The News-Gazette.com:UI professor seeks to harness strength of amoeba into computer cores

UI professor seeks to harness strength of amoeba into computer cores

By Greg Kline

Monday, July 23, 2007 6:56 PM CDT

Sure, you can look down your nose at the lowly amoeba, with its single cell, but the little protozoan has a trick up its sleeve, if it had a sleeve, that people certainly don't.

It can split in two to make a copy of itself, which can make a copy of itself, which can make a copy of itself, which can ...

Before long, you have an amoeba army.

University of Illinois Professor Rakesh Kumar thinks the amoeba may be a good model for new generations of super-powerful computers. Not only in supercomputing centers, but in your desktop, your laptop, the gaming console in your living room, even the palm of your hand.

Kumar, a UI electrical and computer engineering professor, is looking at what he calls "amoebic computing" as a way of taking fuller advantage of the growing number of "cores" on computer processors.

Processors and their cores are kind of like the brains of computers. Traditionally, the chips have had a single core, or been single-brained, so to speak.

Some high-end computers might contain more than one processor to apply multiple brains to a problem at once. Supercomputers can contain hundreds or thousands of processors.

But they've still been, for the most part, single-core processors externally communicating in some fashion to work together.

Now, chip-making giant Intel is making dual-core processors, in essence processor chips containing two brains. IBM's Cell processor, which runs the PlayStation 3, has eight. Those numbers are likely to continue to grow.

"You'll have tens or possibly even hundreds of cores on a chip," Kumar said recently. "It will enable lots of applications."

Kumar sees the technology enabling a better understanding – faster – of the massive amounts of business data, say, Wal-Mart generates every day, and truly realistic games, not to mention computers with which you can converse.

That is, if the industry finds new ways to take fuller advantage of the computing power offered by all those cores.

"We don't know how to program them," said Kumar, whose research focuses on computer processor architecture and programming.

Humans tend to think sequentially and to make computers work that way – finish one step, then move on to another, then another, and so on.

Computers can blow through a series of tasks really fast, but they could work exponentially faster if they were able to employ multiple processors to do several things at the same time, so-called "parallel processing."

Supercomputers do that to some extent. But they're hobbled, among other things, by the relative slowness of the networks linking their processors, which limits them to a large extent to big problems where the advantages of parallel processing are so blatant they outweigh the disadvantages.

Multicore processors with their cores on the same chip don't have that kind of overhead problem, making them attractive for doing a greater variety of things in parallel.

Kumar's idea is to break sequential programs into a bunch of component parts, or services, and farm them out to the available cores.

"You just smatter them all over the chip," he said.

Moreover, the system could, like the amoeba, replicate those parts and run them on more than one core as needed, if the task was particularly large, important or frequent, for instance.

Besides the replication aspect, the system would be like an amoeba in other ways, Kumar said.

Made up of software and, eventually, purpose-designed hardware, it could adapt to its environment, for example, changing how it handles things as new demands arise and old ones drop off, in the same way amoebae change shapes in response to conditions around them.

Among other things, it could shift work where advantageous, to the closest processor space maybe, or one with more free resources. It might even set aside, create, new spaces for handling tasks as needed.

That also happens to be similar to the way the world's most parallel computer works – the human brain.