It's no secret: Computing is moving steadily,
if slowly, toward becoming another pay-as-you-go
service just like water or natural gas.
The concept of utility computing holds
enormous appeal for potential service providers
hoping to create the information technology
equivalent of the electric industry. Other
businesses, too, envision setting up in-house
computing "power plants," capable
of shifting instantly to adapt to changing
demand for computing resources. The vision:
Access to extra processing power or storage
capacity, or bandwidth will be as easy as
turning a tap to fill a tub with water.
Of course, this computing utopia - which
is part of what HP calls the "adaptive
enterprise," --- must overcome some
major hurdles before becoming part of everyday
business life. Among the unresolved questions:
How do you measure usage? And how do you
charge for it?
Creating a computing
HP Senior Fellow Bernardo Huberman and
his team at HP Labs are working toward an
answer. It involves what they call computon,
which is a unit of measurement for the use
of computing resources. (It's pronounced
Essentially, "a computon is a bundle
of processing power, storage, and bandwidth
that can be sold and consumed," says
Huberman, director of HP's Palo Alto-based
Information Dynamics Lab. "
Like many other economic entities, a computon's
value can change constantly, automatically
rising or falling based on factors such
as demand for particular applications, types
of computing resource, time of day and project
In fact, Huberman says, the price of computons
might fluctuate like that of an airline
seat, where the price depends on a variety
of factors: time of departure, seating class,
when it's booked, whether it's a non-stop
flight, whether it involves a weekend stay-over.
Those factors all work together to set the
ultimate ticket price.
Supply and demand set
That's where the computon fits into HP's
Adaptive Enterprise initiative, because
it is a pricing mechanism that would adapt
to shifts in demand, type of usage and other
factors. Huberman's team is examining how
the price and availability of computons
might behave in a market environment, which
is generally governed by the laws of supply
"Markets are very good at doing several
things," Huberman explains. "They
allow people to express their preference
as to the price they're willing to pay for
something. You reveal something about how
much it's worth to you."
In addition, following the "invisible
hand" of economic theory, markets largely
manage themselves, an important asset in
highly automated utility computing. "Markets
don't require a centralized coordinator
to allocate resources," he adds.
Markets also gather and convey information.
For instance, a customer who walks into
a favorite café and finds that the
price for coffee has jumped to $5 per cup
might conclude that either there is less
coffee being produced in the world today
or a lot more people are switching to coffee
instead of drinking something else. Or perhaps
a combination of both factors. Huberman's
team expects to build the same kind of classic
supply-and-demand curves for computing resources
as the ones created for many other commodities.
An IT auction
| One possible market mechanism for allocating
computational resources would be to use electronic
auctions to distribute available computing
resources among users seeking additional power.
Many corporate servers run at only 20 percent
of their capacity, according to Forrester
Research Inc., the Cambridge, Mass.-based
technology research firm. In a market for
IT, anytime such hardware sat idle, it might
automatically solicit bids for the bundles
of information processing encapsulated within
the notion of computons.
"If no one wants it, its price should
be zero," Huberman says. "If a
lot of people want it, it should be higher."
Currently, his team is researching questions
such as how often auctions should be held
and how long they should last. [That research
evolved from Spawn, a market-based system
for allocating unused computing resources
among networked computers, which Huberman
helped develop while working as a research
fellow at the Xerox Palo Alto Research Center
in the late 1990s.]
Huberman hopes to have some solid pricing
schemes developed by the spring of 2004,
and to launch a pilot at HP or a customer
company soon after that. The team is also
developing an auction mechanism inside a
data center that will allocate processing
power among underutilized servers.
Creating a new kind of economy is an ambitious
goal, and Huberman knows it. "It's
not going to be easy, but it's not going
to be totally impossible, either,"
by Anne Stuart