RE: ocean: do not boil

[Margaret Wasserman <mrw@windriver.com>]

Hi Hesham,

> => I believe continuous access is needed. a PDP
> context will be up when the device is turned
> on and remain up pretty much the whole time.
> Some operators will configure things differently
> (e.g. if idle for x hours remove PDP context), but
> I believe people want to be "always on and connected"
> (not meant to be a mrketing stunt!). This is simply
> because once you lose your PDP context you're not
> reachable.
"Reachable" in what sense?  I can see this argument for
the IPv6 PDP Context, since you might run a peer-to-peer
service over IPv6 for messaging, or whatever...

But, what advantage is there to having an IPv4 PDP
Context up all of the time?  If you are behind any type
of NAT (IPv4 NAT or NAT-PT), you won't be reachable
from the outside, anyway.


>   >                  - How many simultaneous IPv4 & IPv6
>   >                          connections are expected?
>
> => Depends on the user, this is orthogonal to the
> PDP context(s).
Right, but this is key to the scaling issues for any NAT
solution.  The "10 million" nodes (or 100,000-500,000 nodes)
number is much less interesting, from a NAT scaling perspective,
then the number of simultaneous communication sessions for
which the NAT box will need to maintain state.

> => NAT-PT can handle them in a single address space. Of course
> not in a single box.
Okay, I agree.  My use of the term "address space" was not correct.
I probably should have said "address pool" or "address range".

We're in agreement, though, that a single NAT box won't be able
to handle 10 million nodes.  Given that fact, I'm not sure that it
matters whether the end-hosts behind the NATs are numbered in
different ranges of the same address space (as they could be with
NAT-PT), or in private address spaces (as with IPv4 NAT).  Both
address spaces would be private from the point of view of the
IPv4 Internet, requiring translation into globally routable IPv4
addresses.

>   >
>   >          (1) There will be internal routers (GGSNs) at a
>   >                  density of one per ~100,000-500,000 nodes.
>
> => Correct.
Okay, now we're talking!  This starts to bring the problem into
a range that I understand how to solve...

>   >          (3) The nodes will only need occasional access to
>   >                  IPv4 & IPv6 services.
>
> => I think continuous accessibility is required.
> We don't want to tear down PDP contexts and start
> them again too often.
Do you know how potential 3GPP operators think about this?  I have
heard different things from different equipment manufacturers...

However, I'll accept that we want a solution that can handle
always-on IPv4 and IPv6 access to every end-node, even if that
isn't how all of the networks are deployed.

Are you willing to accept that it would probably make sense, in
the 3GPP topology to position the NATs (or either type) in or
just behind the GGSNs, rather than having a single (set of) NAT
box(es) between the full 10 million node network and the rest of
the Internet?

Margaret