Sierp base 3 reservations/statuses/primes

[gd_barnes]

Originally Posted by VBCurtis

The fact that BOINC did it doesn't mean it was a good idea. "what is a good idea" is, for me, exactly the intent of his question. You may have taken R3 from 50-100k because that's how it was sieved/prepped, but felt afterward that 50k was too small. I don't recall any commentary about it one way or the other, and I read each thread in this forum regularly.

I'm sad that you "welcome" tiptoes and beginner questions in this way from someone who has contributed to various projects around here for so long. You know this poster is not a crank, but treat him like one.

I think it's fair to assume that if Reb took such a huge range for BOINC for n=50K-100K that took several months for them that n=50K is fine for BOINC. Wouldn't you agree?

Henry is not a beginner and he is not tiptoeing. He questions and comments on the direction and decisions of projects while not contributing or knowing what's going on. He's questioned things at NPLB as well as CRUS and his last 4 comments have not been not helpful. He hasn't contributed to either project in many years. It's time to put some machines up if you want to make comments and suggestions about the direction of a project.

I know that Henry is not a crank. He wouldn't be a super-mod if he was. But it doesn't take a crank to do things that rub people the wrong way. And it's people's right to call them out on it if it keeps happening. Sure a newbie asking why we do things a certain way and needing help getting started...no problem. I'm happy to answer and help. But when an experienced person keeps doing the same with no contribution...well, that's a no go.

Reb, would you care to comment on what a good lower bound for testing is for base 3?

[rogue]

Originally Posted by gd_barnes

Sounds like a plan! Are you sure you're willing to keep all of those machines finding small primes for many months? I know firsthand how boring that can be.

For many of the cores it is a "set it and forget it" situation. I just need to be wary of occasional unplanned updates or outages that would require me to continue the PRP testing.

[rebirther]

Originally Posted by gd_barnes

Reb, would you care to comment on what a good lower bound for testing is for base 3?

90k+, there is a lot of effort and taking time to remove factors or loading on the server including upload issues with my provider every 2 days ^^

[gd_barnes]

Originally Posted by rebirther

90k+, there is a lot of effort and taking time to remove factors or loading on the server including upload issues with my provider every 2 days ^^

I must admit I have egg on my face. To Henry and Curtis in the discussion above: I was wrong and I'm sorry.

Reb, I had no idea. In the future please let me know if something will be a problem for you. I admit to ignorance of the BOINC process. We certainly don't want you to have problems with your provider. We greatly appreciate the BOINC effort on the project and definitely do not want to take advantage of it.

On the current topic of S3. Even n=25K-50K is a huge undertaking for Mark or anyone. I would not want to suggest that all k's be tested any higher just so that it could be taken over by BOINC in a rapid fashion. S3 is something that will stay with individual searchers for the foreseeable future. The tests are just too numerous and fast. We much prefer that BOINC work on the long tests where all the glory is. :-)

[rebirther]

Originally Posted by gd_barnes

Reb, I had no idea. In the future please let me know if something will be a problem for you. I admit to ignorance of the BOINC process. We certainly don't want you to have problems with your provider. We greatly appreciate the BOINC effort on the project and definitely do not want to take advantage of it.

I can only taking smaller sievefiles, could try some standalone if I have the ressources. Iam concentrating now on newer bases at the moment.

[gd_barnes]

Originally Posted by rebirther

I can only taking smaller sievefiles, could try some standalone if I have the resources. I am concentrating now on newer bases at the moment.

Sounds good.

I remember 3-4 years ago BOINC did a whole bunch of bases (~40-50 of them I believe) for n=25K-100K with anywhere from 100-400 k's remaining. They were done without me or anyone initially requesting that BOINC do them. It seemed like the BOINC folks enjoyed those because they were done so quickly. Obviously those are much larger bases than base 3 but it still led me to believe that BOINC could handle tests for n=50K base 3.

Did you have much problem with n=25K-100K for higher bases? Perhaps the fact that there were so many fewer k's remaining made it much easier for you to handle.

[rebirther]

Originally Posted by gd_barnes

Sounds good.

I remember 3-4 years ago BOINC did a whole bunch of bases (~40-50 of them I believe) for n=25K-100K with anywhere from 100-400 k's remaining. They were done without me or anyone initially requesting that BOINC do them. It seemed like the BOINC folks enjoyed those because they were done so quickly. Obviously those are much larger bases than base 3 but it still led me to believe that BOINC could handle tests for n=50K base 3.

Did you have much problem with n=25K-100K for higher bases? Perhaps the fact that there were so many fewer k's remaining made it much easier for you to handle.

No, this is only related for base 3.

[rogue]

Originally Posted by rogue

I'm starting to do a few 1G ranges of S3 to n=50000 trying to gauge the effort of completing the range to n=50000. I suspect that this is doable as a project goal for 2022. I think it should be less work than getting R3 to n=100000.

After only a couple of days, I think that a range of 1G to n=50000 should take about 60 days (both sieving and testing) on a single core. I am using srsieve2cl to sieve to about 2e10 which seems about about right for my hardware. At 2e10 the removal rate is about 1 candidate per second.

If this estimate were to hold, then I can complete S3 on my own to n=50000 in about 8 months (60 days per range * 120 ranges / (30 cores * 30 days per month)). Maybe someone else would be willing to throw a few cores at it.

Compared to R3, a range of 1G to n=100000 takes about 200 days on a single core. Distributed using PRPNet reduces this to about 7 days per 1G range. The tests for S3 are too fast for PRPNet to efficiently manage.

I revise my estimate to about 80 days per 1G range.

[rogue]

I am reserving 1G to 6G up to n=50000.

[rogue]

Range 1G-6G to n=50000 completed. Primes attached.