🚀 go-pugleaf

Barry Margolin wrote:
> 
> In article <Xns944A791483E35juergennieveler@nieveler.org>,
>  Juergen Nieveler <juergen.nieveler.nospam@arcor.de> wrote:
> 
> > Barry Margolin <barmar@alum.mit.edu> wrote:
> >
> > > My guess is they'll need all those routers whether or not they do NAT.
> > > How many downstream devices do you think a non-NAT router can handle?
> >
> > Millions? Billions? With non-NAT routers, there's some huge machines
> > out there that take care of the major internet backbones...
> 
> I thought you were talking about devices that are immediately downstream
> of a particular router, not potentially many hops away.  E.g. a router
> with thousands of devices on its LAN interface.

In GSM/GPRS cellular networks, and with existing equipment, there may be
more than 100'000 phones that are only one IP hop from the access router
(a.k.a. the GGSN). Future 3G nets support over a million. There are
several L2 hops, though...

> > For a non-NAT-router, bandwidth and routing tables are the only issues.
> > For a NAT-router, you need far more memory because it also has to take
> > care of connection states and NAT tables...
> 
> I think you brought this up in the context of IP to cellphones.  I think
> the routers will need to keep track of considerable wireless state to
> implement this, and that may limit the number of cellphones that a
> particular router can handle.

Correct. There's also the firewall part. There may be a kilobyte of data
per session, plus buffers. The access router needs lots of RAM.

An important issue is peer-to-peer services between mobiles. If you
think it is easy, see what the IETF/nsis workgroup has been writing
about enabling services through NAT/FW boxes:
http://www.ietf.org/internet-drafts/draft-ietf-nsis-nslp-natfw-00.txt
(only 147kB). With IPv6, you can avoid most of that hassle.

-- Lassi