8/10/98 article. There's No Stopping IP [Part II]
[Series of articles. ASND references. I've posted just the first article]
nwfusion.com
By Susan Breidenbach, Network World, 08/10/98
The edge is the issue
New and expanded fiber capacity may have
carrier backbones under control, but getting to
and from the backbones is where oversubscription
occurs, causing real problems. If carriers don't
oversubscribe the first hop, you can get PSTN
voice quality from IP telephony. If they do, the
initial router drops packets.
So the big infrastructure challenge is at the edge.
The Telecommunications Act of 1996
hasn't resulted in a lot of competition because the
incumbent local exchange carriers (ILEC) and
interexchange carriers (IXC) have fought each
other to a stalemate. More competition in the
local-access market would attract a lot of private
capital for building out the last mile.
The traditional LECs are doing quite well, largely
because of the growth in second lines to homes.
Their long-term prospects are more questionable,
but local access is a tough market for competitors
to crack.
IXCs face more near-term threats because
long-haul service could become a real commodity
in the future and erode their margins.
"The TCI acquisition shows how worried AT&T
is about all this," Bellcore's Lucky says. The
merger gives the long-distance giant TCI's cable
TV infrastructure, which it can use for the last
mile, cutting out the PSTN. AT&T is expected to
run IP over these connections.
One of the inhibitors that all incumbent carriers
face is their huge installed bases of PSTN
equipment. A lot of it is quite new, ironically
having been purchased to meet the demand for
Internet access. It's not clear whether telcos can
get any real investment life out of this equipment,
which they typically depreciate over many years.
While depreciation schedules historically stretched
to upwards of 30 years, they have accelerated,
down to as few as nine years. But even that is an
eon in Internet time and stands in stark contrast to
the three-year depreciation schedules for many
data switches. In short, if forced to replace circuit
switches before they are fully depreciated, carriers
will essentially be paying for equipment they no
longer use.
Depreciation inequities represent a significant
handicap to incumbents, but some experts say it
will be less of an issue once competition really
bites. "Meanwhile, it's easy for them to say they
can't do certain things because of depreciation,"
says John Matthews, principal consultant for
Ovum, a London-based market research and
consulting firm.
Another inhibitor is that while even modestly sized
businesses often have persistent IP connections,
there is no such thing over the last mile to
residences. There are clearly opportunities for
new access technologies, especially since a lot of
people are getting second lines that could easily -
and even preferably - be IP-based. But a huge
investment is required to switch over this last mile.
"There is no shortage of funds from the investment
community, since people believe that this is going
to happen," ITXC's Evslin says. "If anything, too
much funding is available, because it leads to some
crazy things."
ILECs and competitive LECs are touting a
confusing array of competing access options,
ranging from traditional modems and ISDN to
digital subscriber line (DSL), cable modems and
spread-spectrum wireless technology. The
front-runner for future buildout right now is xDSL.
"Last-mile services to customers must be
integrated, providing both voice and data, because
the backbone is integrated," says Martin Taylor,
chief technology officer for DSL start-up
CopperCom, Inc. in Cupertino, Calif. "Modems
and ISDN are too slow, and cable and wireless
aren't there yet. DSL is the only complete solution
for the last mile."
However, even if DSL meets the most optimistic
expectations, it will hook up a mere one million or
so subscribers over the next couple of years.
There will still be a vast number of people coming
in over analog modems, and those calls will have
to be circuit- switched.
"The winner will be the service provider that
hooks into the legacy edge network, which has an
aggregate $20 billion to $30 billion in assets,
including 130 million copper pairs," says Ron
Vidal, senior vice president of new ventures for
Level 3 Communications in San Francisco. "That's
the big opportunity in the voice business right now,
and the Internet guys just don't get it. They spend
all their time worrying about sharing agreements in
the core, and little or no time thinking about
peering at the edge."
Companies that want to provide IP telephony
services need to get organized as co-carriers so
they can plug into the telco network as peers
rather than customers, Vidal says. As peers they
have access to telephone numbers and the
Signaling System 7 (SS7) call signaling
infrastructure, and can co-locate equipment in
telco central offices (CO).
No price list
Carriers also have to figure out how much it'll cost
to run converged IP nets and what to charge for
services. Nobody claims to have the answers.
Before, pricing was based on time and distance,
but that doesn't make sense in the new order,
where exponential expansion of network capacity
is collapsing time and space. Because certain
overhead elements are constant regardless of call
length, it doesn't cost carriers twice as much to
handle a call that is twice as long - especially if
plenty of available bandwidth is just sitting there
waiting to be used. And the availability of
increasingly cheaper bandwidth means it doesn't
cost that much more to send something across the
country than across town, even on
circuit-switched networks.
Traditional telcos are also caught in a pricing
dilemma. They don't dare cut prices across the
board and thus lose revenue from customers who
aren't price-sensitive, or who don't have time to
think about it. Similarly, the telcos face the danger
of cannibalizing their own installed base, so it's
hard for them to proceed aggressively. The
newcomers don't have this problem.
The cost of customer care is one of the real
imponderables that hangs over the whole
convergence issue. As more customers come
aboard, the customer base gets increasingly naive.
The first users were technologically savvy with
high-end machines, and service providers had
enough trouble even with them.
"People can't stop to figure this stuff out," Lucky
says. "We don't have the time, and no one
understands it in any case. We have a clean sheet
of paper here, and that's a scary thing."
Traditional carriers have huge businesses built on
circuit-switching revenue. The physical
infrastructure, billing systems, order-entry systems,
network management and customer service are all
organized around circuit switching and have to be
re-done. Apply that to 13,000 COs in the U.S.
that differ according to the equipment, copper and
fiber lines they have and the services they offer.
The cost of replacing mere switches pales by
comparison.
Bigger, better switches
Another convergence impediment is the per-port
cost of IP switches. IP telephony equipment costs
about $1,000 per line, compared with $150 for
analog lines. Transmission is cheaper, but the
end-point equipment is still a lot more expensive.
"The cost of the IP switches has to come down
before we can expand our IP telephony services
broadly," says Howard McNally, vice president of
transaction services for AT&T.
Scalability is also an issue, although it is improving
rapidly. PSTN circuit switches typically have
about 10,000 ports, but the highest density found
in IP switches is 96 ports. While IP ports handle
multiple calls, the industry still needs to come up
with much bigger IP switches if they are to replace
traditional CO equipment.
IP equipment also has a ways to go in terms of
software. In voice networks, signaling data runs
on a separate real-time network that provides
added security - the SS7 net. In the IP world,
there is very little notion of signaling, and
everything is in the same hacker-vulnerable
network. If the network drops the message
indicating the end of a session, the customer could
get billed for days instead of minutes.
Vendors are starting to incorporate SS7
capabilities into remote access platforms. Ascend
Communications recently announced a gateway
module that enables its carrier-class MAX TNT
WAN access switches to communicate with the
SS7 net. Service providers can deploy the
Ascend switches to divert data traffic away from
voice switches.
Such products aren't available yet, however.
"IP equipment still lacks a lot of the features and
functions of circuit switches, such as the ability to
put a call on hold, or do call forwarding, credit
card calling or 800 numbers," says Michael Day,
senior director of network evolution planning for
Alcatel Network Systems in Richardson, Texas.
"It will take some time to get a full set of voice
features into IP equipment."
IP equipment also lacks the large-scale directory
databases that run on parallel computers. And, of
course, public IP networks need to have QoS
levels before they can scale to replace the PSTN.
An emerging class of high-speed IP switches can
read the beginning and end of packets at line
speeds, a capability that will go a long way toward
enabling QoS in IP environments.
Marlborough, Mass.-based Nexabit Networks is
breaking new ground with a product that offers
multiterabit switching capacity in a single chassis.
The NX64000 routing switch can forward 6.4
terabits per second and will support up to 64
OC-48 (2.4G bit/sec) connections or 16 OC-192
(9.6G bit/sec) links. Nexabit built in eight QoS
queues per interface so the switch can handle
voice and other time-sensitive traffic.
NX64000 is now in beta testing, with commercial
release scheduled for October. Juniper Networks,
based in Mountain View, Calif., plans to deliver a
similar product before year-end.
Reliability issues also have to be addressed.
Telephone networks are developed to meet a "five
nines" standard of minimum reliability, meaning
that the network has to be up 99.999% of the
time. The telephone networks achieve this in part
through the use of special-purpose, rather static
equipment - such as telephones - at the end
points.
Data networks, on the other hand, have
traditionally traded a certain amount of reliability
for added flexibility. Users interface with them
through general-purpose PCs that are constantly
changing.
"How many people haven't rebooted their
desktops this year?" asks John Hart, chief
technology officer at 3Com in Santa Clara, Calif.
"It doesn't help to get the network to 99.999%
reliability if the desktops are less reliable."
IP improvements wanted
Similarly, the network is only as fast as its slowest
component. The ability to provision end-to-end
QoS is at the top of the list of things IP still needs,
and variable QoS on the Internet is still at least a
couple of years away. Some service providers are
approximating it by using private network
segments to avoid congested areas on the public
Internet. However, such transmissions don't get
the full economic benefit of using the public
backbone.
Some say a lot of the latency and QoS issues can
be mitigated by throwing bandwidth at them.
"Then you don't have to resort to IPv6 and
Resource Reservation Protocol and the like,"
Level 3's Vidal says.
Others insist these problems are best addressed
by managing bandwidth.
"Over-provisioning can work in the campus
environment because bandwidth is so cheap
there," says Lucent's Schriftgieffer. "But in the
WAN, the management approach is going to win.
We have to use ATM to do QoS right now, but in
the long run ATM will probably disappear."
Vendors and service providers also want to see
more standards so more of their equipment
interoperates. Wish lists include standardized
IP-to-ATM transfers, cross-checking between
policy servers and standardized billing methods.
Essential standards work is being done by the
Internet Engineering Task Force's Differentiated
Services Working Group. The committee is
working on standard methods for providing
different classes of service across the public
Internet.
One is a mechanism for using the Lightweight
Directory Access Protocol to map user profiles to
different services. This would enable edge devices
to play the role of ticket agents, identifying various
types of users and data and relegating them to
first-class, coach, steerage or the like.
Industry experts expect different-
iated services to be offered first on "private
Internets" within a single service provider's
infrastructure. Then two service providers will
enter into bilateral signaling agreements that will
enable differentiated services to be offered across
their networks. Gradually, more providers will join
in, and eventually the entire public Internet will be
included.
"It's as if the Internet is a biological organism,"
Bay's Hawe says. "It has no long-term strategy
and moves in tactical steps instead of according to
some master plan."
No turning back
Maybe so, but one thing is clear: The IP
convergence train has left the station. Some of the
passengers are wildly enthusiastic about the
journey, and others are being dragged along
kicking and screaming as they enumerate IP's
many flaws. But whatever its shortcomings, IP is a
done deal - it's the standard that got adopted,
period. It has so much momentum and
development action there is nothing else on the
horizon.
Network managers won't be disconnecting from
the traditional voice network over the next couple
of years. Local IP networks have to connect to
the telco infrastructure in a seamless fashion so
enterprises don't have to buy separate equipment
and employees don't have to punch in a lot of
numbers just to tell the network who they are and
who they're calling.
But you will gradually see more applications that
handle multimedia converge onto IP networks.
Enterprises will start with functions such as
customer service that offer an immediate payoff.
Conver-gence will spread from there as new
applications are developed. Eventually, it will
move from the intranet to the extranet and on to
the public network.
When will this last stage of the transition begin?
Estimates vary widely, but it will certainly be
before the first decade of the 21st century closes,
and very possibly as soon as five or six years from
now.
"Major industry participants are investing in and
rolling out products, and ISPs and telcos are
rolling out services to create a market," says
Neville O'Reilly, director of enterprise consulting
for TeleChoice, a consultancy in North
Brunswick, N.J. "We have to have the whole
system in place, so it's still a cart-before-the-horse
problem. But we're reaching the point where we
can start putting solutions together."
IP lets general-purpose and special-purpose
machines talk together in a way that wasn't
possible before. Right now, we have a specialized
network sitting between those machines. That
network does what it was designed to do - voice
- very well. But we need a general-purpose
network now because the key application isn't
voice anymore.
Drill down
into IP
convergence:
Forum
What's it all mean?
Discuss it in our
convergence forum.
Economics
How convergence
saves you money.
Regulatory
Gov't. may yet turn
the tables on IP.
Customers
Pioneers who are
putting convergence
to work.
Technology
The standards that
make it all possible.
Carriers
Who's planning
what, plus
interviews with
Sprint, UUNet
execs.
Pundits
Opinions from
Network World
columnists:
Anderson
Heckart
Nolle
Links
For even more
information!
Breidenbach is a
consultant and
freelance writer in
San Mateo, Calif.
She can be reached
at sbreidenbach
@usa.net |
