UNIX admin writes:
> The catch is that the ADSL "modem" only has one EtherNet port, so only one 
> system can be connected to the Internet. Well, no problem, IPFilter to the 
> rescue!

Have you tried connecting a hub to that Ethernet port?

That'd have to be an extraordinarily lame device if it doesn't support
more than one client at a time.  If so, I'd probably chuck it out the
nearest window rather than fooling with an extra layer of NAT.

> By NATting all LAN and DMZ packets to the external interface of the
> FW (also a non-routable IP address on the same subnet as the
> "modem"), I trick the "modem" into thinking that there is only one
> host connected, when there is in fact an entire server farm and two
> DMZs behind that one IP address. But it's still wasteful and
> overhead to perform the NAT twice for every packet coming in or out.

Given the relatively low speeds involved here, I wouldn't worry
terribly about the overhead.  I'd worry more about the functionality
of the link.  As long as it's actually working right, more power to
you.

> > At the lowest level, DSL links essentially just
> > provide a synchronous
> > bidirectional (full duplex) communications path.  In
> > order to put
> > packets on there, you need some sort of framing --
> > that can be HDLC or
> > AAL-5 plus ATM.
> 
> If I understood correctly, what they're doing is is transporting TCP/IP over 
> ATM.

Not exactly.  They're carrying multiple layers of encapsulation.  As
long as you put a few "..." in that sentence, it's right.

> > To use routers, you'd have the head-end termination
> > break out the bit
> > stream (as is done today), but instead of running
> > that into an ATM
> > switch, just run it into an access router.  The
> > router can then run
> > HDLC over the link, and PPP works fine on that.
> 
> But could they have run EtherNet all the way, or is the copper line the 
> problem for EtherNet? I mean the last mile. And once at the CO, just normal 
> EtherNet routers and fiber back to the backbone?

I don't see any particular reason they could not use Ethernet, but
it's a completely different protocol, with a completely different set
of engineering issues -- from wiring, to crosstalk, to EMI, and so
on.

The head-end is an important concern there.  The 'ideal' Ethernet
deployment would have routers at each street corner.  DSL doesn't
always look like that; instead, since it is designed to reuse existing
copper from telephone lines, it has to deal with near-end crosstalk
where all these wires are aggregated into a CO.  This is also the
reason for the 'A' in ADSL.

The low-level DSL interface constrains the sort of solutions that make
sense here.

> > be done based on PPP plus backend AAA.  The ATM
> >  mechanism gives
> > you a way to bill separately for each remote
> >  endpoint, even if
> > they're on a single DSL line (should you choose to
> > do so).
> 
> Yes, but why would we need to meter? Yes, there are still fascist-like telcos 
> out there that measure your monthly traffic, but I predict that that 
> billing/business model has no future. In the future all ADSL or SDSL or 
> whichever technology replaces it will be flat rate. If we speculate that my 
> assumption is true, what would need to be metered?

I can't really imagine what problem or constraint metering solves; I'm
merely noting that it's a concern for some providers.

I usually assume that regulatory and business issues just exist
outside of a plane in which engineering takes place.  It's not really
all that worthwhile (at least to me) to try to divine some deeper
meaning in them.

-- 
James Carlson, KISS Network                    <[EMAIL PROTECTED]>
Sun Microsystems / 1 Network Drive         71.232W   Vox +1 781 442 2084
MS UBUR02-212 / Burlington MA 01803-2757   42.496N   Fax +1 781 442 1677
_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to