The proving of 10 Gigabit Ethernet as a viable protocol for long
distance networks is nearing completion - but will it ever get
used?
Ethernet has long been the strongest technology in local networking
and is now threatening to break out into the wide area to create a
single end-to-end transmission standard.
Longer-distance Ethernet services interconnecting enterprise
networks are already available at the metropolitan level and the
impending 10 gigabit version will extend these to greater
distances. But the 10 Gigabit Ethernet standard, due for final
ratification early 2002, will be relevant not just in the context
of wide area network (Wan) and metropolitan area network (Man)
services, but will creep into enterprise networks and become a
consideration for IT network managers.
But this will not happen just yet. Many enterprises are only now
considering moving to single gigabit Ethernet from 100 megabit per
second Fast Ethernet for their backbones. So as Steve Dunwoody,
technical director of Equinox Converged Solutions, a network
installation company, points out, Lan network managers are unlikely
to be holding their breaths for the arrival of 10 Gigabit Ethernet.
"I can't imagine many are stretching their existing 1 gigabit
backbone extensively, particularly with very low cost link
aggregation available as a single upgrade step with most
suppliers," he says. Dunwoody is referring here to the facility of
many existing gigabit Ethernet switches within the campus or Lan to
increase network capacity by aggregating two or more single
gigabit, or indeed 100 megabit, links together to create, in
effect, a larger multiple Gigabit Ethernet backbone.
However, the links that can be aggregated in this way are limited
by the ports available and also by fibres in the ground -
fibre-optic cable is needed to transport Gigabit Ethernet for
distances greater than 100 metres. Therefore as demand for capacity
increases, it is likely that firms will have to upgrade to 10
gigabit Ethernet per link, with the potential to aggregate those as
well in future.
According to Dunwoody, such demand will come first for clustering
large servers and for connecting network attached storage to Lans.
It is likely that network service providers will be the first to
require 10 gigabit Ethernet internally, in Web hosting for example.
In the more immediate future, most firms will be interested in 10
gigabit Ethernet for wider area services which will mean end-to-end
Ethernet transmission for the now ubiquitous Internet Protocol (IP)
traffic. Although IP delivers data through a multi-link network,
there has to be another protocol controlling access and data flow
across each link.
Within the Lan, Ethernet has become the predominant data link
standard. But there are various contenders for the Wan with
asynchronous transfer mode (ATM) being the most common. It provides
transmission, traffic management and quality of service control on
top of a synchronous digital hierarchy (SDH) - or synchronous
optical network (Sonet) as it is called in the US - physical ring
structure which helps with resilience and recovery.
With 10 gigabit Ethernet, companies will be able to connect their
networks over Wans via existing SDH/Sonet infrastructures without
having to worry about any of the other Wan protocols. This will
enable organisations to envelop all their sites in effect with a
large Lan infrastructure, with no sacrifice in throughput over
longer distances.
This fits well with the trend towards wider distribution of
applications and workgroups and outsourcing of core datacentre
services. Julien St John Dennis, director of local Internet
marketing for Nortel, says demand for 10 gigabit speeds is
surfacing among larger datacentres for these reasons.
"Large datacentres and tall shiny buildings will constitute the
first marketplace for long-distance 10 gigabit Ethernet services,"
he says. "But in a year, we may find it has moved into the lower
end of the large corporate marketplace." The availability of 10
gigabit Ethernet services may encourage some firms to hasten plans
to outsource their datacentre to a hosting company and access all
the applications remotely, according to Steve Pettit, technical
director of network integrator Enterasys.
But in the first instance, demand for multiple gigabit services
will be driven by the growing mobility of workers, Pettit argues.
"That, more than anything, will drive demand for multi-gigabit,
along with the fact that, as well as being faster, the new services
will be cheaper."
Cost will provide a case for 10 gigabit Ethernet exploiting
developments in transmission over fibre-optic networks, in
particular dense wave division multiplexing. This enables up to 40
separate 1 gigabit channels to be carried over a single fibre and
the limit will increase to 1,000 soon. It is the availability of
huge capacities that make long-distance multiple gigabit services
possible.
In this context, 10 gigabit Ethernet is just another iteration in
the development of the world's favourite Lan protocol. However,
there is still a limit to the distance between any two nodes on a
pure Ethernet network, depending on the physical transmission
media. With 10 gigabit Ethernet, improvements in optical
transmission technology have been exploited to increase these
distances, but the limit is still 40km over the highest grade
single mode fibre.
Therefore, to extend networks between cities and countries, some
means of connecting local Ethernet networks via SDH backbones was
needed. This has been accomplished with 10 gigabit Ethernet in the
form of a new Wan interface. It is now possible to create a single
global Ethernet Lan, albeit with an intervening SDH network.
Within the Lan, Ethernet defines not just the access and framing of
data at the link level, but also the transmission at the physical
level. The ultimate aim of Ethernet suppliers is to extend the
Ethernet physical layer right across the Wan, eliminating SDH/Sonet
altogether. Then the Ethernet framing protocol would run directly
over the optical fibre with no intervening protocol, just as
happens now within the Lan.
Already there are some single gigabit services providing Ethernet
using optical wires in metropolitan areas over distances of up to a
few kilometres. The 10 Gigabit Ethernet standard specifies minimum
distances of 40km over the highest grade carrier fibre. But as Ed
Hopkins, product marketing manager for the Ethernet system supplier
Extreme Networks, points out, suppliers will soon be exceeding
these distances and it is likely that direct 10 gigabit Ethernet
transmission over fibre will be extended to 200km within the next
year or so and eventually globally.
But before this can happen the world's public networks must reach
the same level of development and some are far from the all-fibre
infrastructure necessary for Ethernet over optical transmission. In
the meantime, digital subscriber line and cable modem technologies
will allow more bandwidth to be extracted from existing copper
networks.
A more relevant issue for network managers is deciding what the
implications will be within their enterprise domain. The first
point to emphasise is that, unlike all previous versions, 10
gigabit Ethernet will only work on optical fibre, a version for
copper will not be available. A number of enterprises have
implemented single gigabit Ethernet using existing copper cabling
over distances of up to 100 metres, encouraged by suppliers such as
3Com, which have introduced affordable products that do not require
fibre.
When the time comes to move to 10 gigabit Ethernet for parts of the
core network, it will be no longer be possible to delay migration
to fibre so 10 gigabit Ethernet will change the balance between
fibre and copper, bringing fibre closer to the users even it does
not bring about a complete meltdown of copper.
But it is not a question of copper versus fibre. There are
different grades of fibre, with various wider bore multimode types
predominating within the enterprise. The distances these cables can
carry high frequency signals is not as great without regeneration,
but the terminating electronic devices cost considerably less.
Narrow bore single mode fibre is more expensive to terminate, but
can sustain greater distances for a given bandwidth, and so is used
in wide area carrier networks. There are different versions of the
10 gigabit Ethernet standard for each type of optical cable, but
there is doubt whether some older corporate networks installed for
the 100mbps fibre distributed data interface backbone rings will be
able to cope with the higher speeds. So existing fibre networks may
have to be replaced as well.
There is also the question of whether existing Ethernet hardware
within the enterprise will be able to cope with the higher speeds
just by swapping modules, or whether it will need to be completely
replaced. The well-worn cliché, about not wanting forklift upgrades
applies here.
Network managers will also be watching the fate of new wide area 10
gigabit services with interest, in the hope that technical teething
problems are ironed out there first. One of the major concerns
relates to the spanning tree algorithm which is used by Ethernet to
recover from failures of links or switches within the network.
Ethernet has shed a lot of its original baggage but spanning tree
remains. The problem is that at the much higher speeds possible
today, the algorithm takes too long to converge.
The need for spanning tree arises because Ethernet is an open
network rather than a closed loop, so to provide alternate routing
between two points, an additional protocol is necessary to flip
over to a secondary path when the primary one fails. Spanning tree
accomplishes this, but can take up to 60 seconds to reconfigure in
the event of a failure. Accordingly a new version of the protocol
is being developed, but it remains to be seen whether this will
solve all the problems on large wide area 10 gigabit networks.
Another issue set to erupt concerns storage area networks (Sans).
These are being implemented with another protocol, fibre channel,
providing the data transmission at 1gbps. Although it requires
another set of skills, fibre channel is being delivered as an
integral part of Sans and is a more efficient protocol than
Ethernet.
According to Nigel Houghton, sales manager for northern Europe a
storage system supplier Gadzoox, Sans is under control of
datacentres rather than networking departments, and so the argument
for Ethernet carries less weight. "Fibre channel is what guys in
datacentres needing high availability data require," says Houghton.
Yet the momentum behind Ethernet has carried all before it in the
past and telecoms suppliers are developing techniques to transport
data within Sans on top of IP and Ethernet.
The main effort in this field is the Internet small computer
systems interface (iSCSI), which is an IP-based version of the SCSI
protocol used for connecting servers with storage subsystems. This
is only just being standardised, but Houghton admits it will
provide a lower cost alternative to Fibre Channel for Sans and is
likely to appeal to smaller firms.
Pettit says the telling factor will be demand from users for access
to Sans. "Users will say, 'We have this fast system for sharing
information. How do we get access to it?'" This will make it
desirable for Sans to be accessed via existing Ethernet-based Lans.
Another factor could be the support of Ethernet for longer
distances, given that many firms will want to disperse Sans over
multiple sites or outsource them.
We must remember Ethernet has yet to prove itself as a
long-distance protocol. It certainly will, but there is room for
doubt given the time it is taking to build in the recovery features
implicit in technologies such as SDH/Sonet.
What is the spanning tree algorithm?
1. An algorithm used in transparent bridges that dynamically
determines the best path from source to destination. It avoids
bridge loops (two or more paths linking one segment to another),
which can cause the bridges to misinterpret results
2. The algorithm creates a hierarchical "tree" that "spans"
the entire network including all switches. It determines all
redundant paths and makes only one of them active at any given time
3. The spanning tree protocol is part of the IEEE 802.1
standard.
Highlights of 10 Gigabit Ethernet
1. Only full duplex - simultaneous data transmission both ways
2. Optical media only (1 gigabit Ethernet had copper
version)
3. Preserves existing Ethernet frame format, with same
maximum and minimum frame sizes
4. Still requires spanning tree algorithm for
reconfiguration and fault recovery but new fast version of
algorithm on way
5. Comprises two families of physical interfaces - local
area network (Lan) family as before and also new wide area network
family, starting with version for accessing synchronous digital
hierarchy and synchronous optical network infrastructures at speeds
of almost 10gbps
6. Lan physical layer specifications include version for
single mode fibre at distances of at least 40km, and for existing
installed multimode fibre of at least 350 metres.
Ethernet's past, present and immediate future battles
Ethernet has seen off competition from numerous rivals, all of
which at the time offered technical advantages. The first challenge
came in the late 1980s from Token Ring within the local area
network.
Then for backbone campus networks, Fast Ethernet at 100mbps
overtook fibre distributed data interface rings operating at the
same speed during the early to mid-1990s. Subsequently, during the
late 1990s and into the millennium, 1 gigabit Ethernet defeated
asynchronous transfer mode (ATM) even more comprehensively at the
campus level.
Now 10 gigabit Ethernet is moving the fight beyond the enterprise
for the first time, taking on firstly ATM (again), and eventually
will overcome the synchronous digital hierarchy infrastructure and
its US counterpart the synchronous optical network.
Then new battle lines will be drawn across the enterprise for the
storage area network, where fibre channel is the rival technology.
Finally, at lower speeds, the wireless version of Ethernet will go
head-to-head with Bluetooth for connecting laptops and personal
digital assistants to peripherals within buildings. The problem
here is that both technologies compete for the same frequencies in
the radio spectrum.