Re: [RRG] Re: Billions of micronets / EID prefixes

[Robin Whittle <rw@firstpr.com.au>]

Short version:  Let's design the core edge separation scheme so
                in technical and business terms it is quite happy
                handling 10^10 micronets.  Then, with a single
                architectural enhancement, this can be the basis for
                global mobility as well as for solving the routing
                scaling problem - by using the TTR approach to
                mobility.  The TTR approach is not in itself an
                architectural enhancement to the Net, since it
                doesn't create a conceptually new kind of address
                arrangement - it simply extends the core-edge
                separation scheme's SPI arrangement to mobile
                devices.

                Making the core-edge separation scheme which can't
                handle more than 10^7 micronets is only a factor of
                10 to 40 more than the capability of the BGP
                architecture.

                We should aim for three or four decades of gain
                if we are going to the trouble of such a major
                architectural enhancement - a factor of 10 to 40
                is way too small.



Hi Brian,

Concerning using a core-edge separation scheme as the basis for
global mobility for 10^9 to 10^10 micronets, you wrote:

>>   http://www.firstpr.com.au/ip/ivip/TTR-Mobility.pdf
> 
> It's my opinion that it would be a fundamental engineering error
> to mix this problem (scale 10 billion) with the problem of
> multihoming medium and large sites (scale 10 million).

I think it is not necessarily an error to do so.  There certainly
needs to be debate about how to solve the various problems.  If we
assume IPv6 - or at least something other than IPv4 - is widely
adopted and assume that a core-edge separation scheme is working, or
at least developed and ready to go, for 10^7 non-mobile micronets,
the question is how best to provide mobility for 10^10 devices as
they roam across different access networks.

In the above-linked paper, Steve Russert and I propose using the
core-edge separation scheme as the top level of a 3 level global
mobility system.  Level 2 is a new layer of Translating Tunnel
Routers (TTRs) and their tunnels to Mobile Nodes (MNs).  The TTRs
behave exactly like ETRs to the core-edge separation scheme.
Two-way tunnels are established by the MN to one or more TTRs.  The
third level is the existing in-network mobility arrangements, such
as those which enable a MN to retain one care-of address while it
connects physically via various base-stations, access points or
other wired or wireless mechanisms.

Consider the following situation:

  Level 1 (core-edge separation scheme of ITRs, ETRs and mapping
  system)  exists, and works OK for 10^7 micronets, with update
  rates driven by portability, multihoming and TE.

  Level 3 exists - Mobile IP - in a variety of forms, in most or
  all wireless access networks.  Many other access networks have
  no mobility functions.

  The goal is to provide global mobility - to any access network -
  for up to 10^10 generally mobile devices.


One approach is the TTR approach, which gives the MN a stable IP
address, or /64 or whatever, no matter what access network(s) it is
currently using.  It makes full use of the existing Mobile IP
techniques, without requiring any direct knowledge of them, or hooks
into those systems.  It uses the ITR-ETR system of the core-edge
separation scheme to get packets from ordinary hosts (both in
networks with ITRs and in networks without ITRs, by using OITRDs -
Open ITRs in the DFZ) to get packets from the Correspondent Nodes
(CNs) to the TTR.  Then the TTR has two-way tunnels from the MN.

The TTR emits packets sent by the MN, which means that the MN
doesn't need to try and sent packets from its own stable, portable,
SPI (Scalable PI) address via whatever local access network it is
currently using.

The TTR approach places no extra burden of complexity on the access
networks - it works with the MN in any access network at all,
including where the MN is behind one or more layers of NAT.

Likewise, the TTR approach places no extra burden of complexity on
the ITR-ETR system of the core-edge separation system - since the
TTR is indistinguishable from an ETR.

The TTR approach could be implemented without any changes to the
core-edge separation system or to the local access networks' Mobile
IP arrangements.  The TTR approach could be done as a single
IETF-standardised arrangements with multiple companies running their
own TTR systems.  For instance, the IETF could standardise one or
more MN<-->TTR tunneling and management protocols.  Alternatively,
multiple TTR companies could implement the MN<-->TTR arrangements in
their own ways.

The burden a successful TTR arrangement would place on the core-edge
separation scheme is indistinguishable from the burden which would
occur if the scheme applied to far more than 10^7 non-mobile micronets.

The current BGP approach has serious problems with the costs of
advertising a prefix, and altering that advertisement, falling on
DFZ router operators worldwide, without any payment from those who
benefit from the advertisement - the end-user network.

*If* the core-edge separation scheme is set up in a similar fashion,
then I think your argument would have more force:  The core-edge
separation scheme would be more efficient at handling micronets than
the current BGP system is at handling advertised prefixes.  In both
cases, the burden of prefixes or micronets falls on parties who do
not gain benefits from these prefixes or micronets - however the
core-edge separation scheme is more efficient and can run OK to 10^7
 micronets, whereas the BGP system can only cope with 10^6 or less.

*If* this was the case - which is not the plan with Ivip - then all
this new architectural arrangement has bought is a factor of 10 to
40 increase in the number of prefixes/micronets which can be
tolerated.  In that case, I would agree that it would be highly
questionable, or unworkable, to say "OK, let's stretch this
core-edge separation scheme a factor of 10^3 more!".


However it is not the intention with Ivip to merely expand the
number of end-user PI address prefixes (in the form of micronets)
which can be tolerated in this unbusinesslike arrangement by a
factor or 10^1 to 10^2.  It is to expand the number which can be
handled by some factor like 10^4 to 10^5, to 10 billion or so, in a
way where every user who has a micronet fully pays their way in
terms of the costs the micronet, or any changes to the micronet's
mapping, places on other parties.  (Likewise, end-users will pay
their share of the costs of running OITRDs.) The plan is to make a
highly efficient technical arrangement for handling mapping changes
which can be run as a profitable business, in a distributed,
competitive, manner, while remaining technically unified thanks to
IETF standards.

In that case, using the core-edge separation scheme for 10^7 mobile
end-user micronets is just good business - the system can be scaled
to cope with it and the more the merrier.   I have written in the
past that global mobility is a highly valuable mass-market service
which could be provided profitably by multiple TTR companies - and
that in fact we can probably pay for most of the core-edge
separation system, over time, by it being a part of the lucrative
mobility industry.

When evaluating core-edge separation schemes, several questions
should be considered, including:

1 - Does the scheme attempt to handle mapping updates like BGP -
    with no way of charging the end-user?  If so, then it
    just perpetuates the current problems which bedevil inter domain
    routing, albeit in a more efficient and larger scale.

2 - Does the system technically gag at some limit like 10^7
    micronets?  If so, then this is only a factor of 10 to 40
    gain over the current BGP system.

3 - Considering that the TTR approach to mobility doesn't require
    frequent mapping changes for most end-users, but would - if
    successful - involve vast numbers of users, could the core-edge
    separation scheme scale to 10^9 or 10^10 micronets?  If so,
    this would be a 10^4 to 10^5 gain over the previous architecture
    - much more befitting a significant architectural enhancement.


I think it would be a failure of the imagination to assume that a
well-designed core-edge separation scheme couldn't cope with 10^10
micronets.  To do so, it must have some arrangements for the users
to pay their way.  Ivip has such arrangements.


I am not saying that the TTR approach, with its portable IP address
or prefix, is necessarily the only way to do global mobility.  I
don't know any other way at present.  I assume there will be a huge
market for being able to retain session continuity - and ideally an
IP address or prefix - when changing from one access network to
another.

I think it would be best to make the core-edge separation scheme
scale nicely to 10^7, so the TTR mobility scheme can solve the
global mobility problem.  The TTR mobility scheme is not an
architectural change to the Internet like the core-edge separation
scheme must be.  It is simply some stuff which is added in the
middle, which doesn't require any changes in provider or edge
networks.  There could be 10 different TTR schemes, all with their
own TTR<-->MN protocols.  As long as the TTRs behave like an
ordinary ETR to the one global core-edge separation scheme, then
this will work nicely.  Ideally there might be just one set of
protocols between TTRs and MNs, but my point is that it is not an
architectural change to the Internet to introduce one or more TTR
schemes, once a suitable core-edge separation scheme is in place.

The core-edge separation scheme is an architectural enhancement,
because we are expecting most end-user to adopt a new form of
address space which requires support from the core-edge separation
scheme to be useful.

My aim is for there to be one architectural enhancement in principle
- in detail, it would be somewhat different for IPv4 and for IPv6.
This would solve the routing scaling problem with a core-edge
separation scheme which could be expanded over time to 10^10
micronets.  Then the global mobility arrangements will follow
naturally, along the lines of the TTR scheme.

If the core-edge separation scheme runs out of puff at 10^7 or for
some reason couldn't be used with a mobility scheme such as the TTR
scheme, then it would probably be necessary to have *two*
architectural enhancements: one for these 10^7 non-mobile end-user
networks and another for the 10^10 mobile end-user devices.

  - Robin


--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg