How exactly are points calculated for phase 2?
I know there's something involving square roots, but that's all I know.
Thanks
Look at http://www.distributedfolding.org/phaseiifaq.html in the section "How does the new scoring work?"Originally posted by Scotttheking
How exactly are points calculated for phase 2?
I know there's something involving square roots, but that's all I know.
Thanks
Cool, thanks.
So one full set of 250 generations is worth 132050 points then?
Sounds about right, though the time a particular machine takes to get through a full set of 250 generations seems to vary greatly. eg. a full set on a p4 3Ghz might take 3 days, then the next full set might take 5 days, so the scoring, does not seem to be a true representation of the calculating effort any given machine actually puts in which is a pity.Originally posted by Scotttheking
Cool, thanks.
So one full set of 250 generations is worth 132050 points then?
It does over time, on average.
Howard Feldman
I would have liked to see less range in the scoring,
with a set being 100 structures, not 250 and gen 0 being 1,000 not 10,000 structures as gen 0 takes too many hours to get back to scoring after a corruption. One of the things that keeps people interested is seeing things tick over at a reasonable rate. At the moment the average score per generation is a bit over 500 points, but you don't get to 500 point scores until you reach generation 100. I would like to se somthing faster like:
eg.
gen 0 is 1,000 structures and 100 points.
gen 1 = 100 * 2^-2
.
.
gen 99 = 100 * 99^-2
gen 100 = 100 * 100^-2
Last edited by erk; 07-30-2003 at 05:11 PM.
They actually have no clue how it's scored. Witness the fluctuation in output over time with a constant number of machines.
Just another reason this project is hosed.
At first I thought you were a real a**hole, but I was wrong. You claim we are all a**holes, yet you still hang around trying to make our life miserable, that makes you just another hemorrhoid!Originally posted by ddn
They actually have no clue how it's scored. Witness the fluctuation in output over time with a constant number of machines.
Just another reason this project is hosed.
Didn't your mother ever tell you, that if you have nothing nice to say then you should just have a nice hot cup of S.T.F.U.?Originally posted by ddn
They actually have no clue how it's scored. Witness the fluctuation in output over time with a constant number of machines.
Just another reason this project is hosed.
Can't stand your criticism of the project. Go slither back in your hole where you once came from and don't come out until you see your shadow.
Stand in front of the mirror and say that out loud and see if it makes any sense when you hear it.Originally posted by ddn
They actually have no clue how it's scored. Witness the fluctuation in output over time with a constant number of machines.
Just another reason this project is hosed.
The problem at hand is not a constant. Each attempt takes varying amounts of time depending on the protein.
So yes, if power were constant, the user never used the machines, anti-virus never kicked in, defrag never ran, the clock was never updated and the internet never paused. Then you would still have varying output from one week to the next.
Were not counting keystrokes and the equation changes with each set of criteria. That's why is uses an algorithm. Start with any number and eventually wind up with the right answer. No guarentee that the right answer will apear on the third try every time.
That's true. But the algorithm is still hosed, or the code is. But no one will believe me until Howard admits it, because what Howard says is the Word. Gee, wonder why Howard won't let anyone else take a stab at the code.Originally posted by PinHead
[B]The problem at hand is not a constant. Each attempt takes varying amounts of time depending on the protein.
/B]
Never argue with a fool. Observers might not be able to tell the difference.
How does the new scoring work?
The scores awarded for the structures depend on the generation number reached.
* 50 points are awarded for the first generation, which consists of 10000 structures.
* each consecutive generation has 50 structures, and the scores for those structures are scaled by a factor of SQRT(generation #).
* e.g.: if you have reached generation 100, you will receive (50 structures) * SQRT(gen.100) = 50 * 10 = 500 points for uploading those 50 structures, and so forth. Fractions are truncated downwards. This scheme is implemented because more time is spent trying to complete each fold in the higher generations. Thus they are visibly slower than the randomly produced generation zero. The higher scoring will also encourage users to reach generation 250.
-----
And yet the client RARELY makes it to generation 250 because of the babysitting it requires.
If the mysterious corruption error occurs, you crash. Start over.
If the client can't contact the server because the server constantly goes down, you crash. Start over.
If the client hangs on a structure, you're screwed. Start over.
If another random bug crops up and you crash. Start over.
And I'm the fool.
Can someone tell me if there is a way to have the forum completely ignore a user? That is, when I read the forums, there is a certain user that I want to not waste my fime with.
Lacking that, could we just have a Free-DC admin just kick this sorry ass out of here???
FreeDC Mercenary
If you go to your control panel, via the little button near the top that says "user cp", there is an option for an ignore list.Originally posted by rsbriggs
Can someone tell me if there is a way to have the forum completely ignore a user? That is, when I read the forums, there is a certain user that I want to not waste my fime with.
Lacking that, could we just have a Free-DC admin just kick this sorry ass out of here???
I just used it and it is pretty sweet. This thread for example got 3 posts shorter.
you guys are funny (and very subtle too). But seriously, I welcome criticism, I never claim to have all the answers or to always be right.
However, the fact that this is the best (or worst?) criticism the world can come up with about our project makes me feel pretty pleased actually.
Anyhow, 2/4 ddn's start over conditions are actually valid and we obviously want to get these fixed ASAP. (I've still had NO problems running on my own machine as a service over the last 3 weeks, go figure).
Now that we have a pretty good sampling of people's screwed up directories it should hopefully give us a better clue of what may be going on. Just for reference, presently 1569/8562 total 'simulations' have reached gen. 250 since the recent changeover, so I can assure you some people have made it there and beyond! (Of the remaining ones I cannot distinguish between ones in progress and ones that have been prematurely teminated).
Howard Feldman
Now, Scott, that is one great piece of trivia to know here!Originally posted by Welnic
If you go to your control panel, via the little button near the top that says "user cp", there is an option for an ignore list.
I just used it and it is pretty sweet. This thread for example got 3 posts shorter.Worked like a charm!
Howard, I will step forward and say that I have 7 machines that have run without a single problem since change over. Well, other than the slow production for the first day or three. None have died. None have failed to transmit. And none have failed to reach the 250 total. Only one is hard connected to the net, the others are all sneaker netted and none are run as a service.
I am pretty much one very happy camper.
Fact #1 Scoring was worked out By Howard and others during the Beta Phase to give a roughly relevant and fair system..roughly fair because in the end it is the results that get the big prize, not the Base 10 number besides your Name on the Stats Pages.
Fact #2 My 6 Computers have run with the 1:43 exe flawlessly, my fastest being on my third Set.
Fact #3 I have never seen 1 User annoy so many people across so many Topics in such a short time
I am not a Stats Ho, it is just more satisfying to see that my numbers are better than yours.
I can't help but wonder what ddn's underlying modivation is? Unless he/she is some kinda Distributed Computing Masochist I can't understand why he/she would stick around a project that is giving him/her these kind of headaches? This reminds me of a drunk heckling a comic in a night club.
Howard you have the patients of Jobe.
oh and one other thing... all my boxen (w2kp and Xpp) have run through serveral sets of 250 since the update to the new old new old client.
Actually we have pulled all our clients. I should mention we had the ability to crunch more proteins than several leaders combined.
Maybe there needs to be more focus on the Linux and other clients. Seems like the Win32 one works alright. The Linux one doesn't, I uploaded 3 broken directories, and that wasn't even close to all the crashes I had. Not to mention there is FAR more power available on Linux/Solaris/IRIX than there ever will be on Windows.
Highly confused right now. Linux was one of the clients that didn't seem to have a problem. It was only windows clients that had issues.
I have 15 or so linux clients and the one thing I like about the client is the set it and forget it.
Now if your talking about nfs mounts and a rogue nic causing excessive network traffic. Yes, it corrupts very easy when it runs out of memory before it can get back to it's network hard drive.
Either way, I think I have been running this project for almost a year now and this is the first trouble of any kind I have run into. And it is only on 2 of 7 windows boxes.
So if you happened to give distributedfolding a first trying during the change from phase I to phase II; then cool off a bit and give it a try in the future. The thing I like most about this project is Howard's attentiveness to problems. But when you venture into new territory, it may take more than a day or two to resolve the issue.
About 80% of our users are running the Windows version - we have no control over that. As a result though, over 80% of the bugs reported are from the Windows version. If there are any UNIX specific bugs (several of which were found by 'shortfinal' and have subsequently been fixed by me with his assistance) they would be fixed as well if reported. If you are receiving any bugs NOT already mentioned in the forum (namely the trajectory distribution missing, or .val file corrupt/missing errors), please report them to us with as much detail of how to reproduce the error where possible and it will get fixed.Originally posted by ddn
Maybe there needs to be more focus on the Linux and other clients. Seems like the Win32 one works alright. The Linux one doesn't, I uploaded 3 broken directories, and that wasn't even close to all the crashes I had. Not to mention there is FAR more power available on Linux/Solaris/IRIX than there ever will be on Windows.
As I said earlier, our tests are done on Windows and sometimes Linux, that is all we can afford.
Howard Feldman
Posting Permissions
Reply With Quote
