[QUOTE=chalsall;307154]In this case, we had four different people doing the work needed -- Jerry (TF), James (P-1), LaurV (LL) and Dubslow (DC). All knew the work being done was concurrent.[/QUOTE]
I was perfectly happy when everybody had an equal chance to get the small plum assignments (directly from GIMPS), even if only "preferred" users got the chance. With a "bot" that reserves plum assignments for selected users, there is essentially no chance for a non GPU272 user to catch one of these. [How are you going to explain it to the 10,000 non GPU272 users if you happen to find a Mersenne Prime? IMHO, your team deserves neither the prize nor the fame, yet the curious public and press will ask "who found it". As I said before you have, IMHO, "poached" the plum assignment from the pool that can be obtained by regular (even preferred) users.

There is a second issue that must be pointed out in the name of good science. Theoretically, the DC should be carried out by a completely independent entity than the original LL test. Different user, different team, no connection between the two, whatsoever. Why does George mask the last 2 digits of the residue?

Please don't misunderstand me. I am NOT suggesting in any way that your users are in collusion with each other. But it isn't good science if they *could be* in collusion with each other.

Again, IMHO, any DC's that are done among any group that *could* be in collusion will need to be triple-checked before George can announce that M(x) has been "proven" to be the nth Mersenne Prime.

When viewed under the microscope of good science and good research, I'm concerned that any attempt to artificially "advance the wave" needs some deep scrutiny to ensure the validity of the GIMPS results is not threatened.

[Note that my concern about independent LL and DC is not exclusive to GPU272. Some users on these forums have asked each other to perform a DC, and I think the same principle applies. Ideally, there should have never been any communication or coordination between the user who performs the LL and the user who performs the DC.]

[QUOTE=rcv;307404]
There is a second issue that must be pointed out in the name of good science. Theoretically, the DC should be carried out by a completely independent entity than the original LL test. Different user, different team, no connection between the two, whatsoever. Why does George mask the last 2 digits of the residue?

Please don't misunderstand me. I am NOT suggesting in any way that your users are in collusion with each other. But it isn't good science if they *could be* in collusion with each other.

Again, IMHO, any DC's that are done among any group that *could* be in collusion will need to be triple-checked before George can announce that M(x) has been "proven" to be the nth Mersenne Prime.

When viewed under the microscope of good science and good reserach, I'm concerned that any attempt to artificially "advance the wave" needs some deep scrutiny to ensure the validity of the GIMPS results is not threatened.

[Note that my concern about independent LL and DC is not exclusive to GPU272. Some users on these forums have asked each other to perform a DC, and I think the same principle applies. Ideally, there should have never been any communication or coordination between the user who performs the LL and the user who performs the DC.][/QUOTE]
If you think they should be triple checked, feel free to triple check them.

There is also more than one instance here of our residues disagreeing, so I think anybody who did investigate in the name of science would find that no collusion has taken place.

[SIZE="1"][QUOTE=rcv;307404]I was perfectly happy when everybody had an equal chance to get the small plum assignments (directly from GIMPS), even if only "preferred" users got the chance. With a "bot" that reserves plum assignments for selected users, there is essentially no chance for a non GPU272 user to catch one of these. [How are you going to explain it to the 10,000 non GPU272 users if you happen to find a Mersenne Prime? IMHO, your team deserves neither the prize nor the fame, yet the curious public and press will ask "who found it". As I said before you have, IMHO, "poached" the plum assignment from the pool that can be obtained by regular (even preferred) users.

There is a second issue that must be pointed out in the name of good science. Theoretically, the DC should be carried out by a completely independent entity than the original LL test. Different user, different team, no connection between the two, whatsoever. Why does George mask the last 2 digits of the residue?

Please don't misunderstand me. I am NOT suggesting in any way that your users are in collusion with each other. But it isn't good science if they *could be* in collusion with each other.

Again, IMHO, any DC's that are done among any group that *could* be in collusion will need to be triple-checked before George can announce that M(x) has been "proven" to be the nth Mersenne Prime.

When viewed under the microscope of good science and good research, I'm concerned that any attempt to artificially "advance the wave" needs some deep scrutiny to ensure the validity of the GIMPS results is not threatened.

[Note that my concern about independent LL and DC is not exclusive to GPU272. Some users on these forums have asked each other to perform a DC, and I think the same principle applies. Ideally, there should have never been any communication or coordination between the user who performs the LL and the user who performs the DC.][/QUOTE]

[/SIZE]

I would like to point out that [B]anyone[/B] is allowed and capable of writing a spider that can grab assignments from PrimeNet.

And, what's more, anyone is free to join GPU272 and reap the fruits of Chalsall's labor that will allow them to grab plum assignments too! Unless there is some compelling reason to avoid using GPU272?

[QUOTE=rcv;307404]There is a second issue that must be pointed out in the name of good science.
<snip>
Ideally, there should have never been any communication or coordination between the user who performs the LL and the user who performs the DC.][/QUOTE]

Utter BS. *Any* two users could be in collusion. In fact, any two ids *could* be the same person. There is absolutely nothing in the GIMPS system that can detect or prevent collusion. This has nothing to do with "good science".

Objectively, isn't it better to have GPU72 get the "plum" assignments and let extremely motivated users manually get the assignments and (most likely) get them done very quickly rather than allowing people to randomly pickup and/or grab and hoard them for extended periods of time with little to no progress?

[QUOTE=sonjohan;307214]I get it that poaching is not supposed to happen, but how can we advance the milestones faster?[/QUOTE]Keep in mind that whenever someone poaches an assignment in order to "speed up the next milestone", that person is shirking his/her contribution to work on the milestones following the next one. Thus a poacher will have slowed down progress on all milestones after the next one by the amount of time spent needlessly duplicating someone else's assignment.

Poaching NEVER speeds up the progress of GIMPS.

GIMPS's progress is not measured by the next milestone, but by all future milestones collectively.

Any apparent "speeding-up" of the nearest milestone is accomplished only by taking effort away from working on milestones that follow that one. Poaching is a near-sighted impatient mistake that slows down GIMPS.

- -

The very fastest way to advance milestones is for each person to work on his own assignments, with no duplication of effort, and with no uninformed judgments about whether someone else is "slow".

People who focus obsessively on only the "next" milestone need to be reminded that GIMPS progress is linked to [U]all[/U] the future milestones, not just the next one.

You really miss from this equation the guys who think like "i do poaching or I do nothing". :razz: (with or without "just to piss you off").

Re: concentrating on the next milestone only, this is not bad either, it depends of what the milestone is. My first milestone was to get in top100 for all worktypes, except firts-time-LL (which was not achievable in the near future at that time). I concentrated on it, and now I fulfill it. The next milestone was to take chalsall and Sid&Andy down from the gpu272 DC top. Fulfilled. Etc. Next milestone may be to piss off cheesehead or rcv... :razz:

[QUOTE=LaurV;307440]You really miss from this equation the guys who think like "i do poaching or I do nothing". :razz: (with or without "just to piss you off"). [/quote]My intended audience was people who sincerely want to help GIMPS, to explain to them how the common poacher's rationalization fails.

People who don't want to help GIMPS will do what they want, regardless.

[QUOTE=rcv;307404][...]There is a second issue that must be pointed out in the name of good science. [...] Ideally, there should have never been any communication or coordination between the user who performs the LL and the user who performs the DC.][/QUOTE]

[QUOTE=axn;307429]Utter BS. *Any* two users could be in collusion. In fact, any two ids *could* be the same person. There is absolutely nothing in the GIMPS system that can detect or prevent collusion. This has nothing to do with "good science".[/QUOTE]
Any two users of PrimeNet could indeed be in collusion, so GPS272 has not introduced any new risk of fraud. However I do think rcv's point is worth highlighting: doing the first-time test and the double-check as a close-knit team of two is a bad idea. I have no idea if it actually occurs amongst the high-throughput community of GPS272, but if it does I would like to see it strongly discouraged by the administrator.

I don't believe any of the group would ever tamper with the results. But if there was ever a question seriously asked about the integrity of the double-checked residues in the database, the whole project should be able to assert that users do not generally have access to the full residue until there have been two matching ones.

There are PLENTY of exponents LL-ed and DC-ed by the same USER. No team, or different colluding guys. I have few too, but check "big guys" and see...

Before CudaLucas era, this wasn't a problem at all: One guy so clever to cheat P95's assignment checksum keys (which include the S0 shifting amount) would not demean himself to fake a DC report just for PrimeNet credit. "Childish" (immature) users which would want to cheat the system were generally not able to do so. Normal user can't cheat P95 reports, as two reports with the same shift won't be accepted. You can't fake two reports with a different shift, as you need to know the shift and generate the return key. People who know how to do that have always better things to do.

We can assume that LL+DC dual reports from the same user, done both with P95 are not fakes. Two reports from P95 can't be easily faked even by a "pro". I pretend to be a "pro" :razz:. I don't know how to generate those things, and I am not interested to know.

For now :razz:.

After CudaLucas era, the problem is that someone can report CL results without that We1 return keys. Theoretically he can get the same assignment two times, as LL and DC, but do it only once [U]with P95[/U], and report it two times, first report only the residue, using CL report form (the line can be written with any text editor), second report with P95. This way he will get credit two times and the database will contain a fake result, assuming the P95 result was wrong. If the P95 result was right, then no harm done, he actually saved us some future work to double-check that exponent, hehe.

Remark that one can not do this without at least a P95 run. You can't clear an expo without any return key, and only P95 generates them.

Viceversa is impossible too: a second CL report (like for both LL and DC done with CL) won't be accepted because of "no shift" implemented in CL (therefore an error in the FFT will generate always the same erroneous result), neither a CL report from the same user after a P95 report by the same user. That is why the test can't be done (fast) with CL only. At least ONE P95 run is need. But a P95 report after a CL report is usually accepted, even if it comes from the same user ([URL="http://www.mersenne.org/report_exponent/?exp_lo=40657343&exp_hi=&B1=Get+status"]see here[/URL]**). So, if you have a P95 run and want to cheat, you theoretically can do it, either using an accomplice (passing him the result and he reports it like his CL test) or doing it by yourself, reporting the residue with CL first, then with P95 (for this, P95 has to be set to "do not automatically contact the server).

[note: ** the mentioned exponent is not a cheat, in fact is was already triple checked, the CL test was run in two different cards in parallel (and even two different FFT sizes were used!) and it gave the same result, then the P95 run was done later. I invite everyone to verify it. I gave the example not to show how you can cheat, but to see that is possible to do reports like that. One user which does this frequently would be very easy to spot. Also, users which work "in pair" or "in team" would be easy to spot. In short, I don't really see a problem with this, but I would welcome any time some "security increase" in CL reports, I even would invest more computing power into GPU-LL (CL) in that case...]

[edit: in the past I tried to bring this up, related to mfaktc and CL reports - they are easy to fake - but some people around here told me I am paranoid and I am overreacting :smile:]