Computerworld Singapore - Vol. 10 Issue No. 7, 19 - 25 November 2003

Side-bar Stories

Mature metacomputing
Legion and Globus are two of the older metacomputing systems that made distributed computing accessible to programmers.

Grid computing is being touted on the premise that large clusters of computers can be bunched together into a larger supercomputer.

But these virtual supercomputers are generally made and used by large organisations with massive funding, such as government agencies or big corporations.

An example is our National Grid Project, which uses a cool $1.2 million just for the network component to tie up the participating campuses and research centres.

Such computing power is suited for crunching parallel threads of data such as geological simulations which make use of massive parallelism – meaning running many small processes simultaneously.

According to Dr Lee Hing Yan, deputy director of National Grid Office (NGO), the term "grid computing", as far as purists are concerned, applies only to supercomputers.

But that does not mean distributed computing is out of reach for the smaller organisations.

Many people would say grid computing is about distributed collaboration; and the resources used can be CPU -cycles, data or databases, and storage, -devices, instruments, or sensors, he said.

So while metacomputing, in the form of grid computing, is targeted at companies with big iron, it can also run on lots of little "ball bearings".

"Public" distributed computing pushes grid computing using the more consumer side of things – individual volunteers with PC units bountiful in schools, companies and homes. An example is the RC5 project by distributed.net to decode the 56, 64 and now 72-bit RSA encryption space, which helped persuade the United States to relax its ban on the export of 128-bit keys.

P2P file sharing have also popularised the public networking model. And participants are voluntary in giving resources from their desktops over the Internet.

Despite the notoriety of P2P thanks to Napster, the public model has been proven to be effective as being cheap and good.

Lee of NGO said metacomputing associated with PCs, or desktop grid computing, has also been successful within enterprises – like a digital media firm harnessing all the computing resources it owns for rendering animation.

It has been harnessed as well for publicly worthy causes, he said, like Seti@Home, an Internet project that processes radio noise for possible signal emissions from extraterrestrials.

David Anderson, director of the Seti@Home programme, said applications from many areas, such as Seti, biochemistry, and mathematics for -example, have used public computing.

Typically, though, most public efforts have been task-oriented. Unlike grid middleware, the Internet programs are hard-coded to do something, and run directly off operating systems like Windows and Linux.

Anderson said public distributed computing has a potential pool of about 150 million Internet-connected computers versus "grid computing", which uses the computers owned by a particular company or university, typically a few hundred or thousand.

"So public distributed computing has the potential to be much bigger: many petaFLOPs of computing power and exabytes of storage."

The Seti@Home team compared itself to IBM's ASCI White, which rated at 12 teraFLOPS, cost about US$110 million ($191 million) in 2001. Seti@Home, meanwhile, gets about 15 teraFLOPs and cost $500,000 at last count.

Also, "Public computing can be reliable even if individual computers aren't, using redundant computing where each computation is done 2 or 3 times and the results are compared. The same technique can be used to achieve reliable storage," he said.

And since there is no real reward to attract client users, contests or intangible promotions are often used. Seti@Home, for example, appealed to end users with its "cool" factor.

The Seti@Home's story goes thus: Seti (Search for Extra Terrestrial Intelligence) programs use large computers to analyse data from the Arecibo observatory. But because of limited funding, none of the computers could process data deeply nor broadly. So the University of California, Berkeley, leveraged on the popular fascination with aliens, and asked for help from the community.

A classic case of a massively parallel application, the "extra" computing power gathered from Internet helpers is now able to process signals about 10 times fainter than its sibling projects.

But with the grid in fashion, full systems are evolving, based on the desktop.

BOINC (Berkeley Open Infrastructure for Network Computing) is an open source spin-off from the Seti@Home project spearheaded by Anderson.

"There's currently no good platform software for public distributed computing. Each project has to make its own. We're trying to change this by developing BOINC which will provide a software platform for Seti@Home and (we hope) many other public distributed computing projects."

Unlike Seti@Home, BOINC will offer a ready-made package to scientists interested in applying distributed computing to their projects.

	Calendar of Events \| Email Newsletters \| About IDG \| Search \| Go to: