It's a shame that not everyone could be at the meeting. IMO, it was
one of those rare IETF moments in history where the room actually had
a moment of clarity and actually was mostly on the same page. And I'm
not saying this only because I personally am happy to see SL
deprecated, but because even part way into the presentation, I was
telling myself the WG is not going to be able to come to any decision
today. But the mood in the room really did seem to shift during the
discussion. I was simply stunned at the end of the meeting at the
apparent clarity and decisiveness that had been expressed in the call
for deprecated SLs. IMO, the room knew exactly what it was doing, and
was also informed about what it was doing. It was not a forced
decision.

I believe the session was broadcast on the mbone, so it must be
archived somewhere. (anyone know where?)  Anyone not at the meeting
who cares about this issue really needs to listen to that discussion
in its fullness; the minutes do not come close to capturing what
happened during the meeting.

Having said that, one of the great difficulties I have with SLs is
that I don't see a viable "compromise" or "middle ground". Every
proposal I have looked at either uses SLs in very restricted settings
(e.g, disconnected sites), or ends up having to implement the full
blown "multi-sited node" (even though try to claim otherwise). I find
scenarios that argue for a middle ground to be an unworkable illusion.

Consider a bottom line. If we say SL addresses are a reasonable thing
for (say) home networks doing autoconfig, and so on, the reality will
be that SOHO routers will find ways of creating zeroconf environments
that enable SL by default. They will then be used in practice in huge
numbers of home networks. Corporations will likewise be tempted to use
them, since they seem to provide "permanent" addresses, and
"stability" across renumbering. So, we have situation not unlike being
half-pregnant.  In practice, SLs will become widely used in many
organizations.

But, once you have wide usage in terms of independent networks
numbered using SLs, you have the inevitable problem of what happens if
a node is in two sites at the same time. I don't buy the argument that
this won't happen much and that the market can fill the gap if there
is a need (and the IETF thus won't need to do this). If widespread
useage of SL in networks becomes the reality, devices will be forced
to work in those enviornments. One simple example: at home, I have my
home network with a DSL link to the internet and also a VPN connection
to my corporate intranet. This is a *very* common operating
configuration. (Does anyone believe this won't be a common
environment?) If both the home and corporate network use SL addressing
(as they will in practice), things just won't work right unless nodes
implement the multi-sited node architecture.

Thus, I continue to come to the conclusion that we can't have it both
ways. We either deprecate SLs, or we do the full blown multi-sited
node. Doing anything in between will not work well enough in practice
(i.e., is equivalent to us putting our heads in the sand) and will
lead to interoperability problems, failed communications, and a less
robust Internet. Exactly the opposite of what I believe IPv6 is all
about.

Does deprecating SL addresses cause problems? Clearly yes. And we need
to put our engineering minds together to take those problems on as
work items and develop solutions. And in particular, better overall
solutions than we have with SLs. But I think we can do this, and we
should do this, as it will lead to a better overall long-term
technical result.

Touching on the most common reasons people seem to have for supporting
SLs:

>       - Site-locals should be retained for disconnected sites.

1) The WG will need to take an an activity to develop a solution for
   dealing with disconnected sites. I know I'll get flamed at some
   level for this, but we need a short problem statement to understand
   the basic constraints. (like, has to work out of the box, has to
   ease the pain of merged networks, etc.) But this can be 1-2 pages
   long, I suspect, and shouldn't take a year to develop. I think we
   collectively already know the constraints, we just need to write
   them down.

   We might also list some possible solutions, just to give folk a
   warm feeling that there are resonable approaches here (others have
   already posted such approaches), and that the WG isn't just putting
   its head in the sand. Personally, I'm not worried about this
   problem because I believe we can find reasonable solutions that
   don't involve SLs. This isn't a fundamentally hard problem like,
   say, multihoming.

>       - Site-locals should be retained for intermittently
>               connected sites.

2) The WG will need to take on an activity to develop a solution for
   intermittantly connected sites. Again, here I'm not that
   worried. For example, we have a lot of IPv6 addresses. There is no
   reason from a technical perspective why intermittantly disconnected
   sites can't continue using a prefix from its ISP after it is
   disconnected. IMO, there is something fundamentally broken with the
   idea that ISPs will be giving intermittantly connected sites
   different /48s or /64s each time they reconnect. 

   But in any case, there are again other approaches as well, e.g.,
   those also developed in 1) above would apply here to. So, again I
   do not think this is an unsolvable problem and am comfortable with
   the idea that we can and will solve it.

>       - Site-locals should be retained for their access control
>               benefits.

3) The WG should continue exploring the simple site boundary security
   approaches (e.g., zill/smb drafts). It's less clear to me
   (personally) how well this work will pan out, but it's definitely
   worth exploring. Besides, there isn't really agreement that SLs
   provide a good security boundary model anyway, so it isn't clear
   that SLs are an obvious answer here.

>       - Site-locals should be retained as a means for internal
>               connections to survive global prefix renumbering.

4) This is strikes me as a nice requirement, but something we don't
   have a solution for, even with SLs. We need to accept that we have
   no solution and SL is doesn't really seem to be a help here. Or, if
   folk really think there is an approach, its high time we had an ID
   to look at that actually fleshed out the details. We've had
   attempts in the past, but they have been abandoned.

Thus, in summary, I support the deprecation of SLs with the
understanding that taking this step will require that we take on some
additional work items. But I don't see lack of specific solutions to
those work items today as being a reason to delay deprecating SLs.

Thomas
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to