April 1995
By Tyron Pike, UB Networks Inc./Stanley Ooi, Advanced
Telecommunications Modules Ltd.
Taking Sides on 155 Mbit/s Over Category 3 UTP
The cabling is the most prevalent on corporate nets. Is it up to
the task of transporting 155-Mbit/s traffic?
A proposal is now before the PHY working group of the ATM Forum
that would standardize 155-Mbit/s ATM over Category 3 UTP. The spec
has already gotten the backing of several vendors of network adapters,
including UB Networks Inc. (Santa Clara, Calif.), which has announced
a product that implements this technology.
According to its proponents, 155 Mbit/s over Category 3 gives
network managers a way to deliver full Sonet (Synchronous Optical
Network) OC3c to the desktop-without having to pull new cabling. They
argue that LAN users will soon need the bandwidth for
videoconferencing and other multimedia applications. They also suggest
that 25-Mbit/s ATM, which was recently ratified by the ATM Forum, may
be insufficient for power users on Unix and Pentium-class
platforms.
Critics of the idea claim that it faces technical difficulties,
including problems with
transmission signal strength, radio frequency
interference, and crosstalk. They also contend that ordinary desktop
PCs would have difficulty coping with 155-Mbit/s traffic and that the
requisite adapters will need extensive-and expensive-buffers.
They also question whether this capacity is actually needed. For
them, the logical place for 155-Mbit/s ATM interfaces is on the
backbone, where Category 5 and fiber optic cable are more common. Even
with delay-sensitive multimedia applications on the way, they argue,
155 Mbit/s to the desktop would be overkill. Instead, they suggest
that 25-Mbit/s ATM, which runs over Category 3 UTP without special
technology, offers a more technically sound and cost-effective way to
deliver ATM.
To explore this debate in greater detail, DATA COMMUNICATIONS has
found experts who understand both the business and the technological
issues. We've asked Tyrone Pike, senior vice president and general
manager of the Network Products Division at UB, to pre
sent the
positive side. And for words of caution about running high speeds on
low-grade wiring, we went to Stanley Ooi, vice president of strategic
marketing at Advanced Telecommunications Modules Ltd. (Cambridge,
U.K.).
Pike:
First off, let me say that I don't think this is a
debate about 25-Mbit/s ATM vs. 155-Mbit/s ATM. Both are viable, and
users will want both-as well as 51 Mbit/s. This is really a debate
about Category 3 unshielded twisted pair and what it can do. Category
3 UTP is far and away the most widely installed cable on today's
corporate networks.
Ooi:
You're right about one thing, Tyrone. Category 3 UTP
is the most prevalent cabling in corporate networks-no one is
disputing that. But the fact is, many next-generation high-speed LAN
technologies-including 155-Mbit/s ATM-require higher signal voltages
than Category 3 can safely support.
Conventional LAN technologies like Ethernet 10Base-T or token ring
have lower power requirements and f
ewer potential problems with RFI
(radio frequency interference). But speeds over 100 Mbit/s really
demand Category 5 UTP or fiber optic cable.
Pike:
That was the case at one time, but there are at
least two transmission technologies, CAP (Carrierless Amplitude and
Phase Modulation) 64 from AT&T and the scheme from Tut Systems Inc.
(Pleasant Hill, Calif.), that make it possible to run full Sonet OC3c
ATM connections over Category 3 UTP. And these breakthroughs make it
possible for network managers to implement full-duplex 155 Mbit/s all
the way to the desktop.
WHO NEEDS IT?
Ooi:
Before we talk about whether 155-Mbit/s ATM over
Category 3 can work, maybe we should ask why network planners might
want it. Category 3 is primarily used on LANs. But Category 5 UTP and
fiber are much more common on the enterprise backbone-where 155-Mbit/s
ATM is most likely to be deployed. Do the majority of commercial users
need 155 Mbit/s to the desktop? I don't think so.
Pike:
It's true that 25- or 51-Mbit/s ATM may be perfectly
adequate for many users right now. That's one reason 25-Mbit/s ATM has
so much momentum. But many users on Unix workstations and
Pentium-class PCs already know they need higher throughput to
establish workgroup and server connections for videoconferencing,
imaging, and other multimedia applications. And I'm certain network
managers would prefer using cabling that's already installed to
shelling out thousands of dollars for new cable pulls.
Running new Category 5 costs about $150; the price jumps to more
than $500 for fiber.
Ooi:
But very few applications, especially on PCs, need
155 Mbit/s. If users require video to the desktop, good-quality
MPEG-2 video can be delivered across networks running at 4 or 5
Mbit/s. Even M-JPEG consumes only around 8 Mbit/s, making 25-Mbit/s
ATM more than sufficient. And slower-speed ATM can run over Category 3
UTP with no problems.
Also, let's consider the impact of a
155-Mbit/s network adapter on
the average desktop PC. The latency of an ISA bus PC-the most common
desktop box-is so great that large buffers will be needed to handle
the traffic a 155-Mbit/s adapter can deliver. More buffering on the
adapter means higher cost. And even then, given the bandwidth
limitation of the PC bus, users will see little improvement in
application performance.
A COOL MILLON
Pike:
Stanley makes some good points-assuming the world
remains static. But demands for application bandwidth are
growing-especially for desktop users. And millions of desktops have
already been upgraded to faster PCI (Peripheral Component Interface)
bus systems that have no problem handling heavy traffic. Let's face
it: A 100-MHz Pentium PC with a PCI bus, local bus video, and 16
Mbytes of memory is a pretty powerful workstation. Intel Corp. (Santa
Clara, Calif.) says its licensees will sell 10 million to 12 million
PCI PCs this year, and they won't all be running Sonic the Hedgehog in
some
body's home entertainment center.
Also, sales of RISC-based Unix workstations and Windows NT systems
create a substantial market for 155-Mbit/s ATM. Hewlett-Packard Co.
(Palo Alto, Calif.), Sun Microsystems Inc. (Mountain View, Calif.),
and Silicon Graphics Inc. (Mountain View, Calif.) will sell more than
1 million Unix workstations this year. And then there are Power PCs
and Alpha systems from Digital Equipment Corp. (Maynard, Mass.). The
percentage of users who could take advantage of 155-Mbit/s throughput
clearly is increasing.
Ooi:
I still think it's highly questionable whether users
and network managers want 155 Mbit/s to every desktop-the cost is
simply too great. Even if adapter vendors can get their prices down,
handling a large number of 155-Mbit/s connections is going to put more
strain on the ATM switch. And the buffers needed to cope with large
bursts could add hundreds of dollars to the per-user cost.
Pike:
There are affordable ways to impleme
nt 155-Mbit/s
ATM over Category 3 UTP. The chips for the adapters are readily
available from a variety of sources. And given the potential size of
the market for adapters that support 155 Mbit/s over Category 3,
high-volume production should be easy to achieve.
Besides, users will already have saved thousands of dollars by not
having to pull Category 5 or fiber.
Ooi:
Let's assume, for the sake of argument, that network
managers are willing to pay the higher prices until the market
achieves commodity volumes-just to deliver bandwidth users may or may
not need. It's still by no means certain that Category 3 UTP can
deliver 155 Mbit/s-at least, not without creating serious
problems.
The first challenge any 155-Mbit/s Category 3 scheme has to face is
conforming with FCC emissions regulations. This is going to demand a
difficult and complex solution (another upward pressure on costs). In
fact, it's almost impossible to use NRZ (nonreturn to zero) encoding
over Category 3 UTP
without violating FCC rules and European
environmental standards.
Pike:
Critics of 155 Mbit/s over Category 3 UTP always
cite the same problem: signal attenuation. But new transceiver
technology can use any binary encoding scheme, including NRZ, for ATM
applications-without exceeding FCC regulations or losing
throughput.
I'm not advocating using Category 3 UTP for distances of over 100
meters. The technology UB Networks has proposed to the ATM Forum makes
no claim to be able to do that. Most commercial office buildings won't
require distances greater than 100 meters, anyway. Cabling schemes
typically follow telecom requirements, which use an intermediate
distribution frame (IDF) every 30 to 50 meters. Each IDF serves one
quadrant of a building, or if it's a larger building, a floor. So
there's not much call to go longer than 100 meters from switch to
desktop in a commercial environment.
Manufacturing environments and labs frequently can involve greater
distances, but ev
en there you might have, say, fiber to the IDF and
the rest of the users on Category 3.
Ooi:
Even within 100 meters, in order to reduce emissions
with the NRZ modulation the ATM Forum requires, the transmission
voltage for a 155-Mbit/s connection would have to be lowered. That, in
turn, is going to require more sensitive receivers at the
destination.
Both of these things can be done, but they make the system much
more susceptible to external noise. And that's likely to increase the
connection's bit error rate so that it no longer conforms to ATM
Forum's PHY specification.
Another significant problem low-power solutions face is crosstalk.
On Category 3 UTP, 155-Mbit/s links need crosstalk cancellation
functions. Yet these systems also are delicate and susceptible to
external noise. Once crosstalk cancellation is disturbed and crosstalk
occurs, it can take the connection several milliseconds to
recover.
Some solutions for running 155 Mbit/s over Category 3 may work
without reducing transmission voltage, but they will be quite
expensive-too expensive to extend to every desktop on a corporate LAN.
For example, AT&T's CAP 64 technology, which packs information into
the low-frequency portion of the cable, stays within the FCC
guidelines by limiting its emission spectrum to 30 MHz, rather than
simply lowering signal power. But this solution isn't simple or
cost-effective to implement.
Pike:
It's true that the problem with CAP is that AT&T
failed to deliver a cost-effective technology with the appropriate
power capabilities. Remember that AT&T originally promised a 51-Mbit/s
solution for Category 3 that would be significantly cheaper than 155
Mbit/s. We advocate finding a cost-effective Category 3 approach that
supports 155 Mbit/s, 51 Mbit/s, and 25 Mbit/s.
But we've had good success with the Tut technology in very hostile
environments. We've demoed it and proven that it works. If we rewound
back seven years, you would hear the same arguments a
bout 10 Base-T.
Back then, people argued that if you ran 10Base-T through a telecom
bundle with ringer current running through it, it wouldn't work. But
they were wrong.
Once the critics are done talking about signal attenuation, they
bring up problems with near-end crosstalk (Next). But it is possible,
using Next cancellators, to keep crosstalk to a minimum. In fact, it
may even be possible to use crosstalk cancellation technology to push
155-Mbit/s ATM over Category 3 beyond the 100-meter limit.
Stanley's right when he says that using cancellator technology
drives up costs. But it's still much cheaper than pulling new
cable.
The proposal in front of the forum uses NRZ coding, which complies
with its spec for Category 5 UTP. This would allow a single adapter to
support both types of cabling-autosensing whether 155, 51, or 25
Mbit/s is being used. Basically, if ATM is going to take off in the
LAN environment and supplant FDDI, 100Base-T, and 100VG as a fast-LAN
technology, network mana
gers will need a single adapter card that can
support various speeds and interfaces. ATM is already much too
complex. If it's going to be pervasive, it's got to be so simple a
Neanderthal can install it.
Ooi:
As long as it's a Neanderthal with very deep pockets.
As I've already said, net managers who opt for 155-Mbit/s ATM over
Category 3 UTP are going to pay much for their network adapters.
Pike:
The goal should be to furnish users with as much
throughput as possible for a given wiring plant-the maximum speed for
the maximum distance. When necessary, the network should be flexible
enough to throttle down to a lower speed. This can be done easily
using 155-Mbit/s technology. Both 51- and 155-Mbit/s ATM use the same
basic framing. It will be simple to create a standards-compliant
155-Mbit/s adapter that also supports 51 Mbit/s-and probably even 25
Mbit/s-for environments where a full 155 Mbit/s can't be delivered to
all end-users. This could even be done using exis
ting virtual circuit
technology, which makes it easy for ATM to dedicate precise amounts of
bandwidth to specific functions.
In addition, 155 Mbit/s over Category 3 UTP, unlike some of the
25-Mbit/s schemes proposed thus far, would be fully compatible with
155-Mbit/s ATM over Category 5 or fiber. Users would not have to worry
about changing throughput specifications when moving from one network
segment to another.
Ooi:
It sounds good, but you haven't explained how to make it
cost-effective. NRZ encoding will likely undergo severe attenuation at
high frequencies, as well as generate significant intersymbol
interference. This probably can be compensated for by using some sort
of adaptive digital signal processor at the receiving end. Again, that
could run into money. There are many, many technical problems that can
be solved if engineers have infinite amounts of money to throw at
them. But that's not the case when you're attempting to develop a
commodity-priced solution for deskto
p systems.
Pike:
It all depends on how you define cost-effective. It
may be easy for a net manager to get the budget to buy new 155-Mbit/s
adapters, or even new workstations and applications for a workgroup of
scientific users. Getting permission to recable could be much harder.
The cost of bringing in electricians and paying them overtime to work
after business hours is much greater than that associated with adding
buffers to cards and switches. It's much more expensive if you bring
down the network and pull cable while the company could be earning
money.
Tyrone Pike is senior vice president and general manager of the
Network Products Division at UB Networks Inc. (Santa Clara, Calif.), a
supplier of network hardware and software and a provider of a range of
integration services.
Stanley Ooi is vice president of strategic marketing at Advanced
Telecommunications Modules Ltd. (ATML, Cambridge, U.K.), a supplier of
ATM switches, adapters, and networked videoservers
for the PC
workgroup.
[
Home
]
[
Registration
|
Subscriptions
]
[
Contact Us
|
E-Mail
]
| 
