March 1998
By Joanna Makris
Gigabit Cabling
The Copper Stopper?
Vendors claim their new cabling systems can carry gigabit data. But with no standard in sight—and products that can't meet the draft spec—net managers should go slow on high-speed copper
|
|
|
Gigabit copper? It's here. Just ask the eight or so vendors who
claim their cables and connectors can carry 1
billion bits per
second—without dropping precious packets. While they're at it,
corporate networkers might want to ask some other pertinent
questions—like how can they be sure that any vendor's gigabit
cabling is up to spec if there is no spec. The ISO/IEC—the
international body that defines electrical standards for a whole range
of networking gear—says that its Category 6 standard is months
(or more) away. Cat 6 will set more stringent requirements for running
high-speed data over copper—be it UTP (unshielded twisted pair)
or ScTP (screened twisted pair).
"Few—if any—of the parameters are agreed upon," comments
Michael Gilmore, a member of the Cat 6 joint technical committee.
And even when Cat 6 is spec'd out, net managers may want to
cross-examine their vendors to see if their schemes can really deliver
the gigagoods. That may sound overly suspicious, but claims about
high-speed copper should be treated very cautiously. After all, some
net managers have al
ready had problems running 100 Mbit/s over
Category 5, the first rung up on the high-performance ladder.
"As far as I'm concerned, UTP is dead," says Ron Catterall,
networking consultant for the Cleveland Clinic Foundation (Cleveland).
He tested 100-Mbit/s Ethernet over Cat 5 and says, "I was getting up
to 50 percent error rates. It clearly cannot cope."
Not surprisingly, the vendors disagree. "We haven't reached the
failure point of UTP yet," says Anthony Sidowitz, director of product
development for Ortronics Corp. (Pawcatuck, Conn.), a connector
manufacturer.
And Lucent Technologies Inc. (Murray Hill, N.J.) argues that its
Cat 6 cabling "pushes UTP to a new level of performance," according to
Steve Bartolutti, distinguished member of the technical staff.
There's another big question looming. Even if Cat 6 can handle
gigabit data, how much farther into the future will it take copper?
Headroom is the issue here—the spare capacity that gives cabling
room to grow as faste
r protocols are rolled out.
Why worry? There's evidence that vendors are cutting the Cat 6
safety margin dangerously thin. At the January ISO/IEC meeting Lucent
successfully lobbied for lower worst-case criteria—the minimum
performance parameters that cable must meet to comply with a spec.
Before the backroom politics, the ISO/IEC was considering metrics
that several vendors—including Lucent—couldn't meet. The
vendors argued that their original calculations hadn't taken into
account patch cords, which degrade performance when they're added to a
cable connection. Ironically, Data Communications has learned that at
least two vendors still can't meet the relaxed draft standard.
Actually, Lucent may have managed to affect both international and
domestic standards, since the EIA/TIA—the vendor consortium that
defines commercial wiring standards in the U.S.—had already
decided to follow the ISO/IEC's lead on Category 6 copper.
"If some people at these sta
ndards meetings were told to support a
protocol over wet string, that's what they would do," says George
Wojtan, a representative to the EIA/TIA (Electronic Industries
Association/Telecommunications Industries Association) and vice
president of major accounts at Ortronics.
Where's the Gig?
|
Gigabit Copper
|
So what's the next move for net managers who need the speed and are
considering copper? Get smart about the Cat 6 spec. Start with the
specific performance parameters. Then grill prospective vendors about
how much headroom their cabling schemes can deliver. To take the agony
out of the exercise, Data Comm has gotten vendors to come across with
their worst-case minimums and assessed their safety margins. That's
good for a start, but corporate networkers should do the math before
they
shell out for Cat 6. After all, vendors are pricing their
proprietary schemes at a premium, anywhere from 10 percent to 50
percent above conventional Cat 5. Finally, check the fine print:
High-performance cabling is only as good as its guarantee.
Stretching the Spec
The ISO/IEC (International Organization for
Standardization/International Electrotechnical Committee), which also
defined the 11801 standard for commercial cabling, is taking the lead
on gigabit copper. Actually, a subcommittee of Joint Technical
Committee 1 is working on two extensions to 11801: Category 6 applies
to UTP and ScTP at frequencies to 200 MHz. (Cat 5 is spec'd to 100
MHz.) Category 7 will most likely cover individually shielded STP
(shielded twisted pair) to 600 MHz. The EIA/TIA also is attending
these meetings and will likely adopt the parameters defined by Cat 6
as an addendum to EIA/TIA 568A, the Commercial Building
Telecommunications Cabling Standard. It's also working on a Cat 5E
spec that covers cable perform
ance that falls between Cat 5 and Cat
6.
Why the new specs? Because the way that data is transmitted over
cable is about to change. All copper cable has four wire pairs, but
token ring, Ethernet, and fast Ethernet only use two of them: one to
carry signals in each direction. But any gigabit encoding scheme will
transmit bidirectional signals on all four wire pairs. For example,
gigabit Ethernet's PAM-5 (pulse-amplitude modulation) shunts 250
Mbit/s over each wire pair. And once four wire pairs are being used,
the performance parameters defined for two-pair transmission are
irrelevant.
Net managers need to understand that higher speed isn't simply a
matter of higher frequency. In fact, there's a popular misconception
that there's a 1:1 ratio between megahertz and megabits. But protocols
like gigabit Ethernet and 155-Mbit/s ATM (asynchronous transfer mode)
cram more than 1 Mbit/s into 1 MHz. For example, PAM-5 achieves a 3:1
ratio. "Differentiating between cabling schemes based simply on
frequen
cy is a red herring," says Mike Hook, product manager at ITT
Cannon Network Systems and Services (Santa Ana, Calif.), a maker of
cables and connectors. He adds that just because a system is tested to
frequencies above the minimum TIA/EIA 568A and ISO 11801 requirements
doesn't mean it will deliver higher data transmission rates.
Performing Arts
So what should net managers look for? The ISO/IEC has tightened up
the performance parameters used to check if Cat 5 is at spec, coming
up with new values specifically tailored to high-speed transmissions.
It also has added a new benchmark to its test suite:
- PS-Next (power sum near-end crosstalk) measures in dB how well one
cable pair resists interference generated by the other three. The
ISO/IEC is talking about a minimum acceptable PS-Next of 37.1 dB for a
Cat 6 channel, which includes cable and connectors. It's 3.1 dB for
Cat 5. (All values for Cat 6 and Cat 5 are given at 100 MHz.)
- PS-ACR (power sum-att
enuation-to-crosstalk ratio) indicates in dB
how much stronger the data signal on one pair is than the noise on the
other three. The minimum for a Cat 5 channel is 3.1 dB; for Cat 6 it's
likely to go to 15.6 dB.
- Return loss measures in dB how well
cabling deals with signal reflections (which interfere with data
transmissions). Higher numbers indicate that only a small amount of
signal is reflected, which is what net managers want. Return loss for
Cat 5 channel is 10 dB; 12 dB will likely be the minimum for Cat 6.
- Propagation delay indicates how long it takes a signal to travel
100 meters. Cat 6 delay will probably be 536 nanoseconds per channel;
it's 538 for Cat 5.
- Delay skew is the difference between the
propagation delay on the fastest and slowest cable pair. Skew is
inherent in the way cable is manufactured. Each cable pair must
exhibit a different twist ratio (to cancel out crosstalk), which means
that each is a different length (depending on the numb
er of twists).
The ISO/IEC is looking at 50 ns as the minimum acceptable delay skew
for a Cat 6 channel; it's 45 ns for Cat 5.
- PS-Elfext (power-sum
equal-level far-end crosstalk) is a new test for Cat 6 that indicates
the ratio of attenuation to far-end crosstalk. It's measured in dB;
higher numbers are better. The minimum acceptable PS-Elfext for a Cat
6 channel will likely be 20.2 dB.
It's worth noting that at press
time, some vendors were predicting that propagation delay and delay
skew might be dropped from the standard.
Relaxing the Rules
Forget the decibels and nanoseconds for a minute. What net managers
really need is a stringent spec, one with enough headroom to handle
faster protocols as they're rolled out. But even though Cat 6 is more
exacting than its predecessor, it's not as strict as it started
out—and it could be relaxed even more.
Here's the deal: The ISO/IEC cabling committee originally came up
with considerably tougher minimum acce
ptable levels for its six
performance parameters. So tough, in fact, that Lucent's new Gigaspeed
scheme—as well as other vendors'—failed to meet all six.
But at January's ISO/IEC meeting various
manufacturers—including Lucent—lobbied for less rigorous
worse-case minimums. And by the end of the meeting, the levels had
been lowered: Five out of six now exactly match Lucent's.
"The results of the ISO meeting were based around a presentation by
Lucent, which forced a change in the formulas," says James Merchant,
electrical design engineer at cable manufacturer Modtap (Harvard,
Mass.).
Actually, it sounds as if Lucent had little difficulty getting its
way. Before the January meeting, Data Comm asked Belden Wire and Cable
(Richmond, Ind.) what it thought the values should be for a Cat 6
standard. "We'll agree with whatever Lucent has told you," said Robert
Kenny, technology development manager.
"Lucent not only has a vested interest in controlling the standards
pr
ocess, but also it has the huge financial resources of Bell Labs at
its disposal," says Robert Pritchett, LAN specialist and senior
designer at SCM Consultants (Kinnewick, Wash.), a Lucent-certified
consultant. "They can afford to send people to the standards meetings;
the mom-and-pop shops can't." Pritchett thinks it would be fairer "if
standards meetings were in an interactive or online forum in which we
all had a voice. That would be democratic. It's doable today, but the
jet set may not want to do that."
Further, the fact that the TIA/EIA is a vendor consortium, not a
government-sanctioned standards body, makes it easier for vendors to
get their way—and to do so in private. The TIA does not publish
results of voting, and it refuses to disclose how many representatives
from which vendor sit on the committees.
The Product Parade
|
BUILD YOUR OWN CUSTOM TABLE
Table 1: Selected Vendors of Gigabit Cabling Schemes
|
So much for the politics. What about the products? Of the eight
vendors offering gigabit copper, Belden only manufactures cable and
Ortronics only makes connectors (see
Table 1
). Belden pairs its cable
with hardware from Krone Inc. (Berlin), Mod-Tap, or Panduit Corp.
(Tinley Park, Ill.)—which it resells. Net managers must buy
everything from Belden if they want the vendor's performance
guarantee.
Ortronics, meanwhile, says its connectors must be installed with
Belden's Mediatwist or with Lanmark 1000 cable from Berk-Tek Inc. (New
Holland, Pa.). Again, corporate networkers who want a performance
warranty must buy everything from Ortronics.
It can get pretty byzantine: Even though Ortronics recommends
Belden cable, Belden doesn't spec or rec
ommend Ortronics connectors.
The cable manufacturer also says net managers—or
resellers—are free to use its cables in any configuration they
want. But Belden won't back them with a guarantee.
The other six vendors all sell end-to-end systems, which include
telecom outlets, patch panels, patch cords, jacks—the whole nine
yards. Predictably, all argue that this end-to-end approach is better
than tacking together components from different manufacturers. They
claim that all their components are specifically engineered to work
together. They're also tested together. "With cabling systems, the
channel is only as strong as its weakest component. So if you have
extremely high-performance cable and lousy connectors, you won't see
good performance when you put it together," says Les Baxter, director
of research and development for Systimax SCS at Lucent.
Banana Splitter
But end-to-end vs. mix-and-match is just one way that vendors
separate their products. For example, Belden'
s Mediatwist has a
banana-shaped cross-section. The vendor says this construction locks
the wire pairs in place, minimizes crosstalk, and reduces problems
resulting from kinks and bends. But critics contend that any
performance gains come at a price. Belden cable, they comment, is less
flexible, making it harder to install around corners or in nooks and
crannies. It also takes more time to terminate (making it more
expensive to pull). "Because the pairs are webbed together, it's more
difficult to strip back the outer jacket without a special tool," says
Peter Heller, vice president of operations at cable installer Extranet
Telecommunications Inc. (New York). Others point out that Belden's
banana does not comply with the 568A standard—which specifies use
of round cable.
Some vendors sell more than straight UTP. BICC Brand Rex
(Glenrothes, U.K.), ITT Cannon, and RIT Technologies Inc. (Tel Aviv,
Israel) offer products with extra sheathing (or screening). The theory
is that the chunkier cable emits l
ess EMI (electromagnetic
interference) and also is more resistant to the EMC (electromagnetic
crosstalk) radiated by power lines, fluorescent lighting, and the
like.
Only ITT Cannon offers ScTP, which encases all four wire pairs in
an aluminum sheath that is itself wrapped in PVC (polyvinyl chloride)
or FEP (fluorinated ethylene propylene).
Catterall thinks that ScTP is the way to go. "I tested screened
alongside UTP, and the screened cable gave no errors in the same
situations." He adds, "We urgently need gigabit to the desktop because
many of our medical researchers are using video imaging, which
consumes a huge amount of bandwidth." And he's chosen ScTP to get the
job done.
BICC sells S-FTP (shielded and foiled twisted pair). This cable
wraps the four wire pairs in foil, which is then encased in a metal
braid. The two layers are even more resistant to EMC.
BICC and RIT sell STP. They claim this approach is the ultimate in
shielding because it encases each wire pair in a meta
l shield, which
is then protected by an insulating sheath.
But there are a few potential problems with shielded cabling.
First, the shield must be fully grounded to the terminators at either
end. If it's not, it's basically useless. In fact, it's worse than
useless: Ungrounded shielding can act as an antenna—just the
thing to attract disruptive radio signals.
"If screened or shielded cabling isn't properly terminated, the
surrounding noise will destroy the integrity of data," says Dominic
Mannina, product manager at Extranet. And terminating STP also takes
about twice as long as terminating UTP, he says.
Who Needs It?
Just to keep things interesting, some folks argue that there's no
need to deploy high-speed copper to handle gigabit traffic.
Tried-and-true UTP will do just fine. After all, the IEEE's 802.3z
subcommittee says it will spec gigabit Ethernet to run over Cat 5.
"Gigabit Ethernet will have an operating frequency of 100 MHz," says
Geoffrey Thomson, chair of t
he IEEE 802.3 working group and manager of
standards development at Bay Networks Inc. (Santa Clara, Calif.).
The majority of cabling vendors concur. "Most Category 5 cable will
support gigabit Ethernet," says Paul Kish, senior manager of IBDN
systems and standards at Nordx/CDT (St. Laurent, Quebec), a maker of
cable and connectors. But they're quick to point out that customers
who want worry-free gigabit performance should deploy high-speed
cabling.
"Right now, most of our customers are running 10 Base-T," says
Lucent's Baxter, "but they want the additional assurance that they'll
be able to comfortably migrate to faster protocols."
The vendors have reason to be concerned. First, the gigabit
Ethernet spec is based on transmission parameters not covered by TIA
568A: PS-ACR, PS-Next, return loss, propagation delay, and delay skew.
Second, some Cat 5 may not structurally support high-speed traffic. In
1994, an FEP shortage forced manufacturers to develop so-called 2+2 or
3+1 cables. Thes
e designs clad either one or two cable pairs in
polyolefin resin. And that could mean trouble. "The difference in
cladding could cause a massive skew difference between wire pairs,
resulting in unacceptably high bit-error rates," says Jonathan Green,
group product manager at ITT Cannon Network Systems and Services.
Help is on the way. "The TIA intends to provide guidelines relating
to the recertification of older Cat 5," says Henriecus Koeman,
principal engineer at Fluke Corp. (Everett, Wash.). The guidelines,
which will be published this year, detail how to use handheld cable
testers to make sure Cat 5 conforms to the new performance
parameters.
|
MORE INFO
Fiber Falling
|
That may not be enough for some net managers who've run into
heartache trying to ship f
ast Ethernet over Cat 5. In fact, they're
worried that gigabit Ethernet, which is 10 times faster, will mean 10
times more trouble. For them, only fiber will do (see "
Fiber
Falling
").
Open to Spec-ulation
Corporate networkers who think gigabit copper is the way to go now
need to make an educated decision about the available cabling schemes.
Since there's no standard in place, they'll have to make do with the
next best thing: an assessment of how products meet the performance
parameters defined by the draft spec.
Caution is the watchword here. First off, the spec itself is not
cast in stone (as demonstrated). Second, since the numbers are
reported by the vendors (and no independent tests have taken place),
it's very important to understand what they signify.
Even something that is as seemingly straightforward as frequency
can be somewhat tricky. For instance, Belden tests its Mediatwist
cable to 350 MHz—and it even calls one of its product
s the
Datatwist 350. But when questioned further, the vendor admits that the
cable can only carry data to roughly 200 MHz. The higher rating
describes the frequency at which the cable still exhibits "stable
electrical performance," according to the vendor's product literature.
Even more important is to understand how a cable was tested.
Vendors use two different configurations: basic link and channel.
Basic link means only the cable has been evaluated. Channel refers to
an end-to-end system: cable, patch panels, patch cords, telecom
outlets, jacks, face plates, and other gear.
By the Numbers
|
BUILD YOUR OWN CUSTOM TABLE
Table 2: Assessing the Safety Margins
|
To take the agony out of the assessments, Data Comm asked the eight
hig
h-speed cable vendors to reveal their ratings for each of the six
parameters defined by the Cat 6 draft spec (see
Table 2
). The numbers
are for worst-case performance at 100 MHz, the maximum operating
frequency of gigabit Ethernet.
The results aren't exactly encouraging. At least four vendors can't meet one or more of the minimum acceptable values. Others barely make it.
In their defense, the vendors point out that their performance
ratings are much better at 200 MHz. That's true, but there currently
are no 200-MHz protocols. Even 155-Mbit/s ATM operates chiefly at 80
MHz (approximately 5 percent of the operating spectrum exceeds 100
MHz).
So how do the vendors stack up? RIT's STP cable had the best
PS-Next rating: 80.5 dB (for basic link; the number would be a little
lower for channel). Cat 6 calls for 37.1 dB. Ortronics has the best
minimum for UTP: 43 dB channel with Belden or Berk-Tek cabling.
Lucent's 37.1 dB was the wors
t showing.
All the STP cables do well with PS-ACR. RIT's 71.4-dB basic link is
more than four times the 15.6-dB minimum. UTP had far more trouble.
Belden's 18-dB basic link makes the cut, but channel measurement for
Nordx/CDT's System 1200 is 9 dB, while its System 2400 scored 14 dB.
Here again, Lucent's 15.6 dB exactly matches the spec minimum.
Ortronics claimed the best return loss: 26 dB channel, with BICC
right behind at 25 dB channel. The draft specifies 12 dB, which is
where Lucent and Nordx finish up.
But knowing return-loss values is only the beginning: Net managers
need to grill vendors on whether they tested with one patch panel
(interconnect configuration) or two (crossconnect). The latter is
tougher since it involves more sources of loss. All the vendors
indicated they used a crossconnect setup, except Belden, which refused
to say.
Most of the vendors had no trouble meeting the minimum acceptable
propagation delay of 538 ns. Lucent and RIT's UTP were the exception
s;
both exhibit 570-ns delay—Lucent for the channel, RIT for the
link. Nordx/CDT has yet to measure this parameter.
All the cable manufacturers can meet the 50-ns minimum delay skew.
At 15 ns on the channel, ITT Cannon has plenty of headroom. RIT claims
the same delay on its basic link. Lucent's channel is once again right
on the mark indicated by the draft spec. It also has the highest delay
skew.
As discussed, the ISO/IEC has added a new performance parameter to
its test suite: PS-Elfext (which measures the effect of far-end
crosstalk on the near-end node). The draft spec calls for a 20.2-dB
minimum. RIT's STP has the highest PS-Elfext: 83.5 dB for the basic
link; its UTP comes in second with 28.5 (also on the link). Nordx's
System 1200 and System 2400 score 19 dB for the channel, below the
minimum.
Playing the Percentages
The foregoing figures are just the beginning. A little applied math
reveals the percentage that each vendor exceeds or fails the minimum
requirement
s—as well as the average margin of safety (headroom).
For example, the PS-Next for AMP Inc. (Harrisburg, Pa.) is 38.3
dB—3 percent above the minimum. Belden's PS-ACR of 18 dB is 15
percent above the minimum requirement of 15.6.
And what about the overall ratings (an average of the average for
each category)? RIT's STP cable fared best at 152 percent—largely
because it does an exceptional job of minimizing crosstalk. Among the
UTP schemes, Belden's 34 percent was the highest average. At -0.4
percent, Nordx's System 1200 has no headroom. The vendor argues that
while its cable can carry gigabit traffic, it's not meant to comply
with Cat 6. And Lucent's -1.0 percent makes it the low scorer. It says
it supplied truly worst-case numbers and says that it "significantly
outperforms" the minimum values.
Insurance Premiums
Getting the straight scoop on prices is about as difficult as
drawing a bead on performance. And don't look to the vendors for much
help. Both Lucent and BIC
C refused to disclose detailed prices,
instead supplying percent premiums over Cat 5. RIT gave per-foot
prices for cable but broke out connectors separately, which makes it
tough to come up with a ballpark figure for end-to-end systems.
Nordx/CDT also supplied price-per-foot, but in this case it represents
an average of components, including cable and connectors. AMP is the
exception: Its "cost per end-user" is simple and straightforward.
What's the bottom line? Corporate networkers are going to pay a
premium for Cat 6. John Piscal, vice president of Extranet, thinks
gigabit copper will average 15 percent to 35 percent more than Cat 5,
with screened cable running 40 percent higher. Nordx is already
pricing its System 2400 50 percent higher.
But high-speed copper doesn't have to be a budget-buster. "Since no
ratified standard is available, customers feel they have no choice but
to pay for the cabling system with the best performance at the highest
frequencies," says Glenn Holmes, a consultant
with Anixco
Communications Systems Ltd. (Edenvale, South Africa), a reseller and
consultancy.
Point well made: When the Cat 5 spec first came out, prices
dropped. What's more, standards make it easier for manufacturers to
get into a market. And competition encourages vendors to revamp their
pricing structures.
Before they part with one thin dime, however, corporate networkers
need to consider what sort of guarantees they'll be getting. And they
need to know that vendors require cabling systems to be set up by
certified installers or resellers for their warranties to be
valid.
All the vendors except AMP offer application guarantees, which
cover specific protocols as well as any forthcoming standards-based
protocol designed to run over UTP. And with the exception of
Ortronics' 25-year limit, all vendors say their guarantees are good
for the life of their cable.
Application guarantees are getting rave reviews from some
customers. "It's not the bandwidth that I really care about,
it's the
comfort of knowing that my protocol is going to run over the
infrastructure," says Paul Barry, director of networking and
communications at Northwestern Memorial Hospital (Chicago). Barry has
already installed Lucent's Gigaspeed cable and is planning to install
an end-to-end system in his new 2.1 million square-foot facility.
AMP argues that it can't offer an application guarantee until the
high-speed copper specs are a done deal. "A cabling specification
still in draft is a moving target," says Tony Beam, director of global
systems marketing for the vendor's Netconnect Systems. Instead, AMP
guarantees that its cabling system will meet advertised performance
parameters for 15 years.
Net managers also need to check out the warranties on cabling and
components. Most products are guaranteed against failures or defects
for 15 to 25 years. Belden is the only manufacturer to offer a
lifetime warranty on its cable.
Some vendors go beyond basics. For example, Lucent and Nordx/CDT
g
uarantee that net managers can mix and match their gigabit cable with
Cat 5 components and still get at least Cat 5 performance. This is a
good deal for corporate networkers interested in phased rollouts.
Other vendors won't promise that these mixed systems will even deliver
Cat 5 performance.
|
CONTACT AUTHOR
jmakris@data.com
|
BICC and ITT Cannon guarantee that their shielded cables can be
safely extended beyond the 100 meters recommended for UTP (though not
for gigabit Ethernet). For example, ITT Cannon's Gigapath can carry
100Base-T to 140 meters without signal loss; 155-Mbit/s ATM can go to
145 meters. And longer cable runs mean fewer wiring closets, hubs, and
so on—all of which adds up to big savings.
Joanna Makris is associate editor, WAN services. She is ba
sed in New York and can be reached at
jmakris@data.com.
|
|
|
 |
|
|
