Change in BA requirements

[kasson]

The conversation about what the role of BA is in the project is a good one to have. The BA program has been of particular scientific benefit to my research group, but the FAH project as a whole has needed to position BA within a broad range of scientific needs.

Data that helped inform where to place the new thresholds:
~5% of active FAH machines with SMP>2 are at least 24 cores
~4% of active FAH machines with SMP>2 are at least 32 cores

Adjusting deadlines to match new hardware requirements will be a necessity. This is something we will have to calibrate carefully, and we unfortunately cannot state what those deadlines will be now. Our goal will be to allow reasonable 24-core and 32-core machines to make the deadlines. So the bottom line is that there will be deadline changes effective Feb 17 and Apr 17, precise nature TBD based on calibration experiments.

[bruce]

...This imbalance of points to long term donors we have already today....

To the other point: not want to make the points meaningless, just react flexible. If SMP is what is need right now more because there is science stuck in the folding pipeline it should receive a higher value to get done.

Let's suppose Stanford takes your advice. Your suggestion is to increase SMP points. That's not rebalancing, that's INCREASING points. While it may keep the bigadv folks happy because their points won't go down, it won't keep the long-time donors happy. Those of us who have been here 10-14 years, remember when each WU was worth ONE point. We may have folded for a year or two and earned maybe 1000 points. If you keep increasing points, as you suggest, our original few years continue to be less and less significant compared to what folks are earning today. You'd be penalizing the old-timers who earned a lot of points a long time ago. It's called "points inflation" when every change increases the points for any particular type of work.

On the other hand, if Stanford decides to rebalance by increasing points for a specific type of projects and they simultaneously decrease points for some other specific type of projects, that may reflect a realistic change in priority while avoiding points inflation but cetain to cause howls from those whose points are decreasing. That's not a good solution, either.

A "balanced" change that rewards somebody while penalizing somebody else is only attractive to those who will be getting their rewards increased -- everybody else is unhappy with the change. Since it's impossible to satisfy everybody, any kind of a change has to be very, very carefully considered -- and as Kasson just said, everybody's input has to be considered and all of the implications need to be considered.

[Grandpa_01]

Thanks Vj and Kasson for the update.

I am not at all sure switching the bigadv rigs over to smp is what is needed a % of the most likley is what they are looking for and from the announcement I would assume they are looking for the lower end of the spectrum. Bigadv is still important I did not see where they said it was not or that they wanted us to switch over. I did see that smp is lagging though. I am guessing that it is some of the larger atom smp projects that are falling behind and that is just guessing by what I get when I do get a smp WU on 1 of the 4P rigs. They are always the very large smp WU's that take several hrs to complete even on a 4P. Stanford could give us guidelines on this but that will not solve the problem of little pay for a high cost of running comparatively.
That is where a scientific mind differs from a donors mind in value and unfortunately I doubt the two will ever meet. A person can have all the good reasons in the world for valuing something one way or another but if the consumer does not buy it, it is just going to sit on the shelf.
Dr. Pande if you want the lower end of the bigadv spectrum to fold smp my guess would be you are going to need to adjust the rewards accordingly otherwise a large % of them will just switch to GPU or just stop folding.

As far as the bigadv change goes that is something that I expected maybe some did not, but I believe most did, I learned that lesson the last go around when my 7 - 1366 rigs could no longer play with the big dogs. It is pretty well known in the folding community that bigadv is an ever moving top end target and if we invest in the lower end of what is capable of doing the work that that investment will have a shorter life expectancy. When the next gen of E5 46xx v2 CPU's come out I would expect that my 4P AMD 61xx rigs will be at EOL for bigadv within a year maybe 2 but I would not bet on getting 2 years out of them (actually I am not sure about 1 year). I would be nice to have a long term road map here and hopefully if they get a PR man things like that will start happening. It could very well be that high end CPU folding is replaced with high end GPU folding but we are most likely talking $2000 to $3000 cards here not consumer grade equipment.

Vj I very much appreciate you taking the time to post in the thread it helps to hear from PG and as always I would not mind a little more info on the future, for a person like myself it is no small task to make future plans and a wrong choice could be very bad not only would I not have the best choice for my folding budget but F@H would not have the best equipment for their needs. And that can be quite discouraging to someone like myself.

[Grandpa_01]

...This imbalance of points to long term donors we have already today....

To the other point: not want to make the points meaningless, just react flexible. If SMP is what is need right now more because there is science stuck in the folding pipeline it should receive a higher value to get done.

Let's suppose Stanford takes your advice. Your suggestion is to increase SMP points. That's not rebalancing, that's INCREASING points. While it may keep the bigadv folks happy because their points won't go down, it won't keep the long-time donors happy. Those of us who have been here 10-14 years, remember when each WU was worth ONE point. We may have folded for a year or two and earned maybe 1000 points. If you keep increasing points, as you suggest, our original few years continue to be less and less significant compared to what folks are earning today. You'd be penalizing the old-timers who earned a lot of points a long time ago. It's called "points inflation" when every change increases the points for any particular type of work.

On the other hand, if Stanford decides to rebalance by increasing points for a specific type of projects and they simultaneously decrease points for some other specific type of projects, that may reflect a realistic change in priority while avoiding points inflation but cetain to cause howls from those whose points are decreasing. That's not a good solution, either.

A "balanced" change that rewards somebody while penalizing somebody else is only attractive to those who will be getting their rewards increased -- everybody else is unhappy with the change. Since it's impossible to satisfy everybody, any kind of a change has to be very, very carefully considered -- and as Kasson just said, everybody's input has to be considered and all of the implications need to be considered.

bruce I tend to believe a balanced approach would work. The other guys provide allot less ppd but are doing quite well as far as donor satisfaction goes and growth. My 4P rigs get 1/15th of what it gets here running there but comparatively, I get lets say a 4P 48 core rig at 3Ghz I get 75% more ppd than a person running a 12 core rig at 3Ghz and I am happy with that because it does accurately reflect my cost. If Stanford wants to add a bonus on top of that for the added initial investment incentive make it 10% or something like that.
If you combine something like that with good PR and a roadmap I think / believe that things can work themselves out perhaps if we ask people what they think about a more even points system and or investment incentive things can be worked out. You can have a incentive program along side a even points program but it needs to be in balance to accurately measure cost vs return.

[k1wi]

Thanks for the follow up post Kasson, I appreciate it.

Certainly some interesting data.

[Bill1024]

The conversation about what the role of BA is in the project is a good one to have. The BA program has been of particular scientific benefit to my research group, but the FAH project as a whole has needed to position BA within a broad range of scientific needs.

Data that helped inform where to place the new thresholds:
~5% of active FAH machines with SMP>2 are at least 24 cores
~4% of active FAH machines with SMP>2 are at least 32 cores

Adjusting deadlines to match new hardware requirements will be a necessity. This is something we will have to calibrate carefully, and we unfortunately cannot state what those deadlines will be now. Our goal will be to allow reasonable 24-core and 32-core machines to make the deadlines. So the bottom line is that there will be deadline changes effective Feb 17 and Apr 17, precise nature TBD based on calibration experiments.

Just to be clear. I understand this to mean 24 core 4P will stay and just 16 core will get eliminated? Thank you Dr. Kasson.

Bruce I remember the old days too, buying a dual core cpu, getting a big 50mhz overclock loading linux, folding a SMP wu for 7 days for a few hundred points.
Those CPUs were not cheap either way back then. I have a pile of obsolete cpus too.
I also remember my Mom giving me a dollar and a note to buy her 3 packs of smokes. I was told to keep the change.
I was able to get a pocket full of hard candy with that change. A dollar or a point, it was hard earned. But a dollar is only worth 20 cents now. Smokes are over 10$ a pack
A computer that can get 300k PPD should be able to get close to that on other WUs too, just as a lesser computer can't.
The power is there to do the work like a 500 HP motor.
It will make a big truck move, or a small car go real fast. A small low HP motor can't do either.
See where I am going with this?

[Punchy]

If you keep increasing points, as you suggest, our original few years continue to be less and less significant compared to what folks are earning today. You'd be penalizing the old-timers who earned a lot of points a long time ago. It's called "points inflation" when every change increases the points for any particular type of work.

Folding@Home entered the realm of points hyperinflation when the artificial QRB was introduced. It is fairly easy for a new donor, in a short amount of time, to eclipse old-timer's many years of contributions by focusing their hardware purchases on maximizing QRB.

In other DC projects, though there is a "natural QRB" simply due to technology improvements, there has been no equivalent exponential points growth, so it is not possible for newcomers to make such quick progress through the ranks.

Since the existing points structure was already destroyed once when the QRB was introduced, there is no reason to not do the same again in attempting to rebalance the system in a way that might make both donors and researchers happier.

[ChristianVirtual]

Thanks Vijay and Kasson for some more insights ... Much needed and appreciated.

The idea to get a communication bridge between both sides is great; hope you find one very soon. I'm sure lots of confusion, frustration and misunderstanding can be resolved by more, detailed and transparent communication; in a language both sides can understand. Starting with a roadmap on projects, cores and the requirements to get those done. Of course it is understood that such a roadmap contain some uncertainties in the long run, risk of life.

I think it is not difficult to keep BA running; just have the basic requirements on a roadmap will allow donors to plan ahead. For me concrete: I'm still in the decision process to get a 2P xeon E5 2680v2; dedicated for folding Thats 20/40c with 2.8GHz nominal speed. Sounds ok for now, sounds ok for the change in Feburary , kind of ok for April. Not sure for end of 2014. Due to the constraint of getting those chips at retail price only (and not spicy price) that's a significant chunk of money which need to run for a while. A 4P system at retail is not @home anymore; at least not for me.

I'm happy with Core 17 and GPU folding, I believe the job is well done from your team. My two 780 working smooth with few issues not in your responsibility. To hear there are more projects coming after 7810/7811 run out is good; again a hint in some roadmap would be great and reduce noise. When roughly and how many projects ?

Another aspect for me to optimize is also: space. Living in Tokyo is not leaving enough room to put multiple CPU boxes in my apartment. So I need to find folding solutions optimized for PPD/sqm. That's my individual problem, not many might have that. But that said: GPU and BA folding are for me the only two ways to create the feeling of a significant contribution.

Happy holidays to you, too.

[P5-133XL]

The conversation about what the role of BA is in the project is a good one to have. The BA program has been of particular scientific benefit to my research group, but the FAH project as a whole has needed to position BA within a broad range of scientific needs.

Data that helped inform where to place the new thresholds:
~5% of active FAH machines with SMP>2 are at least 24 cores
~4% of active FAH machines with SMP>2 are at least 32 cores

Adjusting deadlines to match new hardware requirements will be a necessity. This is something we will have to calibrate carefully, and we unfortunately cannot state what those deadlines will be now. Our goal will be to allow reasonable 24-core and 32-core machines to make the deadlines. So the bottom line is that there will be deadline changes effective Feb 17 and Apr 17, precise nature TBD based on calibration experiments.

Those numbers are useful to put a scale of bigadv against all SMP but I think more useful would be:

total # of active FAH machines running bigadv
# of active FAH machines running bigadv >24 cores
# of active FAH machines running bigadv >32 cores

Then they can be converted to percentages.

The reason those numbers are more useful is that it tells you better how many machines you are actually affecting when you institute your new policy. What you need to know is if you are affecting 5% of bigadv's or 50% of those machines when you go from 16 to 24/32.

[tear]

Those numbers are useful to put a scale of bigadv against all SMP but I think more useful would be:

total # of active FAH machines running bigadv
# of active FAH machines running bigadv >24 cores
# of active FAH machines running bigadv >32 cores

Then they can be converted to percentages.

The reason those numbers are more useful is that it tells you better how many machines you are actually affecting when you institute your new policy. What you need to know is if you are affecting 5% of bigadv's or 50% of those machines when you go from 16 to 24/32.

^^ +1
Performance (average bigadv WU processing time) vs number-of-hosts histogram would be even better.

[HaloJones]

With regards to points inflation I echo the comments that we already have it. We have got to stop looking backwards, constantly worrying about "old-timers". I'm one of them having been in the original genome@home program. Personally, I'd be in favour of resetting the clock, declaring the current points race finished, publishing them somewhere permanent and moving on to Folding@Homev2 with all new rationalised points across all units. QRB for all!

[billford]

Personally, I'd be in favour of resetting the clock, declaring the current points race finished, publishing them somewhere permanent and moving on to Folding@Homev2 with all new rationalised points across all units.

As a comparative newcomer, I'd not argue with that.

QRB for all!

But I might argue with that- I think QRB for none would be better. If you've got a fast system you'll process a WU in less time than a slow one, so the PPD goes up proportionately.

The current system effectively gives a double bonus- the "formal" QRB and another due to getting through more WUs in a given period of time.

[EXT64]

That is there for a specific reason - they want to encourage fewer fast machines over more slow ones. Without QRB 2 computers at 0.5x speed are equal to one computer at 1x speed. But to the scientists the 1x speed computer is a little more valuable as it can process a longer Gen in a reasonable period of time and thus get a project done faster (whereas the 0.5x will require more projects and get them done slower). But if points were really radically altered all aspects of the points system need to be scrutinized.

[billford]

I can see the logic in that… but (if I understand the PRCG "tree" correctly) the two 0.5x machines can be processing generations in two clones, thus exploring two trajectories.

Whether that's better (faster overall?) from the pov of the scientists only they can say…

[k1wi]

Whether that's better (faster overall?) from the pov of the scientists only they can say…

They have (when they outlined the QRB)

A computer that completes a WU (or trajectory) in a given period of time is more valuable than the same computer that half folds two in the same time frame.

ADD: bruce likes to use the relay-race analogy.