Pay-As-You-Grow Supercomputing

IBM made its first
on-demand computing play of the new year Thursday when it extended its
vaunted ubiquitous computing services to supercomputing.


As with many of its on-demand endeavors, Big Blue is offering customers
the option of scaling software requirements out as much as needed, but
without the fixed costs associated with most software licensing
schemes. Specifically, IBM customers now may choose between buying POWER or
Intel processor-based supercomputer clusters or access the power on
demand, paying for processing power based on the required capacity and
duration of use.


IBM hopes to convince the enterprise that its flexible payment approach
to scalability closely resembles the way the Internet works. Dave
Jursik, vice president of WorldWide Linux Cluster Sales, told
internetnews.com his company is providing a virtualized resource that
customers can draw as needed.


“Customers in some sectors want access to large scale computing power
in short bursts,” Jursik said. “Supercomputing on demand promises to
help turn fixed costs into variable costs, matching supercomputing power
exactly to customer demand.”


Jursik noted that certain market segments — such as digital media and
life sciences — require supercomputer-like power, but only at certain
times in their product cycles. There is also downtime in these sectors,
meaning the supercomputers sit idle. Computing power depreciates after
it goes unused for a year or more. IT managers may feel they are
wasting money if they pay a high price for powerful machines that don’t get
used with regularity. IBM’s strategy, Jursik said, is to give the IT
manager the power to boost or lower the processing power they need.


IBM’s infrastructure of choice for this initiative? IBM will build a
grid of Intel and POWER processors, to be made of hundreds of IBM eServer
p655 systems and a massive Linux cluster with IBM eServer x335 and x345
systems. The first IBM supercomputing hosting facility will be based in
Poughkeepsie, N.Y., with other national and international facilities to
follow.


Jursik said the philosophy is geared for firms who undertake complex,
computing-intensive projects that are short term –months as opposed to
years. IBM’s first customer for this was PGS Data Processing, a
division of Petroleum Geo-Services, which needed supercomputing on demand for
a seismic imaging project to locate reservoirs of oil.


“Seismic imaging services employ the latest numerically intensive
applications, but are also highly cost competitive. PGS has been looking
for a more flexible business model which addresses peak computing
requirements, assures rapid response to our customers, but minimizes long
term, incremental cost commitments to PGS,” said Chris Usher, President of
Global Data Processing.


Rush, who estimated contracting with IBM for supercomputing on demand
could save his firm $1.5 million, said PGS can now scale real-time to
handle requests for important deep water imaging solutions.


“There is a fair amount of demand for this type of capacity — for
large clusters — and there isn’t anywhere else a customer can go to get
this type of power to execute a short-term job,” Jursik said.


In fierce competition with Sun Microsystems’ Sun ONE and Microsoft’s
.NET, IBM’s on-demand
strategy
attempts to offer an easy path to integrating a company’s
disparate and diverse computing applications and operations.

News Around the Web