points per WU
entropy
Yah-Der-Hey (Wisconsin)
are points determined strictly by how long it takes to compute? because i've gotten a few 400s that take a day or so, but right now i'm working on a 2500 that seems to go wayyy faster, cuz it's in a chain instead of a cluster. so is the 2500 worth 6x more than the 400? or is the size and time averaged or what..?
0
Comments
KingFish
KingFish
tinkers, the 400 frame WUs, can have lots of molecules or few. Gromacs and double gromacs WUs can also. Pointage is based on partly time to compute, and partly complexity.
Folding has lots of boxes of all the capacity in range currently in the field folding right now, to balance all to faster folding WUs for any one box would be unfair to owners of other gens of computer that have owners who are willing to let theiur boxes fold but cannot afford newest gear for whatever reason.
So, folding has to balance boxes versus points across generations of computing devices. That is why pointages are at best a compromise between time and capacity of box and complexity of thing being studied.
The idea is that the average box out there should get same poiintage as another average box. Right now, lots of benching is being switched to a P4 2.8 GHz box, AFAIK. But, there still are probably WUs out there that have not been rebenched, still ebing worked on, as this is happening now.
So, what was true before is now not quite true, as the fastest gens do not fold the Tinkers (the 400 frame WUs) in as timely a manner as the older boxes do, but fold the newer WUs proportionately faster. To keep older folding and not lose more folders than are gained, Fodlign admins have tio balance changes. So, since tinkers take longer for all boxes, Tinkers suddenly got points raised per WU based on times for a faster box to bench them versu what same box can do with complex WUs. In soem cases, Gromacs stayed the same for a lot of Gromacs projects, and Double Gromacs started with more points as they are usually more complex and need more complex calcs done to finish.
Balanced points valuing based on average extant folding box's ability to fold a WU, in time and complexity, or value of output to research. Idea was to balance gens across board as to how many points they CAN get per hour, for a single processing pipe.
To answer your two questions, it is not either or, but a dynamically adjusted balancing of boht that is being reworked now and started being so reworked in last 30-45 days wuith impact being felt a lot in last 10-12 days and probably going to be felt in future also.
Folding is not only using multiple cores, or calc engines for certain braod classes of WU, but they are also finding better ways to measure how a box handles things and monitor boxes without privacy intrusioning by gettign data only related to folding and your box ID and your User ID.
For now, while they tune that, expect points per hour on any one box to flux from day to day-- also expect points per day to flux on a multiday cycle, as many new WUs take even my P4 boxes 1\3 to almost three days each to complete. As folding gets better at autoadjusting to boxes folding of WUs, expect more and more boxes to get WUs they are "good" at and fewer they are not so good at-- expect the distributed network to get better at this over time for each new box brought into folding, and for exisiting boxes to get better and better tuned as folding's admins find better and faster ways to estimate how your box should fold a WU. This takes a feel-in approach, objective results versus varying boxes. Then, you need measurements that allow for individualization of WU feed withotu overloading some of the WU acceptance and sending out servers of whihc ffolding has a limited number.
If the best WU for your box is on a busy or offline server, expect to get a "second-best" for your box WU, adn if second best is on a server that is alsop busy, expect to get a third best WU rather than none for a long time and zero points while things are backed up for the wait for the "best" WU. But, if by chance you get a server open with bandwidth that has the "best WU, you might get best next time. Random chance, and you will get a blend in any small tiem frame of days or two weeks or even three, idea is over longer tiem to makje things better and better for as many as possible while not frustrating too many with things thta drive them away from folding. I do not fold purely or even majorly for points, but points encourage me to build better boxes to fold things faster prtly because of my underlying motivations:
First, I cannot afford to run many boxes, so what I do run has to be performers. Second, both boxes have to do what I otherwise need to do also, as I cannot afford to just fold 24\7 with high powered b oxes nor even to power protect and power feed 10-20 boxes with all the other gadgets I cannot get along without.
Since computers are indidividual as much as they are "standardized," when we get down to details of hardware and software interactions, do not expect perfectly ideal pointages now for your single box or couple of boxes. But, the longer you fold the more likely things are to get better for your boxes to gen points more effectively as many aspects of folding are being worked on at once. Trick is to balance that with scientific study, some things are more probable than others at any given time and knowledge in depth level, and that can change and get more and more complex as things narrow down.
Second major thing, Folding is dynamically changing what projects are active more and more. Mostly due to the things I scratched the surface levels of above.
AND lots of the grunt admin work is done by folks not paid much to do it and instead in it for some of the same reasons we are-- they are trying to speed up results in finding cures for things that we have no reliable sures for yet, and while they may be getting benefits, those can be indirect and longterm for many of them and not much in immediate money.
John D.
steps are subunits for calculating, in terms of the calcs needed to be done by CPU. Frames are groups of steps, and need not always be same number of steps per frame.
Tinkers have 400 frames, variable numbers of steps.
Gromacs do also, but with Gromacs one frame is one percent of completion.
BUT, different CPUs calc differnet kinds of steps more or less effectively compared to other CPUs. One computer's best WU can be another computer's WU that just plain will not complete.
We are talking about a global calc population, in terms of computers calcing world-wide for Folding, of old pentium boxes up through a few Opteron and P4 Prescott boxes. The earliest of these cannot handle SSE2, while trhe best at handling SSE2 do nto handle old 16 bit code worth a relative snot to good drink if you end up having to swallow both at least once to not be very rude. Different calc styles result as being more effective in long term for different boxes. differnt KINDS of cores are needed for each basic style of WU, and all core major kinds are being developed and enhanced at least some right now--we expect, those of us who fold and study folding, that eventually Tinkers will be fully retired mostly when not enough volunteers have boxes that can only calc Tinkers or not calc other kinds of WU like Gromacs or Double Gromacs at all well in large enough amounts to make creating Tinkers worthwhile.
Some want that tinker elimination sooner, but others can only afford to volunteer boxes with limited-complexity folding ability as to complexity of what they can do over finite enough time for teh human sponsor to see some reward that is more immediate than a hopeful cure-- we have found better ways, but not total cures yet because of extreme research complexity needed. The WU offerings will be dynamic based on the distributed network's working machine population and what kind of WU best suits what kind of study of what disease.
John D.
http://vspx27.stanford.edu/psummary.html
You should be able to gather the following info from the chart:
Project Number
Server IP
Work Unit Name
Number of Atoms
Preferred (days)
Final deadline (days)
Credit
Frames
Code
Description
Contact
The points and deadlines are determined on a 2.8GHz P4 machine.
I'm not sure if I answered the question! LOL
Project Number Server IP Work Unit Name Number of Atoms Preferred (days) Final deadline (days) Credit Frames Code
1049 171.67.89.149 p1049_unf96_273 6399 13.00 18.00 59.00 100 GROMACS
see how many frames it says? 100. how many does my SS say it has? 2500. also, just now i did preview (in windows to check it) and all the sticks from ball and stick mode just disappeared :S
SS???
Second, the 2500 number is probably in thousands of steps or in steps per frame. For Gromacs, frame=percentage complete for progress in numer in ( ) but steps can be not related to number of frames in aGromacs-- in fact the two do not relate together except as WU specific thing mostly.
Let me give you a sample progess data snippet and take it apart for you, I think I have WU similar enough in logs to give you an understanding of what you are seeing and asking about.
Default display is not ball and stick, by the way, it is molecule tight to molecule, and you can accidentally get this restored to that.
but, snippet:
I am going to break it up and show how it is used and relates after I show it. This is from a p1047, which my Linux box is working on now, and which is same gross number of steps as the p1049 you have.
[03:20:31] Completed 1600000 out of 2500000 steps (64%)
[03:27:24] Writing local files
[03:27:26] Completed 1625000 out of 2500000 steps (65%)
[03:34:12] Writing local files
In this case, the percentage complete is frame last completed.
1600000/2500000 is number of steps over total steps, and if you were to didive that out you would get 64% or 64 frames complete if your box were my box and working on same WU.
The graphical thing will show 1600/2500 and abbreviate the info to thousands of steps of calculation fro this big a WU in sens eof the number of steps it has. But, it should below that show also a percentage, and that is number of frames complete. also, right clcik the cog next to the clock, and see if you have a percentage show there or repeat of the 1600/2500 or boht, OK??? I am posting from a Linux box and right now
am running all console clients. I do not have a graphical image to attach here that would show what you are looking at near enough to be good for explaining what you see.
John D.
what is the SS you are referring to?
100 frames is correct ...so 2500 steps @ 100 frames will be 250000 steps when the wu is complete.
This is from my fahlog.txt:
[01:37:28] Folding@home Gromacs Core
[01:37:28] Version 1.62 (April 2, 2004)
[01:37:28]
[01:37:28] Preparing to commence simulation
[01:37:28] - Assembly optimizations manually forced on.
[01:37:28] - Not checking prior termination.
[01:37:29] - Expanded 941148 -> 4855789 (decompressed 515.9 percent)
[01:37:29] - Starting from initial work packet
[01:37:29]
[01:37:29] Project: 1013 (Run 6, Clone 30, Gen 26)
[01:37:29]
[01:37:29] Assembly optimizations on if available.
[01:37:29] Entering M.D.
[01:37:35] Protein: p1013_lzip_273
[01:37:35]
[01:37:35] Writing local files
[01:40:23] Extra SSE boost OK.
[01:40:24] Writing local files
[01:40:24] Completed 0 out of 125000 steps (0)
[01:48:18] Writing local files
[01:48:18] Completed 1250 out of 125000 steps (1)
[01:55:50] Writing local files
[01:55:50] Completed 2500 out of 125000 steps (2)
[02:02:54] Writing local files
[02:02:55] Completed 3750 out of 125000 steps (3)
[02:09:58] Writing local files
[02:09:58] Completed 5000 out of 125000 steps (4)
[02:17:02] Writing local files
[02:17:02] Completed 6250 out of 125000 steps (5)
[02:24:06] Writing local files
[02:24:06] Completed 7500 out of 125000 steps (6)
[02:31:11] Writing local files
[02:31:11] Completed 8750 out of 125000 steps (7)
[02:38:15] Writing local files
[02:38:15] Completed 10000 out of 125000 steps (8)
[02:45:19] Writing local files
[02:45:19] Completed 11250 out of 125000 steps (9)
[02:52:24] Writing local files
[02:52:24] Completed 12500 out of 125000 steps (10)
[02:59:43] Writing local files
[02:59:43] Completed 13750 out of 125000 steps (11)
[03:07:02] Writing local files
[03:07:02] Completed 15000 out of 125000 steps (12)
[03:14:18] Writing local files
[03:14:18] Completed 16250 out of 125000 steps (13)
[03:22:11] Writing local files
[03:22:11] Completed 17500 out of 125000 steps (14)
[03:29:25] Writing local files
[03:29:25] Completed 18750 out of 125000 steps (15)
[03:36:40] Writing local files
[03:36:40] Completed 20000 out of 125000 steps (16)
[03:43:57] Writing local files
[03:43:57] Completed 21250 out of 125000 steps (17)
so each frame is 1250 steps.
some wu's have 200 frames and some have 400 and occasionally you'll find and oddball with something like 20.
To check your percentage of completion just open the unitinfo.txt file and it will look similar to this:
Current Work Unit
Name: p1013_lzip_273
Download time: May 5 01:37:27
Due time: May 23 01:37:27
Progress: 16% [|_________]
# Windows Screensaver Edition #################################################
###############################################################################
Folding@home Client Version 4.00
http://folding.stanford.edu
###############################################################################
###############################################################################
[03:42:14] - Ask before connecting: No
[03:42:14] - Use IE connection settings: Yes
[03:42:14] - User name: SFHS (Team 93)
[03:42:14] - User ID = 7F88EA3D013C7E99
[03:42:14] - Machine ID: 1
[03:42:14]
[03:42:14] Loaded queue successfully.
[03:42:14] Initialization complete
[03:42:14] + Benchmarking ...
[03:42:16]
[03:42:16] + Processing work unit
[03:42:16] Core required: FahCore_78.exe
[03:42:16] Core found.
[03:42:16] Working on Unit 01 [May 5 03:42:16]
[03:42:16] + Working ...
[03:42:17]
[03:42:17] *
*
[03:42:17] Folding@home Gromacs Core
[03:42:17] Version 1.62 (April 2, 2004)
[03:42:17]
[03:42:17] Preparing to commence simulation
[03:42:17] - Looking at optimizations...
[03:42:17] - Files status OK
[03:42:17] - Expanded 195925 -> 946533 (decompressed 483.1 percent)
[03:42:17]
[03:42:17] Project: 1049 (Run 3, Clone 49, Gen 3)
[03:42:17]
[03:42:17] Assembly optimizations on if available.
[03:42:17] Entering M.D.
[03:42:37] (Starting from checkpoint)
[03:42:37] Protein: p1049_unf96_273
[03:42:37]
[03:42:37] Writing local files
[03:42:37] Completed 700000 out of 2500000 steps (28)
[03:42:37] Extra SSE boost OK.
Folding@home Client Shutdown.
almost looks as if the frames are just taking off the last 3 zeros from the steps...
Thanks, that is an excellent summary, but I think he also wants to know how to make sense of teh data and is vERY interested in details as am I. so, I ma gonna be a bit more verbose and explain soem for him and others while saying you are right but I think he wants more than that in this thread.
BTW, entr0py, mmonnin is our team captain, and I am noe of the site detail-explainers who likes staying as an ordinary user for some things I like to play with and take apart and put back together. Folding is one of them, that I take apart intellectually to an excruciating degree.
John D.
\\EDIT: i know that the molecule filling up next to molecule is default, but personally i think that's ugly lol. so i changed it to ball and stick, but it somehow lost the sticks...just have to wait for it to finish and see if it changes on the next wu i guess
Thats one of the reasons why this forum is so great. There are members all over the place that will help someone if they have a question.
the core you have, and the client you have, are leaving off the percentage after the 28 that my Linux client logs. Last frame complete is frame 28, your machine is 28% through the WU plus soem hundredths of a percent that will not get logged. Electron Microscope III does a pretty good job of guessing exactly where the core is between percentages ciomplete.
For calcing frame, the WU steps are rounded down by 1,000, yes, fro boht total and what is complete, for most Gromacs now being given out-- I have seen smaller ones where the rounding was by 100's for both top and bottom. Frame and percentage are calced from number of complete steps, but log only shows even frames or percentages (SAME thing with a Gromacs). For Tinkers, log shows frames, 400 frames per WU, 4 frames per percent-- take Frame number from log, divide by four to get percentage. For Gromacs, one of which your box is working on, all Gromacs have 100 frames and 100 percentage points. That makes calcs easier for us humans. Tinkers were issued first, then Gromacs were developed, and now Double Gromacs are being brought into the public areas of distribution.
As far as point gen, marc is perfectly right, steps do not enter in for Gromacs or tinkers, at all-- the only way steps relate is to show how much calcing is needed roughly and that lets folding admins guess at what time the WU should take. But between Tinkers and Gromacs, frames and percentages relate differently as to what is logged, and as you can see even Linux client shows different things than Windows in log to a small degree-- in this case, the percentage point shows up literally in log from current Linux client.
John D.