Bases 501-1030 reservations/statuses/primes

[rebirther]

Reserving R822 as new base using the new-base script up to 2.5k and sieving to 10k

[MisterBitcoin]

Originally Posted by rebirther

R895 tested to n=2.5k, sieved to 10k

2090 remain

Results emailed - Base released

Sounds pretty decent. Can you send the sieve file to me?

[rebirther]

Originally Posted by MisterBitcoin

Sounds pretty decent. Can you send the sieve file to me?

attached.

[gd_barnes]

This is starting to get a little out of hand.

I prefer not to have open bases only tested to n=2500 and then stopped. Even if a sieve file is provided for n=2.5K-10K. It is better if they are reserved and fully tested to n=10K. Otherwise we will have a whole lot of bases tested to only n=2500, which is what I would like to avoid because it becomes too cumbersome to continually remove large numbers of k's.

There are already several bases that BOINC is testing for n=2500 to 10K. Can we please finish testing those to n=10K before starting any more new bases? I'll go ahead and reserve and show status for R822 and R895 at this point.

MisterBitcoin, are you reserving R895 to n=10K ?

[rebirther]

Originally Posted by gd_barnes

This is starting to get a little out of hand.

I prefer not to have open bases only tested to n=2500 and then stopped. Even if a sieve file is provided for n=2.5K-10K. It is better if they are reserved and fully tested to n=10K. Otherwise we will have a whole lot of bases tested to only n=2500, which is what I would like to avoid because it becomes too cumbersome to continually remove large numbers of k's.

There are already several bases that BOINC is testing for n=2500 to 10K. Can we please finish testing those to n=10K before starting any more new bases? I'll go ahead and reserve and show status for R822 and R895 at this point.

MisterBitcoin, are you reserving R895 to n=10K ?

I want to prepare a lot of new bases for the future up to 2.5k. At the moment the progress for 2.5-10k is very slow. I will change some things to speed it up. For the future aspects its good to have already the sievefiles. I cant see any negative aspects to have bases up to 2.5k.

[gd_barnes]

Originally Posted by rebirther

I want to prepare a lot of new bases for the future up to 2.5k. At the moment the progress for 2.5-10k is very slow. I will change some things to speed it up. For the future aspects its good to have already the sievefiles. I cant see any negative aspects to have bases up to 2.5k.

Ugh. OK. Well I'll attempt to keep up. Please limit the new bases to 2 per week at most.

It's just my preference that we not have a whole lot of bases tested to such a low n-level. There are so many k's remaining so I have to post large sieve files and properly remove k's that are found prime. It also takes a little bit for me to determine which k's that can be removed by algebraic factors.

As you probably know testing n=2.5K-10K takes likely 50-100 times as long as testing n<=2.5K. If you're going to do this, please provide sieve files sieved to at least P=10G for n=2500-10K. I saw that the recent one that you provided was only sieved to P=100M. That is not deep enough for the typical user. At P=100M, factors are being removed much faster than the average test length for n=2500-10K. Testing n=2.5K-10K is quite a bit of work for the average single user when there are more than 2000 k's remaining. Thanks.

[rebirther]

Originally Posted by gd_barnes

Ugh. OK. Well I'll attempt to keep up. Please limit the new bases to 2 per week at most.

It's just my preference that we not have a whole lot of bases tested to such a low n-level. There are so many k's remaining so I have to post large sieve files and properly remove k's that are found prime. It also takes a little bit for me to determine which k's that can be removed by algebraic factors.

As you probably know testing n=2.5K-10K takes likely 50-100 times as long as testing n<=2.5K. If you're going to do this, please provide sieve files sieved to at least P=10G for n=2500-10K. I saw that the recent one that you provided was only sieved to P=100M. That is not deep enough for the typical user. At P=100M, factors are being removed much faster than the average test length for n=2500-10K. Testing n=2.5K-10K is quite a bit of work for the average single user when there are more than 2000 k's remaining. Thanks.

Sieving at 100M is enough. The tests a rather small (1-6s). My thoughts were to run the bases later on BOINC but cannot hold many on the server without any progress for month. Iam splitting the files in 2 pieces, 2.5-5k run the range, remove the primes from the sievefile, run 5-10k.

[VBCurtis]

At 100M, the factors are found less than 1 per second? That seems hard to believe. I've had sieves where the factor-found rate is 1/second at 10G, let alone 100M!

There is also a benefit to having (and posting) smaller sieve files; if the sieve-factors rate is the same as the primality-testing rate, I'd sieve more to shrink the file even there is no net gain in efficiency.

[MisterBitcoin]

Yes, I reserve R895 up to n=10K. Will sieve deeper, for sure.



Also, S406 reached n=400K, no prime found, releasing.

[rebirther]

Originally Posted by VBCurtis

At 100M, the factors are found less than 1 per second? That seems hard to believe. I've had sieves where the factor-found rate is 1/second at 10G, let alone 100M!

There is also a benefit to having (and posting) smaller sieve files; if the sieve-factors rate is the same as the primality-testing rate, I'd sieve more to shrink the file even there is no net gain in efficiency.

No, thats not the factor found rate thats the runtime. To sieve deeper is worthless in my eyes because you have then less 4000 tests from over 1M.

[gd_barnes]

Originally Posted by rebirther

Sieving at 100M is enough. The tests a rather small (1-6s). My thoughts were to run the bases later on BOINC but cannot hold many on the server without any progress for month. Iam splitting the files in 2 pieces, 2.5-5k run the range, remove the primes from the sievefile, run 5-10k.

How are you making the determination that sieving to 100M is enough? I put the R895 on my modern I7 at your sieve depth of P=100M and it removed over 2000 factors in the first minute alone. That's 33 factors per second or one factor for every 0.03 seconds.

Analysis:
Removal rate 1 every 0.03 seconds.

Testing rate per your statement: 1-6 secs. We will round down and call it an average of one test every 3 seconds.

3 / 0.03 = 100

So you need to sieve 100 times as deeply for it to be removing factors at the same rate as the tests. (This is not entirely accurate since factors are removed and hence removal rate reduction is not linear but it is close enough for our purposes.)

Per this analysis, if you are going to provide a sieve file for others outside of BOINC, I think it is reasonable that you sieve them to 100M * 100 = 10G. (What I previously requested.)

Even if you discount this some because of primes that will be found, the 1-6 sec test rate doesn't include tests for n=5K-10K. In reality P=10G is likely still well undersieved for the whole n=2500-10K range. (I think I have typically sieve to P=20G or 25G if sieving the entire range if I don't want to sieve any more the rest of the way after removing k's found prime.)

You can feel free to sieve it to whatever depth you want for BOINC. Everyone likes more tests for their credit. I get that. But when sieving for everyone else we would like to optimize total testing time.

Question: When you remove the k's found prime for n=2500-5000 when beginning testing for the n=5000-10K range, would it make sense to do additional sieving on the remaining file for n=5000-10K? After all the average test time is about 4x greater. Just thought I'd point that out.

Bottom line: Providing us with files sieved to P=100M apparently only takes about a half hour of CPU time and is well undersieved so I'd prefer not to post them. I would be doing others a disservice. But I'll keep up with whatever you do here as best as I can.