As the cost of running increasing quantities of data along ageing
copper networks spirals, migrating to optical technology may be the
most cost-effective option for network managers. Philip Hunter
reports
Traditional copper-based network solutions such as 10/100 Base-T
Ethernet are slowly but inexorably being squeezed between the
advancing waves of fibre and wireless solutions. Fibre is invading
enterprise backbones for campus-wide inter-workgroup communications
and server consolidation, while wireless Lans are taking over in
places where mobile users congregate.
For the rapidly growing area of storage area networks (Sans), fibre
is the only show in town. Based on the Fibre Channel standard, Sans
use the same fibre physical access mechanism as Gigabit Ethernet.
The growth of Sans is giving fibre a leg up into corporate networks
and this is persuading some IT managers that it is the only viable
solution for future broadband communications - at least within the
backbone.
Fibre has not yet made significant inroads into mainstream
workgroups for desktop connections, but the rapid increase in CPU
power and the imminent arrival of the Infiniband bus technology are
leading to PCs capable of transfer rates of 1gbps - the point at
which copper cabling begins to run out of steam. Some enterprises,
particularly in the financial sector, are installing fibre as a
matter of course to provide greater capacity to support emerging
technologies, even if they do not need the bandwidth now.
Companies whose business embraces multimedia technology already
need gigabit bandwidths and for other enterprises fibre is certain
to receive consideration soon. Decisions on whether to migrate
cannot be delayed much longer.
IT managers encountering fibre for the first time will have to
dispel several myths. The first is that by cabling with fibre you
are automatically future-proofing your network. The truth is that
fibre optic cable has gone through almost as many iterations as
copper in the journey to higher bandwidths. Manufacturing standards
have improved, as has the type of fibre preferred for given
applications.
According to Peter Kjeldsen, an analyst at Gartner Group, the
situation with optical transmission has been clouded by the fact
that the economics of fibre change as you move down from long-haul
carrier trunk networks running hundreds of miles underground to
campus networks within buildings. In between are metropolitan area
networks (Mans), which cover up to 20 miles.
Optical transmission over long distances originally required
electrical signal regeneration every few miles. This was done by
using repeaters that amplified the degraded signal and then
reissued it into a new set of fibres. Over a long-haul network, a
large number of these repeaters were needed, adding enormously to
the cost, so it was worth investing in narrower-gauge fibres,
called multimode, that could sustain the signal for greater
distances. Fewer repeaters were then needed, but the cable itself
was more expensive, and so this was still a significant cost.
Then, about 10 years ago, a technology called DWDM (Dense Wave
Division Multiplexing) entered the long-haul network arena,
bringing costs down substantially.
DWDM saves on the number of fibres needed by multiplexing multiple
channels along a single fibre. Each channel is encoded onto a given
wavelength of light so, by carrying a number of different
wavelengths, a single fibre can carry multiple channels. The
current limit is about 100 channels, which is a total bandwidth per
fibre of 250gbps if each channel is 2.5gbps.
In a few years it is likely that a single fibre will be able to
carry 1,000 channels this way, each at 10gbps, bringing the total
capacity to 10tbps. Lucent Technologies has already demonstrated
transmission at 30tbps along a single fibre in its
laboratories.
Additional cost savings come from the fact that DWDM allows a
single repeater to be used for multiple channels, rather than
needing a separate repeater for each fibre.
An alternative would be to have one large channel, rather than
carrying multiple channels along a single fibre, that is, to have
one 400tbps channel instead of 40 10gbps channels. This would
require just one sophisticated line card at each end of the link to
drive the single channel.
Whereas DWDM, while needing single repeaters for each fibre, has to
have a line card at the source and destination of every channel
encoded over that fibre. With a single channel solution, the
repeaters are more expensive but the total line card cost is less
than for an equivalent DWDM capacity.
Therefore, the economics depend on the distance being traversed,
with DWDM more likely to score over the longer haul.
A third option is to have multiple fibres, each carrying a single
channel at current speeds. This is cheapest in terms of the
terminating line cards, but more expensive than either of the other
options in terms of repeaters and cable. This solution is more
likely to win at the enterprise level and, in some cases, for Mans.
A further consideration is the grade of the fibre used.
Another myth is that DWDM technology will sweep all before it and
eventually take over on an end-to-end basis. There is a
seductiveness about DWDM's elegant configuration, whereby
applications and user groups can be segregated onto their own
wavelengths, that appears to make administration much simpler. But
none of this matters if the economics are not right. And they are
unlikely to be in the case of DWDM at the enterprise level, at
least for the foreseeable future - it will be just too
expensive.
Although there are uncertainties at the physical level, there is
only one winner for the link-level transmission within the optical
network at the enterprise level - Ethernet. Gigabit Ethernet is
purely optical and is now installed in a number of enterprises. The
next version of the standard for 10gbps transmission is about to be
ratified, and will be implemented at the metropolitan level, and
later in enterprise networks.
The significant aspect of 10 Gigabit Ethernet, apart from its
greater speed, is the incorporation for the first time within
Ethernet of a Wan interface. This enables enterprise or
metropolitan Ethernet networks to be extended over long distances
via public Sonet/SDH networks.
SDH is the physical transmission technology used in public carrier
networks within Europe. Sonet is the US version. In effect,
Ethernet is bridged over the Sonet/SDH network, with virtually no
drop in speed, so end-to-end transmission at 10gbps becomes
possible.
According to Mark Iliffe, European support manager at networking
equipment supplier Foundry, 10 Gigabit Ethernet is a highly
significant development for enterprises, even if they do not
appreciate it yet. As soon as they run out of capacity at the
single gigabit level,
10 Gigabit Ethernet will provide a more cost-effective upgrade path
than the alternative of having multiple single gigabit links.
"A lot of enterprise customers today are aggregating single gigabit
ports, which is alright until you run out of fibres in the ground,"
he said.
This comes back to the trade off between installing more fibre or
upgrading to a higher speed transmission technology.
A similar trade off calculation is having to be made by many
enterprises as they agonise over whether to migrate existing campus
networks from copper to fibre. There comes a point when a migration
to optical technology cannot be delayed any longer and even becomes
the cheaper option, as the cost of pumping more data down ageing
copper networks can become prohibitive.