I was just about to retract what I had said when I read your post. I have to agree with you - Linux and ppc boxes claim much less on the whole than wintels do.

I'm going to have some breakfast - A full stomach has a tendency to make people think more effectively :-)

---

@MajorKong

Nicely said (and I especially agree with your point regarding needing 3 results)!

I would add one thing though...technically, validation is not part of the credit system, it is the new scientific methodology being employed (and the credit system is necessarily derived from this method). For example, no credit at all could be given to anyone, but validation would continue since it is necessary to verify the results to avoid various types of error (i.e., the reason you stated regarding the need for three intial hosts per unit). Indeed, validation could be added to the SETI classic system without changing the actual credit awarded other than providing no credit for invalid results.

@all

It appears that the seeming unfairness of the credit system in BOINC is what drives most of the complaints, and this unfairness appears to come primarily from two sources: inconsistent credit across platforms (espcially *nix vs. WIN boxes as well as the processor variations with fast intel chips being lower credit claimers compared to slower chips, AMD's, and especially celeron cpu's) and receiving no credit for work that was completed but declared invalid despite not doing anything that would directly cause such invalidation (e.g., outright cheating, pushing the overclock too far, etc.). I believe that the solution is in the works for much of the first issue in the upcoming 4.5x version of BOINC (especially the *nix vs. WIN issue).

As for the second issue, it might be wise to impose a base credit for all completed work units. Consider, for example, two individuals who volunteer at a suicide hotline. Both come in on separate days and work the same number of hours. Person A handles 6 calls successfully on his night while Person B didn't receive any calls (or maybe recieved 6 calls but 3 were not successful). If they were to receive credit for their work based on the current SETI BOINC model (and yes, such places do provide certificates of appreciation and some even certifcates of time volunteered to satisfy the needs of social work students, etc.), then Person B would receive no credit at all (or no credit for the 3 unsuccessful calls). However, both Person A and Person B donated the same amount of work. In such a case, Person B might not continue contributing for long (or even worse, might complain to other potential volunteers resulting in fewer of them signing up).

While providing full credit (as in SETI classic) would be absurd, it might prove useful to provide a very small base credit for that invalid work (null results are still worthwhile to the scientific endeavor). For example, a .1 credit might be applied as a token of appreciation for work completed which might reduce the dissatisfaction with this part of the credit system. Such a low credit would also be of little use to cheaters since one would need to return more than 200 invalid units to get comparable credit for one valid unit (and a minimum cpu time could be used to better insure this).

Well, just my two cents...

> Sure, but check out the wus, you'll see that there are no such big differences
> (at most around 40 credits). But look at that. My laptop claimed 70 credits,
> the other 2 were linux boxes and claimed 23 and 24 credits. My granted credit
> was 46 credits lower than claimed, even though I'm not a cheater. If the
> average was calculated, I'd get around 40 credits, much more acceptable
> amount.
>

At most 40? Are you serious?

http://setiweb.ssl.berkeley.edu/workunit.php?wuid=5889457 (85.05)

http://setiweb.ssl.berkeley.edu/workunit.php?wuid=6583382 (52.48)

http://setiweb.ssl.berkeley.edu/workunit.php?wuid=6583369 (65.78)

http://setiweb.ssl.berkeley.edu/workunit.php?wuid=6429199 (59.86)

That's a bit more than 40.

You better go back and look again.

http://setiweb.ssl.berkeley.edu/results.php?hostid=396878&offset=20

I should also point out that in your example above you even stated that your granted credit was 46 lower! And notice that MANY of these hosts are completing the work twice as fast as your laptop!

I also noticed a couple days ago your measured integer speed on the laptop was well over 4000. That seems a bit inflated. Right now it is sitting at 3285.75. I see plenty of hosts completing the same work units in half the time with lower bench scores. Something seems wrong with that.

---

<img src="http://boinc.mundayweb.com/seti2/stats.php?userID=2259&amp;team=off">

> I respectfully disagree with your statement that two is the minimum required
> 'for the science'. In the case of a disagreement in the results, how can you
> tell who is more correct?

Two is the minimum required by the science, not the maximum. There is no maximum -- the requirement is that two results match, and there is no (theoretical) limit on the number of non-matching results that could occur.

I think that a work unit should be sent to two hosts. If the two results agree (which should occur in the vast majority of cases) the work unit is finished, with the minimum number of calculations (two). If the two do not agree, it gets sent out to a third host. When that result comes in, it is compared to the previous results. If there is a match, the two matching units get credit and the work unit is finished. If not, the process is repeated until either a match occurs or some maximum number of hosts have failed to produce a match.

> Also, I do agree that initially sending out 4 copies is a waste. 3 is the
> number of initial copies I would send out.

I don't see any point in sending out three copies initially. It may decrease the expected time until a match is found (or it may increase it -- depends on the distribution of turnaround times and the probability of needing a third result), but the data do not have any time value. It really makes no difference if a unit is validated this week or three months from now.

> But, in any case, 4 is WAY fewer
> than S@H-Classic sends out. Analysis shows that Classic sent out each work
> unit AT LEAST 7 or 8 times, and I have heard Dr. Anderson say (In a video of a
> speach he gave) that it was 'at least a dozen times'. So, S@H/BOINC is 2 to 3
> times more efficient than Classic.

Yes, this duplication has been commented about, back in the classic forums. It is acknowledged that duplicate work units were being sent out for no purpose other than to keep the machines busy.

With SETI classic this makes some sense. SETI classic did not coexist very well with any other distributed processing projects, so they could make the assumption that the choice was between keeping the machines busy doing useless work and letting them go idle. That is not the expected model for BOINC, where the expectation is that there will be a number of projects available, and the total demand for processing power over all projects will far outstrip the supply of processing power. Busy work by one project is therefore stealing resources from other projects.

> I am so anti-cheater because of the harm that cheating can do to the
> science, not because some cheating loser got their nickname up on top of a
> leader-board.

Yes, the project needs to protect the data against cheaters. But I think that the requirement for two matches largely does that. If somebody cheats by turning in bogus results, they are unlikely to match (they will match only if somebody else is generating exactly the same bogus results, and they are paired) and the science is unaffected. Indeed, the project can keep track of how often a host returns a result that fails to match, and any host that returns a high percentage of such results could be flagged as suspect and either cut off from the project or require a match to two other results for credit.

But somebody who cheats merely by claiming excessive credit makes no difference to the science -- only to the user statistics.

>
> The credits system for BOINC projects was designed the way it was for two
> reasons. First, the old 'wu count' had to be dropped in favor of a numerical
> score. Different type work units on the same project, as well as different
> BOINC projects entirely, are not going to have anywhere near the same CPU
> requirements, and a 'count' would not be fair to people that get a bunch of
> 'long' ones. The 'credits' score makes it possible to combine work from
> multiple work unit types and even multiple BOINC projects into ONE total,
> overall score. See my friend Toby's stats site if you are curious about
> combined project stats. http://stats.kwsn.net Check the last line in the
> table.

I agree that a credit system based on the actual work needed for each unit, rather than a simple unit count, is desirable. However I don't think that the current system does a reasonable job of that. In fact, I don't think that any system that attempts to base the credit on measured CPU time is going to work well.

The problem is that the model assumes that for each machine there is some number (the "benchmark") that one can multiply by the CPU time and get a reasonably accurate measure of the actual work done -- a measure that is consistent across different machines and across replications on the same machine. I maintain that with current machines there simply is no such number. The actual production rate of a machine varies so much based on the environment (HT machines will vary dramatically depending on whether the other thread is working and whether it is competing for the same execution units, laptops will change their speed depending on the power source, machines may change their speed depending on the ambient temperature, multiprocessors will be strongly affected by the amount of contention for memory access, etc.) that any benchmark is at best a gross estimate. Good enough for planning purposes (how many units to download into the queue), but not for an equitable credit system.

The only way you are going to see consistent and reproduceable credit is to make the credit calculations a part of the application. Count the number of times through the major loops, and apply work factors that have been determined by the application writers. Many functions, like an FFT or an integrator step, are very standard and require a fixed number of operations (amount of work) regardless of the data. And these calculations do not have to be very accurate to be more accurate than the current benchmark results. And regardless of their absolute accuracy, they will be consistent -- indeed, the loop counts could be treated as part of the returned results and subject to the consistency checking.

---

------- Tom Gutman

Please excuse my asking this, but what's all this about 'claimed' credit and actual credit?

I receive work units - I crunch 'em - I return 'em! I don't 'claim' credit! In fact I'm a bit behind on credits.........

http://setiweb.ssl.berkeley.edu/results.php?userid=7819296

---

> Please excuse my asking this, but what's all this about 'claimed' credit and
> actual credit?
>
> I receive work units - I crunch 'em - I return 'em! I don't 'claim' credit! In
> fact I'm a bit behind on credits.........
>
> http://setiweb.ssl.berkeley.edu/results.php?userid=7819296
>

It's even worse than that, John. You're not getting work units - you're getting the results of somebody else's calculation on a work unit. All the crunching you're doing is on something they already know. Otherwise it wouldn't be a "result", it would be a "work unit". The "result" is the end of the computation, not the beginning. See Merriam-Webster OnLine which defines "result" as: "2 : something obtained by calculation or investigation". A "result" is the _output_, not the input.

In other words, they're not sending you a raw work unit and asking you to process it. It's clear that they already have that output, or they'd call it a "work unit", not a "result". Instead, they're sending you a "result" from somebody else's work and asking you to verify it.

The good news, of course, is that they are up-front about it and use the correct term, acknowledging that it is a result (output) and not a work unit (input).

Seems a big waste to me, but they're the experts. I'm just doing my little part to find ET. But I don't see the utility in verifying other results. I'd rather be working on raw data: work units.

---

--Bill

You raise a valid point. .o0(yes, pun intended) Can we differentiate between crunching work units and confirming valid results?

---

> It's even worse than that, John. You're not getting work units - you're
> getting the results of somebody else's calculation on a work unit. All the
> crunching you're doing is on something they already know. Otherwise it
> wouldn't be a "result", it would be a "work unit". The "result" is the end of
> the computation, not the beginning. See Merriam-Webster OnLine which defines
> "result" as: "2 : something obtained by calculation or investigation". A
> "result" is the _output_, not the input.
>
> In other words, they're not sending you a raw work unit and asking you to
> process it. It's clear that they already have that output, or they'd call it
> a "work unit", not a "result". Instead, they're sending you a "result" from
> somebody else's work and asking you to verify it.
>
> The good news, of course, it that they are up-front about it and use the
> correct term, acknowledging that it is a result (output) and not a work unit
> (input).
>
> Seems a big waste to me, but they're the experts. I'm just doing my little
> part to find ET. But I don't see the utility in verifying other results. I'd
> rather be working on raw data: work units.
>
>

Bill & Patsy Zahrt,

Hey, don't tease the newbies like this!

A 'result' (in BOINC lingo) is a copy of a work unit that is issued to one particular host. Perhaps a short description of the process is in order.

First, the raw data from the tapes are 'split' (by a splitter) into work units. When the BOINC servers prepare to issue the work unit, it is copied into 4 (until recently it was 3) 'results', which are then placed into the 'ready for downloading' queue. Yes, I know the term 'result' can be confusing, but it is the term that Berkeley chose to apply to these copies. A 'result', in the BOINC context refers to a copy of a work unit before it is sent out and processed, while it is being processed, as well as after it is completed and sent back.

Berkeley did not need to make this construction visible on S@H-Classic because the individual work units were not assigned to specific, individual computers (hosts), as is the case in BOINC. BOINC keeps track of exactly which computer a 'result' was issued to, and will only accept its return if it was processed on and returned from THAT exact computer. Berkeley implemented things this way in large part to prevent a type of cheating that S@H-Classic has been plagued with for the last several years (and is still going on... I have seen evidence that it is still being done in Classic as late as yesterday).

Now, it *IS* true that S@H/BOINC is processing a lot of the same work units that S@H-Classic is/has processed. This was done to verify that the new S@H/BOINC science application is producing output that is comparable to what the S@H-Classic science code produced. I don't know if it is still going on, but after S@H-Classic is turned off in the near future, obviously it will not be happening anymore.

So, John Hunt, relax... You are not wasteing your time here.

---

https://youtu.be/iY57ErBkFFE

#Texit

Don't blame me, I voted for Johnson(L) in 2016.

Truth is dangerous... especially when it challenges those in power.

> You raise a valid point. .o0(yes, pun intended) Can we differentiate between
> crunching work units and confirming valid results?
>

Simple, computing a result is done on your computer... Confirming a valid result is done on one (or more) of the Berkeley S@H/BOINC servers... ;)

>
> Hey, don't tease the newbies like this!
>
>
> Now, it *IS* true that S@H/BOINC is processing a lot of the same work units
> that S@H-Classic is/has processed. This was done to verify that the new
> S@H/BOINC science application is producing output that is comparable to what
> the S@H-Classic science code produced. I don't know if it is still going on,
> but after S@H-Classic is turned off in the near future, obviously it will not
> be happening anymore.
>
> So, John Hunt, relax... You are not wasteing your time here.
>
No problem ! If that's the way the 'egg-heads' want to do it, who am I to argue?

---

Much obliged! But just so I'm clear on this, BOINC's splitter calls "results" what we get as "workunits" - and our results are to tell the BOINC thingamajiggy whether or not a particular splitter-result/user-workunit is worthy of further processing.

Did I get that right?

---

> Much obliged! But just so I'm clear on this, BOINC's splitter calls "results"
> what we get as "workunits" - and our results are to tell the BOINC
> thingamajiggy whether or not a particular splitter-result/user-workunit is
> worthy of further processing.
>
> Did I get that right?
>
I'm still struggling with this too. The BOINC/SETI folks aren't grade school dropouts. They're part of The University of Berkeley, one of the premier universities in the world. And, on top of that, they're engineers - people known for precision, accuracy, and attention to detail. So it's fair to assume that they know the meaning of "result" and are using the term deliberately. I.e., they're not a group of people who would be that careless...

Which makes the term pretty confusing to me. When they send me a "result" for my computer to compute, that is an input, not an output. For the reasons above, I don't think that's a grammatical mistake on their part; I think it's intentional.

So, what do they really mean by this? It's confusing!

Thanks for indulging me... I hope this isn't a stupid question...

---

--Bill

> Which makes the term pretty confusing to me. When they send me a "result" for
> my computer to compute, that is an input, not an output. For the reasons
> above, I don't think that's a grammatical mistake on their part; I think it's
> intentional.
>
> So, what do they really mean by this? It's confusing!
>
> Thanks for indulging me... I hope this isn't a stupid question...
>
>

If you click on a "Work unit ID" link in your stats page, you'll see the results belonging to this WU.

The seti server creates those result records before sending a WU to you, and uses one of them to store the result you'll send back later.
So the result records already exist before you started crunching.

Regards Hans

---

> Please excuse my asking this, but what's all this about 'claimed' credit and
> actual credit?

John,

THere is a handy dandy glossary that you can use to look up terms. And beat up on the definer if there is not a term that you want a definition for...

If you look close at this message there is a clue in here somewhere ...

:)

---

> > Which makes the term pretty confusing to me. When they send me a
> "result" for
> > my computer to compute, that is an input, not an output. For the
> reasons
> > above, I don't think that's a grammatical mistake on their part; I think
> it's
> > intentional.
> >
> > So, what do they really mean by this? It's confusing!
> >
> > Thanks for indulging me... I hope this isn't a stupid question...
> >
> >
>
> If you click on a "Work unit ID" link in your stats page, you'll see the
> results belonging to this WU.
>
> The seti server creates those result records before sending a WU to you, and
> uses one of them to store the result you'll send back later.
> So the result records already exist before you started crunching.
>
> Regards Hans
>
>
Thanks, Hans! I get it now, and this makes a lot more sense. It's not other crunchers who do the initial analysis, but it's SETI themselves who do the initial analysis when the SETI server creates each result record before sending the WU out. That's why the result records already exist before we start crunching them (as you said) because SETI does the initial analysis themselves. Makes perfect sense now. Then all we're doing is validating the initial result that the SETI server created.

Thanks. I get it now!

---

--Bill

> So, what do they really mean by this? It's confusing!

Being an engineers/scientist can mean that the words used are slightly different in meaning than expected. One of the problems we have is that there is no one enforcing a standard set of definitions and regulating word usage.

Case in point "result", in the documentation this word usually means the thing that the participant calculates and returns when the processing is completed. However, is is also used in reference to the information, currently incomplete, on the server during the issue process.

So, as a consequence, it appears from the documentation that a result is issued to you (by their explanations this does happen) along with the work unit which you then fill out and return.

There are other terms that are equally confusing and so far I have be singularly without success in getting significant changes made. As a former systems engineer I learned that termonology that comes from "techie-talk" is just confusing to the casual participant.

At any rate, I have definitions with cross references to the preferred terms in the glossary so you can look up the confusing one and find out what SHOULD have been used.

---

> Thanks, Hans! I get it now, and this makes a lot more sense. It's not other
> crunchers who do the initial analysis, but it's SETI themselves who do the
> initial analysis when the SETI server creates each result record before
> sending the WU out. That's why the result records already exist before we
> start crunching them (as you said) because SETI does the initial analysis
> themselves. Makes perfect sense now. Then all we're doing is validating the
> initial result that the SETI server created.
>
> Thanks. I get it now!
>

Erm, that's not how I understand it. I don't think SETI actually do any initial analysis as you said, Bill Z, they just create the 3 or 4 records that they're going to send out to us. We 3 (or 4!) do the analysis - after the 1st one is sent back, the other 2 or 3 results are compared against the first result to make sure they're valid.
The terminology's slightly confusing, but I think I understand the process behind it all.

Chris.

---

<img border="0" src="http://boinc.mundayweb.com/one/stats.php?userID=268&amp;prj=1&amp;trans=off" /><img border="0" src="http://boinc.mundayweb.com/one/stats.php?userID=268&amp;prj=4&amp;trans=off" />

Being an engineers/scientist can mean that the words used are slightly different in meaning than expected.
The most abused word is "vector". To engineers, nurses, programmers, climatologists, et al, "vector" means something completely different.

As for "engineers", a little cross-continental campus comedy…
♪Godiva was a lady who through Coventry did ride
To show the royal villagers her fine and pure white hide
The most observant man of all,
an engineer of course,
Was the only one who noticed that Godiva rode a horse♫

♪A maiden and an Engineer were sitting in the park
The Engineer was working on some research after dark
His scientific method was
a marvel to observe
While his right hand held the figures, his left hand traced the curves♫

♪A [three-syllable college, like NCC or MIT] computer nerd got drunk one fateful night,
he opened up a console and smashed everything in sight,
and after they subdued him,
the judge he stood before said,
"Lock him up for twenty years - he's rotten to the core!"♫

Happy new year, you engineers! :-D

---

I got my first work unit today which was sent to 4 hosts.

Here it is 7090428.

Let's see if the credits are going to come faster.

---

Red Bull Air Racing

Gas price by zip at Seti

> I got my first work unit today which was sent to 4 hosts.
>
> Here it is <a> href="http://setiweb.ssl.berkeley.edu/workunit.php?wuid=7090428">7090428[/url].
>
> Let's see if the credits are going to come faster.
>

How are you getting 4885 MIPS out of that computer? My 2.0 4M is only doing 3161 :(

---

<img src="http://boinc.mundayweb.com/seti2/stats.php?userID=2259&amp;team=off">