Information Technology Reference
In-Depth Information
protein structure determination, which are part of the drug development process.
Back in 2003, Novartis had about 65'000 desktop PCs (Salamone 2003). It was
assumed that about 90% of the computing cycles of each PC were unused. At the
same time, the needs for computing power at Novartis research were constantly
increasing. In general, computing power is increasingly essential to drug develop-
ment. Pharmaceutical researchers rely on “in silico” experiments to explore drug
actions, speed the development cycle, and reduce the need for expensive, robotic-
controlled physical experiments (Intel 2003).
Novartis made Grid Computing a strategic part of its five-year plan for adding
R&D computing resources. The goal was to link existing PCs into a Grid and share
the unused computing cycles for compute-intensive research tasks.
The building of the Novartis PC Enterprise Grid started with a pilot involving
50 PCs in Basel, Switzerland. The 50 PCs were Grid-enabled and connected with a
standard LAN in a star-like setting within a day of time. The PCs were Grid-enabled
by installing a client agent on each machine that checks for idle compute cycles and
donates them to the Grid while ensuring the Grid does not impact the work of the PC
user (Intel 2003). The connection via simple LAN with a bandwidth of 100MBit/s
was sufficient for tasks that are compute- but not data-intensive. The average size of
computing tasks communicated to the Grid nodes was about 1 Mbyte (NZZ 2003).
The Grid was controlled by monitoring software running on a dedicated server that
managed the workflow, assigned tasks to the nodes and assembled the results. The
results of the pilot exceeded expectations: within a week the Enterprise Grid of
50 PCs provided 3.18 years of additional aggregate processing time (Intel 2003).
Based on this encouraging result from the pilot, the Grid was quickly extended to
include 2700 PCs located in Basel, Vienna (Austria) and Cambridge (USA) (Intel
2003). The total computing power of the extended Novartis PC Grid reached up to
5 Teraflops. At that time, this equalled the computing power of the European Centre
for Medium-Range Weather Forecasts in Great Britain, which was ranked no. 15 on
the official list of super computers in the world (NZZ 2003).
The Novartis Enterprise Grid also illustrates how taking advantage of existing
computing resources can deliver additional performance at a fraction of the cost
for purchasing, deploying, and managing new systems. Instead of buying an HPC
system, building another computer centre, and employing the people to support it,
Novartis made an investment of roughly $400'000 in Grid software licenses and
saved at least $2 million of investment that would have been necessary if new infra-
structure was bought in order to gain the same computing power (Intel 2003).
The Novartis Enterprise Grid created business value by improving the innova-
tion power and competitiveness of the company. On the one hand, with the Grid
it was possible to speed up time-to-market in the competitive drug development
process. On the other hand, with the Grid it was possible to extend research activi-
ties and to perform research tasks that have not been possible before. For example,
research that would take computation of six years on a single computer can be run
on the Grid in 12 hours. The Novartis Enterprise Grid was leveraged also by other
business functions, for example for advanced data mining by business analysts
(Intel 2003).
