I attended an interesting workshop this week. It was one of the series that Peter James has put together for his SusteIT project; this one focussed on desktop PCs. The talks and panels looked at measurement procurement of energy use, procurement options, desktop grids, power management and thin clients. In any sizeable organistion, Desktop PCs use a large amount of electricity and there are many options available for reducing this consumption - and saving money too. This is the second time that I've seen a British university do the sums and expect to save £250,000 a year.
The panel on power management tools was interesting. These seem to be coming of age at last. Operating systems have had support for managing individual computers but a large organisation needs a system for managing thousands of PCs, with different policies for different groups, and of course the important facility to wake up in time for the distribution of updates.
James Osborne gave an interesting analysis of the FLOPS/Watt providing by Cardiff University's Condor pool, as compared to their old cluster (inefficient) and the one they installed last year using modern cooling systems. The new cluster is much more efficient, so it seems that if you have the funds to install modern dedicated technology for high-throughput computing, that may be the way to go. Of course, if you don't have that sort of money, a desktop grid will make good use of your existing infrastructure.
Unless, of course, you go for the thin client option. As Queen Margaret University have shown, a thin client approach can dovetail very neatly with energy-efficient buildings. Conversely, Merrill Lynch found that they used more energy in the server room than they saved on the desktop. The moral, I believe, is that you have to measure carefully the gains and losses for your particular installation. Case studies from elsewhere can give hints about what to look for and approaches to consider, but they are no substitute for performing your own analysis.
The panel on power management tools was interesting. These seem to be coming of age at last. Operating systems have had support for managing individual computers but a large organisation needs a system for managing thousands of PCs, with different policies for different groups, and of course the important facility to wake up in time for the distribution of updates.
James Osborne gave an interesting analysis of the FLOPS/Watt providing by Cardiff University's Condor pool, as compared to their old cluster (inefficient) and the one they installed last year using modern cooling systems. The new cluster is much more efficient, so it seems that if you have the funds to install modern dedicated technology for high-throughput computing, that may be the way to go. Of course, if you don't have that sort of money, a desktop grid will make good use of your existing infrastructure.
Unless, of course, you go for the thin client option. As Queen Margaret University have shown, a thin client approach can dovetail very neatly with energy-efficient buildings. Conversely, Merrill Lynch found that they used more energy in the server room than they saved on the desktop. The moral, I believe, is that you have to measure carefully the gains and losses for your particular installation. Case studies from elsewhere can give hints about what to look for and approaches to consider, but they are no substitute for performing your own analysis.
Comments