Short version: An attempt to clarify the current, too-loose, use
of the term "map-and-encaps" and to depict the
three current approaches to tunneling packets
in a Core-Edge Separation scheme.
I am happy to go along with the suggestion of Eric and others that
HIP is a genuine ID/Loc Split system and that LISP, APT, Ivip, TRRP
and Six/One Router are not.
Here is how I see things:
Elimination:
HIP, Six/One, SHIM6 and ILNP. (GSE?)
Separation:
Map-Encap:
LISP, APT, TRRP and Ivip using encapsulation.
Map-Forward:
Ivip's two Forwarding modes.
Map-Rewrite:
Six/One Router.
HIP (as part of a routing scaling solution) is an example of the
"Elimination" approach, as defined in:
Towards a Future Internet Architecture: Arguments for
Separating Edges from Transit Core
Dan Jen, Lixia Zhang, Lan Wang, Beichuan Zhang
http://conferences.sigcomm.org/hotnets/2008/papers/18.pdf
Elimination involves new responsibilities for each host. Bill's
"Strategy B" - or a somewhat broader version of it - seems to cover
the Elimination approach:
http://bill.herrin.us/network/rrgarchitectures.html
I think that ILNP, SHIM6 and Six/One are also part of the Elimination
class. All these involve ever host which participates in multihoming
(including those which are not multihomed themselves) managing
multihoming via the use of multiple addresses at each multihomed host.
My critique of all such schemes - an argument against pushing this
responsibility to the hosts - is:
Fundamental objections to a host-based scalable routing solution
http://www.irtf.org/pipermail/rrg/2008-November/000233.html
However, see also my next message about the scaling problems of such
Elimination schemes as every host probes reachability via each of its
addresses to the one, two or more addresses of every host it is
communicating with.
The other major approach to the routing scaling problem (not counting
trying to soup up BGP, or to change the interdomain routing system in
a more drastic manner, such as with geo-aggregation) is "Separation".
This is the approach preferred by Dan Jen et al.
Initially, the only proposals which fell into this class were map and
encapsulate proposals: APT and LISP (which came first?), then Ivip
and TRRP. The term "Core-Edge Separation" came later, I think.
Then, in early 2008, Six/One Router was proposed. ("Six One" on its
own, was an earlier, host-based, Separation scheme.) This uses
address rewriting rather than encapsulation. The current version is:
http://users.piuha.net/chvogt/pub/2008/vogt-2008-six-one-router-design.pdf
In late 2008 I proposed two related but separate "Forwarding"
approaches, as an alternative to encapsulation or address rewriting.
Both these use modified formats for the IP header, and require
upgrades to all DFZ routers. This may well be feasible, and the
benefits include eliminating encapsulation overhead and the PMTUD
problems of encapsulation:
ETR Address Forwarding (EAF) - for IPv4
http://tools.ietf.org/html/draft-whittle-ivip4-etr-addr-forw-01
Prefix Label Forwarding (PLF) - for IPv6
http://www.firstpr.com.au/ip/ivip/ivip6/
Initially, the terms "ITR" and "ETR" were used with the map-encap
approaches solely - for Ingress and Egress Tunnel Routers. This made
sense since tunnelling has traditionally been done by encapsulation
and was in all these proposals.
With the two Forwarding approaches, I keep the ITR and ETR terms for
the devices which rewrite the header and which restore it
respectively. It is still like a tunnel - a one-way tunnel as with
map-encap - although there is no additional header.
The terms "ITR" and "ETR" are not used in Six/One Router, but it is
easy to think of the rewriting roles of the Tunnel Routers in these
terms.
All these Separation schemes involve:
No changes to hosts.
A mapping system.
New devices (or additional functionality in existing devices) such
as ITRs and ETRs in the network (although with Ivip at least, it is
possible to integrate the ITR function in some sending hosts).
So Separation schemes are "Network-based" while Elimination schemes
are "Host-based".
Some host-based systems may not require changing the API and
applications, while others will require this. All these host based
systems require IPv6 - since there is not enough address space in
IPv4 for hosts to have multiple addresses.
No host changes are required for the Separation approaches, except that:
1 - Six/One Router is IPv6 only.
2 - Ivip's IPv4 EAF Forwarding approach and Ivip's IPv4 map-encap
approach to PMTUD both involve not sending fragmentable
packets longer than a certain length, such as somewhat less
than 1500 bytes. EAF doesn't send fragments either.
- Robin
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
