Require P-1 and other factoring work to be done on reliable machines

[tha]

I've been doing some extra factoring in the 12M and 10M range lately using mfaktc. I've found a few P-1 errors in the 10M range and many in the 12M range. This leaves me to believe one or some machines have done and may still do P-1 factoring that are highly unreliable. This causes others to have to do a whole lot more work which is completely unnecessary. Can and should we demand P-1 machines to do reliability tests that include heavy memory integrity checks?

[petrw1]

Originally Posted by tha

I've been doing some extra factoring in the 12M and 10M range lately using mfaktc. I've found a few P-1 errors in the 10M range and many in the 12M range. This leaves me to believe one or some machines have done and may still do P-1 factoring that are highly unreliable. This causes others to have to do a whole lot more work which is completely unnecessary. Can and should we demand P-1 machines to do reliability tests that include heavy memory integrity checks?

Can you elaborate "P-1 errors".Do you mean factors missed by P-1?

Or invalid factors submitted....I don't think PrimeNet allows this...that is they are all verified as valid factors.

[tha]

Originally Posted by petrw1

Can you elaborate "P-1 errors".Do you mean factors missed by P-1?

Correct.

Faulty dedicated P-1 clients miss a lot of factors which requires all P-1 work to have to be double checked or, as happens now, doing unnecessary LL tests.

[R.D. Silverman]

Originally Posted by tha

Correct.

Faulty dedicated P-1 clients miss a lot of factors which requires all P-1 work to have to be double checked or, as happens now, doing unnecessary LL tests.

Do we have a list of LL tests that could have been avoided if P-1 had worked properly?

[tha]

Originally Posted by R.D. Silverman

Do we have a list of LL tests that could have been avoided if P-1 had worked properly?

The data is very wobbly, and far from claimed to be complete or accurate, but a list of some of the P-1 runs that were found to be wrong can be found at http://www.mersenne.ca/p1missed.php?s=x&o=a

[Madpoo]

Originally Posted by R.D. Silverman

Do we have a list of LL tests that could have been avoided if P-1 had worked properly?

It's probably a facetious question, but I'll start out by noting that:

• 11,320 exponents have later been factored by P-1 after at least one LL test had been done
• 3,093 of those had some other P-1 work done where no factor was found.

I didn't go through the exercise of seeing which previous P-1 runs were actually done prior to the LL test, or whether or not it *should have* found the factor that was eventually found. So out of context like that, the #s above are kind of meaningless in terms of answering the question (whether sincere or not).

[R.D. Silverman]

Originally Posted by tha

The data is very wobbly, and far from claimed to be complete or accurate, but a list of some of the P-1 runs that were found to be wrong can be found at http://www.mersenne.ca/p1missed.php?s=x&o=a

Some of these "missed factors" show a gross misunderstanding of how P-1 works.

[chalsall]

Originally Posted by R.D. Silverman

Some of these "missed factors" show a gross misunderstanding of how P-1 works.

And the world ends....

[R.D. Silverman]

Originally Posted by Madpoo

It's probably a facetious question,

It was not.

[*]11,320 exponents have later been factored by P-1 after at least one LL test had been done

People have continued to look for factors even when factors were already known.
Are these instances where both TD and P-1 FAILED, prior to running LL, even though
P-1 later found a factor??


People also need to understand. For P-1 to succeed when (say) P-1 is divisible by a prime
to a degree higher than 1, the software must include the higher power in the smoothness
bound.

[Madpoo]

Originally Posted by R.D. Silverman

People also need to understand. For P-1 to succeed when (say) P-1 is divisible by a prime
to a degree higher than 1, the software must include the higher power in the smoothness
bound.

In truth it would be hard to say with any certainty whether some factors *should* have been found by a previous P-1 check, simply because not all clients give full information on the factor that it did find, like what B1/B2 it was using, and also because some older data is missing the "date received" so it wouldn't be possible to say an LL test came before/after a P-1 run.

I won't lose sleep over it those. Missing a factor that *should have* been found is inconvenient since it results in one or more LL checks (depending on if a factor is found later), but it's still just a mere inconvenience.

Still, if someone wants to go through and double-check previous P-1 runs by a computer that may have been flaky, more power to them. It has happened, where a machine doing factoring work of some kind simply missed things. Whether it was caught before or after an LL test is the part I'm not sure we'd always have an answer to.

I spot checked a handful of exponents where:

1. Some P-1 work was done, no factor found
2. An LL test was run
3. More P-1 work was done and found a factor

(in that order)

From my spot checks, everything looked okay, but I had to artificially limit myself to LL runs where the date was known. It was a pretty small subset of around 100 exponents, and for some a successful factoring job the first time around would have only saved a double-check since the first-time LL test had already been done.

My understanding of P-1 and probabilities of finding a factor are limited though so it may not have been obvious to me if the factor that was found would have actually been found by a previous check. My guess on the spot checks were "no" because the bounds were kind of small on the first run. But honestly I spent maybe 10 minutes checking and then moved on so take that for whatever it's worth.

[R.D. Silverman]

Originally Posted by Madpoo

I won't lose sleep over it those. Missing a factor that *should have* been found is inconvenient since it results in one or more LL checks (depending on if a factor is found later), but it's still just a mere inconvenience.

We are in strong agreement.