Today, there are two approaches to increase the
performance of computer hardware. The first, scale-up, involves the
installation of increasingly powerful, multi-processor server
hardware, to enable IT directors to achieve the greatest
performance from a single server.
The
HP Superdome,
Unisys ES7000 and
Sun Fire E20K are examples of machines that
allow IT directors to get the most from a single server,
enabling them to simplify their datacentre by consolidating
hardware.
The second approach is scale-out or grid computing. Here, rack
upon rack of low-cost commodity PC servers running
Intel or
AMD processors can be configured as
computing grids. Under the right circumstances, a given piece of
computational work can then be divided up so that each processor
in the grid is able to run a portion of the overall task. As a
result, the more processors available to the grid, the faster
the computational task can be completed.
One example of such a grid is run at investment bank Merrill
Lynch, where a follow-the-moon policy means that complex trading
and risk analysis algorithms can take advantage of spare computer
capacity as the working day ends in different regions of the
world.
Analyst firm Gartner defines a grid as the application of server
resources from more than one owner to address single, large tasks.
For some organisations, this means that they run and manage part of
the grid, and a partner organisation runs and manages a different
part. For others, a grid might be run and managed in-house but span
global locations.
Grid computing also covers external providers that lease
computer resources that can be combined with in-house assets in
order to boost the processing of certain large tasks, says Carl
Claunch, research vice-president of servers and storage at
Gartner.
Grid technology is gaining recognition among businesses because
grids can be used as a pool of computing resources, regardless of
the location of the servers. This allows operators either to
schedule or to allocate jobs according to which areas of the grid
are free.
Since their inception in 1995, grids have offered the scientific
and academic communities a way to solve taxing problems such as
protein folding, earthquake simulation and climate modelling. Grids
are also being used to manipulate DNA data and to search for
extraterrestrial life.
Now grids are infiltrating the commercial world, and users are
harnessing huge amounts of computing power to boost their
business.
Global banking group HSBC uses grids to carry out derivatives
trades, which rely on making numerous calculations based on future
events. The bank also uses grids for risk analysis.
Defence and aeronautics firm BAE Systems is using a
Microsoft-based grid, hosted by the University of Southampton, to
improve the aircraft design process by modelling aerodynamic flight
data.
Grids are not only linked to number-crunching tasks. Automobile
magazine and website AutoTrader is using grids to support the rapid
expansion of its online motor exchange operations. Were it not for
the investment in grid technologies, a Unix server upgrade would
have been necessary, which would have meant a capital investment of
more than £1m, AutoTrader said.
The company’s grid runs an Oracle 10g database and Real
Application Clusters on Sun v40z Opteron servers running Red Hat
3AS.
Oracle, Microsoft and other suppliers have entered the grid
software market with products such as Oracle 10g and Microsoft
Windows Compute Cluster Server 2003.
These systems run on commodity Intel and AMD server hardware
platforms such as Hewlett-Packard Proliant or Bladesystem
servers.
In addition, grid software specialists such as DataSynapse and
Platform Computing have helped businesses to develop grids through
the use of their grid software platforms and consultancy
expertise.
Such platforms offer IT management and application devlopment
support for grids. This enables the application to scale across
the grid to take advantage of free servers.
Business use of grids is still low compared with academic and
scientific use, but it is growing, says Michael Speyer, senior
analyst at Forrester Research. Forrester’s 2006 data shows grid
usage for US enterprises at 8% of total grid use, and this 8%
contains a wide spread of industries.
The biggest users of commercial grids are finance organisations
and insurance firms, according to Speyer. These sectors lead the
way in commercial computer grid usage, at 12% of total enterprise
users.
Insurance companies need to carry out risk analysis to comply
with industry regulations, and grids offer a way to carry out
large-scale computations of risk.
Next are public sector enterprises, which account for 8% of
business users. These include government departments as well as
education and healthcare services.
Manufacturing firms are the third highest users of grids, at 6%.
The aerospace, automotive and high-tech manufacturing industries
use computer grids for design automation chores.
Oil and gas firms use the technology for seismic exploration and
reservoir analysis, and pharmaceutical firms use grids for the
discovery and modelling of proteins, new drugs and genetics.
Pharmaceutical firms face a competitive market as they race to
get new drugs out early, and each day gained means millions of
pounds in additional revenue and profits, says Claunch.
Business services and retail and wholesale organisations are
also starting to adopt enterprise grids, according to
Forrester.
Computer grid technology is also being used for the rapid
analysis of new consumer packaging and logistics planning, as well
as for data transformation for datawarehousing and analytics.
But grid technology is still considered to be in its infancy and
has several problems associated with it. “Grid computing has many
issues that arise directly from the multiple ownership domains that
participate,” says Claunch.
Software licensing methods used by third-party software
suppliers can also be a problem. Like virtualisation, grid
technology might split an application over multiple servers, so if
a product is licensed on a fixed number of machines, or by specific
server serial numbers, this may break the licence agreement.
“Most of the new and more challenging problems that arise with
grids are a consequence of the multiple management points:
conflicts over priorities, difficulty with cost allocation,
security questions, and challenges over funding upgrades to the
grid, among others,” says Claunch.
Some users have also admitted that when problems arise in the
grid, trawling through the server logs to locate the problem can be
a complex and time-consuming exercise.
But Claunch argues, “For those early adopters of grid computing,
the business benefits are so large that they quickly overshadow the
issues.”
Read the Merrill
Lynch case study
Read the HSBC case
study
Licences could
hinder grids
http://grid.globalwatchonline.com
Comment on this article:
computer.weekly@rbi.co.uk