Re: [RRG] Re: Practical Proposals vs. endless theoretical discussions

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

Hi Ran,

You wrote:

> Earlier, Robin Whittle said:

> % Most talk is of clean-slate designs, flat routing, ILNP etc.
> % which have nothing to do with our need to have something
> % attractive to end-users in the next 4 to 8 years to solve
> % the scaling problem in today's IPv4 Internet.

I responded to the ILNP-specific parts of your message in a separate
thread.

>   You would be so much more credible in your quite numerous
> notes to the RG list if you would actually would read and study
> topics before you make bold and incorrect assertions about them.

This is a broad-brush condemnation of my efforts, apparently
extrapolating from your disappointment with my perceived
misunderstanding of your own proposal.  I think it is bad form to
make such accusations on a mailing list without providing suitable
evidence.  You don't bother to correct my apparent misunderstanding
- your complaint would be more credible and useful if you did.

I think I do a pretty good job of contributing to progress in
scalable routing, even ignoring my own proposal.  My critiques of
APT and Six-One Router have been welcomed by the designers of these
proposals.  The current Six/One Router proposal was prompted in part
by my critique of the initial version.  I understand that my
suggestion that there be only one APT island is being considered.

I have similarly made extensive critiques of LISP - for instance of
the long-path problem of LISP-ALT - which other people have
concurred with.

You also assert that I don't read the documents of proposals I
discuss.  I most certainly do.

I didn't read your documents in detail because your proposal
involves host changes, and because it only provides benefits when
both hosts are upgraded.  In my view, this makes your proposal a
non-starter in the time-frame within which I think we need to
provide a solution, such as 4 to 8 years.

LISP with PTRs or Ivip are both vastly better than your proposal in
these regards - they don't require host changes and they can provide
multihoming and portability for packets sent by ordinary hosts in
non-upgraded networks.  APT can in principle do the same - although
I think would be best if there was only one APT island.
Furthermore, these can work with IPv4 and IPv6, while ILNP is IPv6-only


> A large volume of notes is not an effective substitute for a
> very few, short, well-written, and technically correct notes.

I reckon my signal to noise ratio on this list is pretty high.  As
long as that is the case, I don't have to apologise for the quantity
of the signal, since it is provided without charge.  As for writing
longer messages than most folks, that is better than writing the
same material in a bunch of shorter messages.

To me, constructive critiques of of actual proposals is a very
important form of progress - however, for many months, maybe since
the end of last year, I can't remember, all such discussions have
been out of scope on this list.  I and others persist, since there
is nowhere else to discuss these things, but we are sometimes told
off for "discussing proposals":

  http://psg.com/lists/rrg/2008/msg01285.html

and when I complain about these restrictions, I am told that is out
of scope too:

  http://psg.com/lists/rrg/2008/msg01864.html

I reject suggestions such as you once made to me on the list that I
should ignore engineering details and focus on "architecture".  One
without the other is useless.  This is Internet scalable routing, so
one should expect a lot of detail in most meaningful discussions.


>   Perhaps I am guilty for not having engaged in as much
> "advocacy" as you have been doing, but it seems at best
> negligent on your part to make the statement quoted above.

I don't think it is negligent of me to state:

> % Most talk is of clean-slate designs, flat routing, ILNP etc.
> % which have nothing to do with our need to have something
> % attractive to end-users in the next 4 to 8 years to solve
> % the scaling problem in today's IPv4 Internet.

I stand by this - as long as it is assumed we need to solve the
routing scaling problem in IPv4, or whatever problem arises in IPv6,
within 8 years.

Do you think we have more time than this?  If we do, then that makes
it easier to introduce my Forwarding approaches, since over a that
time-scale it would probably be quite practical to upgrade those
current routers which would still be in service in 2016, and to have
the new capabilities built into new routers.

If you think we need the new solution to be widely adopted in 8
years or less, do you seriously suggest that ILNP could be that
solution?  You would have to convince everyone to move to a new kind
of IPv6.  There are (I guess) a billion end-users.  IPv4 is the
world's greatest IT system and people use it because everyone else
uses it.  IPv6 isn't much use to most people until everyone else is
using it - so I can't see how you can assume you can change the
whole world over to IPv6 inside 8 years.


>   Separately, it is utterly unrealistic to believe that the
> deployed IPv4 Internet is going to accept *any* major changes
> at this point.

I don't think it is unrealistic at all.  The routers are being
replaced every 4 to 6 years or so.  LISP, APT, Ivip, TRRP etc. don't
require host changes.

> I spend a great deal of my time talking with
> users on various continents of the globe.  They have a very
> consistent message that major changes to their IPv4 deployments
> (e.g. site border router reconfigurations to enable any new
> routing protocol or to enable any sort of new tunnelling) are
> not going to happen.  The notion of "fixing IPv4", frankly,
> is indeed a lost cause and we would better serve the Internet
> by trying to ensure that IPv6 deployments can move to some
> better architecture, whatever that might turn out to be.

Sure, but there's no way they can sell an IPv6-only service to most
end-users, because it has no value, since it doesn't allow them to
communicate directly with all other Internet users.

The long-held fantasy of IPv6 proponents is that their imaginary
IPv6 egg will transform into a real IPv6 chicken, and/or their
imaginary IPv6 chicken will lay a real egg.  This magical
transformation is now supposed to happen in the real world because
networks are facing increasing pressure to use IPv4 address space
more efficiently, or because it the IPv4 DFZ routers need more RAM
and CPU power to cope with the bloating DFZ routing table.

IPv4's routing scaling problem can be solved nicely with LISP, APT,
Ivip, TRRP etc. - with a lot less expense and trouble than moving
everyone to IPv6.   These will also enable much better use of
address space.   I agree its not ideal, elegant or going to last
forever.  However, I think we are still a long way away from
applications and operating systems so fully supporting IPv6, and
from there being any real reason to adopt IPv6, that IPv6-*only*
adoption will become large enough to reduce IPv4 usage.

>   Finally, for some long while now it has seemed to me that
> you confuse the concept of an "IRTF Research Group" (which
> this is) with an "IETF Working Group" (which this is not).
>
>   Research Groups are supposed to be deliberate, careful,
> thorough, and are supposed to look at clean-slate architectures
> in the course of their *research*.   Theoretical discussions
> are explicitly within scope for any IRTF Research Group.

I wasn't suggesting that theoretical discussions were out of scope.

What I was pointing out is that by discouraging discussion of
potentially practical proposals, the RRG has not helped much in the
development of anything which will actually be useful in the next 10
to 15 years.

I am not suggesting ILNP shouldn't be discussed.  However, I think
there needs to be some kind of agreement on a time-frame, or
multiple time-frames.

Maybe your ILNP proposal would only be practical in a very long
time-frame, like 10 or 20 years.  That's fine to discuss ILNP as if
it was practical, if we agree we have 10 or 20 years.  I think it is
a fundamentally good idea to split the address space into network
and host identifiers as you do.  Could you do a comparison between
SHIM-6, ILNP, GSE and 8+8 schemes?

If we agree we don't have as long as 10 or 20 years, then we also
need to discuss proposals which are *practical* in the conditions of
the next few years.  My complaint is that this has been sidelined.


>   By contrast, IETF WGs are developing engineering specifications
> to be considered for standardisation.  Both roles are important,
> but those two roles are very very different.  It seems to me that
> your frustration is primarily that this RG is not a WG.  To the
> extent that is correct, the issue is with one's incorrect
> expectations, and is not a legitimate issue with the operation
> or behaviour of the RG.

It is not my understanding that the RRG should only consider
complete new Internet architectures.  My understanding is that we
also need to consider modifications to the current architecture of
IPv4 and IPv6.  Since we can't force anyone to adopt a new
technology, we have to make the new system firstly easy to adopt and
secondly attractive to adopt.  It is only going to solve the routing
scaling problem if the great majority of end-users adopt it.

There's no way end-users are going to abandon the Internet which
works for them and invest in an Internet which doesn't yet (and
won't for a long time) just because of any concern they might have
about the memory and CPU burdens of DFZ routers.

If this was an academic question: "Design a new scalable Internet
architecture free of constraints about backwards compatibility" then
it would be fine to ignore mere tweaks such as LISP, Ivip etc. and
the practical constraints which prompted these proposals.

However, my understanding is that there is a big problem coming up
involving the world's biggest IT system and that it is our job to
find a solution.  This involves the solution being freely adopted by
existing end-users.  For that reason, I suggest that any such
solution, such as yours, or IPv6, which involves end-users changing
their host operating systems and applications is a non-starter.
This is especially the case for IPv6 or your modified form of IPv6,
in which the average end-user would find the new system pretty
useless until such time as essentially every other end-user adopted
it as well.

  - 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