xDSL Internet Access

Mon Jan 26 00:09:51 UTC 1998

NANOG Folks;

Some of you engineers need to take a stint in product development to see
how alternatives like xDSL play out.  :-)  This note is a bit long, but if
you are interested in how DSL services work, read on. I'll make it worth
your while.

First, Microsoft/Intel/Compaq don't care who provides the xDSL and Internet
services -- they will have nothing to do with the service delivery. They
simply want to sell one type of xDSL modem and they want to be able to
integrate these modems into their systems and they want simple installation
(hence the interest in "splitterless" xDSL modems). This troika is simply
trying to get the phone companies to pick ONE DSL standard or set of
compatible standards NOW so they can sell hardware SOON. Money will be
forthcoming to support these commitments, much as MS invested in Comcast
and has offered financial incentives to use WinCE in set-top boxes.

xDSL-based Internet services have three components:

1) the copper cabling
2) L2 aggregation
3) Internet service

Only the incumbent local exchange carriers (ILECs), or as they are often
called, the regional bells (including GTE), can provide the copper circuits
through tariffed unbundled services. I suppose you could pull copper to
every residence and small business, but I don't know of any really serious
proposals to do so.

The L2 aggregation can be done by the ILECs or by the competitive local
exchange carriers (CLECs). The non-ILEC DSL service providers MUST be
CLECs, else they can't interconnect with the ILECs. CLECs can have a
variety of interconnection agreements, but *any* CLEC can get *some* kind
of interconnect agreement with *any* ILEC, by law. If not, sue and make
your business plan that way (what I call The Lawlor Plan B). ISPs often
cannot get co-lo space, but CLECs always can, just not always on the most
favorable terms.

The Internet service can today only be provided by non-ILECs, since the
ILECs are, for now, prohibited from carrying data traffic between LATAs by
the Telecomm Act and the MFJ before that. But that will change soon.
In the case where the ILEC provides the DSL service but not the Internet
service, the ILEC installs DSL access multiplexors (DSLAMs) in their
central offices, aggregates DSL traffic from their circuits and pipes it to
the various ISPs. The subscribing ISPs are buying the aggregated DSL
service from the ILECs, typically via one or two ports.

In the case where the CLEC provides the DSL service, the CLEC rents
unbundled conditioned copper circuits (not alarm tariffs) from the ILEC.
The CLEC rents co-lo space for its mux equipment or else it backhauls the
DSL circuits to a nearby facility (thereby reducing the DSL service radius
from that central office). The CLEC backhauls the traffic from the many COs
via a local network. The CLEC then sells aggregated DSL service to the ISPs
via one or more interconnects to the local CLEC network. Of course, if the
CLEC is national, then they might offer access to a national DSL service
via one or more access ports. (It could happen.)

Now, an ISP can become a CLEC and provide DSL-based Internet access, but
the ISP needs to lease the copper circuits from the ILEC and install the
mux equipment in COs like a CLEC. Realistically, you can't backhaul DSL
very far, so trying to avoid the co-lo space is difficult, unless you can
rent facilities next door to the CO.

Depending on whether the ILEC or CLEC is offering the aggregated DSL
service (or you are doing it yourself as a newly-franchised CLEC), there
are choices as to access method. The ILECs are trying to preserve their
investments in ATM networks, so they would prefer you to connect via the
ATM (or SMDS) networks that they already have in place. Then they can bill
out the depreciation of the copper and ATM networks to you, keeping the DSL
price up. The CLECs just want to satisfy the customer, so they may offer
you an Ethernet port or a DS-1 or DS-3 cisco HDLC circuit. You usually have
to pay a port fee and pay for the access circuit and then pay for the DSL
circuits as ordered.
Not every ILEC offers the aggregated DSL service to any and all ISPs.
Pacific Bell started out that way, but SBC has different ideas, so I would
expect the DSL services in SBC/PacBell/SNET territory to change over time.
GTE has no legal restrictions at all, so they shouldn't be expected to
offer DSL services to competitive ISPs. If the ILEC doesn't believe in
"open and equal access" for their DSL services, they will use DSL as access
for their own ISP subsidiaries and will not offer DSL to competitive ISPs,
who will have to buy aggregated DSL services from CLECs, if any. 

In this case, the ILECs don't need to use ATM or any other L2 muxing
technology if they don't want to, since they will aggregate all of their
DSL circuits to only one IP network. However, with the inter-LATA
restriction still in effect (for now), they need a wholesaler ISP to carry
the traffic inter-LATA and pay for all the backbone expenses. If the
"repeal" of the '96 Telecommunications Act holds, then this inter-LATA
restriction will be relaxed, the ILECs will enter the long distance voice
market (just as the long distance voice market peaks and starts to
decline), and they will build or buy an IP backbone and do the whole
DSL/ISP thing end to end. At that point, the ISP's only choice is via a
CLEC, if any, via DS-1s, or as a CLEC doing self-xDSL.

xDSL has other problems, including bandwidth v distance curves that aren't
very good for Internet access margins and a relatively high cost compared
to POTS and cable, but still somewhat lower than T-1 (which is usually HDSL
technology). Prices range from $40 to $50 per month for so-called
telecommuter services (no IP backbone included) up to $250 per month or so
for 1Mbps asymmetric, getting up into the T-1 price range.

Once a large number of residential and small business sites are getting
anywhere from 384k up to 10M, the World Wide Wait is going to be a real
deflator of consumer expectation regarding the performance of their new
high-speed access service. Milo Medin of @Home properly anticipated this
scenario and long ago designed a proxy-caching scheme for the @Home
eyeballs that the content providers can live with. Most other ISPs are
either still throwing expensive backbone bandwidth at the problem or
struggling to get 25% to 35% cache hit rates out of proxy-caches. 

As Paul Vixie pointed out, you need to interconnect many proxies with some
hierarchical inter-cache communication and cover large populations of
browsers to get hit rates up to perhaps as high as 50%. You also need to
install proxies transparently in the data stream to avoid "minor
configuration changes" as Intel calls it and that takes expensive high
performance hardware. It is a rather poor business model, since almost no
one (Intel excepted) expects the consumers to pay for better performance,
so ISPs have to pay for proxy-caches with expected backbone bandwidth
savings. Still in all, MirrorImage, Inktomi, Cisco, and CacheFlow should
make some money on selling proxy-caches. Enterprise proxy-caches should
work better if everyone on the intranet is looking at the same small data
space (perhaps an internal web site).

Better performance will have to wait for someone to shim some sort of real
data architecture into the Web either through protocols (did I hear someone
talking about URNs five years back?) or some other high-performance web
hosting service (with mirrors or mirror-like servers).

--Kent

Kent W. England                   Wireline->            fon:650.596.6321
VP of Technology                  Receptionist->        fon:650.596.1700
GeoNet Communications, Inc.       Facsimile->           fax:650.596.1701
555 Twin Dolphin Drive            Email->             mailto:kwe at geo.net
Redwood City, CA  94065           Web Site->          http://www.geo.net
                                  Wireless Text-> mailto:pagekwe at geo.net
PGP Key->  http://keys.pgp.com:11371/pks/lookup?op=get&search=0x6C0CDE69