Just a quick note to say I agree 100% with everything that Tony (mmciastro) has written here.

I also have no objection to optimised applications, and have used them since Tetsuji introduced them last year. But like Tony I was very skeptical of the BOINC client alterations, and have frequently raised my objections to the assumption that the credits for an average unit prior to version 5.n was approx. 32 credits. And like Tony I have not in the past accused people who use optimised clients of cheating, but do believe they are, especially those using them with other projects.

If you read the Beta NC board you will probably have noticed I was, in most cases, the first to point out the credits/time were not in line with other projects, and that until the final phase the credits were not linear with AR. Initially when the crunch times were 30+ hrs they were too low. My Beta account # is 163, so you can see I signed up within the first few days.

On this subject I think the project was released here too soon, as I don't think enough low and very low angle range units were done to make the correct adjustments to the credit calculation. I also have some reservations on the credit calculation at high AR with respect to different cpu's, especially as I understand most of the analysis was done on one make of cpu, and that is not the most common make of cpu.

Andy

Just a quick note to say I agree 100% with everything that Tony (mmciastro) has written here.

I also have no objection to optimised applications, and have used them since Tetsuji introduced them last year. But like Tony I was very skeptical of the BOINC client alterations, and have frequently raised my objections to the assumption that the credits for an average unit prior to version 5.n was approx. 32 credits. And like Tony I have not in the past accused people who use optimised clients of cheating, but do believe they are, especially those using them with other projects.

If you read the Beta NC board you will probably have noticed I was, in most cases, the first to point out the credits/time were not in line with other projects, and that until the final phase the credits were not linear with AR. Initially when the crunch times were 30+ hrs they were too low. My Beta account # is 163, so you can see I signed up within the first few days.

On this subject I think the project was released here too soon, as I don't think enough low and very low angle range units were done to make the correct adjustments to the credit calculation. I also have some reservations on the credit calculation at high AR with respect to different cpu's, especially as I understand most of the analysis was done on one make of cpu, and that is not the most common make of cpu.

Andy

Which type of cpu was that?

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's

Thinking back to when enhanced was being developed (Seti Beta), I remember Matt L. saying something to the effect that under the current Boinc system and growth that it was becoming more and more difficult for the servers to maintain that growth. Enhanced and eventually Astorpulse would dramatically reduce the load on their hardware and allow the project to continue their growth.

I believe that many of us in the Seti Community can accept these changes. Very few would argue that if the development of these projects did indeed allow for that growth.

Of course I'm stating the obvious that the whole argument is the credit system. Could it have been possible that rather than the resulting system that has been packaged with enhanced..but to have left it similar to what the original boinc app produced?

If the purpose was to reduce the load on the servers..why would one alter the credit system as well? I think that more people would have been happy if that Enhanced and Astropulse would have only done so..(reducing server load)

Of course I'm stating the obvious that the whole argument is the credit system. Could it have been possible that rather than the resulting system that has been packaged with enhanced..but to have left it similar to what the original boinc app produced?

It would be entirely possible to have left the credit system alone.

You may want to read the message boards and look at all the complaints about S@H, benchmarks, how the benchmarks are "inaccurate" and how different CPUs benchmark low but produce work quickly, or benchmark high and produce low.

4.18 estimated the number of (mainly floating point) operations based on the benchmark * CPU time. 5.12 actually counts floating-point ops.

If you look at your own results, you can see that there was almost always a difference between claimed and granted credits, and now they are almost always the same.

That's better, even if some "win" and some "lose" as a result.

One way to stop the complaints would be to just raise the credits by 50% but that starts the whole credit inflation thing, and that's exactly what the project was trying to avoid.

---

Just a question from someone who has only been at this for a few months. If credits don't matter why are they even kept and recorded? If they are not important why list and post them everywhere? Why all those little stat boxes i see all around? I understand this whole BOINC thing is about the science but anyone can plainly see it's not only about the science. That may have been the project's original intent but it has obviously evolved into something more. Pure scientific exploration started this mission, but giving it's participants something tangible to experience is what has made it grow to include some 650,000 people world wide. Science is at the core of why we are all here. I see nothing wrong with dressing it up a little if it leads to more work getting done. However frivilous it may seem to some the element of competition is what will get BOINC it's 1,000,000th and 2,000,000th user.

---

So that a workunit claimed X amount of credit, no matter what machine it is processed on. And that is fine. I think it's great. But to do that, then also slash the amount of credit granted by each workunit is WRONG. At least running v5.11, the granted credit WAS close to the break-even point. But when each one of us looks at what they have accomplished here, then sees that accomplishment cut by a factor of THREE, then what do we have to look forward to?

Well, let's see, someone running the "normal" application on a specific computer gets example 10 Cobblestones/hour, and this "pay" haven't changed significantly from v4.18 & v5.12.

Someone running on an exact "clone" of the same computer, is running an optimized seti-application, and doesn't understand why he's suddenly not getting 30 Cobblestones/hour any longer...


Well, why does an optimized v4.18-seti-application give 3x more Cobblestones/hour compared to "normal" application?
Because it's 3x faster doing the same work.
But, an optimized Seti_Enhanced is maybe only 25% faster, and will therefore only get "paid" 25% more than "normal" application.

So, the fact is, no-one is paid "less" for doing the work, but it's rather the optimizers isn't doing 3x more work/hour and gets 3x more Cobblestones/hour, but instead with Seti_Enhanced they're only doing 25% more work/hour and gets only 25% more Cobblestones/hour.





Let's look on a non-SETI@Home-example, let's say Jack & Joe is filling 1-litre-milk-bottles, and is paid according to how many milk-bottles they've filled in a day.

Jack walks and collects 1 bottle at a time, fills it with milk, and walks and puts it in a case. He manages 1000 bottles/day.

Joe after a little thinking decides walking with only 1 bottle wastes much time, so instead he moves the whole case, fills-it, and moves the case back. Due to cutting-down on walking, he manages 3000 bottles/day.

In this example, Joe gets paid 3x more than Jack, since Joe has filled 3x more bottles/day.


So, after some time, a new "high-speed" pump that is 5x faster filling a bottle is installed, and Jack & Joe gets slightly different work-descriptions, they'll get 2x as much for each milk-bottle they manages to fill, but each bottle is larger, 10 litre.

Jack works just like before, walks and collects 1 bottle at a time, fills it with milk, and walks and puts it in a case. He manages 500 bottles/day. Since each bottle is now worth 2x, he gets paid the same as he did for 1000 1-litre-bottles.

Joe still collects the whole case, fills-it, and moves the case back. But, since the bottles is now much heavier, he uses much longer to move the full case, and manages only 625 bottles/day. At 2x, this means he is paid the same as if he did only fill 1250 1-litre-bottles.

Now, Joe is still filling more bottles/day than Jack, and he's still getting paid more than Jack. But, since Joe isn't filling 3x more bottles than Jack any longer, he isn't paid 3x more than Jack any longer either. Instead, Joe now gets 25% more paid than Jack, since Joe is filling 25% more bottles than Jack.

We can live without Seti, but the project cannot exist without US. And what about all the other projects that would not even be here, if not for the success of Seti. Like it or not, credits are the "wages" of the D.C. world.

Daniel:

Sounds like extortion to me. I don't think that's in the spirit of a community dedicated to a grand search for ET. I think you might need to reconsider your motivation to participate in this project. I don't intend that statement to be mean-spirited. All of us who are truly interested in this search for ET, however improbable the search may be, greatly appreciate the contribution to this project that you and others like you have made.

Regarding Benchmark vs. FPOP:
The benchmark system was/is intended to encourage users of old slow computers to contribute their computing time to Boinc projects. Under the benchmark system the 'pay' was for hours of time crunching. Under the FPOP system the 'pay' is for actual work done. Another way of saying this would be 'pay' for product produced.

If my old computer produces 1 unit of product in 3 hours and my new computer produces 3 units of product in 3 hours, which computer is doing more work? Which computer should get more 'pay'? If I run an optimized app and my other new computer produces 4 units of product in 3 hours, wouldn't you say that it is doing more work? Shouldn't it get more 'pay' than the old slow computer?

The danger is that the FPOP system WILL CAUSE people to remove their old slow computers from the project. The advantage is that it will encourage people to upgrade the computing power of the computers that remain in the project. In the end I believe this will benefit all Boinc projects by providing a higher powered more competent base of computers which will increase the actual amount of work being done.

Of course I'm stating the obvious that the whole argument is the credit system. Could it have been possible that rather than the resulting system that has been packaged with enhanced..but to have left it similar to what the original boinc app produced?

It would be entirely possible to have left the credit system alone.

You may want to read the message boards and look at all the complaints about S@H, benchmarks, how the benchmarks are "inaccurate" and how different CPUs benchmark low but produce work quickly, or benchmark high and produce low.

4.18 estimated the number of (mainly floating point) operations based on the benchmark * CPU time. 5.12 actually counts floating-point ops.

If you look at your own results, you can see that there was almost always a difference between claimed and granted credits, and now they are almost always the same.

That's better, even if some "win" and some "lose" as a result.

One way to stop the complaints would be to just raise the credits by 50% but that starts the whole credit inflation thing, and that's exactly what the project was trying to avoid.

Problem is I think Intel cpus and AMD cpus have a difference in speed here as Intel is supposedly faster in floating point, Of course If the app has an intel bias now then It doesn't seem anymore fair than the previous system as Now AMD systems are penalized for having slower floating point, The Megahertz myth is alive and well It seems. And I don't and won't any buy Intel stuff I might add.

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's

Problem is I think Intel cpus and AMD cpus have a difference in speed here as Intel is supposedly faster in floating point, Of course If the app has an intel bias now then It doesn't seem anymore fair than the previous system as Now AMD systems are penalized for having slower floating point, The Megahertz myth is alive and well It seems. And I don't and won't any buy Intel stuff I might add.

The machine from my example is an Athlon XP.

The old scoring system gave various results depending on the CPU types and efficiencies of the quorum -- if AMD was better, having two AMD parts would bias the score up, while having two Intel parts would bias it down.

---

Well, let's see, someone running the "normal" application on a specific computer gets example 10 Cobblestones/hour, and this "pay" haven't changed significantly from v4.18 & v5.12.

If I understood correctly, the 'standard' 5.12 itself is more optimized than the 'standard' 4.18. And that would be the reason why the optimized 5.12 isn't 3x quicker than the optimized 4.18...

So indeed the optimized users get less credit than before... that's a pity for anyone fond of credits. All in all, more science is done. That should weigh up to it for most of us. After all, it's only a hobby. :)

Greetings,
Sander

---

Problem is I think Intel cpus and AMD cpus have a difference in speed here as Intel is supposedly faster in floating point, Of course If the app has an intel bias now then It doesn't seem anymore fair than the previous system as Now AMD systems are penalized for having slower floating point, The Megahertz myth is alive and well It seems. And I don't and won't any buy Intel stuff I might add.

The machine from my example is an Athlon XP.

The old scoring system gave various results depending on the CPU types and efficiencies of the quorum -- if AMD was better, having two AMD parts would bias the score up, while having two Intel parts would bias it down.

Well I had asked and then You didn't reply, For reasons unknown to Me. In the past software was set for Intels, Of course I don't and haven't used Athlon XP cpus in a couple of Years, Nor do I have any.

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's

Problem is I think Intel cpus and AMD cpus have a difference in speed here as Intel is supposedly faster in floating point, Of course If the app has an intel bias now then It doesn't seem anymore fair than the previous system as Now AMD systems are penalized for having slower floating point, The Megahertz myth is alive and well It seems. And I don't and won't any buy Intel stuff I might add.

The machine from my example is an Athlon XP.

The old scoring system gave various results depending on the CPU types and efficiencies of the quorum -- if AMD was better, having two AMD parts would bias the score up, while having two Intel parts would bias it down.

Well I had asked and then You didn't reply, For reasons unknown to Me. In the past software was set for Intels, Of course I don't and haven't used Athlon XP cpus in a couple of Years, Nor do I have any.

I think I responded promptly.

Statements like "software was set for Intels" always bother me. Optimizing code for a specific set of features (cache size, number of floating point units, avoiding pipeline stalls, taking advantage of parallel execution) is always possible, for the project it isn't worth the bother. It requires a lot of time and a certain amount of actual talent.

SETI isn't trying to make Intel (or AMD) look good, they're trying to run on as many different architectures as possible -- and get correct results.

Instead, the project can look for optimizations like caching the results of SIN and COS operations (which are slow on all CPUs). Those help everyone (if I'm not mistaken 4.18 does not cache trig ops, and 5.13 does).

They can look for faster FFT implementations that run across all platforms.

Mostly, different processors are, well, different.

As for my choice of processor: the Athlon XP was a good choice when I bought it, not the fastest thing around, but this machine spends most of its time waiting for me to type (Crane's law: all computers wait at the same speed). A dual Opteron would do the same.

When I can buy a new machine that it at least three times faster, I'll spend my money. Until then, fast enough really is fast enough.

---

If the previous credit system was so unfair, why have I not seen discussions to that affect?

I don't know how you could miss it. It's been ongoing for as long as i've been here (12-24 months?).

---

Grant
Darwin NT

Just a question from someone who has only been at this for a few months. If credits don't matter why are they even kept and recorded? If they are not important why list and post them everywhere? Why all those little stat boxes i see all around? I understand this whole BOINC thing is about the science but anyone can plainly see it's not only about the science. That may have been the project's original intent but it has obviously evolved into something more. Pure scientific exploration started this mission, but giving it's participants something tangible to experience is what has made it grow to include some 650,000 people world wide. Science is at the core of why we are all here. I see nothing wrong with dressing it up a little if it leads to more work getting done. However frivilous it may seem to some the element of competition is what will get BOINC it's 1,000,000th and 2,000,000th user.

Only as a way to measure progress.

For some, credits become an overriding quest -- to post the biggest number.

For others (like myself) it's just interesting to see what my (one) cruncher has done, on a shared basis with other tasks.

---

Suppose you show up at work tomorrow. Your boss tells you that he is cutting your wages by a factor of three...

Yes, but this isn't work.
It something you chose to do voluntarily. If you choose not to then so be it.
It would be sad that people take their bat & ball & go home becasue they're not getting as much as what they're used to; but i guess that's just the way some people are.

---

Grant
Darwin NT

...If you look at your own results, you can see that there was almost always a difference between claimed and granted credits, and now they are almost always the same.

That's better, even if some "win" and some "lose" as a result.

One way to stop the complaints would be to just raise the credits by 50% but that starts the whole credit inflation thing, and that's exactly what the project was trying to avoid.

Agreed Ned

understood that the faster, more powerful machines still would earn more credit than slower less powerful. People just want to see their machines earn the same overall credit.

Question is...is it unreasonable to ask to earn the same amount of total credit?

We can live without Seti, but the project cannot exist without US. And what about all the other projects that would not even be here, if not for the success of Seti. Like it or not, credits are the "wages" of the D.C. world.

Daniel:

Sounds like extortion to me. I don't think that's in the spirit of a community dedicated to a grand search for ET. I think you might need to reconsider your motivation to participate in this project. I don't intend that statement to be mean-spirited. All of us who are truly interested in this search for ET, however improbable the search may be, greatly appreciate the contribution to this project that you and others like you have made.

Regarding Benchmark vs. FPOP:
The benchmark system was/is intended to encourage users of old slow computers to contribute their computing time to Boinc projects. Under the benchmark system the 'pay' was for hours of time crunching. Under the FPOP system the 'pay' is for actual work done. Another way of saying this would be 'pay' for product produced.

If my old computer produces 1 unit of product in 3 hours and my new computer produces 3 units of product in 3 hours, which computer is doing more work? Which computer should get more 'pay'? If I run an optimized app and my other new computer produces 4 units of product in 3 hours, wouldn't you say that it is doing more work? Shouldn't it get more 'pay' than the old slow computer?

The danger is that the FPOP system WILL CAUSE people to remove their old slow computers from the project. The advantage is that it will encourage people to upgrade the computing power of the computers that remain in the project. In the end I believe this will benefit all Boinc projects by providing a higher powered more competent base of computers which will increase the actual amount of work being done.

Not if it causes those who joined for the competition to leave! Most of the upgrading that has been taking place has been due to the competition between individuals/teams in the race to be the best. Do you really think everybody is spending money and running multiple machines just for the science?? If you do you are in for a rude awakening. One team alone has crunched approximately 2-3 million WUs in just over a year mainly due to the competition (and that is probably a low estimate). I would hazard to say that probably 50% or more of the "volunteers" crunching today are in it as deep as they are because of the competition. Without the competitive side of it I would probably be quite happy running it on the 3 main machines in my house, but then you would be losing the other 21 machines I am currently using, and I would have spent the money I have used for upgrades to these machines on other things. If you kill the competition with a diluted credit system, in the long run you will be hurting the project rather than helping it.

---

I would say that the system is fair. I ran an optimised client & seti app but found that I was normally claiming 30-35 credits but getting 20-25 (or even less). I am now running the optimised enhanced app and find that I'm now claiming as near as damn it what I was getting before hand.

The only reason my RAC has droped is because I'm waiting on other computers to return work so what I've done can be validated.

---

I really hope someone at Berkeley is reading this thread and paying attention to all the people who are being told "just leave if you do not like it cuz we don't need you"!! That is really a good way to solve the issue. Just run off the ones who are complaining.

---

I think the analogy with workers and wages is good, but I have a different view.

I have been crushing for S@H since 1999. With Boinc, S@H became part of a shopping mall, each project selling the same basic product, with different themes. They are all competing for scarse assets (i.e. clients). As in all marketing activity, the sellers have to cativate their clients, or they will go away, and do business in the next shop. We, the cruncher, are the clients. Now we have the option to choose what science is worth our time (pernonal and computer).

At this mall, some stores are offering rewards for crushers who improve their products (like E@H is doing with Akros), other are letting the client decide the size of the WU they whant (like Rosetta).

I see S@H still thinking as the monopoly of science, with a motto "My way is the right way". I even read someone calling users of optimized apps as "cheating". I hope this is not the official view of the project...

This is a great project, but others also are. If there are some chunchers that want more credit, give them. It's much easier than raising money thru donnations.

---