First time LL is no more

[Prime95]

The server no longer hands out first-time LL assignments. I've noted the news on the main page as well as the Manual Assignments page. Requests are converted to double-check assignments.

Let me know if I broke anything.

[chalsall]

Originally Posted by Prime95

The server no longer hands out first-time LL assignments.

Thanks, George!!!

Although this means the LL'ers are going to lose the occasional P-1 run done before most Cat 3 and 4 assignments are abandoned. No biggie; the P-1s are usually Stage 1 only...

[ATH]

Lucas-Lehmer tests have served us well for over 24 years. They can rest now, but will be hopefully be called upon again soon for the honor of verifying the next Mersenne Prime.

[kriesel]

Next in line:

1. when to cease awarding computing credit for LL first test results
2. when will an mprime/prime95 version supporting Primenet API issued PRP/GEC/proof as DC of unfortunate LL first tests be available? (I am phasing out the less efficient LL DC.)
3. when will LL DC be no longer issued as assignments
4. when will LL DC results no longer receive computing credit

With due consideration for how long end user transitions take, of course.
Systems that haven't the OS compatible with current or near future prime95/mprime releases, probably are too old to be effective at first primality testing or DC wavefront, anyway and ought be switched to P-1 factoring, or retired as primarily space heaters.
It's been over a quarter century (early 1996 to early 2021). Time to move on from routine LL, to only PRP tests in production runs. We'll still need some code capable of redundantly performing LL tests to confirm any reported PRP is in fact a newly discovered Mersenne prime. At exponent ~100M, that's a <1ppm probability per prime exponent of a first LL test being useful. https://www.mersenneforum.org/showpo...4&postcount=16

[M344587487]

This is a big deal, there should be cake.

[Uncwilly]

Originally Posted by kriesel

1. when to cease awarding computing credit for LL first test results
2. when will an mprime/prime95 version supporting Primenet API issued PRP/GEC/proof as DC of unfortunate LL first tests be available? (I am phasing out the less efficient LL DC.)
3. when will LL DC be no longer issued as assignments
4. when will LL DC results no longer receive computing credit

1 and 4 should be never. Never stop giving credit for work done. George has resisted pressure to stop giving credit for CPU done TF.
3 is when the last LL's need need DC.
2 should be 'no time soon' in my estimation.

[kriesel]

My point #2 was consistent with your own guidance posted earlier.

Originally Posted by Uncwilly

This is the location for the DC list:

Any exponent that has had a single First Time LL should no longer have an LL DC. It should get a PRP on a machine running Prime95 v30+ or GpuOWL v 6.11-318+ that produces VDF (proof files). The extra effort to run the certification from the proof files vs the need for TC tips the balance in favour of the PRP (with the better error checking code.)

Since LLDC is an inferior check compared to PRP/GEC/proof, it should be phased out. So should LL first test. The error rate is too high and increases with exponent. (~20.% at 100Mdigit) https://www.mersenneforum.org/showpo...6&postcount=14
The utility of a new LL first test is almost always less than zero. It uses time and resources at least the equal of a PRP/proof run. But the efficient way of verifying an exponent is composite is not LL & LLDC or LL & then PRP/proof, it is PRP/proof from the start. An LL first test has a 999,999.5+ ppm chance of no benefit. We don't give computing credit for repeated TF of the same bit level work done either, because it provides no benefit.

[Uncwilly]

Originally Posted by kriesel

My point #2 was consistent with your own guidance posted earlier.

That was specifically for the strategic DC's, not run of the mill DC's. The Strategic DC's were more error prone.

[Chuck]

Do you think people will ask for PRP tests and then edit the worktodo changing them to LL?

[kriesel]

Run of the mill LL wavefront first tests had 2% error rate per test before the addition of the Jacobi check to some of the testing software. CUDALucas still lacks it. So average LL error rate will still be ~1% or more at the wavefront. (Or ~20.% error rate at 100Mdigit, or ~90% error rate at 1G without Jacobi check or dual runs with frequent manual interim residue comparisons.) The current DC wavefront is from several years before the introduction of the Jacobi check, so the 2% error probability applies.

That 1-2% is high enough that it is more efficient to PRP/GEC/proof/CERT, at ~1.01 effort, than to LLDC at 1.01-1.02 effort and lower reliability. The higher the exponent, the higher the error rate in LL, and the greater the savings by PRP/GEC/proof as DC.

It would be good if the most popular client software supported a user's desire to use the more reliable and efficient algorithm with PrimeNet API assignment, like we can with manual assignments.
George, please update prime95 to allow the end user to specify automatic PRP/GEC/Proof in place of LLDC on exponents that had an LL first test.

[LaurV]

The truth is somewhere in the middle. No need to kill each-other for it. You are both right, each in his own way. LL credits will still have to be given, otherwise some people will stop participating. Most people will not care and many won't know the difference either. But 101 miles per hour are always better than 100 miles per hours, and credits are "free", I mean, George doesn't buy them from a bank or so. There will be always some guys trying to take advantage of the system, but that shouldn't impede the "honest" guys to work toward their goals. Also, in the past people used LLDC work to keep their computers "in check" (in fact, if you look in the Lifetime Top, some of the largest LLDC contributors are the largest because of that! Chris? Anybody?). Now, this "opportunity" is lost, and CERTs are few, never enough for continuous work (and they also don't stress the hardware so much, due to often interruptions). Also, some big contributors (like CurtisC) have an army of computers lost around the campuses or factories, which won't switch the work type (or upgrade to a PRP+CERT-aware P95 version) soon.

Originally Posted by Chuck

Do you think people will ask for PRP tests and then edit the worktodo changing them to LL?

Some will do. I see one every morning in the mirror
Of course, motivations may vary.