Suggested Change to the PPD System

Moderators: Site Moderators, FAHC Science Team

MtM
Posts: 1579
Joined: Fri Jun 27, 2008 2:20 pm
Hardware configuration: Q6600 - 8gb - p5q deluxe - gtx275 - hd4350 ( not folding ) win7 x64 - smp:4 - gpu slot
E6600 - 4gb - p5wdh deluxe - 9600gt - 9600gso - win7 x64 - smp:2 - 2 gpu slots
E2160 - 2gb - ?? - onboard gpu - win7 x32 - 2 uniprocessor slots
T5450 - 4gb - ?? - 8600M GT 512 ( DDR2 ) - win7 x64 - smp:2 - gpu slot
Location: The Netherlands
Contact:

Re: Suggested Change to the PPD System

Post by MtM »

@vbironchef, that's one part of the game, the other part is the amount of people who use points to correlate them to science completed.

Participation is something which is more important to the one, science is more important to the other.

We tried using polls to get answers to which group supporting either stance is larger, but polls never seem to work ( and I say that while there been enough examples where such a poll giving the same result as I been trying to advance ). It's never a final word. It's not even a solid approximation, as you can not be sure if those posting in poll are representative of the entire community.

I'm here for the cause, if I knew for sure more computational power would be available to f@h using system b even while I think system a is better, I would support system B because it would mean f@h will get results quicker.
k1wi
Posts: 909
Joined: Tue Sep 22, 2009 10:48 pm

Re: Suggested Change to the PPD System

Post by k1wi »

MtM,

I would like to reiterate that I only used "my proposal" in the context of the proposal as I saw it in the post of mine where I referred to it as "my proposal" that is, I saw the formula, and my understanding of the formula as the representation of the concept I was attempting to explain in the first post/this thread. I do think we differed in how we calcuated Y (my solution is to let the experts (PG) determine that, but to suggest ideas as to how it could be benchmarked so that it is possible to also explain why it is important to benchmark it). Things like using the monetary formula, or dollar values of hardware are not how I'm attempting to calcuate Y, but they are a way of explaining the reasons behind adjusting for Y.

As I said, I don't hide the fact that others have had possibly the same idea. I do see it as unfortunate that their threads died before a workable formula was created :(

Now regarding how we want the curve to be not sure if I'm repeating something I've already explained properly :)...

I would like to reiterate that a situation where there is a horizontal line over time will not remove the incentive to fold faster. What a horizontal curve is saying is that we are perfectly accounting for the technological improvement over time.

My concept and the formula as I have it currently aims to preserve the current points curve at a given point in time - someone completing work twice as fast at a given point in time will still earn exponentially greater points, it's just that exponentially greater points won't increase in absolute terms as time increases.

That is, if we succeed in removing technological improvement from the equation, we take the graph with the flat line and consider time to be moving left at half the distance to the Y intercept every 18 months (or however often PG determines it). At any point on any of those lines, there will still be the same curve below: ImageThat is, a computer folding 4x faster will earn the same proportion higher points.

What people are asking is "how can we change the point in time curve" - I want to acknowledge that as it stands my proposal does not attempt to ask that question, but I to believe that once we have resolved the initial question, we could perhaps take it into account and I think that the mind shift required for accepting my proposal might make answers to that question easier.

I think when you asked what happens when you graph it - I will need time to do so, if I can do so at all! But I think it will look like I said above - over time the line remains constant, but at any given point in time the point distribution retains the same curve as is present.
k1wi
Posts: 909
Joined: Tue Sep 22, 2009 10:48 pm

Re: Suggested Change to the PPD System

Post by k1wi »

Also I would like to add if we did differ in how we calcuate Y, we are working together to solve the formulaic problem :)
k1wi
Posts: 909
Joined: Tue Sep 22, 2009 10:48 pm

Re: Suggested Change to the PPD System

Post by k1wi »

Also, by "a computer folding 4x faster will earn the same proportion higher points."

I actually mean:

"a computer folding 4x faster will earn the same proportion higher points as they currently do."
vbironchef
Posts: 23
Joined: Mon Jan 19, 2009 10:41 pm

Re: Suggested Change to the PPD System

Post by vbironchef »

MtM I can only speak for the team I used to fold for. People that bought the latest and greatest graphic cards or cpu's did not buy them for F@H as the one and only reason they made the purchase. Most are gamers and the others need them for work. F@H is a second or third thought, if at all. To base your theory that people go out and buy the latest and greatest just for F@H is incorrect. Most likely you will get a couple of hours a week if your lucky. Not to mention most people that buy those graphic cards do not pay for the electric bill. Not taking account for a dedicated folder that folds 24/7 is a mistake.

I do think I get where you are coming from. F@H doesn't care how or where the WU's are completed, just as long as it's fast and the more the better. F@H is not paying the electric bill or spending thousands of dollars building new rigs. So really F@H doesn't care a bit as long as the WU's are done. I'm glad I quit!
Image
Zagen30
Posts: 823
Joined: Tue Mar 25, 2008 12:45 am
Hardware configuration: Core i7 3770K @3.5 GHz (not folding), 8 GB DDR3 @2133 MHz, 2xGTX 780 @1215 MHz, Windows 7 Pro 64-bit running 7.3.6 w/ 1xSMP, 2xGPU

4P E5-4650 @3.1 GHz, 64 GB DDR3 @1333MHz, Ubuntu Desktop 13.10 64-bit

Re: Suggested Change to the PPD System

Post by Zagen30 »

vbironchef wrote:Let's take a new video game for example. A new player doesn't expect to be on level 50 from the beginning, right? Same goes for F@H. you have to put your time in. Even if you have the latest and greatest rig you start at level 1. So you play the game for a couple of months and a new player shows up with a faster rig than yours, so what happens? He chops you into little pieces! :lol: Did he put in the time and effort you did? Of course not! Are you upset? Answer, yes! Do you go out and buy a faster computer or do you stop playing the game because it's not fair. I choose not to play anymore. Simple as that. What's the point, there will always be something new and improved that costs to much. Do you really want to continue to play, I think not. Same applies to F@H. People should have to level up and pay their dues before they can slay the dragon.
That analogy makes no sense, because outside of you having a rig that's not capable of playing a game at a playable framerate, your machine shouldn't have any effect on how good you are at a game.
Image
Grandpa_01
Posts: 1122
Joined: Wed Mar 04, 2009 7:36 am
Hardware configuration: 3 - Supermicro H8QGi-F AMD MC 6174=144 cores 2.5Ghz, 96GB G.Skill DDR3 1333Mhz Ubuntu 10.10
2 - Asus P6X58D-E i7 980X 4.4Ghz 6GB DDR3 2000 A-Data 64GB SSD Ubuntu 10.10
1 - Asus Rampage Gene III 17 970 4.3Ghz DDR3 2000 2-500GB Segate 7200.11 0-Raid Ubuntu 10.10
1 - Asus G73JH Laptop i7 740QM 1.86Ghz ATI 5870M

Re: Suggested Change to the PPD System

Post by Grandpa_01 »

Hey MtM / K1wi I just got done reading the last 3 pages for the 3rd time trying to figure out where you guys are at in this discussion. It appears you have found some common ground in areas but quite frankly I am a bit lost as to where you guys are as far as to what your actual proposals are. It is quit obvious to this old fella that the 2 of you have allot more brain power than I. So could the 2 of you condense your thoughts and theories into individual post so I can catch up. I see quite a bit of discussion between the 2 of you on what the other is saying and I am having to jump back and forth between post to try and figure things out. (and I am failing at that task) So without discussion of what the other has said can you guys post your ideas.
Image
2 - SM H8QGi-F AMD 6xxx=112 cores @ 3.2 & 3.9Ghz
5 - SM X9QRI-f+ Intel 4650 = 320 cores @ 3.15Ghz
2 - I7 980X 4.4Ghz 2-GTX680
1 - 2700k 4.4Ghz GTX680
Total = 464 cores folding
k1wi
Posts: 909
Joined: Tue Sep 22, 2009 10:48 pm

Re: Suggested Change to the PPD System

Post by k1wi »

Grandpa_01 wrote:Hey MtM / K1wi I just got done reading the last 3 pages for the 3rd time trying to figure out where you guys are at in this discussion. It appears you have found some common ground in areas but quite frankly I am a bit lost as to where you guys are as far as to what your actual proposals are. It is quit obvious to this old fella that the 2 of you have allot more brain power than I. So could the 2 of you condense your thoughts and theories into individual post so I can catch up. I see quite a bit of discussion between the 2 of you on what the other is saying and I am having to jump back and forth between post to try and figure things out. (and I am failing at that task) So without discussion of what the other has said can you guys post your ideas.
I'm going to repeat the formula from a previous post and precede it with an attempt at describing the idea :)

Theory & Concept

In a nutshell, I am proposing to remove technological improvement as a source of point inflation over time. My reasoning behind that is that as it stands, the point system rewards technological improvement over time - you can get more points in the future simply because in the future someone will have made computers more advanced. Inflation due to technological improvement also causes issues where PPD (and therefore total points) increase exponentially over time. PG can measure the level of advancement in computer performance/technology/power, relative to a single point in time, and we can represent this as Y. That is, 3 years after the starting point, computers have gotten 400% faster, therefore Y should equal 400% (or 4). It has the added benefit that a user can compare their relative performance in one time period with another, based on their PPDs in each time period - if your ppd dropped from time period a to time period c, it was because your collection of computers/computer became relatively less powerful compared to others.

This eliminates PPD growth over time that is attributed to technological improvement but what I am proposing has the benefit in that if it is done properly, it will not influence the points curve at any one point in time, which is the graph:

Image

Of importance is the shape of the curve, which is retained, not the TPF along the bottom (because these numbers will decrease as we account for technological improvement. What we end up is as follows:

Image

Here the red line shows the current growth in PPD over time, the curve is not horizontal because over time computer power improves. (I think going from right to left each x step is equal to the halving of TPF - under the current system the current points curve is the same as the over time curve...)

The green line is what I aim to achieve (for a minute ignore the legend) - over time there is no point inflation due to technological improvement. But what is also important is the black lines. They represent the 'point in time' curves, which my best attempt at replicating the curve of the graph of 8004, in paint ;). The black lines show that at a given time period, there will still be incentive to fold faster, because the curve retains the same incentive curve as we have today.

So, if you were to look at the current time period, the current point structure still applies and a computer that is 4x more powerful than another computer, it will earn the same proportion more points as it does now. My reasoning being that if we need to discuss this later, my model preserves the ability to do so (and if people can have the mindshift to accept this proposal, maybe it will make having a mindshift to find an equatible solution to that the next question easier). The more I look at it, the more I realise that it will solve a problem where the QRB compounds point inflation due to technological improvement because this inflation is being applied twice (once to the base model, and again to the speed ratio).

Therefore, the revision will be 'appropriate' if the growth in PPD over time remains constant, because this means we are theoretically accounting completely for the effects of technological improvement. That horizontal line would basically be the overall average for the whole of FAH. It would be possible to have a different horizontal line for each platform. I.e. if PG determines that BA should have a 20% premium over SMP then the horizontal line for BA should be 20% higher than the SMP line. But over time, it should remain horizontal. This premium would also be present for the "point in time" point curve - the curves would be the same, but the BA curve would be shifted UP by 20% across TPS where computers can complete both SMP and BA within deadlines.

It also preserves the ability for PG to shift both these lines over time if they see fit - perhaps there is a major advancement in the GPU core which increases their relative power to CPUs, but the absolute increase in 'scientific value' to PG is smaller - they can shift the GPU curve down so that it remains in line with their idea of relative scientific value. Again, this is just conjecture and could just as likely happen under the current system.

That sums up the justification and explanation for the ideas behind my proposal (hopefully), I now look at the process I use to get to a formula that I think fufils this:

Working, towards an applicable formula:
Revised PPDn = PPDn / Yn (also written as "Revised PPD = PPD / Y")

where Y is determined by PG based on whatever methodology that they think suits, and n is the time period. Changes in Y can be publicised (depending on what PG decides) - so that from time n to time n+1 etc people can understand why their current PPD has been normalised to the level it has.

This then makes:

Total Points = sum(PPD1/Y1 + PPD2/Y2 + ... +PPDn/Yn).

However, I don't think that PG can use PPDn/Yn on it's own. There needs to be a way of incorportating PPDn / Yn into the following formula:

ppd = base_ppd * speed_ratio * max(1,sqrt(x*speed_ratio))

However the following:

ppd = base_ppd * speed_ratio * max(1,sqrt(x*speed_ratio)) / Yn

does not solve the effect and this can be seen on the following graph, where over time the purple line is still increasing:

Image

I put this down to the fact that what I said earlier and that it is because 'technological improvement' is having a compounding effect. Once in increasing the baseline PPD and then again in increasing the speed ratio.

Alternatively, and now I think about it, perhaps the following formula needs to be adjusted so that the concept of PPDn / Yn remains constant over time:

final_points = base_points * max(1,sqrt(k*deadline_length/elapsed_time))

Over time the value 'final points' for a given WU should actually decrease, because every 18 months the rate at which a given WU is completed doubles and this should cancel it out if we account for technological improvement.

I wonder, therefore, whether the following holds that true?

final_points = base_points * max(1,sqrt(k*deadline_length/elapsed_time/ Yn)) / Yn

Conclusion:

This is attempting to change as little of the formula as possible. It still allows PG to set relative importance on different platforms (by adjusting either the base points or the k value I think), but it simply removes the effect of technological improvement.
Jesse_V
Site Moderator
Posts: 2850
Joined: Mon Jul 18, 2011 4:44 am
Hardware configuration: OS: Windows 10, Kubuntu 19.04
CPU: i7-6700k
GPU: GTX 970, GTX 1080 TI
RAM: 24 GB DDR4
Location: Western Washington

Re: Suggested Change to the PPD System

Post by Jesse_V »

Thank you. I've been following this elaborate and very lengthy conversation and I appreciate the time you spent to rehash your proposal. :D
Perhaps you've already answered it, but how exactly would the PG determine the average hardware at a given point in time? Your idea is interesting and I can certainly see its advantage. You still have the incentive to get better hardware to fold faster, but while its destroying point inflation if I understand your idea correctly a donor's otherwise steady PPD will decrease over time. This could be a problem and this must be weighted very carefully against combating inflation. Currently I'm tentatively leaning towards the "it does more good than harm" side of the fence, but the question remains how the PG will move the curve. I would guess that they keep track of the hardware a donor has, and I know that they know how fast a WU was returned (adding the knowledge of how much computation is in a WU gives the FLOP count). Perhaps they can also see total PPD. In your opinion, would it be a good idea to take the total PPD and divide it by the number of active processors to get the average PPD/processor? Or are you letting it totally up to them as to how to find the average hardware? Remember that not everyone folds 24/7 so that will influence average hardware estimations unless they're careful.
F@h is now the top computing platform on the planet and nothing unites people like a dedicated fight against a common enemy. This virus affects all of us. Lets end it together.
k1wi
Posts: 909
Joined: Tue Sep 22, 2009 10:48 pm

Re: Suggested Change to the PPD System

Post by k1wi »

Thanks Jesse,

I have given examples of how PG could measure relative increases in technological performance in earlier posts, but have stepped back from doing so now because people then critique the suggest examples of how, rather than analyse and improve the framework/formula. Take the response I got to the 'average computer' or $'1000 dollar computer' examples I gave to explain my hypothesis for example.

My reason for simply saying "How to measure technological improvement is something that PG decides" is because at the end of the day they have a hell of a lot more data and statistics to play with than anyone in the folding community does! As a folding member I have publicly available FAH stats and I have non-FAH stats, but I would suggest that there is a lot of extra information that only PG has.

Each of your suggestions outlines the pros and cons for each method, but I would imagine that they could come to a figure based on a basket of measurements. The benefit of a basket being that they can look at a range of different statistics and come to a educated decision that is beneficial to all/majority. Perhaps they could look at quantiles and medians and means and perhaps also look at the relative growth in particular platforms.

I have thought about how accurate a measurement needs to be, and I would argue that even a rough estimation would still be better than the status quo. (The more accurate the better we would be able to compare relative performance at different time periods).

In my 'rehashing' I also didn't discuss the interval of adjustments. I do feel that it should be regular enough that it becomes 'a regular feature of the FAH calendar' and regular enough that the adjustments are not too big, but infrequent enough that people like beta testers can still compare new projects to existing projects under their current PPD curve. If it were done every quarter, in my own feeling this would be a good middle ground. Doing it every 18 months would (if performance does indeed double every 18 months) result in 100% decreases. Every year would be maybe 50%? (not sure on that), every 6 months would be 20%? (again not sure). If it were done regularly every three months, people could look forward to the "quarterly Pande Group technological improvement update and adjustment"!
MtM
Posts: 1579
Joined: Fri Jun 27, 2008 2:20 pm
Hardware configuration: Q6600 - 8gb - p5q deluxe - gtx275 - hd4350 ( not folding ) win7 x64 - smp:4 - gpu slot
E6600 - 4gb - p5wdh deluxe - 9600gt - 9600gso - win7 x64 - smp:2 - 2 gpu slots
E2160 - 2gb - ?? - onboard gpu - win7 x32 - 2 uniprocessor slots
T5450 - 4gb - ?? - 8600M GT 512 ( DDR2 ) - win7 x64 - smp:2 - gpu slot
Location: The Netherlands
Contact:

Re: Suggested Change to the PPD System

Post by MtM »

vbironchef wrote:MtM I can only speak for the team I used to fold for. People that bought the latest and greatest graphic cards or cpu's did not buy them for F@H as the one and only reason they made the purchase. Most are gamers and the others need them for work. F@H is a second or third thought, if at all. To base your theory that people go out and buy the latest and greatest just for F@H is incorrect. Most likely you will get a couple of hours a week if your lucky. Not to mention most people that buy those graphic cards do not pay for the electric bill. Not taking account for a dedicated folder that folds 24/7 is a mistake.

I do think I get where you are coming from. F@H doesn't care how or where the WU's are completed, just as long as it's fast and the more the better. F@H is not paying the electric bill or spending thousands of dollars building new rigs. So really F@H doesn't care a bit as long as the WU's are done. I'm glad I quit!
Wait, you're glad you quit because you don't want f@h to be as efficient as possible? Strange reasoning :(

I buy my own cards, I pay the electricity bill. When I bought them, most of my hw all was quite good in ppw/w, and yes that was a big consideration. But there is no way I would want f@h to become inefficient to protect my epeen, that's the most selfish idea I even heard. What makes you more important then the science, unless 'you' would be a representative of a group large enough to influence efficiency/throughput.
MtM
Posts: 1579
Joined: Fri Jun 27, 2008 2:20 pm
Hardware configuration: Q6600 - 8gb - p5q deluxe - gtx275 - hd4350 ( not folding ) win7 x64 - smp:4 - gpu slot
E6600 - 4gb - p5wdh deluxe - 9600gt - 9600gso - win7 x64 - smp:2 - 2 gpu slots
E2160 - 2gb - ?? - onboard gpu - win7 x32 - 2 uniprocessor slots
T5450 - 4gb - ?? - 8600M GT 512 ( DDR2 ) - win7 x64 - smp:2 - gpu slot
Location: The Netherlands
Contact:

Re: Suggested Change to the PPD System

Post by MtM »

Image

makes things easier to visualize. Saying 'flat line' when it will be a jigsaw confused me. As I said I didn't feel like putting the formula in a spreadsheet right then so I have to take your data for facts.

It would be easier to replace the line graph with a bar chart, where you have a sub chart which represent a single bar's actual curve. What happens if I read it right; if you keep the same hw, ppd will keep the same ( per adjustment period, not overall ), if you get hardware which is quicker you'll both return the work unit quicker = more throughput = more points, and your bonus will be higher = even more points, and you're not changing the exponential base of the QRB. At each adjustment, you will drop in ppd in a manner relative to the performance increase in computation speed.

Is this an accurate reflection?
Jorge1950
Posts: 113
Joined: Fri Apr 08, 2011 3:57 pm
Contact:

Re: Suggested Change to the PPD System

Post by Jorge1950 »

Hello all. I think if there are some who buy equipment for FAH, and pay electricity. The points used them to play on my mind. NO are essential. :D

We have talked about the players, they also pay their electricity, and if they are important points, since it is part of the "game".

The K1wi proposal is reasonable. Well supported and balanced. :!:

PG would be good to hear and opine. :?:
Sorry for the translation. This aspect, obliges me to very short and concrete sentences. Little by little he will go forward, also in this field.
Thank you for your understanding. Jorge Barrientos
Ravage7779
Posts: 144
Joined: Sun Jun 21, 2009 10:42 pm
Contact:

Re: Suggested Change to the PPD System

Post by Ravage7779 »

So we trade points inflation for points deflation. I have to upgrade my hardware to maintain the same points production. An expensive proposition.

Or as it is today, I have to upgrade my hardware to keep up with the joneses point production. Also an expensive proposition since you are locked into 4p amd server gear to manage that.

Which option provides more incentive to the points hounds and your average joe at the same time without turning people off to the project?
Get more Mac OS X folding help from the friendly experts @ http://www.foldingathome.org/index.html
Image
vbironchef
Posts: 23
Joined: Mon Jan 19, 2009 10:41 pm

Re: Suggested Change to the PPD System

Post by vbironchef »

Ravage7779 wrote:So we trade points inflation for points deflation. I have to upgrade my hardware to maintain the same points production. An expensive proposition.

Or as it is today, I have to upgrade my hardware to keep up with the joneses point production. Also an expensive proposition since you are locked into 4p amd server gear to manage that.

Which option provides more incentive to the points hounds and your average joe at the same time without turning people off to the project?
+1 Exactly.
Image
Post Reply