Bad sieve files found for 2020 and 2021

[gd_barnes]

Originally Posted by pepi37

Yes , under Linux srsieve2 is multi-threaded, Windows version never support MT ( it was left on TODO list)
But in any case one instance per core give bet output as VBCurtsi say

He was referring to sr2sieve, not srsieve2. I wasn't aware that sr2sieve was multi-threaded in Linux.

Srsieve2 is multi-threaded in both Windows and Linux. I sometimes run multi-threads on srsieve2 in Windows.

[gd_barnes]

I have now sent files to Yoyo for re-sieving for our reserved and recommended bases per #1 and #2 under prioritizing the sieves in the announcement post. I've also clarified those in that post. The bases sent were:

R111, R534, R667, S529, & S536

[gd_barnes]

The 5 sieve files from the last post have now been fully re-sieved by Yoyo. All bases with recommendations or reservations now have good sieve files. I have run additional tests as needed below the current test limit or below where the reservations began. No additional primes were found. Updated files have been posted on the pages and people with reservations (who I could get a hold of) have received additional tests to run.

I have finished analyzing all of the applicable sieve files as specified in this thread. Out of 202 files sieved to 5e14 in 2020, 2021, and early 2022, 109 (54%) were identified as potentially bad. Out of the 109, 104 are very obviously bad. The other 5 may have an incorrect sieve depth or other unusual error such as some factors that may not have been removed or some unusual algebraic factors that I have yet to identify.

All 104 files that are clearly bad occurred between Aug. 2020 and Sept. 2021. They were definitely impacted by the scripting error. The other 5 occurred a little outside of this time-frame and it is not so clear what is going on with them. They do deviate from the number of tests that they should have so we will see what happens when Yoyo re-sieves them.

All of the above 109 files have been re-sieved to 1e12 and will be sent to Yoyo to sieve to 5e14 (or 1e15 in some cases) a few at a time to avoid overwhelming them. The next set of 6 files have been sent. They are the bad files from SRBase's 2k recommended effort from 2022. Most of the bases with 2-3 k's remaining with bad files will only have a few hundred additional tests that need to be run so are very unlikely to yield a prime. The more k's remaining, generally the bigger the problem.

All potentially bad files have been removed from the reservations pages with applicable notes about what is going on with them.

[gd_barnes]

All bases that originally had 2 k's remaining and were identified with potentially bad sieve files have been re-sieved. 1667 previously missing tests were run. No primes were found. Corrected files have been posted on the reservations pages.

Re-sieves and 4180 previously missing tests were also run for 3 reserved bases that originally had 3, 7, & 10 k's remaining. No primes were found. Coordination was done with people who had the reservations.

Files with 3, 4, & 5 k's remaining have been sent to Yoyo for re-sieving. Sieves with 3 k's remaining are nearing completion.

[gd_barnes]

All bases that originally had <= 4 k's remaining and were identified with bad sieve files have been re-sieved and missing tests have been tested. No primes found.

Applicable corrected sieve files have been posted on the reservations pages.

Continuing on...

[gd_barnes]

All bases that originally had <= 8 k's remaining and were identified with bad sieve files have been re-sieved and missing tests have been completed.

Quite a few bases with > 8 k's remaining have also had their missing tests run. Pepi pointed me to some old factor files from Yoyo that are sufficient for what is needed to avoid having them re-sieve quite a few bases. Many of these had 10 to 17 k's remaining. This saves us a lot of time going forward. In between the times that Yoyo is doing its re-sieving, I have been able to test many such bases.

1 prime found and previously reported. A total of 51,894 missing tests have been completed.

Applicable corrected sieve files have been posted on the reservations pages.

General status:

17 bases needed no missing tests run; reasons:
   The original file was good but the sieve depth was wrong.
   Some factors were not originally removed, enough so that the file appeared bad.
   No testing had yet been done on the file.
46 bases had missing tests and have now been fully tested.
46 bases remain to be sieved and/or tested; breakdown:
   1 is currently being tested.
   11 have been sent to Yoyo for re-sieving.
   21 have not yet been sent to Yoyo and will need re-sieving.
   13 have an old but good factor file from Yoyo so will need no re-sieving.
109 total

So we've completed testing on as many bases as remain to be tested.

Continuing on...

[gd_barnes]

Updated status:

All bases that originally had <= 10 k's remaining and were identified with bad sieve files have been re-sieved and missing tests have been completed. Also many bases with > 10 k's remaining were completed where no re-sieving was needed as a result of the old factor files from Yoyo described in the last post.

All hands were on deck since the last update so a significant dent was made in the number of bases completed.

3 primes found and previously reported. A total of 164,009 missing tests have been completed.

All applicable files have now been sent to Yoyo for re-sieving.

Applicable corrected sieve files have been posted on the reservations pages.

General status:

17 bases needed no missing tests run; reasons as stated in the last post
72 bases had missing tests and have now been fully tested.
20 bases remain to be sieved and/or tested; breakdown:
   1 is currently being tested.
   1 has been fully re-sieved by Yoyo and will eventually be added to my testing queue.
   11 have an old but good factor file from Yoyo and will eventually be added to my testing queue.
   7 have been sent to Yoyo for re-sieving but sieving is not complete.
109 total

Previously I had been working my way down the list by k's remaining. Going forward, I'll be working down the list by easiest to most difficult to complete. For the most difficult final few, I may ask for help depending on how tough they look when the sieving is done.

Continuing on...

[mdettweiler]

Originally Posted by gd_barnes

Updated status:
Previously I had been working my way down the list by k's remaining. Going forward, I'll be working down the list by easiest to most difficult to complete. For the most difficult final few, I may ask for help depending on how tough they look when the sieving is done.

Feel free to load the toughest bases into my semi-private PRPNet server (port 1300) if you'd like my fleet to take a detour for a few days to knock them out quickly. Unglamorous doublecheck work is one of my specialties.

If you wanted, for instance, to prioritize a batch of doublechecking work ahead of the main effort we currently have loaded in the server, you could use the trick I mentioned before in an email to make the "a" (age) sortoption hand out candidates in any desired order. The LastUpdateTime column in SQL table Candidate records when a given candidate was loaded into the server in the form of a Unix timestamp (integer number of seconds since the Unix epoch of January 1, 1970). If you set it for a particular group of candidates to be "older" (i.e., a smaller integer) than anything else loaded in the server, the server will see those candidates as the oldest and prioritize them until they're done.

[gd_barnes]

Originally Posted by mdettweiler

Feel free to load the toughest bases into my semi-private PRPNet server (port 1300) if you'd like my fleet to take a detour for a few days to knock them out quickly. Unglamorous doublecheck work is one of my specialties.

If you wanted, for instance, to prioritize a batch of doublechecking work ahead of the main effort we currently have loaded in the server, you could use the trick I mentioned before in an email to make the "a" (age) sortoption hand out candidates in any desired order. The LastUpdateTime column in SQL table Candidate records when a given candidate was loaded into the server in the form of a Unix timestamp (integer number of seconds since the Unix epoch of January 1, 1970). If you set it for a particular group of candidates to be "older" (i.e., a smaller integer) than anything else loaded in the server, the server will see those candidates as the oldest and prioritize them until they're done.

Cool. Thanks for the offer. I'm not there yet. I'll have a better idea in a month or so. By then, all of the sieving will be done and I'll have time do all the final compares to see how many missing tests there are on each of the remaining bases.

In the mean time, you'll have a chance to knock out a few of our 1k bases to n=1M.

The tests may be a little quick for so many machines but I think they'd be OK. The server would likely bog down some on the lower end of the tests for you. 90% of the testing is in the n=100K-300K range but there are 3 bases left for n=100K-500K, one of which I'm working on now (R107) that doesn't have too many tests (~7500). The n=300K-500K range would be more ideal for your machines. The good news is that I've found 4 threads for each instance to be pretty much optimal for that entire test range except maybe slightly inefficient for R107 at the very low end, since that is the smallest base that I've tested. Your machines would need no re-configuring.

By the way, these are not double-check work. They are first-time tests that were missed. There won't be any already known primes in them.

[gd_barnes]

Updated status:

All bases with bad sieve files that had < 10,000 missing tests have been fully tested.

9 bases were completed since the last update. 11 bases remain.

Overall, 5 primes have been found and previously reported. A total of 240,531 missing tests have been completed.

All remaining files have now been re-sieved.

All bases completed have had their corrected sieve files posted on the reservations pages.

General status:

17 bases needed no missing tests run; reasons were previously stated
81 bases had missing tests and have now been fully tested.
11 bases remain to be tested; breakdown:
   2 have been partially tested and will be sent to Max later on.
   1 is currently being partially tested and will be sent to Max later on.
   2 have an old but good factor file from Yoyo and will eventually be added to my testing queue.
   6 have been fully re-sieved by Yoyo and will eventually be added to my testing queue.
109 total

Upcoming:
The remaining 11 bases have ~400,000 tests left to complete. I am testing them for n=100K to n=150K or 200K. I will then send the remaining of n=150K-300K or n=200K-500K to Max to complete the testing after he is done with his current reservation. Anticipated completion is the end of 2023.

Continuing on...

[gd_barnes]

Below is additional info on what remains to be tested for the bad sieve file effort. They are listed in order of difficulty, which is the order that Max and I will complete them.

base / original missing tests / n-range needed / status / # primes

S991  13,938 100K-300K complete; 2 primes
S396  10,628 100K-500K done to 200K; Max testing 200K-500K; 1 prime
S672  17,338 100K-300K done to 150K; in Max's queue; 2 primes
S381  19,180 100K-300K done to 150K
S378  24,376 100K-300K done to 150K
S693  21,312 100K-300K done to 150K
S466  29,715 100K-300K done to 150K; 1 prime
S408  39,473 100K-300K Gary testing to 150K
S421  59,252 100K-300K in Gary's queue
R25   99,911 150K-300K
S352  67,124 100K-500K

All files have been fully sieved.

I will update this as the effort progresses.