What if accessing the computing power of the planet was as easy
as turning on the tap for a glass of water?
What if you could add computing or storage capacity in minutes
- instead of days or weeks? What if you could pay for technology
resources the same way you pay for utilities like electricity or
water - based on what you use?
Computing as a utility? A team at HP Labs is working to make it
The vision: a world of service-centric computing, where scalable,
cost-effective information technology capabilities are delivered,
metered, managed and purchased as a service.
A worldwide network
An enormous amount of processing power and data storage will be
required to meet future computing demands. Today's data centers
contain thousands of servers. But they could be 10 times larger
in the future -- as many as 50,000 servers -- and consist of commodity
servers and storage connected via a high-speed IP-switched fabric.
At the same time, data centers around the world would be networked
so that IT resources are allocated where they're
That's where HP Labs comes in. Researchers are creating an entirely
new model of computing to develop and manage this vast infrastructure.
They call their model planetary computing. The goal: infrastructure
on demand. Infrastructure that's scalable, flexible, economical
and always on.
"The key is to provide a shared resource pool that allocates
resources to applications on demand," says researcher Rich
Friedrich. "If you need more servers, we'll take them from
the pool. When you no longer need them, we'll put them back. We
think that's an essential element in the future of Internet computing."
Currently, businesses must acquire more processing
and storage than they need simply to cover peak times. Amazon.com
and other web retailers must ensure they can handle peak demand in
November and December, while the IRS has a peak in demand in April.
For both entities, the hardware is under-utilized the rest of the
Adding or reconfiguring infrastructure capacity is slow and costly.
And any regular Web user knows what often happens when a Web site
is flooded with unexpected demand. At worst, the site crashes, and
at best, there's the "world wide wait."
Infrastructure on demand
Researchers' work to solve these problems was recently incorporated
into HP's new Utility Data Center, architecture and software for
the first scalable, programmable data center. The goal is to provide
automated infrastructure on demand with little or no operator intervention.
"You don't have
to call the electric company before you plug in a new refrigerator
and say you need another kilowatt of power," says Friedrich.
"Adding information technology should be just as easy as plugging
in an appliance."
In a programmable data center, the infrastructure
is physically wired once, but can be rewired programmatically, to
meet the changing needs of customers and services. This includes
linking data centers so that resources can be optimized for a region,
a nation or around the world. When it's "off-hours" in
one area, data centers there could be used more fully by providing
services for users in other parts of the world, and vice-versa.
The smart data center
Another challenge is managing all these
resources, especially as data centers approach 50,000 servers.
"There are only so many gurus around who can keep the infrastructure
running," Friedrich says. "As the demand on Internet computing
grows, the experts become fewer and farther between."
Sooner or later, infrastructure growth will outstrip the ability
of people (even gurus) to operate it cost-effectively.
that problem, researchers are developing a data center control system
that will broker between application demand and resource capacity.
The system will determine what hardware or software assets are available
and will then install, configure, deploy, monitor and assure services
on a global scale.
This system will be self-monitoring, self-healing and self-adapting.
That is, services and resources will monitor their own health in
much the way people do, making changes to the system when demand
changes or trouble arises -- calling on experts only when absolutely
Data that follows you
HP Labs is also addressing the need for a planetary storage system,
one that provides anywhere, anytime access to data.
The goal is to allow information to follow people and their devices
so that they can get the content and services they need wherever
they go. Researchers are developing automatically managed storage
systems that would ensure secure data storage, access and delivery.
Finally, researchers are tackling the difficult problem managing
the energy demands of massive data centers. Increased computing
power means more heat and, consequently, higher demand for air conditioning.
The old model of cooling each section of the data center equally
Researchers are pioneering more sophisticated numerical, measurement
and control techniques that allow for the determination of the temperature
and flow distribution within a room, and provisioning the air conditioning
For details about this work, see the technical paper, "Computational
Fluid Dynamics Modeling of High Compute Density Data Centers to
Assure System Inlet Air Specifications." This paper was written
by Chandrakant D. Patel, Cullen E. Bash and Christian Belady of
HP Labs and Lennart Stahl and Danny Sullivan of Emerson Energy Systems
and was included in the proceedings of IPACK 2001, the Pacific Rim
/ASME International Electronic Packaging Technical Conference and
© ASME, 2001
by Jamie Beckett