Until mid-November 2002, you’d have been right to assume that the fastest
speed of Ethernet achievable over twisted-pair copper was 1 Gbps–capable of
being run over Cat 5e or Cat 6 cabling systems. Although there had been talks
that Cat 6, which has its electrical characteristics specified up to 250 MHz
(although GigE actually uses only about 80 MHz), would be theoretically capable
of 2.5 Gbps, IEEE had no plans for anything faster than 1G over copper. Even the
10 Gigabit Ethernet Alliance’s website stated that 10 Gbps needed to run over
fiber!
What led to change of heart?
The
Shannon Principle
Way back in 1948, Claude Shannon, a research mathematician with Bell Labs,
became the father of information theory with his "Mathematical theory of
communication". Far from simple, his theory tells us that there’s only
‘so much’ information that you can push down a pipe of a certain diameter
pipe. The effect is often called the ‘Shannon Wall’.
What Shannon actually said, comes in two parts. First, if the ‘entropy’
rate or the rate of transmission of information from a source is higher than the
capacity of the channel (pipe), there will be unavoidable and uncorrectable
errors i.e. it won’t work–which makes sense and leads us to believe in ‘Shannon
Walls’.
Second, and more important, is that if the entropy rate (and by the way
entropy rate isn’t bits per second–it’s a very complex statistical
measure) is just a fraction less than the channel (pipe) capacity then,
mathematically at least, there must be a solution whereby input information can
be recovered or reconstructed at the output–even if it is distorted and
subject to noise en-route. In practical terms, this means that the more complex
and clever compression techniques technologists can invent the more information
they can squeeze through a given size pipe without breaking the first rule.
Breaking Shannon Walls
If we look back over history then successively:
10 Mbps Ethernet used no compression coding (1 bit represented 1 bit) and was
near to the then apparent Shannon Wall. But then new technological strides were
made in both bandwidth and coding (or compression) from binary to tertiary
levels (with 6 tertiaries representing 8 bits and reducing the number of symbols
needed to convey the same information by 25 percent)–and the Shannon Wall
moved magically to somewhere just beyond 100 Mbps. There were tradeoffs, of
course; the increased frequencies on the pairs meant that signal-to-noise ratios
became more of a problem with increasing physical length of the channel. But, in
practice, 100 meter was relatively easily achieved and all was well for a while.
In spring 1997, the IEEE formed a study group to look at new technology ideas
that could once again move the ‘apparent Shannon Wall’ to just beyond the 1
Gbps point by using pulse amplitude modulation with five levels (PAM5) and some
very clever forward error correction. And as we all now know, it proved that by
specifying and measuring (controlling but not actually canceling or
counteracting) a couple of new ‘noise’ components which up-rated Cat 5 to
Cat 5e it suddenly became possible to run 1 Gbps Ethernet over Cat 5e, or Cat 6
twisted-pair copper.
And just as they had at every stage along the way, engineers said, "Well
that’s as far as the Shannon Wall will let us go!"
Will Shannon Wall Move Again?
Various companies are claiming that they are dismantling the 1 Gbps Shannon
Wall and can move it to beyond 10 Gbps. In fact, they’ve already done it–to
a point. 10 Gbps has already been demonstrated over twisted-pair copper. But
only over short distances.
Some of the electronics companies are claiming that they can make 10 Gbps
work at up to 20/25 meter over twisted-pair copper cable. One company claims to
have demonstrated 60 meter. The problem that they all have to grapple with is to
recover a wanted signal that is actually much weaker than the noise signal
coming into the receiver–the electronic equivalent of finding a needle in a
haystack.
If the technology has been pushed to 60 meter, then most likely it’s only a
matter of time that it can be extended to 100 meter–though some experts would
currently disagree with this. The big question is whether it can be done
economically and practically. After all, 10 Gbps over copper is only of use to
us in the structured cabling industry. For example, we expect the channel length
in the horizontal to be 100 meter (90 meter fixed cabling and up to 10 meter
cords) and so far we’ve been used to a metric of ten times the speed for three
times the cost in the electronics of Ethernet. And in practical terms, the
success of any 10G solution will depend heavily on whether it can be run over at
least some of the existing installed base of ‘future proof’ Cat 5e and Cat 6
cabling systems.
So What’s Happening?
Experiments and tests undertaken by a number of silicon vendors–the people
who develop the chipsets that can be produced in zillions for £1 or £2 without
which network interface cards (NICs) just wouldn’t be economically feasible–have
given them the belief that 10 Gbps over twisted-pair copper can be done. They’ve
moved the ‘apparent’ Shannon Wall.
Next, they’ve made such a strong argument for their case that the IEEE–developer
and keeper of the Ethernet standards–with the backing of active equipment
manufacturers and cabling systems manufacturers, has agreed to back two study
groups into 10 Gbps over twisted-pair copper.
We need to be clear at this point, however, that these study groups have been
established to consider and report on the feasibility of developing new 10 Gbps
over twisted pair copper standards. There is at this time, no commitment or
guarantee that either of the two approaches will ever become real IEEE Ethernet
standards.
The Twin Projects
The first, 10GBASE-CX4, is targeted at short-range copper interconnects
between active equipment in the Comms Room or DataCentre and for backplanes in
stackable/bladed actives.
Nothing like Cat 5e or Cat 6, 10GBASE-CX4 uses a shielded 8 twisted-pair cord
called InfiniBand. It works by utilizing four sets of two pairs, with each
two-pair set operating in full-duplex and carrying 3.125 Gbps–a total of 10
Gbps. The solution is simple (relatively speaking), and makes use of simple
electronics, signal conditioning and the same 8B/10B coding currently used for
10 Gbps over fiber and at lower speeds for the XAUI interface. The result is a
system that works over distances of between 10m and 15m only–and, in order to
work at all, the bit error rate needs to be less than 1 per 1015 bits–which is
10,000 times fewer errors than permitted in 100BASE-T and 1000BASE-T Ethernet
over Cat 5e/Cat 6 systems.
All said, it’s an important development because if 1 Gbps is to be
delivered to the desktop, switches will need to be interconnected at 10 Gbps at
least. And even though the cost of 10 Gbps fiber interfaces (called GBICs) have
halved over the last year, they’re still around £20,000 each–well outside
the reach of most enterprises. The IEEE study group is looking to produce a
solution that’s only 3—5 times the cost of 1000BASE-T electronics and
cabling–a massive, necessary, reduction.
This project is at a highly advanced stage and there is little doubt that it
will soon turn from feasibility study into a full-scale standard-writing
project.
The second study group 10GBASE-T, which is looking at 10 Gbps over
twisted-pair copper in the horizontal, of course, interests those in the
structured cabling business. But as you can guess by now, getting 10 Gbps
through 100 meter of twisted-pair copper is a mammoth undertaking compared to
the 10/15 meter we’ve just been discussing!
Can 10G over Cat 5e/6 Work?
At the November 2002 IEEE meeting, six electronics vendors reported on their
approach and findings. Four were confident that they would achieve 10 Gbps on
Cat 5e cabling for a distance of 25 meter. Three thought that they would achieve
100m. Interestingly, at this stage, only one thought that Cat 6 would be
necessary. Of those who thought that 100m would be possible, two shared the
similar approach of adding far-end crosstalk (FEXT) cancellation to an extended
1000BASE-T system. Others took radically different approaches with varying
levels of digital signal processing (DSP). All, however, are looking at
massively increased bandwidths, and much more complex PAM coding systems with
eight, nine or even 10 discrete levels, which in turn means that far better
signal-to-noise ratios will be needed than for 1 Gbps systems. Techniques like
far-end crosstalk and echo cancellation are likely to be needed.
To paraphrase Shannon, "It’s not impossible–just very, very,
difficult!"
As an initial objective, the 10GBASE-T study group stated an aim to find a
solution that would work on Cat 5e systems. But before any reader thinks of
making any decisions based on this, here are a few words of caution:
One, exciting though the prospect of 10 Gbps over Cat 5e/Cat 6 is, the IEEE
work is currently only to determine whether 10 Gbps is feasible. No more. No
less.
Two, the various technologies outlined at the November 2002 IEEE meeting
require bandwidths of 250—625 MHz. Currently, Cat 5e is only characterized to
100 MHz, and Cat 6 to 250 MHz; so any solution is going to require Cat 5e and/or
Cat 6 to be characterized to much higher frequencies. Whether this can be done
at all is still unproven, whether it can be done in such a way that all existing
manufacturers solutions become 10G capable is highly improbable. Interestingly,
those existing products with the most headroom at 100 MHz and 250 MHz
respectively are not necessarily the ones that will give the best performance at
higher frequencies.
A major problem will be the vagaries of patch-cords in installed systems,
which can readily exhibit +/-30 Ohm impedance variations when moved. This is
likely to introduce far more echo than the proposed solutions can cope with.
What Happens Next?
The IEEE 10GBASE-T working group will be meeting again to commence its
in-depth study of the various technical proposals for the electronics. This
process will take many months.
The IEEE is likely to request input from ISO/IEC (the international
structured cabling standards body) on the implications of specifying existing or
new transmission performance parameters on Class D (Cat 5e) and/or Class E (Cat
6) cabling systems beyond the existing 100 MHz/250 MHz to, say, 450 MHz. This
response will significantly guide the IEEE in the writing of any subsequent 10G
BASE-T standard.
Other organizations like the TIA (where much of the technical work of
standardizing structured cabling systems is done), CENELEC and the Category 6
Consortium will all provide input and influence too.
Industry sources feel that the market is unlikely to accept another
class/category and that, from a cabling perspective, the most acceptable and
likely way forward would be an addendum to existing Cat 5e and Cat 6 standards
to extend the frequency range to, say, 450 MHz, without any improvements to
transmission performance at the current 100MHz/250MHz limits. This would not
adversely affect existing Cat 5e and Cat 6 products’ and solutions’ ability
to carry 1 Gbps Ethernet.
Most likely, there’ll be a lot of marketing-hype about 10 Gbps over
twisted-pair. Probably even more than there was over Cat 6. Buyers should keep a
strong watch on the developments at the IEEE and the standards bodies, lest they
get carried away unduly.
There’s talk of a solution being available in 2005. However, experience
with Cat 5, Cat 5e and Cat 6 shows that in each case it has taken four years to
reach a true ratified standard.
We can but wait and see!
Keith Ford, market development manager (commercial premises) Krone (UK)
Technique