As some of you may know, close to two years ago the server for the original PSP project went down. Until that time PSP and PrimeGrid had an agreement to cooperatively work on the Prime Sierpinski Problem together.

The original PSP project isn't coming back. We're going to continue the search to prove this conjecture on our own. Unfortunately, all the data from the original server appears to be unavailable.

In a conjecture project such as PSP, it's vitally important that no primes are missed since a missed prime would incur a huge penalty in extra work. We do not know which candidates were single checked on the original PSP, which were double checked, and which may have never been checked at all. We are therefore going to recreate all the missing work. Fortunately, the vast majority of the work was done on PrimeGrid, so it's not as bad as it seems. (And it's not anywhere close to as bad as the situation we face with SoB.)

Here's the work that will have to be tested in the double check. Most of the following will need two tests, although a small number of these may have an imported residue and will only need one additional check.

Will the double check be mixed with leading edge work, or will it all be double check work until the double check is done?
____________

1. I have residues up to 2M, are you interested?

Will the double check be mixed with leading edge work, or will it all be double check work until the double check is done?

1. I have residues up to 2M, are you interested?

# Being checked by PSP project, no info about doublecheck 79309 -1 # my: 2M 79817 -1 # my: 2M 152267 -1 # my: 2M 156511 -1 # my: 2M 168451 -1 # my: 2M 222113 -1 # my: 2M 225931 -1 # my: 2M 237019 -1 # my: 2M

2. Do we have confirmation, and if not, will we confirm that following known primes are optimal (minimal) ones:

_90527 9162167 # my: 2M _222361 2854840 # my: 2M _258317 5450519 # my: 2M _265711 4858008 # my: 2M

These 4 are biggest ones (other primes are small and could be easily confirmed optimal manually). It's not much work to do for full-scale PG project.

Are you looking for something like this? :)
50309803353016993 79817 32998439
50750214895543277 222113 48428381
50826961283402153 222113 19803693
50855700998451133 168451 34502256
50942205515241673 225931 41499752
or this?
193057765750527416730031 156511 6570336

The double check has begun! A few moments ago we loaded all the double check tasks into the system. Please note that we expect the server to send out approximately 400 leading edge tasks before it starts sending out the much smaller double check tasks. And, of course, even once the small tasks start getting sent out, there's always a chance of getting a resend of one of the leading edge tasks.


____________
My lucky number is 75898⁵²⁴²⁸⁸+1

I should warn that leading edge tasks are huge, they're second after SoB with average CPU time of 6 days. So if anybody planned to use some old slow PCs for small double check tasks, it's better wait until most leading edge tasks are purged out of the system. And, as Michael already noted, it still be a chance that you'll get resent leading edge task later.

Just put one system on to try it out... of 7 tasks downloaded, 6 estimate at 2 minutes, 1 estimates closer to 2 days. Edit: make that 3 days.

I completed some double check tasks this morning. I plan to complete a lot more tonight. Hopefully at least one missed prime gets turned up by the double check.
____________

Speaking of which... what about the 35% credit bonus from tasks being "big"?

I just want to note that residues provided by stream were fed into our system and for candidates with n < 2M only one task will be generated per workunit. The other (fake) task is automatically created using stream's residue. The workunit is then handled the same as any other one: If the residues match, the workunit is validated. If they don't match, more tasks will be generated until we get two matching residues.

There's a possibility that we have candidates where no residue was provided in which case two normal users tasks will be created. Stream's residues ended at n=2M, so all candidates above that will be handled normally with two user tasks created.

Speaking of which... what about the 35% credit bonus from tasks being "big"?

The three systems I put on this are bogged down with big units now, which I'll leave running.

In a quick look I see some selection of the smaller units going on, and the big ones are getting aborted.

In a quick look I see some selection of the smaller units going on, and the big ones are getting aborted.

As one would expect, we're making excellent progress through the very small tasks at the beginning of the double check.

The double check started at n=500K with candidates of about 150K digits. We're up to n=2M which is about 600K digits.

Here's a breakdown of the double check work by FFT size:

Almost 1 million tasks of SoB server were lost - 14 years of work.

It is interesting will PrimeGrid recalculate of those tasks?

Leading edge of double check passed 2M range.
It means two results are needed per work unit.

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

I think we've made good progress, but there is still tons of progress left to go. Hopefully we have all of this wrapped up in a few years.
____________

I think we've made good progress, but there is still tons of progress left to go. Hopefully we have all of this wrapped up in a few years.

Status:

Yes.

Almost 1 million tasks of SoB server were lost - 14 years of work.

It is interesting will PrimeGrid recalculate of those tasks?

I am now on PrimeGrid SoB.

Thanks.

I think we've made good progress, but there is still tons of progress left to go. Hopefully we have all of this wrapped up in a few years.

If by "wrapped up", you mean "done with the double check", I suspect it will take less than a year. It depends on how much participation there is.

If the same computing power is put into this as PPS-MEGA, it should take somewhere in the neighborhood of 3 months to complete.

If you meant "solve the Prime Sierpinski Problem", that will require a bit of luck to wrap it up anytime soon. :)

I think we've made good progress, but there is still tons of progress left to go. Hopefully we have all of this wrapped up in a few years.

If by "wrapped up", you mean "done with the double check", I suspect it will take less than a year. It depends on how much participation there is.

If the same computing power is put into this as PPS-MEGA, it should take somewhere in the neighborhood of 3 months to complete.

If you meant "solve the Prime Sierpinski Problem", that will require a bit of luck to wrap it up anytime soon. :)

"Done with the double check" is indeed what I meant. Thanks for clearing that up, I had previously misread the original statement about comparing the double check to PPS-MEGA. The double check now sounds that much more promising. I'll leave some resources on it as long as I can.

It really all depends on how much resources people put into the double check. If there were as many computers crunching the PSP double check as there are crunching PPS-MEGA, we'd be done in about 90 days. If we use only the computers that were crunching PSP before the double check, it might be more like 900 days. So it's really all up to you guys. :)

I'm just getting up to my 30-days-to-go point of PSP sv ready for if/when there's an announcement on that front then I'll help too.

I'm just getting up to my 30-days-to-go point of PSP sv ready for if/when there's an announcement on that front then I'll help too.

I'm not really sure what you mean. There's nothing planned for PSP (and ESP/SoB) sieving. We just suspended that project in the summer because it had reached optimal.

I mean TRP sv LOL! O dear...

I would also love to help but we have a bottleneck situation.

Similar to Dave, I need about 40 days to finish up with TRP Sieve but I will immediately thereafter be looking after the next deadline situation with GCW Sieve ("3-6 months" stated elsewhere) and, presumably, followed by GCW LLR since the latter two will not be running concurrently.

@Dave: I realized your typo when first seeing it. I am in the same boat.

Deadlines

Normal PSP tasks have a deadline of 21 days. The new short double check tasks have a deadline of 4 days.

When n reaches 8,000,000, I will change the deadline to 6 days. This is about 8% of the way through the double check.

When n reaches 10,000,000, I will change the deadline to 10 days. This is about 19% of the way through the double check.

When n reaches 13,000,000, I will change the deadline to 14 days. This is about 36% of the way through the double check.

When n reaches 16,000,000, I will change the deadline back to 21 days. This is about 80% of the way through the double check.

This plan is subject to change.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

I spread the word around, and someone brought up something.
They wondered if a badge would come with helping redo lost work or double-checking this kind of work.

This PSP and SOB and anything else like that.
What are the chances of say a "PG helper badge" of some sort?
Maybe with a special badge more people will jump in and help out.
Just a thought.
Glad to help out when we can.
Bill

Status:

Thanks for the update.

Just wondering about couple things...none of them terribly important :-)

How do we calculated FFT size since Min/Max N are overlapping for each FFT size?

How many large (non-DC leading edge) tasks are still in progress? Ie. is vast majority of computer power already doing DC effort? I guess so since a week passed...
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Just wondering about couple things...none of them terribly important :-)

How do we calculated FFT size since Min/Max N are overlapping for each FFT size?

How many large (non-DC leading edge) tasks are still in progress? Ie. is vast majority of computer power already doing DC effort? I guess so since a week passed...

As part of the new credit system we put in place a few years ago, in order to accurately predict the run time of each task, and therefore the correct amount of credit, the server starts running LLR on each candidate when it's loaded into the server. It runs the task just long enough for LLR to determine and output the FFT size, and then it aborts the LLR task.

As part of the new credit system we put in place a few years ago, in order to accurately predict the run time of each task, and therefore the correct amount of credit, the server starts running LLR on each candidate when it's loaded into the server. It runs the task just long enough for LLR to determine and output the FFT size, and then it aborts the LLR task.

You really do it THIS way?

llr -oSetupOnly=1 ...

Give it a single candidate and you'll get what you need.
Give it a sorted list of candidates and you'll get "wrap points" where FFT size changes.

Hello,
someone (I think its not Primegrid, because the prime is not on the homepage and no news until now) found an SoB Prime, 9.38M Digits, Rank 7.
http://primes.utm.edu/top20/page.php?id=3
http://primes.utm.edu/primes/page.php?id=122473

Has this find affect of our DoubleCheck?

Hello,
someone (I think its not Primegrid, because the prime is not on the homepage and no news until now) found an SoB Prime, 9.38M Digits, Rank 7.
http://primes.utm.edu/top20/page.php?id=3
http://primes.utm.edu/primes/page.php?id=122473

Has this find affect of our DoubleCheck?

That prime score is ginormus too! 1,130,218.71
It dwarfs Scott Brown's score.
Just amazing.

Hello,
someone (I think its not Primegrid, because the prime is not on the homepage and no news until now) found an SoB Prime, 9.38M Digits, Rank 7.
http://primes.utm.edu/top20/page.php?id=3
http://primes.utm.edu/primes/page.php?id=122473

Has this find affect of our DoubleCheck?

Yes, that's us. http://www.primegrid.com/forum_thread.php?id=7110 :)

We're still working on the official announcement and stuff, but the prime does show up here.

This discovery doesn't affect the PSP double check.

That prime score is ginormus too! 1,130,218.71
It dwarfs Scott Brown's score.
Just amazing.

How many SoB candidates were checked between this find and the previous one?

How many SoB candidates were checked between this find and the previous one?

Status:

Status:

Hello,
someone (I think its not Primegrid, because the prime is not on the homepage and no news until now) found an SoB Prime, 9.38M Digits, Rank 7.
http://primes.utm.edu/top20/page.php?id=3
http://primes.utm.edu/primes/page.php?id=122473

Has this find affect of our DoubleCheck?

Found on PrimeGid

Check this thread.

BTW, in your second link, in the Proof-code you can see that 'PrimeGrid' is part of the list. That means found by a PrimeGrid user.

Cheers

Thanks for the clarification.

____________
Badge Score: 1*2 + 1*5 + 12*6 + 4*7 + 2*8 + 1*9 = 132

The double check has surpassed n=8M, and as stated previously, the deadline is now 6 days. The next deadline change is expected at approximately n=10M at which time I will increase the deadline to 10 days.

Status:

We have passed the 10% completed mark!

The double check is 15% completed. N is at 9.1 million.

We have reached n=10 million. As discussed previously, the deadline is now 10 days.

The double check is almost 17% complete.

We have passed the 20% completed mark. N is over 11 million, and the candidates are now 3.3 million digits.

There are 14 leading edge candidates still being processed. Most of them are simply running on very slow computers.

We have passed the 25% completed mark. N is over 11.6 million, and the candidates are now 3.5 million digits.

There are 4 leading edge candidates still being processed.

The last of the remaining leading edge PSP tasks has now finished, so all PSP tasks sent out from this point on will be the shorter double check tasks. There's no possibility of getting a long "normal" task until the double check is completed.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Do you mean that all tasks from "Waiting" column of your table has been calculated at least once?

Or I misunderstood something?

The double check is now 30% complete.

Deadlines

Normal PSP tasks have a deadline of 21 days. The new short double check tasks have a deadline of 4 days.

When n reaches 8,000,000, I will change the deadline to 6 days. This is about 8% of the way through the double check.

When n reaches 10,000,000, I will change the deadline to 10 days. This is about 19% of the way through the double check.

When n reaches 13,000,000, I will change the deadline to 14 days. This is about 36% of the way through the double check.

When n reaches 16,000,000, I will change the deadline back to 21 days. This is about 80% of the way through the double check.

This plan is subject to change.

We've hit the 35% complete mark. Candidates being tested are just below 4 million digits.

We've hit the 40% complete mark. Candidates being tested are just below 4 million digits.

How is the work column value determined, or, perhaps more usefully, what does it represent?
____________
Eating more cheese on Thursdays.

How is the work column value determined, or, perhaps more usefully, what does it represent?

We've hit the 40% complete mark. Candidates being tested are just below 4 million digits.

We've hit the 40% complete mark. Candidates being tested are just below 4 million digits.

4 million digits?
14 million digits?

2^13.5M is 4,063,905 digits. So while n is between 13M and 14M, the digit count is 4M.

What kind of error rates are you finding with the double check? How accurate was the work that was not done through primegrid?