I think there is a slow subtle shift of emphasis happening in Grid Computing. When I first heard about it, the essential principles were:-

1) The grid task taps into the unused spare computing capacity of a machine while the user is using it anyway.
2) Thousands or even millions of machines as in (1) share the required computing load.
3) Because of (2), the power or degree of availablility of each individual machine does not really matter, it is the SUM that matters.

So, by the original principle, while I am writing this email, my spare computing power (most of it at the moment) is being used to crunch WCG's worthy workunits. When I finish this, if I have no other work to do I shut my PC down, having done my work and made a useful contribution to important medical research at the same time. Most people would find it hard to find reasons NOT to help in this way!

Increasingly, grid computing does not work like this. Many of us leave machines on when not actually using them, or even all the time, because we believe in the causes. Some crunchers are even prepared to upgrade machines (memory etc) to meet requirements. This is fine, because in effect the costs of the extra computing effort becomes a donation to charity. A grid composed entirely of such "Premium" machines would be a grid scientist's dream. It would be so easy to control and use efficiently - like the best supercomputer!

Obviously it is diffcult to program and use a grid according to the strict principles, and in practice compromises and constraints are necessary. But it worries me that the trend may be moving too far. Didactylos mentioned 10 hours as an average completion time (I know that the absolute time limit is much longer). As the limits get tighter, I can see that the cruncher who chooses to contribute using the "strict" principles, may eventually be squeezed out. At the moment, we even have an option to "Do work only when computer is Idle", for those who want reassurance that their computer performance is unnaffected. This option, and the option to shut down a computer may become unusable (even for fast machines) if the power hungry trend continues. The situation gets worse for Laptops - which are on average less powerful and powered up for shorter periods. Laptops are very popular!

I think we should strive therefore to treat the original principles as an important design consideration for WCG workunits, and make them as small as practicable. I know that there are limits on how much a task can be broken down. What about the solution they adopted at Climate Prediction? Here, the workunits were necessarily HUGE, but they had a system of INCREMENTAL reporting and crediting during the processing of a WU.

David Hobbs - Thanks for your contributions, and please do not be discouraged. You might like to look at another worthy project for your "little" machines - http://www.malariacontrol.net/ - I don't know what the min requirements are but as they are in Beta at the moment they are releasing VERY SMALL WORKUNITS which process like lightning.

nice work