The story of computing is no stranger to the tale of the exotic
turning into the mundane. Furthermore, technology that was
developed for the specialist applications needed by scientists has
repeatedly migrated into the world of corporate computing, writes
Julia Vowler.
"Think of parallel processing and reduced instruction set
computing," reminds John Barr of Sun Microsystems. "Once they were
weird and wacky. Now, they are a general part of our technology."
What will be next to move across to corporate IT? "The most obvious
technology is grid computing," says Barr.
Grid computing - or metacomputing or utility computing as it is
sometimes known - works on the premise that "spare" capacity and
resources can be siphoned off as required, rather like drawing
electricity off a grid.
"Fundamentally, [grid computing] enables a virtual organisation to
use distributed computing resources in a fairly ambiguous way,"
says Barr.
"The nirvana would be for you to go to a device, access a portal to
run your application 'somewhere' and get billed for it on your
credit card."
Before the nirvana of a national computing grid can be achieved,
users can implement three stages of grid computing: local, campus
or global. All three stages exploit spare cycles, or unused
capacity, on existing machines, but with varying levels of tasks
that can be performed.
The lowest stage is a cluster, such as all the desktops on a
network. This collection of resources is owned by one single
department and the task is to match the workload to the available
power.
The next stage is a campus or site. This, says Barr, is bigger than
the cluster, but no more complex. You will, however, need tools to
manage the workload and agree inter-departmental use of spare
capacity.
The final stage is to go global, where a task could potentially run
anywhere in the world, which means that data has to be safely
transported and securely authenticated.
Computer farms are unexceptional within the world of technical
computing. Not just because technical users find it easier to use
novel technology, but because the jobs they are processing tend to
be computationally intensive.
Most importantly, these tasks are easily divisible into standalone
units of work that can be farmed out and completed remotely using
spare capacity on a machine, overnight, for example.
Barr acknowledges that business applications may be less suited to
grid computing of this nature.
"Technical computing usually requires you to run 'this job with
this data' - or 100 jobs simultaneously - and the process of moving
jobs around [the grid] is straightforward. But the business
environment is more complex - you have many interlinked
applications and operations that are dependent on large databases
attached to one particular machine," he says.
Nevertheless, given the current global economic downturn, corporate
IT is increasingly looking to contain costs, or better still reduce
them, so the challenge to get the most out of existing capacity is
very real. And the amount of unused capacity in corporate IT could
be considerable.
"The typical utilisation of technical computing is high, but in
commercial IT it can be as low as 30% utilisation," argues Barr.
"The flexible use of resources can really drive down costs. One
user I know got a three-fold increase in throughput on what he
already had, because he was using unused cycles."
Barr argues that if companies want to manage grid computing,
certainly at cluster level, the necessary are already in place. At
the global level, however, companies will probably have to use an
open source tool kit.
In the realm of corporate IT the idea of grid computing is still
novel, but it may be the next technology to move out of technical
computing into the business world.