The idea of 10 Gigabit Ethernet was seen as the glittering solution
to network problems: a bigger, faster, even fatter pipe that
maximises a network's performance. But does the average IT
department really need such capacity, asks Philip Hunter
If someone had scripted the Ethernet story 16 years ago when the
technology first hit the commercial IT world, no one would have
believed it. Time and again it has withstood challenges from
apparently more efficient, better performing, network transmission
technologies, such as Token Ring, fibre distributed data interface
and asynchronous transfer mode (ATM). And in all cases Ethernet has
seen them off by reinventing itself.
Some would say the success is illusory, arguing that the
original 10 megabits per second (mbps) shared Ethernet running over
coaxial cable, and the emerging 10 gigabits per second (gbps) for
fibre transmission, which is 1,000 times faster, resemble each
other in little more than name. The underlying physical
transmission has been transformed to meet far more stringent timing
requirements, and it was once argued that Gigabit Ethernet was less
like Ethernet than ATM.
But the question is, should users care? It is becoming clear
that the answer is no, given that the industry has solved most of
the technical problems, and indeed ensured that there is at least
some continuity in management between the various iterations of
Ethernet.
For IT and network managers, the onus now is on following the
various developments with the emerging 10 Gigabit Ethernet
standard. While they may not need these speeds now, it is going to
have a big role in emerging broadband networks, at least in the
mid-term to immediate future. It may, in fact, turn out that 10
Gigabit Ethernet is the last of the long line of Ethernet
improvements and that further increases will be achieved by
developments at the physical optical layer, notably with dense wave
division multiplexing.
But, meanwhile, Ethernet has been lucky, benefiting from
parallel improvements in transmission technologies, plus commercial
developments that mean it is now poised to take over in the wide
area network (Wan) as well, having already won the battle in the
local area network (Lan). In this way Ethernet at last promises to
provide an end-to-end transmission protocol carrying data in the
form of IP packets. ATM emerged as a likely candidate to provide
this ubiquity during the early 1990s, but technical and commercial
developments since then have worked against it.
As Fred Engel, chief technical officer of the US networking
system supplier Concord Communications points out, ATM was
developed for an era when the Wan was a serious bottleneck with
limited bandwidth, and there was a need for a protocol that could
optimise traffic efficiently, and guarantee quality of service for
those real-time applications such as voice and video that could not
tolerate much delay. To provide this optimisation, ATM had to be
complex and difficult to manage. And now, with Wan bandwidth
becoming plentiful because of all the fibre being laid, this is a
price no longer worth paying, says Engel.
Their complexity made ATM switches, and the network interface
cards for attaching systems to the network, expensive. This issue
of cost also hindered ATM's penetration into the Wan, according to
Roger Hockaday, vice-president of strategic marketing for Alcatel.
Hockaday's view is that ATM could only have prevailed if suppliers
of ATM systems had adopted the same sort of enlightened approach
that the mobile phone industry has taken with handsets, subsidising
them to make them affordable and create the critical mass that
would have led to subsequent economies of scale.
"The business model was wrong, with ATM costing five times as
much as Ethernet for comparable performance," says Hockaday. "The
general rule is that people will pay for added value, but only at a
premium of 20% or so."
Similar considerations are now giving Ethernet a leg-up into the
Wan, and at 10 gbps the next variant looks likely to be adopted by
carriers and service providers that currently use ATM widely within
their wide area switching infrastructures. This will enable them to
change the shape of networking by providing high-speed Ethernet
interfaces allowing customers to extend their existing Lans over
long distances with no bottleneck and no sacrifice in
performance.
At first sight the only difference this appears to make is one
of speed. After all Lans have been interconnected for well over 12
years, with routers from suppliers such as Cisco usually used. But
the routers have always provided a point of discontinuity, with
different protocols such as frame relay and ATM often providing the
wide area transmission. The physical Lans at each site had to be
managed as separate entities, and although it was possible for a
user on one Lan to access say a server on another at a remote site,
this required a layer of processing and management that both
increased complexity and reduced performance.
With the growing dispersion of teams and workgroups over
multiple sites, this increased the number of complaints from users
about the network being slow. The idea of the virtual Lan emerged
to describe a fluid workgroup that could be dispersed in this way,
but this comprised little more than a set of tools to ease
administration. Nothing changed beneath the covers.
But if it is possible to extend Ethernet Lans with no break
either in protocol or bandwidth over long distances, than the word
Lan really does become a misnomer. And the long-awaited dream of
the unified end-to-end network will have become a reality. This is
a little way off, but progress is rapid, and to judge by analysts'
predictions, will come to pass within two to three years. According
to Gartner, the worldwide market for the top end 10 Gigabit
Ethernet switches, which has barely started as the standard is only
just firming up, will kick off properly in 2001 with an estimated
$71m (£44m) in sales. The market is then projected to increase
tenfold to be worth $700m in 2002 and reach $3.6bn in 2004.
Assuming that such projections are right, and that 10 Gigabit
Ethernet does take over the networking world, providing end-to-end
data transmission, some issues will be raised and questions begged
for IT managers.
An obvious question is how to migrate from existing enterprise
networks, and whether equipment such as routers will need
replacing. At first sight routers, whose role has been to connect
multiple Lans together over either short or long distances, will
become redundant as networks flatten out over what in effect would
be one huge geographically dispersed Lan.
But at second sight, the role of routers, while changing in some
respects, will be as important as ever. There will still be a need
for a wide range of control, management and security functions,
many provided by routers which, after all, have expanded enormously
in scope since the early days when they just ensured that data got
to the right destination Lan.
Routers of some form will also need to be retained for another
reason, which is to provide buffering capacity at each site to hold
data about to be transmitted in case it needs to be resent. The
point here is that, occasionally, packets are dropped during
transmission and need to be resent. For this to be possible copies
of the packets need to be held temporarily within a buffer until
the destination confirms that they have been received. The required
size for such buffers increases in proportion to both the
transmission speed and the time taken to reach the destination.
The latter also increases over longer distances and so will be
greater with the new wave of wide area Gigabit Ethernet services.
And 10 gbps is very fast. Therefore, according to Bill St. Arnaud,
senior director of network projects for Canada's advanced Internet
development organisation, CANARIE, current router/switches, which
already have large buffer memories, will predominate in the 10
Gigabit Ethernet market for customer premises equipment.
There will continue to be some complexity to deal with in
managing end-to-end 10 Gigabit Ethernet networks, and partly for
this reason there is likely to be a shift towards greater
outsourcing, often to the carriers providing the long-distance
Gigabit Ethernet services.
This beg the questions of how soon most enterprises will need 10
Gigabit Ethernet. To an extent this projected surge in network
transmission capacity rekindles the old debate over whether the
best response to that perennial moan from users, "the network's too
slow" is to throw bandwidth at it or improve efficiency. As Marc
Droulez, chief technical officer of Avaya, formerly the enterprise
network group of Lucent Technologies, points out, there is a
tendency to fill the bandwidth available.
But even this tendency will take time to consume even the single
gigabit capacity backbone networks that are in place within many
enterprises today, and for this reason Droulez expects there will
be a relatively small niche for 10 Gigabit Ethernet in the next
year or so, very much in line with Gartner's initial
projections.
"There's always been a subset of customers throwing bandwidth at
the problem," says Droulez.
A somewhat different view is taken by Kevin Johnson, divisional
director for IP business solutions at network system supplier
Getronics. Johnson argues that precisely because applications tend
to fill the available capacity within a purely managed network, it
is important to try to thwart that tendency and avoid continually
throwing money at the problem.
"Today's 100 mbps or 1 gbps Ethernet networks will probably be
pressurised with voice, data and video, and we will need to look at
how you manage that data," says Johnson. "Yes, if we had 10 Gigabit
Ethernet, we'd be able to back off that for short time, but the way
the trends are going, the focus needs to be on management.
Concord's Engel vehemently disagrees with this line, but
irrespective of how this debate goes, there is one group of
enterprises that will definitely soon need Gigabit Ethernet. This
is the Internet service provider (ISP) sector and particularly
concerns the Internet Exchanges that act as interconnection points
between different ISPs. For example the London Internet Exchange
(Linx) will soon be implementing 10 Gigabit Ethernet as a matter of
urgency, according to its executive chairman Keith Mitchell.
"Currently, we run at 3.2 gbps within our network
infrastructure, and with demand doubling every 100 hours, it
doesn't take a mathematical genius to see that we will quickly run
out of capacity," says Mitchell.
ISPs whose networks are interconnected by Linx do not yet need
10 Gigabit Ethernet, but it will only be a matter of time before
many do, Mitchell reckoned.
One point worth clarifying is the fact that 10 Gigabit Ethernet
will not immediately provide the full 10 gbps capacity over a
single channel. Initially, 10 gigabit backbones will in fact
provide multiple lower speed channels whose sum total is 10 gbps,
just as happened at lower speeds with single Gigabit Ethernet. At
first, therefore, the full 10 gbps will only be available as an
aggregate capacity rather than to any single system. Not that
anyone would need 10 gbps to their desktop anyway, but it will take
a year or two before it will be possible to pump data at 10 gbps
down a single channel.
It then remains to be seen whether there will be yet another
extension of Ethernet to 100 gbps. It may well be that further
increases in bandwidth will be provided by optical technologies,
transporting the IP packets directly bypassing the Ethernet
layer.
But it is just as likely that there will be yet another twist to
the Ethernet tale.
Into the blue
How soon will you need 10 Gigabit Ethernet?
The case for Gigabit Ethernet is undeniable for Internet
exchanges, and increasingly for other service providers as an
alternative to ATM. But for most enterprises, the need for it can
be delayed by making more efficient use of the bandwidth already in
place. According to Matthew Bell, product manager at Fluke, one of
the world's largest suppliers of network test equipment, many IT
managers respond to the familiar call of users, "the network's too
slow" by throwing more bandwidth at the problem, when in reality
the bottleneck lies elsewhere.
The problem is, they lack the tools to prove it.
"I think there is a real argument to be made that typical
current bandwidth levels are more than adequate for the vast
majority of situations," said Bell. "What seems to be lacking in
some situations is that the network manager doesn't have the time
or tools available to get a good idea about how well the network is
actually running.
Bell admits that hot spots do occur within networks, where extra
bandwidth is needed. But even these hot spots will not need 10
Gigabit Ethernet, and indeed will require single Gigabit Ethernet
at the workgroup level.
So really at the enterprise level, 10 Gigabit Ethernet will only
figure as a backbone technology interconnecting a significant
number of power users with top-end Unix workstations.