I have a question about DC in general.

D

dbu8554

Supreme [H]ardness
Joined
May 7, 2002
Messages
5,159
Ok so 1 GPU folding is awesome
2 is better, and as other people have shown us 4 is like woah!

Same with CPU a p4 single core is better then nothing
2 cores and 4 cores are better.

But in the end its only one computer at a time working on its own work units, maybe its me but would it not be better if they had a program that lets say took all of our CPU's and GPU's and put them to work all together at one time.

I dont know enough about netowrks or how DC works maybe but the way I see it is this.



Right now stanford has a mountain they need to move, they are using shovels.
Then they said woah why don't we get other people to help us.
Thats an awesome idea so people show up for work and they say you are on your own dude, you dig with what you got, some people like the dude with 4x280's has the crew from a mining operation, other people with just a CPU are digging with those little plastic shovels they give kids.

What I see them doing is instead of hiring one large organized company to move the mountian they hired anyone and everyone who wants to show up, large corporations showed as well as little people.

Back to what I am asking, why don't they design a program for all of us to log into and work together it seems like it would be faster does anyone get what I am saying?

Is the technology not there or is it there but its just not efficient just wondering thanks alot.
 
Well actually you've hit on the essence of the project. The work units themselves are just a small part of much larger equations. Stanford spins small sections (work units) off to individual folders and then combines them once they're completed. Even our quad-gpu guys aren't anywhere near powerful enough to run the calculations needed for a full simulations. By distributing the work out in sections, we're basically each tied into what amounts to a large super-computer.
 
What you had in mind is a kind of cluster and this doesn't work exactly like that. Instead of putting everything together doing a single calculation, they are basically breaking the calculations in tiny parts then distribute them to every computer to crunch it. In the end, when the bits get returned, they reassemble it together to get the big picture. The shovels analogy is good and while a shovel is a slow way to move a mountain, the sheer amount is what count the most.

IIRC, F@H crunching power is over 2.5 Petaflops, much more than the best 5 supercomputers. There isn't even a supercomputer breaking a petaflop barrier yet.

 
Has anyone seen an article explaining what Stanford does with our returned WUs?
When I first started folding, I searched and couldn't find anything. I would like to
know the process they go through to stich our completed WUs into something
useful.

For example, I've wondered if the process is all automated, or does it require some
human intervention? And, once the WU data are stiched together to show the
folding process, what process detects mistakes in protein folding?

If anyone knows of a link that explains this, please point me in the right direction.

Folding on ...

 
The only thing that bothers me about the Stanford project, is that they use thousands of volunteers to actually create the data (even if they do alot of work on there own getting it to the point they can use those people) but yet don't make the data openly available. The release papers about findings in journals, but don't ever release the large amounts of raw data they have available. I can't help but think that there is possibly something they are missing that someone else may discover if they would just release the raw data.
 
Without a background in biophysics and biochemistry, I can only venture a guess but here's a good one :

Each WU returned will be added in a huge database (we are talking petaflops of data) and when they setup a new project, they already calculated a "expected" trajectory then each WU will give as result a certain trajectory so if it works as expected, it should be trajectory++ until the end. If it does end up diverting from the trajectory, they will see it then create new projects with a few variables to determine what cause this. It's like if you launch a ball for 1 Km... You know that normal theorical physics will tell it's a specific curve from A to B but something in the way (wind, bird, etc...) will divert this and that's what they are trying to find.

In the end, if protein misfolding happen under some circumstances, a drug could be created to fix the circumstance causing the misfolding.

I hope this very very rough explanation will shed the light on how F@H works.

 
MasterOfPupets said:
The only thing that bothers me about the Stanford project, is that they use thousands of volunteers to actually create the data (even if they do alot of work on there own getting it to the point they can use those people) but yet don't make the data openly available. The release papers about findings in journals, but don't ever release the large amounts of raw data they have available. I can't help but think that there is possibly something they are missing that someone else may discover if they would just release the raw data.
Click to expand...

It's not quite that simple. If you recall the discussion of the potential Storage@Home project, this is an initial idea of how to share the vast amount of data collected thus far. We're talking petabytes of data. Even if you had the disk space to store a copy, thats a lot of bandwidth to be asking Stanford to foot the bill for every time someone else wants a copy. And with the rate at which we're crunching WU's today, think about how fast that database grows, even on a day to day basis.
 
MasterOfPupets said:
The only thing that bothers me about the Stanford project, is that they use thousands of volunteers to actually create the data (even if they do alot of work on there own getting it to the point they can use those people) but yet don't make the data openly available. The release papers about findings in journals, but don't ever release the large amounts of raw data they have available. I can't help but think that there is possibly something they are missing that someone else may discover if they would just release the raw data.
Click to expand...


Looks like they are starting to do what you were thinking about. I would suspect that for anyone that actually wanted the data and had the means and know how to manipulate it, all they would have to do would be to ask Vijay Pande and I'm sure something could be worked out. Academic collaborations happen all the time.

http://folding.typepad.com/news/2008/07/sharing-fah-results----raw-data-now-available.html

 
You must log in or register to reply here.
Back
Top