Poaching

[ewmayer]

Originally Posted by Deamiter

It is the great downfall of American science that credit is everything, and progress becomes less and less important.

...before WWII, scientists traded information freely looking for answers rather than credit.

And centuries before that, Newton wasted years of his life in a bitter
priority dispute with Leibniz over who should get credit for having
invented the modern calculus. And that is just one example of countless
many in the history of science. Trust me on this, the image of legions
of altruistic savants freely exchanging ideas without regard for fame
or acknowledgement is the aberrant one here. We'd often like it to be so,
and on rare occasions it actually is, but most of the time that's simply
not reflective of human nature.

In my mind one of the great things about the scientific enterprise is that
it functions as well as it does despite human nature being what it is.
To paraphrase a line from a popular movie, ultimately it really is "the
question that drives us," even if we *would* love the acclaim of being
acknowledged as the person who found the answer.

[cheesehead]

Originally Posted by Deamiter

You do not need credit!

The primary goal is to stop poaching, because of its deleterious effects. Credit is discussed in this context only because it is a means or effect, not the goal itself. Denial of credit for L-L contribution might be used as a means to discourage poaching. Among the multiple effects of poaching is interference with the orderly assignment of legitimate opportunities for credit both for effort contributed and for discovery of primes.

Though some participants don't care about credit, others are, quite legitimately, partly motivated by the rewarding of credit for contribution of effort or for the remote possibility of discovery of another Mersenne prime. GIMPS's contribution to the advancement of basic scientific knowledge would suffer if the latter group lost some of its motivation to participate, just as it would suffer if the former group lost some of its (different) motivation.

It is the great downfall of American science that credit is everything, and progress becomes less and less important.

Since the progress of GIMPS suffers as a result of poaching, poaching needs to be discouraged or prevented if possible.

How much more can we accomplish if we *gasp* work together for a scientific goal rather than working against each other in a race for attention?

Poaching interferes with the former, and is analogous to the latter (substitute "impatiently early achievement of near-future milestones" for "attention").

Should we value our human nature above what we might be able to learn through working together?

No, but disregarding human nature is a big mistake.

[Gary Edstrom]

Here's one way to handle poachers[link removed].


The link is offensive and in bad taste. I left it up for a little while but have got complaints since so I am removing it -- Mod

[pakaran]

Originally posted by Gary Edstrom
Here's one way to handle poachers[link removed].


The link is offensive and in bad taste. I left it up for a little while but have got complaints since so I am removing it -- Mod

Can someone message it to me?

[garo]

By the way, I wanted to make this clear but didn't. Gary, please do not take my removal of the link personally. It was a hard sell and I did leave it up there first. But then people complained so I took it off.

[xilman]

Originally posted by outlnder
I'm sorry Quint, but I would not waste the electricity on a machine that took more than 30 days on a first time LL test.

I am an unemployed pharmer and my slowest machine is a 1.1 GHz Duron.

Sometimes you just hafta let the old machines die.

Some of us live in climates where fan heaters are useful devices. My collection of old fan heaters keep my study warm and, as an added bonus, I get free computation as a byproduct.

Some of the older fan heaters are still doing sterling work for ECMNET, despite being in the high-end 486 to low-end classical Pentium class for computation.


Paul

[QuintLeo]

*chuckle* a lot of CURRENT machines take more than 30 days to do a first time LL test - on the 10,000,000+ digit LL tests my machines do (other than the ones that do factoring).

In the wintertime, in particular, it's atually cheaper to heat using the older machines electric usage than with natural gas - especially factoring in that the most heating needed is at night, when my Time Of Day rate makes the electric REAL cheap (appx 4.3 cents per KWH)....

Summertime, I may shut some of the older machines down - expecially the K5s and the 1'st gen Pentium and the C6. Wintertime, there's no reason to do so and lotsa reason to NOT do so.

Your milage may vary.


Now, if any of those ancient machines DO die, I won't waste $ (much less $$ or even $$$) getting them back up. If I have the part I need to fix them on hand, great - if not, the machine goes into permanent retirement and gets parted out.

[Uncwilly]

I know that this is an old thread, but it is the longest on poaching, so it should be the best place to discuss this.

With the new little tool that George has posted here we can peek behind the veil and see the progrss that others are making. With this: http://v5www.mersenne.org/report_factoring_effort/ we can see the bit level the the exponents are at.

Being able to see that a number is actively being LL'ed, but the TF and P-1 have been bypassed, has created a new issue for those that are not on the inside. A person with a GPU could run a number up to the goal bit depth quite quickly and save a 'costly' LL (and DC). Up in the 100M digit range, LL's are taking over 15 GHz years (IIRC).

There are 2 things I would like to bring up for consideration:

First: If a user has been assigned an exponent to LL, but has bypassed the TF and P-1 (as determined using the tools listed above), and is not expected to complete the number for quite some time: would it be ok for another user to do the TF and P-1 on the number concurrently??

Second: If that is ok, what about a way for PrimeNet to communicate to the client (next time the client checks in), "Ok, you can stop that number now, it has been factored, and here is a service credit for the time that you have spent. And here is a new assignment of your prefered work type."?


Cheesehead, George's new tool has changed things from before. We can now be sure that they have bypassed the TF, if they are reporting LL progress. What do you think about this? I would fully expect the server to reject any attempt to get the expo assigned to the TF'ing machine.

I realise that a new version of the client might be required, in order for it to be waved off. Also, PrimeNet would have to be changed.

I am not trying to pick a fight, but raise a new and thoughtful area of discussion.

[cheesehead]

Originally Posted by Uncwilly

First: If a user has been assigned an exponent to LL, but has bypassed the TF and P-1 (as determined using the tools listed above), and is not expected to complete the number for quite some time: would it be ok for another user to do the TF and P-1 on the number concurrently??

First reaction: no, but I need to think more about it.

Second: If that is ok, what about a way for PrimeNet to communicate to the client (next time the client checks in), "Ok, you can stop that number now, it has been factored, and here is a service credit for the time that you have spent. And here is a new assignment of your prefered work type."?

I can see how that scenario could be better for the LL assignee, and for GIMPS in general, than the present situation.

I think there would need to be a warning to the LL assignee that skipping TF/P-1 could result in the above scenario.

Cheesehead, George's new tool has changed things from before. We can now be sure that they have bypassed the TF, if they are reporting LL progress.

Can we be sure? How foolproof would that deduction really be (consider possible foulups in the assignment report)?

Also, consider possible idiosyncratic not-your-standard-order-of-processing scenarios that sufficiently creative participants might follow (e.g., by tweaking worktodo after assignments were added), which would eventually result in good and standard results, but which might not be fairly or accurately described/implied in the new assignments report while the assignee was in the middle of processing the assignments.

Consider a "Mad/Foolish/Misguided/Eccentric Assignee" who deliberately and knowingly performs LL first, but then also performs the skipped-earlier TF and P-1 as soon as the LL is finished. What is the fairest scenario for that person?

What do you think about this?

I am concerned about erroneous and/or unfair conclusions being drawn about assignees in unusual or abnormal situations, followed by unfair actions being taken by vigilantes.

I realize that such situations would most probably be only a small fraction of 1% of all assignments, but I'd like to see some safeguards against vigilante necktie parties for kooks like, or unlike, me.

I would fully expect the server to reject any attempt to get the expo assigned to the TF'ing machine.

??

You mean at present, or under the new plan? And could you be more specific and detailed about the scenario for this?

a new and thoughtful area of discussion.

Thank you for the thoughtful suggestion!

[ckdo]

Originally Posted by Uncwilly

Cheesehead, George's new tool has changed things from before. We can now be sure that they have bypassed the TF, if they are reporting LL progress.

You can not.

I have, in the past, completed various parts of assignments in parallel. For example 6 different bit levels of TF on 3 cores and P-1 on the fourth. PrimeNet would only list the P-1 progress.

I don't see why this shouldn't be extended to LL tests...

-- Carsten, thinking different (as usual).

[Uncwilly]

Originally Posted by ckdo

I have, in the past, completed various parts of assignments in parallel. For example 6 different bit levels of TF on 3 cores and P-1 on the fourth. PrimeNet would only list the P-1 progress.

I don't see why this shouldn't be extended to LL tests...

If the person has picked up a number and has reported in 30% LL complete, yet there are 4 bit levels and P-1 that have not been done, is that not clear? The bit levels in question might take 15 GHz days for the first 2. If the 30% LL reported in ~= 5 GHz years, would it not make tremendous sense to do the TF first?