure that I have everything covered, but
I'm trying to work on draft unless there is major showstopper that has
been discussed earlier. Technically, this is quite straightforward but
it will split the Internet into RIR domains and divide service
providers into categories - then it could get ugly because
"politicians" get involved. But I guess that is the price of moving
from a flat to a hierarchical architecture.

-- patte
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg

Reply via email to