-
03-27-2005, 10:11 PM
#151
Originally posted by happybunny
so how long have ya'll been folding?
I have been doing it for only about 10 days and am racking up some fast numbers.
i have an advantage, i think. I am in charge of our VMWare environment and can secretly build Guest OS's to fold on. I'm gona build a 2 CPU box tomorow to add to my totals.
No wonder your gaining so quickly
:wq
-
03-28-2005, 04:53 AM
#152
For those who missed it...(like me :D)
Need help in realtime? Visit us at #linuxnewbie on irc.libera.chat
Few of us will do as much for our fellow man as he has done.
--Andrew Morton on RMS
-
03-28-2005, 06:53 AM
#153
Hmm, after reading the article I slowly begin to grasp why they need such a huge computational power: they are using a Monte Carlo design combined with cluster analysis... these guys are crazies! (probably only matched by those who compute the whole thing) 
And yet we are still folding at 470K, I'm really curious if we could fold at 370K if there are enough boxes running for a year.
-
03-28-2005, 10:41 AM
#154
-
03-28-2005, 06:12 PM
#155
Slow but sure:
PC1 - 204 out 400 frames
PC2 - 63 out of 400 frames
PC3 - 54 out of 500 frames
-------------
Folding is Fun
I thought I made a mistake once, but, of course, I was mistaken. 
-
03-28-2005, 06:27 PM
#156
sort of...
Originally posted by Parcival
Hmm, after reading the article I slowly begin to grasp why they need such a huge computational power: they are using a Monte Carlo design combined with cluster analysis...
From my limited understanding, Monte Carlo is actually the _least_ computationally expensive method they could have chosen. Around here we usually sample 10-20 million states, and that takes about 3 hours on a 15-node P-III cluster.
Need help in realtime? Visit us at #linuxnewbie on irc.libera.chat
Few of us will do as much for our fellow man as he has done.
--Andrew Morton on RMS
-
03-28-2005, 07:27 PM
#157
speaking of cluster, we got bored at work and are playing with openmosix, i have a 8 node p4 setup and testing balancing but having some issues with the network
when we get it all lined out its going to be moved to a 40 node p3 lab on weekends
-
03-28-2005, 08:04 PM
#158
When running F@H on a cluster does it distribute a single protein fold across the entire cluster or, do you just run multipal clients on one machine and each of the clients workload is transfered to a differnt node?
I've wondered that for a while....
Last edited by blueINK; 03-28-2005 at 08:07 PM.
-
03-28-2005, 09:00 PM
#159
I believe that it is multiple instances of the FAH program
:wq
-
03-29-2005, 03:59 AM
#160
Re: sort of...
Originally posted by je_fro
From my limited understanding, Monte Carlo is actually the _least_ computationally expensive method they could have chosen.
Technically this is true, yes. As there is no way (or at least none within acceptable range) to compute the whole population of proteins one has to extract a representative sample. The Monte Carlo method is one way to do it by randomly chosing the proteins we run our tests on out of the whole crowd of proteins there is. Hence the method reduces the computational expenses, yes.
However, it also has a major weakness: uncertainty. Who's telling you you aren't just being lucky and compute the sample of proteins that proves your theory right while any other sample would have proven it wrong? That's right, the confidence interval does. Now my problem is that I don't understand enough about the biology behind it to figure out what sample size is required to reach a satisfactory confidence interval.
Furthermore, it surprises me they are useing the randomized method at all. When I started folding I believed this research would be a lot more theory driven which would allow to compute a very small yet scientifically meaningful sample. Well, I guess that's just the bias in me speaking depending on my scientific background. We psychologists deal a lot with very abstract hypothetic constructs, so we extensively rely on randomized data and have a hard time to imagine folks looking at more "real" objects of investigation have to do the same as well.
-
03-29-2005, 11:57 AM
#161
but on the bright side....
MC simulations will allow you to reach conformations that might lie over an insurmountable energy barrier that another technique (like steepest decent energy minimization) would miss. It's a trade-off no matter which way you go. I think they are using an algorithm similar to Metropolis MC, which is a bit more directed in seeking a path to a low energy state, but I don't have time to check their paper right now.
Later!
Need help in realtime? Visit us at #linuxnewbie on irc.libera.chat
Few of us will do as much for our fellow man as he has done.
--Andrew Morton on RMS
-
03-29-2005, 12:51 PM
#162
Yay us! Although it's probably just part of the ebb and flow of points being reported we are currently poised to take over AV123 in 57 days!! As of March 29 11:50AM EST we are at this point anyway. We are only averaging slightly more than they are at this point so that's why it's gonna take that long. Assuming their points won't spike up. But they will.
Regardless good job guys!
-
03-29-2005, 02:35 PM
#163
First off, who is AV123? Is there some competition there that I don't know about? I mean, I understand the Microsoft one (48 days as of this writing), but I don't understand the significance of AV123.
Second, it might not be the ebb, or whatever. I know happybunny came blazing in, and I'm working my way up a little also. I'm working on getting a few more computers going. I started it up my sister's college computer yesterday, which she leaves on all the time, but I still haven't been out to my brother's dorm, where he has an athlon64 machine running, and I have a friend who I has 3 computers that I will be getting it on tomorrow.
On the Microsoft thing, it looks to me that passing them up is inevitable. It looks like our 24 hour average is 3-1/2 times theirs, and growing. Like I said before, 48 more days...
Linux has changed me...
I'm a GNU man now!
-
03-29-2005, 02:54 PM
#164
I've stated before why AV123 is a goal. But I will again. They went blazing past us some time in January. I picked a little competition with them because they're a more immediate goal than MS. And by matching AV123 for points we'll reach MS that much faster. We're likely ahead of them at this point due to the way points are reported. AV123 has been known to report much higher daily averages than we ever have and that will likely return. Ebb. Although once cs.wiu.edu starts reporting more consistently we may stay ahead.
Great to hear about the new boxes being added! You'll be averaging over 400/day with those additions!
-
03-29-2005, 04:41 PM
#165
i figured out the problem that ghosting killed the dkeav machines, i should remedy it soon
for anywho who ghosts and wants to keep FAH from dieing delete the MachineDependent.dat file before you create your image (make sure F@H isnt running when you do) it will regenerate when F@H fires up again
Posting Permissions
- You may not post new threads
- You may not post replies
- You may not post attachments
- You may not edit your posts
-
Forum Rules
|
|