You're a military commander, with tens of different teams
scattered throughout a mountainous area. The enemy is difficult to
see, because it's foggy and raining, and your teams may need to
call in an air strike over their GPS-enabled digital links at any
time. But with all these mountains in the way, will you get the
signal?
The easiest way to know what is going to happen to something is
to build a copy of it to experiment with, but some models are
easier to build than others. Building a second server to test how
software will behave if a component fails is one thing, but how do
you replicate an entire digital network? Forget a mobile radio
network spread around a mountain range, few if any companies have
the wherewithal to recreate what's in their building.
Luckily, there is an answer. Companies have taken to building
logical models of physical networks to help them understand how
they will perform in the future. Network emulation and modelling
tools have been available for several years. Some of them, like
Scalable Network Technologies' QualNet, came from military
research. SNT emerged from a DARPA-funded project to explore ways
of predicting network operation during just this sort of military
situation, and are now being applied in the commercial space
too.
Others are provided by companies such as Shunra Software and
Opnet Technologies. They can help companies conduct "what if"
scenarios, analysing what will happen to their network if traffic
peaks beyond a certain point, for example, or projecting the result
of consolidating several links into one for cost-saving
purposes.
Pradeep Singh, senior vice-president of engineering in the
network management solutions group at Opnet, argues there is a
difference between network modelling, and network emulation. The
former involves recreating a logical copy of your network and
analysing traffic flow. "You have to build a simulation model that
faithfully represents a production network to do any kind of
accurate analysis," says Singh. A network model can be useful for
analysing packet flows across a network and finding security holes
such as misconfigured firewalls, for example.
Emulation takes things a step further by running real-world
traffic in real time across the virtual model of your
infrastructure to see how it copes. "The traffic representation is
very important for many of the performance-related modelling
practices," Singh says.
Opnet's competitor Shunra sells modelling and emulation software
under its virtual enterprise (VE) banner. VE Modeler discovers and
models a network topology. VE Network Capture is a software-based
tool that captures the network behaviour, and both of these tools
fuel the VE Appliance, which emulates the network.
"You can run real traffic through our network in a box, and
experience applications in a lab as if they were running thousands
of miles apart," says senior product manager Gene Litt.
Because end users are interested in applications and not network
packets, it's important to consider application performance and
latency rather than simple bandwidth when emulating a network,
warns Chris Knowles, head of solutions at network consulting
company Dimension Data. "If a customer has three datacentres, and
they want to close two and move all the applications into the
third, how well will the application perform in the remote country?
That would be a typical scenario that a customer might want to
model," he says.
Producing a paper-based analysis won't necessarily answer that
problem. Some of the more successful projects that Knowles has seen
involve connecting an application to the virtual network along with
several end-user workstations, and seeing what it feels like.
With the rising importance of wireless technologies in the last
few years, the need to model and emulate these networks has also
increased. However, as Rajive Bagrodia, CEO of Scalable Network
Technologies explains, the parameters affecting the performance of
a wireless network can be even more complex. In addition to the
technological considerations inside a wired network, you have to
consider everything from weather conditions to the changing number
of wireless devices in a given area, and how all of this may affect
radio signals.
"Because there are so many variables affecting the physical
environment, the most important part of computing a wireless
network is computing the signal path as a result of them. That is
what makes it computationally intensive," he says, adding that the
firm takes advantage of parallel computing to speed up
simulations.
Whether you're halfway up a mountain about to call in an air
strike, or rigging up a VoIP system and trying to predict what it
will do to your existing LAN, network modelling and emulation can
be a good way to find out. Better know now, and avoid getting into
a combat situation with your managing director.