automate computer sharing
Ted Smalley Bowen,
Technology Research News
How many economists does it take to inaccurately
forecast a recession?
To answer that impertinent puzzler, you might try tapping into a grid
of computing power made up of spare cycles from sources as disparate as
a university supercomputer across town, a cluster of servers in another
state, and a scattering of workstations around the world. You'll also
need some stray disk storage and spare networking
resources to tie it all together.
Grid computing started as a response to scientific users’ need to pull
together large amounts of computing power to tackle complex applications.
These ad hoc assemblages of distributed resources are coordinated by software
that mediates different computer operating systems and manages things
like scheduling and security to create sophisticated, virtual computers.
Grid computing, still generally confined to the research community, is
one manifestation of utility-style data processing services made possible
by the Internet. Peer-to-peer computing, which allows disparate users
to dedicate portions of their computers to cooperative processing via
the Internet, is a related phenomenon used mostly by consumers and businesses.
Both models harness a potentially vast amount of computing power in the
form of excess, spare or dedicated system resources from the entire range
of computers spread out across the Internet. The University of California
at Berkeley, for example, coordinates one popular scientific example of
grid computing -- an Internet community application that uses background
or downtime resources from thousands of systems, many of them home PCs,
to analyze telescope data for the search for extraterrestrial intelligence
A group of researchers at Monash University in Australia and the European
Council For Nuclear Research (CERN) in Switzerland has proposed a scheme
that has the potential to increase the reach of grid computing by applying
traditional economic models - from barter to monopoly - to manage grid
resource supply and demand.
The researchers have built a software architecture and mapped out policies
for managing grid computing resources; these could also work with peer-to-peer
applications, according to Rajkumar Buyya, a graduate student in the computer
science department at Monash University.
The methods could facilitate a broad range of computing services applications,
“They can be used in executing science, engineering, industrial, and commercial
applications such as drug design, automobile design, crash simulation,
aerospace modeling, high energy physics, astrophysics, earth modeling,
electronic CAD, ray tracing, data mining, financial modeling, and so on,”
Although peer-to-peer and grid computing are not new, there hasn't been
an overarching scheme for handling the massive amount of bargaining and
staging required to carry out such on-demand jobs with reliable levels
of quality, and pricing to match, Buyya said.
The researchers' scheme is aiming to fill that gap, he said. “We are focusing
on the use of economics as a metaphor for management of resources and
scheduling in peer-to-peer and grid computing, as... a mechanism for regulating
supply-and-demand for resources depending on users'... requirements.”
The researchers scheme allows consumers and computing service providers
to connect and hammer out pricing and service levels. It would allow the
parties to agree on one price for quick delivery of services during times
of peak demand, and another for less urgent delivery, for example.
Resource brokering/sharing tools analogous to Napster will eventually
handle the trade in access to computers, content, scientific and technical
instruments, databases, and software, Buyya said.
“With new technologies, the users need not own expensive [computer] resources.
Resource brokers [can] lease services that are necessary to meet... requirements
such as deadline, spending limit, and importance of the work. Our technologies
help both resource consumers and providers to manage the whole scenario
automatically,” he said.
In a grid computing scheme, consumers usually enlist brokers to procure
computing resources for a given project. Grid service providers make their
systems available by running specialized applications and resource trading
services. A grid market directory links brokers and providers.
The researchers' grid architecture goes a step further, using standard
economic pricing models, such as commodity market, posted price, bargaining,
tendering and auctions, to hash out the terms of broker-provider deals.
The researchers' tools, Nimrod-G Computational Resource Broker, DataGrid
broker, Grid Trading Services, Grid Market Directory, and Grid Bank, work
with existing grid middleware like the Globus toolkit.
The researchers have tested the tools on the World Wide Grid (WWG), a
global network testbed of different types of computers including PCs,
workstations and servers.
Two types of tests simulated brokering, scheduling and execution computing
jobs, and emphasized speed and cost, respectively. The tests used a commodity
market pricing, or fixed-price model. One application scheduled computations
needed for a drug design application that screened molecules, he said.
The researchers used Nimrod-G to aggregate the systems resources as they
were needed. “The resource broker automatically leases necessary resources
competitively, depending on the [users'] requirements, such as deadline
and budget constraints," Buyya said.
Using a more common systems-centric approach would make it more difficult
to provide service levels that can vary from user to user and application
to application, depending on the importance of the problem at the time
of execution, he said.
As the tools get established, they could be deployed for use in production
systems such as Australian Partnership for Advanced Computing (APAC) and
Victorian Partnership for Advance Computing (VPAC) resources for routine
use, said Buyya. “Depending on market forces, we believe that it will
take two or three years for widespread use of economic models for Grid
and [peer-to-peer] computing,” said Buyya.
The researchers plan next to test the methods’ scalability, improve scheduling
algorithms, and update the Nimrod/G broker software to handle more sophisticated
task allocation and management, Buyya said.
The study makes a good start at hashing out ways in which disparate computing
resources can be made available and consumed, according to Lee McKnight,
a professor at Tufts University’s Fletcher School of Law & Diplomacy.
The researchers' contribution is "imagining and testing a standards or
protocol-based framework through which computing resources may be accessed
or shared on the basis of one of a variety of different models for brokering
or trading resources,” he said.
But the way the researchers used the models is artificially limited to
narrowly defined grid computing resources and doesn't address networked
computing services like application hosting and bandwidth brokering, and
quality controls like service level agreements, said McKnight.
The work "is but one element of a yet-to-be defined economic model of
pervasive computing and communications environments,” he said. "The 'data
economy' as the authors call it will ultimately include both [peer-to-peer
and] a variety of other interaction and resource access modes.”
Buyya’s research colleagues were Jonathan Giddy and David Abramson of
Monash University and Heinze Stockinger of CERN.
The work was funded by the Australian Government, Monash University, Cooperative
Research Center (CRC) for the Enterprise Distributed Systems Technology
(DSTC), and the Institute of Electrical and Electronics Engineers (IEEE)
Computer Society. Heinz Stokinger's work was funded by CERN and the European
The researchers are scheduled to present their work at the International
Society for Optical Engineering (SPIE) International Symposium on The
Convergence of Information Technologies and Communications (ITCom 2001)
in Denver, August 20-24, 2001.
Timeline: 2-3 years
Funding: nbsp; Government, University
TRN Categories: Internet
Story Type: News
Related Elements: Technical paper “Economic Models for Management
of Resources in Grid Computing,” Proceedings of the International Society
for Optical Engineering (SPIE) Conference on Commercial Applications for
High-Performance, August, 2001.
Internet stays small world
Tools automate computer
Nanotube kinks control
Hydrogen chip to fuel
Research News Roundup
Research Watch blog
View from the High Ground Q&A
How It Works
News | Blog
Buy an ad link