Distributed computing

Posted at: Tuesday 29th June 2004 by Ian Betteridge

You might just use your bog-standard 2.4GHz PC for writing inane messages to your friends, but it could become the most powerful computer in the world, or at least part of it. Ian Betteridge explains how

You probably reckon your Vapochilled 4GHz Pentium 4 PC is pretty powerful, but not even the most ardent overclockers could class their PC as a supercomputer. But with the right software, and a few thousand other machines lending a hand, that's exactly what it could be - part of a supercomputer that uses 'distributed computing' principles to crunch some serious numbers. Not only that, but it could also contribute to important scientific projects that range all the way from medical research to the hunt for a real-life ET.

What's it all about?
The principle of distributed computing is fairly simple. Most computing tasks can be split up into smaller chunks, and these can theoretically be calculated on different machines at the same time. So, if you split a task into enough pieces, you can then execute it on a large number of machines in parallel, achieving results that are well beyond the reach of all but the most expensive supercomputers.

Of course, this principle of executing application parts in parallel isn't just confined to distributed computing; it's similar in many ways to the massive amounts of parallel processing used by supercomputers. The Terascale Cluster at Virginia Tech, for example, uses 1,100 Apple Power Mac G5 machines connected with InfiniBand cards over Gigabit Ethernet.

Supercomputers, however, use dedicated machines and processors to maximise performance. What makes distributed computing different is that, instead of dedicated machines, it uses spare CPU power scattered around a network, which can be as large as the Internet itself. And with the average desktop computer having a processor with far more welly than the average user needs, there's a lot of spare power around.

The history
As with most good ideas, the concept of distributed computing isn't new. The first working distributed computing system emerged in 1973, with a project at the University of California in Irvine, USA, known as DCS (Distributed Computing System, naturally). This used a purpose-built operating system running on DEC minicomputers, including one of the first micro-kernels, and a special file system that allowed you to run any computing task across a set of connected computers. As a general-purpose distributed computing system, it had a far more advanced architecture than most of the systems used now. Yet because of the vast amount of power in modern PCs, even a small distributed computing project today has many more computing resources available.

What's more, DCS required a dedicated network, called The Ring, to work properly. Fast-forward to the mid-1990s, and thanks to the Internet, the requirement for a dedicated network has vanished. Instead, a variety of computers are now working across the Internet on projects such as GIMPS, which is an effort to find the largest prime number, while distributed.net was launched in 1997 to break a 56-bit encryption code. In 1998, distributed.net took only 39 days to break a message encrypted using 56-bit DES - an encryption algorithm that was previously unbroken, which demonstrates the awesome potential of distributed systems.