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

However, few of the clients available can make direct use of multiple processors, which, on the face of it, means that you're not going to get much extra performance out of a dual- or quad-processor system. However, there are two strong caveats to this. First of all, multiple processors mean you can achieve higher sustained performance while you're working on something else, as your client will be able to use the entire resources of one processor, while the other is devoted to email or whatever your other task is. This, of course, depends on how efficient your operating system is at spreading applications across processors, but with any modern Windows or Linux OS, this shouldn't be too much of a problem.

The second way a multiprocessor system can help is by running several instances of the same client in parallel, with each assigned to a different processor. For example, if you have a dual Opteron machine, running two SETI@home clients will allow you to effectively double your performance with almost no performance hit. And, because SETI@home uses small (340KB) chunks of data, you won't get a bottleneck between your memory and CPU either.

You can also ensure that your client is running at its absolute peak by making sure it stays in your Level 2 cache at all times, giving it very fast access to the CPU. Any screen activity - including a screen saver - is likely to hammer the CPU and push your client out of cache, so ensure that all screen savers, including any ones that are part of the client, are turned off.

Then, if you're leaving your machine running, unplug the mouse to make sure that it doesn't trigger any activity, and ensure that there are no background processes running, including the 'fast search' indexing in Windows XP and Office. Finally, if there's a command line client available, use that instead of any GUI program - believe it or not, the GUI can take up to 60 per cent of the available CPU time by simply drawing fancy pictures and contributing nothing to your processing performance.

Oh, and one small word of warning - if you're running a distributed client on a laptop, then it will significantly reduce your battery life when it's not running from the mains. This is because applications that constantly use up processing cycles will make any power management software go insane.

The environmental impact
Of course, the best way to achieve results in a distributed computing project is to leave your machine on 24 hours a day, seven days a week. But there are drawbacks to this, and one of the key ones is the impact on the environment.

Leave your PC on all that time, and you'll use up around £66 worth of electricity per year, consuming about 0.12kW per hour. Of course, if it's just one machine on its own, then the environmental impact is fairly negligible, and certainly a lot less than the impact of making the machine in the first place.

However, once you've multiplied that by all the machines in an average distributed computing project, then it starts to become more serious. In one study for Tufts University, researchers worked out that leaving its 4,300 PCs on 24 hours a day would require over three million kilowatts per year, while producing the required electricity would release 2,100 tonnes of carbon dioxide (the main greenhouse gas) into the environment. Absorbing these emissions safely would require around one million extra trees - something that, in times of declining forests, is extremely unlikely.