| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: FightAIDS@Home
Thread: FightAIDS@Home performance |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 6
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I read a post that indicated that it had been learned that sending a work unit to 3 machines instead of 4 improves productivity by 25%.
Good idea. But I think there may be another opportunity to improve performance. I have 3 machines working on these calculations. Every morning, wihen I get up, I look at the screen saver for a few minutes. The calculated bonding energy of some of the compounds fit very well with the best bonding energy curve predicted, while others don't fit well at all. For those that don't fit, the calculations go on mindlessly, wasting machine cycles. It seems that it would be a good idea to do a least squares fit when a portion of the calculations are completed on a particular compound, virus combination. If the fit is poor the calculations on that combination could stop. The really bad fits are obvious when only a small portion of the calculations are complete. Stopping the calculations when the fit is poor could greatly increase the efficiency of the analysis. Ciao |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi Porky,
A compound with a terrible fit at one point might make contact like a well-fitting glove if it is rotated or translated a little. Until we have presented all angles of the compound to the protease molecule, we cannot be sure that we are wasting our time. This is just my personal opinion. The Scripps scientists are the only ones who really know. Lawrence |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Lawrence, I believe you have hit the nail on the head.
Atomic modelling is by no means as simple as the graphics in the screensaver may suggest. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Porky, on a different tangent, i presume you only crunch when the Screensaver is on. That eats alot of CPU juice, even when you have a good graphics card. If concerned you get held up by the UD agent running 24/7, particular when you're gaming e.g., you can set in the WCG UD agent profile on the website the times you want it to crunch. Also in the agent you can set a Snooze period, up to 120 minutes. See the options under 'V'.
----------------------------------------Can assure you from an unbiased 'client' perspective, The process is entirely unobtrusive. Even if you set it 24/7, when gaming or anything else it just stops. It only uses your idle CPU ticks and a little front, but only during result transmissions......and does that graph not get boring after a while? ciao
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Porky, on a different tangent, i presume you only crunch when the Screensaver is on. That eats alot of CPU juice, even when you have a good graphics card. If concerned you get held up by the UD agent running 24/7, particular when you're gaming e.g., you can set in the WCG UD agent profile on the website the times you want it to crunch. Also in the agent you can set a Snooze period, up to 120 minutes. See the options under 'V'. Can assure you from an unbiased 'client' perspective, The process is entirely unobtrusive. Even if you set it 24/7, when gaming or anything else it just stops. It only uses your idle CPU ticks and a little front, but only during result transmissions......and does that graph not get boring after a while? ciao Actually, I only look at the screen saver when it's on. The software is crunching whenever it can. Of the 3 machines running, two are at work and are busy during the day. I sometimes see a performance hit when using these machines for other purposes and have to exit the program. I'll look into your suggestions. At night those machines are idle, so they are working all the time. My home machine doesn't get used much so it's crunching all the time mostly. When the machine's not doing something else, I don't really care what it's doing. I just like to know that it's doing something usefull, and not just calculating because it can. I guess the simplicity of the graph is misleading. ciao |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I appreciate all of your thoughts about making our docking computations more efficient. We have been pondering this issue for some time. The problem is that the "search" that we are conducting in a very high dimensional space (in some cases upwards of 20 dimensions!). The process that we use for this search is stochastic -- that is we make random move in this large parameter space. How do we know that we've covered this space effectively? By running many dockings of the same ligand target pair with different random starting positions, and different random seeds for the docking, we can look at all of the results (usuallly 100 dockings) and cluster them to see if we hit the best energy consistently. If we do see significant clustering from the different runs, then we have some confidence that we have found the global optimum.
We have done some experiments to see if during a given docking we can terminate early if the best energy does not change over a number of "generations." We have not found a magic number yet that says that it it no longer useful to continue the docking -- but we are still looking. Thank you all for your thoughtful remarks. Art Olson |
||
|
|
|