Folks

My reply was a bit over the top...

When I again picked, the Donations and the Hardware Donations... I became privy to a large amount of the hardware needs and what in the "process" was required to get MultiBeam into Seti Main... With the advent of the Staff Blog and Tech News Forums people have been both forewarned and given more insight into the process that happens on the other side of Port 80. A portion of that curtain has been lifted... It was quite an interesting time sitting over coffee with Eric in Seattle...

I also Thank, the numerous people that have found a way to send money and hardware to keep Seti running to this day...

That said, I apologize for the manner and tone of my statement... I am one of people believes Seti should survive and when things are right Results of all of "OUR Work" will start being known. I still have a large amount of faith that we will find "something," whether it is ET or uncover other important things about our universe... It will happen...

---

Please consider a Donation to the Seti Project.

Folks, I was trying to be sincere. Sorry you took it differently. (Except my crack about Pappa, who seems to me to have started this e-tussle.)

<snip>

A better board would be filled with objective analysis of the issues, rather than amateur psycho-analysis, emotional demagoguery, and hand-waving argumentation. Let's all just try to be better engineers and scientists.

Sorry if you take exception to this, but again, question of style.

In this post you say "It just seems like good project management to go to the effort to engineer the change" and I don't know why you'd make that statement unless you thought that the change was not engineered.

In this post you said "So if all we have done is look at data, we likely are not making progress." The purpose of our crunching is to find the signals that need further measurement. It's a litmus test. The (capital S) Science starts after we find a possible candidate.

In this post "something at CommandCentral has a loose nut and that it isn't my slug's fault."

Maybe it's just poor choice of words, but I don't think I'm alone when I see an implied "and it's a failure of SETI's management team" in many such posts.

Frankly, I'm impressed that so much is being done by so few with so little, and if in fact this turns out to be a mistake, it's one they can reverse fairly quickly.

From the answers you received from Matt, it does not appear that you needed to get on your "change control soapbox" at all.

---

Hummmm... I think a more appropriate point of view is that the power efficiency is considerably improved. In the most common case, each WU will consume the system power to be computed three times instead of four. The extras beyond the first generate confidence. Confidence beyond that needed for the scientific result should not be weighed equally in a power efficiency calculation.

Obviously I am assuming the project folks are making a good choice on the value (or lack of value) of quorum-driven confidence.

Einstein has been sending out two and requiring two for some time. That combined with a very non-random quorum partner assignment gives a much more bursty pattern of validation and credit assignment, but most of the regulars posting there seem to have found their peace with it. It also is a considerable additional power efficiency and total output gain over 3/2.

I disagree for a couple of reasons:

1.) Whereas you can say from the Projects POV it's more power efficient in that they are getting more WU's done per unit time, this fails to take into account they aren't paying for the electricity to do the actual calculations. You simply cannot get around the fact that from the hosts POV, one third of the results are now going to be doing no science rather than one quarter.

2.) Results which are not included in the validation process add nothing to the confidence that the ones which were submitted accurately represented the WU.

Proof:

Consider the case where a quorum has been formed. All the results submitted are compared and a representative result is chosen to be canonical and inserted in the MSD.

Now lets consider what happens when the "trailing" result returns. At this point if it's at least weakly similar to the canonical result it will be granted credit. But what happens if the trailing result is aborted or never gets returned?

If the result was actually needed, then the WU would not have validated in the first place and would have remained in the "Checked, but no concensus" state, and if it has validated then the trailing result wasn't really needed at all. In fact if a replacement has already been queued up by the project server, it's canceled before it's sent if possible.

Therefore, extra results sent out above what the minimum quorum is set to are superfluous in all cases from a science viewpoint, and merely waste electricity on hosts unfortunate enough to be running them.

Alinator

I disagree for a couple of reasons:

1.) Whereas you can say from the Projects POV it's more power efficient in that they are getting more WU's done per unit time, this fails to take into account they aren't paying for the electricity to do the actual calculations. You simply cannot get around the fact that from the hosts POV, one third of the results are now going to be doing no science rather than one quarter.

2.) Results which are not included in the validation process add nothing to the confidence that the ones which were submitted accurately represented the WU.

Proof:

Consider the case where a quorum has been formed. All the results submitted are compared and a representative result is chosen to be canonical and inserted in the MSD.

Now lets consider what happens when the "trailing" result returns. At this point if it's at least weakly similar to the canonical result it will be granted credit. But what happens if the trailing result is aborted or never gets returned?

If the result was actually needed, then the WU would not have validated in the first place and would have remained in the "Checked, but no concensus" state, and if it has validated then the trailing result wasn't really needed at all. In fact if a replacement has already been queued up by the project server, it's canceled before it's sent if possible.

Therefore, extra results sent out above what the minimum quorum is set to are superfluous in all cases from a science viewpoint, and merely waste electricity on hosts unfortunate enough to be running them.

Alinator

Perhaps you should consider the situation where only 1 result is returned. The other 2 expire in perhaps 30 days, then 2 more are issued with still a 30 day turn around time. Do you want want to wait up to 60 days for your work to be granted credit? This was the reason behind the extra work being sent out!

[edit]They, the extra work units, are NOT wasted work as many believe![/edit]

---

Boinc....Boinc....Boinc....Boinc....

Perhaps you should consider the situation where only 1 result is returned. The others expire in perhaps 30 days, then 2 more are issued with still a 30 day turn around time. Do you want want to wait up to 60 days for you work to be granted credit?

I have, and is why I moved my slugs over to EAH exclusively.

I couldn't care less if a result stays pending until the cows come home, as long as I am reasonably sure that they actually were doing something useful. It makes absolutely no difference in the grand scheme of things if it takes 30 seconds or thirty years to get credit granted for a WU since it's not like the utility company will accept Cobblestones for payment of power used, even for something as worthy as science experiments. ;-)

Alinator

I couldn't care less if a result stays pending until the cows come home, as long as I am reasonably sure that they actually were doing something useful. It makes absolutely no difference in the grand scheme of things if it takes 30 seconds or thirty years to get credit granted

I'm in total agreement.

---

Join TeamACC

Sometimes I think we are alone in the universe, and sometimes I think we are not. In either case the idea is quite staggering.

Whereas you can say from the Projects POV it's more power efficient in that they are getting more WU's done per unit time, this fails to take into account they aren't paying for the electricity to do the actual calculations. You simply cannot get around the fact that from the hosts POV, one third of the results are now going to be doing no science rather than one quarter.

With all due respect, there's one major flaw with this logic. You are assuming there are infinite workunits to process. This number is finite. Let's call it N. So with four-fold redundancy there will be 4N results. With three-fold redundancy there will be 3N results.

As of last week, 25% of 4N results were "extraneous" (i.e. above min quorum). So N results "add nothing" to the science.

As of now, 33% of 3N results are "extraneous" (i.e. above min quorum). So once again N results "add nothing" to the science.

So nothing really has changed except that the world is doing 25% less work to process a workunit, which vastly reduces user electricity expenditure which I think we can all be happy about.

Even better will be when target results == minimum quorum == 2, in which case 0N results will be "extraneous." We'll aim for that once we prove our servers can handle it.

- Matt

---

-- BOINC/SETI@home network/web/science/development person
-- "Any idiot can have a good idea. What is hard is to do it." - Jeanne-Claude

LOL, agreed, the givens should have been stated:

1.) SAH is not in immediate danger of going POOF (hopefully).

2.) The source of raw data to crunch is not going to go POOF (hopefully).

Granted these could be dangerous assumptions. ;-)

Also, I should have said that there are other concerns like the effect of allowing 4x CC's to participate, as well as covering possible problems due to the app itself being open source (ala the Coop 2 app bug). Both of these make a pretty good argument for a 3/3 scenario, but that has to be your call since you have the best look at the overall big picture.

:-)

Alinator

Even better will be when target results == minimum quorum == 2, in which case 0N results will be "extraneous." We'll aim for that once we prove our servers can handle it.

- Matt

Matt,

What would happen if target results == minimum quorum == 3?

Thanks -- Ned

---

Sounds all well good staged steps to see what increase in WU processing can be made and to ensure that the servers don't go "POOF" from WU returns overload... I guess s@h is still the leader in all this for the absolute numbers?...

Is the target to go for 2/2 for the quorum?


Matt, you've certainly got more patience than me, both for the sys admin and the PR!

PhonAcq, don't forget that Pappa has (voluntarily) worked hard to try to help for s@h...


Regards to All,
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Is the target to go for 2/2 for the quorum?

Yes, but probably not until we have multibeam workunits (or astropulse workunits) to send out as well. 3/2 is a good "bridge" until then.

Matt, you've certainly got more patience than me, both for the sys admin and the PR!

I wish I had all the time in the world to address all the questions. One thing is for certain: If I'm writing on the message boards, I'm procrastinating. All least it's productive procrastination.

- Matt

---

-- BOINC/SETI@home network/web/science/development person
-- "Any idiot can have a good idea. What is hard is to do it." - Jeanne-Claude

There is one thing not mentioned here. The fairness of credits granted.

With so many still using older 4.x BOINC this is going to create a problem imo.

Someone with bad luck and fast computer might end up being second to report and get low credit due to a 4.x user claiming low. Basically a worst case scenario is a low clock P4 beating a C2D in RAC due to pure luck.

I know this is a science project, but I do belive a lot of people do this for the competition aswell. Lowering the quorum is ok, but make sure its fair by not accepting results from 4.x users. As it is now you might end up getting less done if people move to other projects.

It was brought up, but maybe a better solution to not accepting results from 4x clients at all would be to not use the claimed credit from them if a 5x result is available in the quorum.

Alinator

You really have to hand it to the folks at SETI on the whole. The volume of data that they keep track of must be truly enormous. I don't think people really fully grasp that. So kudos to the keepers of Seti@Home.

I'm kind of glad for the 2/3 change. I have been keeping track of the fellow computers that I work on WUs with. I am up to over 2000 in just two months. That was tracking 3 other computers per WU. Now I will only have to track 2 others per WU.

Alinator, In a previous post you said that there was 33% waste with the 3/2 and 25% with the 4/3. Here's your quote

However, under "ideal" conditions the efficiency from a power consumption POV is worse, since now 33% of the total results sent out will not be contributing to the science instead of 25%. Remember once a WU validates and the canonical is chosen every other result returned after that is fluff.

Since only two "strongly" similar results are needed to validate for both, shouldn't that be:

33% for 3/2, and 50% for 4/3 (instead of 25)?

I mean 1 for the 3/2 isn't always needed and 2 of the 4/3 aren't always needed.

There is one thing not mentioned here. The fairness of credits granted.

With so many still using older 4.x BOINC this is going to create a problem imo.

Someone with bad luck and fast computer might end up being second to report and get low credit due to a 4.x user claiming low...

All you can do is to maximise your chances of getting in first by running a minimum sized cache.

A small cache should help all round for keeping the credits moving swiftly.

Happy crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Alinator, In a previous post you said that there was 33% waste with the 3/2 and 25% with the 4/3. Since only two "strongly" similar results are needed to validate for both, shouldn't that be:

33% for 3/2, and 50% for 4/3 (instead of 25)?

I mean 1 for the 3/2 isn't always needed and 2 of the 4/3 aren't always needed.

[edit] Oooooooops... I should actually read what numbers are written! [/edit]

As described by Matt, the flawed logic is in ignoring the absolute numbers.

As you describe above, there's less 'wastage' overall in that fewer WUs are getting validated overall, and yet we get the same usefulness of science return.

I still suspect that there will be a little extra database overhead for weeding out the 1% or less bad results that manage to sneak in...


Happy crunchin',
Martin

[edit] Sense changed to 'agreed' after actually reading it! [/edit]

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Alinator, In a previous post you said that there was 33% waste with the 3/2 and 25% with the 4/3. Since only two "strongly" similar results are needed to validate for both, shouldn't that be:

33% for 3/2, and 50% for 4/3 (instead of 25)?

I mean 1 for the 3/2 isn't always needed and 2 of the 4/3 aren't always needed.

That's true, at 4/3 it is possible to validate on two, but three still have to be returned before the result is submitted to the validator for consideration, and I have seen some where this has happened. IOW, three went to validation, one failed, the fourth didn't get returned, but the WU completed overall on the 2 strongly similar with no reissue.

Alinator

Ofcourse, we could take the accountants view and say that 3/2 uses 50% more than needed and 4/3 uses 100% more than needed. LOL

I guess that would be the "cup's half empty way" of looking at it. ;-)

In any event, I've been tracking this in my personal host database, so I'll leave my settings alone and see what happens now at 3/2.

I don't think it's so much a matter of absolute numbers or the constraints you put on the problem as much as it is one of your frame of reference. IOW, whether your talking about the project as whole, or from the viewpoint of any given host.

Alinator