Making outlandish early claims of a large X% drop without acknowledging the fact that we have had a very nice "bonus" period throughout a lot of S5R3 is not really the best way to win people over to your point of view.
Well human nature being what it is... government clearly understands the psycology of instituting policy by giving something to the citizenry to gain favor and power because once given you have a war on your hands if the other party later tries to take it away. The so-called "S5R3 bonus" has become the defacto standard!
But what was the standard S5R3 app, the Linux power app or the standard app that most set it and leave it participants use?
You volunteer to contribute to the science, credits may give you a buzz, credit competitions, the science, whatever, you volunteer .. you volunteer ...
Credit junkies - If you don't like the points given to the new WUs then move onwards and find another BOINC project/s that credits more - you have that choice. [http://boincstats.com/stats/project_cpcs.php BOINC Credit Comparison]
If you 'like' the science you will stay and crunch onwards regardless of the credits - you have that choice.
Hopefully the powers to be at Einstein will monitor the joiners and leavers closely.
Hopefully there will be a healthy hard core to help continue their science.
I think that cross project equalisation is important, if for no other reason that to be able to assume that my 2.2 million BOINC credits are appoxiamately equal to Mike's 2.3 million.
Without a freeze of the current cross-project / BOINC-wide stats and having everyone start at zero, then that goal is unachievable due to the variances already being factored in.
Quote:
We knew as Gary pointed out that Einstein's cr/time was going to come down, about 25% at the end of S5R3, and to be in line with Seti.
I can't say for my Intel system, as I've never had it attached to here until just now, but for my AMD system, *ONLY* when the SSE app came out did my AMD system see any kind of advantage over the optimized SETI apps. That "advantage" generally ranged from 10-20%. Before that time, I saw a severe penalty (30-50%) during S5R2 for even running Einstein as opposed to SETI. Once it was figured out that the compiler was causing those of us with AMD systems some issues, we hacked around the problem until a different compilation method was used. Even that did not bring my system into "parity" with SETI, from my vantage point. Even the early non-SSE optimizations in S5R3 did not bring "parity".
What BOINC needs to do, since David Anderson clearly has this as an EXTREMELY HIGH PRIORITY, is to alter the stat dump to include a field that has an indicator of "stock" vs. "optimized" application. Bruce and Bernd then need to decide whether the "0" applications roughly equate to SETI's "stock" application and the "1" applications equate to SETI's 3rd party applications and calibrate accordingly.
Quote:
And those of us that also do Seti know that their rate is going to come down as they have calculated they are overpaying with the default app by about 15%.
It was first stated that projects that were under, like LHC for example, should move up to meet SETI. Now SETI is moving down to meet them. The most plausable explanation is that David was met with resistance from the projects in regards to implementing flop counting due to their own budgets, thus his continued quest to tilt at the "credit war" / "credit chaser" windmill needed a new plan of attack.
In the final analysis, there are still plenty of ways that this "New Deal" will fail to address actual "parity". It is something that has been hastily put in place and will not be implemented at all projects (I have already been told that there are technical roadblocks that will prevent implementation at some projects).
As the "New Deal" is proven to have failed in the quest for "parity", more cries for more "fixing" will come about.
Quote:
So one has to be prepared for an average reduction of over 40%, probably a lot more if your OS is Linux.
Linux apps have SSE2. Windows apps have SSE. This continues the tradition here of the Windows apps being at a disadvantage even on a system that is a dual-boot system (same hardware). In fact, it widens that gap in performance. Bernd has said that there will be an attempt to rectify that situation, but as it stands right now, even intra-project "parity" has not been achieved, thus even if this plan of Eric's (and probably influenced heavily by David) is the best thing in the whole wide world, the choice of me participating in this project which has a penalty for the OS choice, means that there is no "parity" for my systems BOINC-wide either...
Quote:
Based on performance of my computers, I think the present granted is low and my guestimate is that S5R4 units should get 260 +/- 20 credits.
I suggest abolishment of credit. At this point, it is the only thing that has "parity".
I came to the forum today because I noticed the credit had dropped significantly with the new wu's. I saw the thread, and as usual, saw that the "only in it for the science" purists were already suggesting that credit is pointless and that people interested in credit should go elsewhere. Great marketting.
The thing with BOINC is that the credit races, challenges, leagues etc. are basically fun.
Sitting doing nothing but have your computer using a lot more power then it otherwise would is not fun. The projects might be fun if there was enough feedback and interesting input from the staff, but how many projects are like that? How many have the resources?
You want lots of people to crunch lots of science, then it has to be worth it for the average punter. If it isn't, then you'll get people sign up out of curiosity then drift off after a short while. While it is fun, they'll stay.
With this specific case, i.e. a fixed credit/wu project, I have to assume that the value of a CPU hour is lower with the new wu's then the old. That indicates that either the science has less value, or that the wu's are doing less work in the same time period.
Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.
You know, whenever this cross-project credit equalization topic comes up, there is a small devil on my shoulder that tells me this: "Why don't you rent a few servers, and set up a BOINC project of your own. It would be named "1-2-10@Home" , and what the app would do is actually counting from 1 to 10, all over again, for, (say) ca 24 hours when run on a standard C2D 2.0 GHz. Each result would be worth 1.234.567.890 credits". The effort to setup such a system would be quite moderate (validation should be easy..). The aim of this project would be to heat up the atmosphere so I don't need winter tires anymore.
I mean just as an "experiment of thought": I could finance this by accepting money in exchange for higher credits for certain workunits (still less expensive to pay me to add more credit than buying an extra computer plus electricity to run it).
Would people flock to 1-2-10@home just to increase their combined stats (provided I can trick one of the stats sites to include my project)? Would they write SSE2 versions of the app which can count in parallel and much faster, for even more credits?
Well, probably not (hopefully not), but you must admit that this concept would annihilate the whole cross project credit equality idea in an instant. 1-2-10@Home would be nice to demonstrate that credits are worth nothing and scientific value is what really counts.
I totally agree with Mike. In my perfect BOINC-world, Einstein@Home should be crediting "Alberts", Seti should pay in "E.T.s", CPDN can call their credits ... whatever. Just so that nobody thinks that adding 10 Alberts plus 500 E.T.s should be equal to 510 something.
The stats-sites could then make a science out of converting from "Alberts" to "E.T.s" etc on a daily exchange rate basis, like "on Aug 6 2008, we think based on [whatever method], 10 Alberts are roughly equivalent to 125.5 E.T.s., and the credits of 1-2-10@Home aren't worth anything measurable at all, so that the total credit delta for user X from all projects on this day combined is roughly equivalent to [choose your preferred BOINC currency]".
The thing with BOINC is that the credit races, challenges, leagues etc. are basically fun.
I'm 100 % with you here. But are there "cross-project" races or challenges ? The races and challenges and team battles I've seen so far about ranking in a project stats or getting a certain RAC in a certain project. I really suspect that the combined score by adding all credits across projects is overrated. I#m not sure so many people care about this.
EDIT:
Quote:
With this specific case, i.e. a fixed credit/wu project, I have to assume that the value of a CPU hour is lower with the new wu's then the old. That indicates that either the science has less value, or that the wu's are doing less work in the same time period.
As for E@H, it's quite the opposite: The scientific value is the same between runs, but the apps are now doing more science per CPU hour than they used to do. If you don't lower credit, then a CPU hour of E@H will be worth more credits than a CPU hour at SETI. That's not so fair. Why is looking for E.T. worth less than looking for gravity waves. So the credit is lowered, leading to the situation you described: a CPU hour in the new run appears to be worth less than in the old run. It's a dilemma: Credit per WU is just a scalar number and you can't make it fair in all dimensions that you can consider fairness: cross-project, cross-architecture, and historically. No way.
People that care about their BOINC Combined figure are deluding themselves. When a high BOINC Combined RAC'er comes along, you simply look at their portfolio then decide.
Andersen has crusaded many times to try to bring the projects into line, but it is simply not going to happen. Why should it? We all know some projects "pay" more then others. We also all know that some projects are vastly more demanding then others. We also all know that wu failure rate is high at some projects.
If you are going to harmonise credit, then there must be a way of taking into account these other aspects, (and there are many others of course - I just hoghlight a few). That, I suspect, would be even more fraught with trouble then the credit differentials.
Call an apple an apple and 10 apples 10 apples but don't try to equate 10 apples with 10 spades - they are intrinsically different.
It is ironic that DA also campaigns to keep the projects independent of any kind of central authority?
Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.
I really suspect that the combined score by adding all credits across projects is overrated. I#m not sure so many people care about this.
I used to put more stock in it than I do now. I know that specific stat is worthless now.
However, a large percentage of people do care about that stat. It is the primary driving force behind all the machinations that we go through in the name of "cross-project parity".
Linux apps have SSE2. Windows apps have SSE. This continues the tradition here of the Windows apps being at a disadvantage even on a system that is a dual-boot system (same hardware). In fact, it widens that gap in performance. Bernd has said that there will be an attempt to rectify that situation, but as it stands right now, even intra-project "parity" has not been achieved, thus even if this plan of Eric's (and probably influenced heavily by David) is the best thing in the whole wide world, the choice of me participating in this project which has a penalty for the OS choice, means that there is no "parity" for my systems BOINC-wide either...
If you can't even get intra-project parity how in the world are you ever going to get inter-project parity on credits? I've been crunching here with all AMD's even with this "discrimination" because I enjoy the project and I though with the optimized client and getting 237+ credits for the work was reasonable. I don't really care what any other projects pay in credits. Seti is a disaster and always has been with hardware, reliablity, just to mention a few.
I have to wonder what it all means that now I spend twice as much time doing a wu and getting less credits. I think before we start worrying about inter-project parity we need to get everyone in the same project on the same page. Once you've accomplished that then you can worry about everyone else and decide if its even worth it. If your users are satisfied that everyone here is getting the same amount of credit for the same amount of work and they are happy with the way the project is run and the value of the science is worth while they aren't going to be running around chasing dreams anyway.
RE: RE: Making
)
But what was the standard S5R3 app, the Linux power app or the standard app that most set it and leave it participants use?
My observations for what its
)
My observations for what its worth ..
Its a personal choice is it not?
You volunteer to contribute to the science, credits may give you a buzz, credit competitions, the science, whatever, you volunteer .. you volunteer ...
Credit junkies - If you don't like the points given to the new WUs then move onwards and find another BOINC project/s that credits more - you have that choice. [http://boincstats.com/stats/project_cpcs.php BOINC Credit Comparison]
If you 'like' the science you will stay and crunch onwards regardless of the credits - you have that choice.
Hopefully the powers to be at Einstein will monitor the joiners and leavers closely.
Hopefully there will be a healthy hard core to help continue their science.
RE: I think that cross
)
Without a freeze of the current cross-project / BOINC-wide stats and having everyone start at zero, then that goal is unachievable due to the variances already being factored in.
I can't say for my Intel system, as I've never had it attached to here until just now, but for my AMD system, *ONLY* when the SSE app came out did my AMD system see any kind of advantage over the optimized SETI apps. That "advantage" generally ranged from 10-20%. Before that time, I saw a severe penalty (30-50%) during S5R2 for even running Einstein as opposed to SETI. Once it was figured out that the compiler was causing those of us with AMD systems some issues, we hacked around the problem until a different compilation method was used. Even that did not bring my system into "parity" with SETI, from my vantage point. Even the early non-SSE optimizations in S5R3 did not bring "parity".
What BOINC needs to do, since David Anderson clearly has this as an EXTREMELY HIGH PRIORITY, is to alter the stat dump to include a field that has an indicator of "stock" vs. "optimized" application. Bruce and Bernd then need to decide whether the "0" applications roughly equate to SETI's "stock" application and the "1" applications equate to SETI's 3rd party applications and calibrate accordingly.
It was first stated that projects that were under, like LHC for example, should move up to meet SETI. Now SETI is moving down to meet them. The most plausable explanation is that David was met with resistance from the projects in regards to implementing flop counting due to their own budgets, thus his continued quest to tilt at the "credit war" / "credit chaser" windmill needed a new plan of attack.
In the final analysis, there are still plenty of ways that this "New Deal" will fail to address actual "parity". It is something that has been hastily put in place and will not be implemented at all projects (I have already been told that there are technical roadblocks that will prevent implementation at some projects).
As the "New Deal" is proven to have failed in the quest for "parity", more cries for more "fixing" will come about.
Linux apps have SSE2. Windows apps have SSE. This continues the tradition here of the Windows apps being at a disadvantage even on a system that is a dual-boot system (same hardware). In fact, it widens that gap in performance. Bernd has said that there will be an attempt to rectify that situation, but as it stands right now, even intra-project "parity" has not been achieved, thus even if this plan of Eric's (and probably influenced heavily by David) is the best thing in the whole wide world, the choice of me participating in this project which has a penalty for the OS choice, means that there is no "parity" for my systems BOINC-wide either...
I suggest abolishment of credit. At this point, it is the only thing that has "parity".
I like this project and my
)
I like this project and my host continue to crunch Einstein@Home WU's without any conditions. :)
I came to the forum today
)
I came to the forum today because I noticed the credit had dropped significantly with the new wu's. I saw the thread, and as usual, saw that the "only in it for the science" purists were already suggesting that credit is pointless and that people interested in credit should go elsewhere. Great marketting.
The thing with BOINC is that the credit races, challenges, leagues etc. are basically fun.
Sitting doing nothing but have your computer using a lot more power then it otherwise would is not fun. The projects might be fun if there was enough feedback and interesting input from the staff, but how many projects are like that? How many have the resources?
You want lots of people to crunch lots of science, then it has to be worth it for the average punter. If it isn't, then you'll get people sign up out of curiosity then drift off after a short while. While it is fun, they'll stay.
With this specific case, i.e. a fixed credit/wu project, I have to assume that the value of a CPU hour is lower with the new wu's then the old. That indicates that either the science has less value, or that the wu's are doing less work in the same time period.
Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.
You know, whenever this
)
You know, whenever this cross-project credit equalization topic comes up, there is a small devil on my shoulder that tells me this: "Why don't you rent a few servers, and set up a BOINC project of your own. It would be named "1-2-10@Home" , and what the app would do is actually counting from 1 to 10, all over again, for, (say) ca 24 hours when run on a standard C2D 2.0 GHz. Each result would be worth 1.234.567.890 credits". The effort to setup such a system would be quite moderate (validation should be easy..). The aim of this project would be to heat up the atmosphere so I don't need winter tires anymore.
I mean just as an "experiment of thought": I could finance this by accepting money in exchange for higher credits for certain workunits (still less expensive to pay me to add more credit than buying an extra computer plus electricity to run it).
Would people flock to 1-2-10@home just to increase their combined stats (provided I can trick one of the stats sites to include my project)? Would they write SSE2 versions of the app which can count in parallel and much faster, for even more credits?
Well, probably not (hopefully not), but you must admit that this concept would annihilate the whole cross project credit equality idea in an instant. 1-2-10@Home would be nice to demonstrate that credits are worth nothing and scientific value is what really counts.
I totally agree with Mike. In my perfect BOINC-world, Einstein@Home should be crediting "Alberts", Seti should pay in "E.T.s", CPDN can call their credits ... whatever. Just so that nobody thinks that adding 10 Alberts plus 500 E.T.s should be equal to 510 something.
The stats-sites could then make a science out of converting from "Alberts" to "E.T.s" etc on a daily exchange rate basis, like "on Aug 6 2008, we think based on [whatever method], 10 Alberts are roughly equivalent to 125.5 E.T.s., and the credits of 1-2-10@Home aren't worth anything measurable at all, so that the total credit delta for user X from all projects on this day combined is roughly equivalent to [choose your preferred BOINC currency]".
Cheers
Bikeman
RE: The thing with BOINC
)
I'm 100 % with you here. But are there "cross-project" races or challenges ? The races and challenges and team battles I've seen so far about ranking in a project stats or getting a certain RAC in a certain project. I really suspect that the combined score by adding all credits across projects is overrated. I#m not sure so many people care about this.
EDIT:
As for E@H, it's quite the opposite: The scientific value is the same between runs, but the apps are now doing more science per CPU hour than they used to do. If you don't lower credit, then a CPU hour of E@H will be worth more credits than a CPU hour at SETI. That's not so fair. Why is looking for E.T. worth less than looking for gravity waves. So the credit is lowered, leading to the situation you described: a CPU hour in the new run appears to be worth less than in the old run. It's a dilemma: Credit per WU is just a scalar number and you can't make it fair in all dimensions that you can consider fairness: cross-project, cross-architecture, and historically. No way.
CU
Bikeman
RE: I#m not sure so many
)
People that care about their BOINC Combined figure are deluding themselves. When a high BOINC Combined RAC'er comes along, you simply look at their portfolio then decide.
Andersen has crusaded many times to try to bring the projects into line, but it is simply not going to happen. Why should it? We all know some projects "pay" more then others. We also all know that some projects are vastly more demanding then others. We also all know that wu failure rate is high at some projects.
If you are going to harmonise credit, then there must be a way of taking into account these other aspects, (and there are many others of course - I just hoghlight a few). That, I suspect, would be even more fraught with trouble then the credit differentials.
Call an apple an apple and 10 apples 10 apples but don't try to equate 10 apples with 10 spades - they are intrinsically different.
It is ironic that DA also campaigns to keep the projects independent of any kind of central authority?
Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.
RE: I really suspect that
)
I used to put more stock in it than I do now. I know that specific stat is worthless now.
However, a large percentage of people do care about that stat. It is the primary driving force behind all the machinations that we go through in the name of "cross-project parity".
RE: Linux apps have SSE2.
)
If you can't even get intra-project parity how in the world are you ever going to get inter-project parity on credits? I've been crunching here with all AMD's even with this "discrimination" because I enjoy the project and I though with the optimized client and getting 237+ credits for the work was reasonable. I don't really care what any other projects pay in credits. Seti is a disaster and always has been with hardware, reliablity, just to mention a few.
I have to wonder what it all means that now I spend twice as much time doing a wu and getting less credits. I think before we start worrying about inter-project parity we need to get everyone in the same project on the same page. Once you've accomplished that then you can worry about everyone else and decide if its even worth it. If your users are satisfied that everyone here is getting the same amount of credit for the same amount of work and they are happy with the way the project is run and the value of the science is worth while they aren't going to be running around chasing dreams anyway.
Just my 2 cents worth.