Grid Computing

[Josiah]

I read this article today and I found it to be very interesting as to F@H.

http://www.foxnews.com/story/0,2933,347212,00.html

You could bind together thousands of computers at once acting as if they were one computer. How long does it take to transfer a 3 gig file from one hard drive to the next? A minute or two right? So essentially what they are saying is that the grid would work faster then normal computer users hard drives can at times.

All F@H computers could be bound together and perform one process or there could be different sub sets. The sum of the parts could be greater then what they were before.

You could also get in to the whole idea of a virtual computer. If F@H needed more RAM it could grab ram from this computer and that computer and use the CPU from the other computers. It could optimize computer calculations by grabbing from the strengths of each computer. If the speed barrier of the internet was broken, well some amazing things could happen!

I just found it to be an interesting idea for F@H. You think F@H could find a use for this?

[7im]

I found this part interesting...

Additionally, the grid is being made available to dozens of other academic researchers including astronomers and molecular biologists.

It has already been used to help design new drugs against malaria, the mosquito-borne disease that kills 1m people worldwide each year. Researchers used the grid to analyse 140m compounds - a task that would have taken a standard internet-linked PC 420 years.

The FAH project has 200,000 internet linked PCs, so it would only take FAH 18 hours to finish that study. The connection speed isn't so much of the problem as is the processing power for FAH. But more speed certainly couldn't hurt.

[science man]

7im is right. As far as these projects go it's not how faster our computers send the info back and forth but how fast our processors can compelete WU's. If you want fast internet though find an internet provider that support fiber optic wiring that should be fast enough you for a while.

[bruce]

7im is right. As far as these projects go it's not how faster our computers send the info back and forth but how fast our processors can compelete WU's. If you want fast internet though find an internet provider that support fiber optic wiring that should be fast enough you for a while.

Like anything in Engineering or Science, there are always trade-offs. For FAH to work, you have to be able do divide a big problem into a reasonable number of smaller problems that can be worked on independently. Each of those smaller problems must be able to contain enough work that a PC can work on it independently for a certain number of hours or days before returning the result. Clearly that's what FAH is doing. After each smaller problem is completed, there's additional time spent uploading, resynchronizing that result with others (if necessary) and downloading the next assignment. The time spent uploading, synchronizing, and downloading must be small, compared to the time spent computing.

A different trade-off is found inside of the SMP client. In this case, a downloaded WU is divided into four smaller jobs and each of those jobs is assigned to a different CPU. How is that different, you may ask. Well, the four sub-jobs cannot be completed independently of each other. Each of the four jobs processes a (very) small amount of work and then it must stop and synchronize with the other three jobs before they can proceed. The speed at which these four tasks can exchange information is orders of magnitude faster than the internet, but it's fast enough that the four jobs can be completed a lot more quickly than if only one CPU was working on the job. The amount of information that is exchanged between these four tasks multiplied by the number of times the data must be synchronized gives a HUGE total amount of data moved, but that's a trade-off that produces better/faster results than is possible with classic FAH.

It's a matter of having the right tool for the right job.