Tomorrow's Metacomputer

Expo/Computation/The Metacomputer

Metacomputing Tomorrow

There's no doubt that the speed of microprocessors will continue to increase by leaps and bounds, and that computer scientists and engineers will find better ways of storing, sharing and moving data. But metacomputing is already with us. Much of the metacomputer's hardware--compute machines, mass storage systems and information servers--is basically in place.

Nevertheless, just as it took time to build power stations and to create a network that could distribute electricity to every home, it will take a quite a while longer before the large-scale metacomputers become commonplace. And, like any emerging technologies, those of metacomputing--and our expectations of them--are constantly evolving.

Metacomputing's Holy Grail
Roadblocks to Metacomputing
Metacomputing for all?

Metacomputing's Holy Grail

From today's standpoint, the ideal metacomputer would

Capitalize on the strengths of individual computers
Distribute any given computational problem appropriately over the available resources
Pull the results to an appropriate visualization display environment
Take additional input and alter the computation "on the fly"
Perform all of this "transparently" to the user

Of all the capabilities promised by metacomputing, the transparent user interface is the most eagerly awaited by researchers. "That's the holy grail," says NCSA Director Larry Smarr, "to make it seem like the only computer you're using is the one your fingers are touching."

Roadblocks to Metacomputing

NCSA and other supercomputing centers have already shown the potential of metacomputing on both local and, in a limited sense, a national scale. But many hurdles remain before metacomputing is no longer just a grand concept but becomes an everyday reality.

Software

Compatible families of computers, like those within each of NCSA's three clusters, speak the same computer language. But because configurations of metacomputers on a national scale will more likely be heterogeneous, there is an urgent need to develop more robust, "universal" computer languages like Message Passing Interface. Then there's the the problem of optimizing the use of diverse machines for a given problem. Compilers must be smart enough to recognize which parts of a program are best tackled by distinct sets of processors.

Moreover, there needs to be improved communication between the different processors constituting a metacomputer, so that each processor "knows" when the others are done with their parts of the calculation. All this will also require better software to efficiently automate scheduling of the different parts of a program for computation. In short, there's no doubt that efficient metacomputing will require major advances in software.

Networks

The national metacomputer demands faster networks with more access points. The speed with which data moves over wide-area networks is but a fraction of local area networks; local area networks. High-speed networks must be available not only to a few supercomputing centers but to every school, business and home.

User Interfaces

Metacomputing demands superior user interfaces enabling easy, rapid access to diverse computational resources in a fully "transparent" manner; that is the user need have no knowledge of the myriad "widgets" in hardware and software lying behind the interfaces. Today, the World Web Web and the multitude of browsers it has spawned, particularly the newer, more interactive browsers such as "Hot Java" , could provide a foundation for such interfaces to emerge.

Already, Web viewers can interactively run simulations at remote sites and display the results locally. Coupled to virtual environments and the latest collaborative software, the Web will enable tomorrow's metacomputer users to control faraway instruments and computers, navigate through their data with ease and, in realtime, share their insights with colleagues across continents or oceans -- all this at the click of a mouse or its equivalent in virtual reality.

Metacomputing for All?

Engineers and scientists will likely be first large-scale users of continental-scale metacomputing resources. They'll spend more time doing research and less time tinkering with the nuts and bolts of computation.

But what about the rest of us? How will metacomputing help the small-town physician, the high school student, the farmer?

Here are three scenarios:

A small-town doctor in may practice for years without seeing an unusual case. When a patient walks into the clinic with a hard-to-diagnose illness, the doctor turns to a specialist. But when the nearest major hospital is 50 miles away, that specialist may be digital. The physician runs some diagnostic tests and sends the results and the patient's profile over a high-speed network to a computerized database at the National Institutes of Health in Bethesda, Maryland. In minutes, the physician receives information about the probable cause of illness, pertinent electronic reprints of papers from medical journals, and a treatment plan.

Classroom computers could open a window to high performance computing resources. Students may learn about tornadoes by running a simulation on a supercomputer hundreds of miles away and seeing the results displayed on a classroom screen moments later. In scaled-down virtual environments, they'll be able to compute and walk around molecules, or immerse themselves in three-dimensional mathematical objects which can be rotated, distorted, or compacted at will.

Farmers may use their personal computers (perhaps wireless laptops) to access public databases containing the latest information about commodities markets, long-term weather forecasts and crop prices worldwide. At the click of a mouse, remote supercomputers on the network turn all this information into a long-term economic forecast, sending the results back to the farmer, thus enabling him or her to make informed decisions about next year's planting strategy.

When might these scenarios become reality? The continued pace of innovation in computing and communications technology suggests that it won't be very long. In fact, if you've perused this exhibit, we think you'll agree that metacomputing is no longer the stuff of dreams but will soon become a fact of life.