Folding@Home... Help find a cure!

[CrankyDragon]

http://folding.stanford.edu/

Unlike Seti@Home, searching for something that isnt there, check out Folding@Home, researching
protine folding in an effort to find cures for cancer and other diseases! This one is the real McCoy,
Stanford University is using megaclusters to solve complex problems... Download the little program,
run it in the background, and help find a cure! I myself leave my computer on 24X7, might as well
let it do some good while Im not playing games or buringing DVDs!

Please, pass this on to others you know,
Thanks!
Andrew

PS
If you download and run, please post here so we can measure the post's success! Thanks!

 

[mrhnau]

http://folding.stanford.edu/

Unlike Seti@Home, searching for something that isnt there, check out Folding@Home, researching
protine folding in an effort to find cures for cancer and other diseases! This one is the real McCoy,
Stanford University is using megaclusters to solve complex problems... Download the little program,
run it in the background, and help find a cure! I myself leave my computer on 24X7, might as well
let it do some good while Im not playing games or buringing DVDs!

Please, pass this on to others you know,
Thanks!
Andrew

PS
If you download and run, please post here so we can measure the post's success! Thanks!

Thats pretty cool. Glad alot of people are doing it, especially since I'm in the field! wish I could use some of those computers LOL

From my experience though, I don't know how efficient it is. Its not a linear increase in efficiency. Even among local supercomputer clusters, when you get too many nodes/CPU's, you start losing efficiency. Communication costs become prohibitive, especially when doing things like protein folding/molecular dynamics. I'd be interested in seeing their stats...

BTW, they claim to have smashed the millisecond barrier. I'd love to see their results. I'll have to check out some of the papers. From experience using supercomputing clusters w/ my research, getting close to 100 nanoseconds is still difficult. I've heard of groups breaching 1 microsecond w/ small systems, but nothing too large yet.

MrH

 

[CrankyDragon]

Thats pretty cool. Glad alot of people are doing it, especially since I'm in the field! wish I could use some of those computers LOL

From my experience though, I don't know how efficient it is. Its not a linear increase in efficiency. Even among local supercomputer clusters, when you get too many nodes/CPU's, you start losing efficiency. Communication costs become prohibitive, especially when doing things like protein folding/molecular dynamics. I'd be interested in seeing their stats...

BTW, they claim to have smashed the millisecond barrier. I'd love to see their results. I'll have to check out some of the papers. From experience using supercomputing clusters w/ my research, getting close to 100 nanoseconds is still difficult. I've heard of groups breaching 1 microsecond w/ small systems, but nothing too large yet.

MrH

I understand most of the computer jargon you used, and thats about it, I have no clue what there are, or you are, talking about when it comes to the folding and protine stuff! Otherwise, I know its for a good reason so Im running it!

 

[arnisador]

From my experience though, I don't know how efficient it is. Its not a linear increase in efficiency. Even among local supercomputer clusters, when you get too many nodes/CPU's, you start losing efficiency. Communication costs become prohibitive, especially when doing things like protein folding/molecular dynamics. I'd be interested in seeing their stats...

If it's being run like this, as SETI and www.mersenne.org do, then it must be highly parallelizable...I assume it runs a separate simulation(s) on each machine, with minimal communication needs, more like doing many runs of one experiment in parallel than doing one huge computation that requires intercommunication. If so, the speed-up would be for real.

 

[mrhnau]

If it's being run like this, as SETI and www.mersenne.org do, then it must be highly parallelizable...I assume it runs a separate simulation(s) on each machine, with minimal communication needs, more like doing many runs of one experiment in parallel than doing one huge computation that requires intercommunication. If so, the speed-up would be for real.

well, then you need -small- systems.I tried running some small simulations (~2000 atoms) on my desktop at home and at work. It TOTALLY crawled. Now, MPI on alot of nice clusters is very fast, because the communication time between nodes is typically in microsecond range. if you are talking about computers in different states, or even across a city, you are dealing with serious lag time (speed of your connection plus time for preparing/sending packets). Try using something like MPI on computers over your personal network at home. not much of an efficiency gain...

now, if you are talking about small molecules (100-200 atoms), this might be reasonable... I did a cursory look at their site, I'll have to look closer. Still, its always nice to have more computational power, even if for small systems. But systems that small are not all that meaningful (need ~20 amino acids to have the smallest biologically active proteins, thats < 200 atoms, depending on your system). Also, these computations are real memory hogs!

only a few systems are highly paralalizeble. The only one I'm immediatly familiar with is some astrophysics calculation. in the work I do (similar to protein folding, sort of), we get optimal speed with 8 nodes w/ 2 CPU each. With 16, we gain about 15%, with 32, we have no more gain and with 64 we actually lose.

now, something you -could- do with this would be using monte carlo type simulations... get different starting states of same system, you could get a very good idea of the energy surface profile... Also, here is a horrible example of using seperate systems to communicate. imagine you have two computers sharing a program. Someone comes online and needs their CPU time. What happens? does the simulation go on hold until both are back at full use? does the not busy one get the full job or look for another computer on the network? that can get messy...

So, we agree that its probably running small jobs on single computers. would this be feasible for millisecond scale expirements?

regardless, I'd love to see their control program and efficiency records. would be quite interesting

boy, I'm on a roll

MrH

 

[mrhnau]

I understand most of the computer jargon you used, and thats about it, I have no clue what there are, or you are, talking about when it comes to the folding and protine stuff! Otherwise, I know its for a good reason so Im running it!

Feel free to use your computer if you want it is a good cause, and I'm sure they are doing good science! I'd love to learn more..

MrH

 

[arnisador]

Monte Carlo simulations was exactly what I was thinking about.

The site says they send one 'work unit' to you and don't get communication back until it's done. Whatever part of the computation it's doing it must be pretty self-contained.

I've used MPI--I gave an MPI-based parallel computing course a year back--and I just can't believe it's trying to fake a cluster in that way. It must have the PCs do intensive computations, gather them to the master machine, and do a simple calculation based on those results.

 

[mrhnau]

I gave an MPI-based parallel computing course a year back.

:bow:

 

[arnisador]

It was a lot of fun--all juniors and seniors in math. and CS--but finding a decent book was tough! If I give it again I'll use my own notes.

 

[mrhnau]

It was a lot of fun--all juniors and seniors in math. and CS--but finding a decent book was tough! If I give it again I'll use my own notes.

I took MPI courses at the local super-computing center. if you ever care to check out the notes, I can dig them out it might literally require digging though LOL Though I'm sure your own notes are good.

Indeed, there are not alot of good books on the topic? Want one written by Hnau and Arni? heheh

MrH

 

[arnisador]

We'll be rich!

 

[OUMoose]

Not to ward anyone off, but I had Folding@home running on a spare system, and it didn't seem to care if it was in the background or not. It just hogged the CPU all the time and made the machine run like dirt.

So, I'll stick with SETI@home and waste my time apparently. That or find the other grid processing program again that was searching for pulsars and black holes.

 

[arnisador]

The primes program was pretty friendly, as I recall. Thanks for the warning about this one.

 

[CrankyDragon]

Not to ward anyone off, but I had Folding@home running on a spare system, and it didn't seem to care if it was in the background or not. It just hogged the CPU all the time and made the machine run like dirt.

Its a shame you didnt look further. Using the latest version software, in the sys tray right click mouse over the folding icon, bring up Configure. Click Advanced tab. Adjust slider bar on CPU Usage Percent, the program thread will have a lower CPU priority and will allow other more demanding programs to execute with higher priority. You can set the bar to half way or a quarter of the way and still yield good results. During idle time when no other programs are running it will use 100% of CPU time. When you sit down to do something, it will yield processor time to your other applications.

HTH,
Andrew

 

[OUMoose]

That or find the other grid processing program again that was searching for pulsars and black holes.

For those that are still curious: http://www.einsteinathome.org/

 

[Carol]

From my experience though, I don't know how efficient it is. Its not a linear increase in efficiency. Even among local supercomputer clusters, when you get too many nodes/CPU's, you start losing efficiency. Communication costs become prohibitive, especially when doing things like protein folding/molecular dynamics. I'd be interested in seeing their stats...

Distributed computing is not necessarily efficient, true, but there is more going on here than just number crunching. There is also outreach, public awareness, and a way to reach an educated audience.

I too woud be interested in seeing their stats