Riesel base 16 - team drive #2

[mdettweiler]

LLRnet reserving 90K-100K.

[gd_barnes]

Originally Posted by Anonymous

LLRnet has completed 87.5K-90K. lresults are attached. (The LLRnet results file has been converted to a normal lresults file by a Perl script that I wrote.)

Very good. The run times are kind of scary. I assume those will improve as things stabilize.

Thanks for changing the status above Anon.


Gary

[tnerual]

Originally Posted by gd_barnes

Very good. The run times are kind of scary. I assume those will improve as things stabilize.

Thanks for changing the status above Anon.


Gary

run times are not based on processor times but on "out of server time"

so if you ask for a cache of 5 k/n pair, your processing times will be multiplied by 5 ...
i work with a cache of 20 k/n pairs for 5 cores (i have a local LLR proxy server)

[mdettweiler]

Originally Posted by tnerual

run times are not based on processor times but on "out of server time"

so if you ask for a cache of 5 k/n pair, your processing times will be multiplied by 5 ...
i work with a cache of 20 k/n pairs for 5 cores (i have a local LLR proxy server)

Okay, I didn't know that. Thanks!

Anyway, I guess, long story short, when dealing with lresults files made out of LLRnet results files, take the runtimes with a grain of salt.

[gd_barnes]

Sieving will complete tonight for n=100K-200K for all k-values for the continuation of this team drive.

I'll post files up to ~n=120K on Saturday or early Sunday.


Gary

[gd_barnes]

The Riesel base 16 team drive has restarted. Files have been posted for n=100K-140K.

With little interest in individual-k reservations, I added 4 k's to the drive this time. This still leaves 6 individual k's, 5 of which are already reserved (3 by me), as a result of them also being base 4 k's and in some cases base 2 odd-n or even-n conjecture k's.

For k's that are for more than one base, we have to test them for the lowest base else risk doing partial double-work or adding the complexity of testing them at a higher base and then removing either all of the odd n's or even n's when testing the lower base, which would be error-prone. Hence, I will always leave them out of team drives on the higher base as is the case here.


Gary

[mdettweiler]

LLRnet has completed 90K-93.4K. (It's actually had some results come in farther up, but I'm only submitting up to the minimum outstanding n--i.e. there's no holes in this lresults file.) lresults is attached.

Note to Gary: The reason why the higher bound of the completed portion of LLRnet's range is lower than the one you marked is because my earlier rough estimate of n-range was based on the leading edge of LLRnet, not what's been completely done with no holes in between.

[gd_barnes]

Originally Posted by Anonymous

LLRnet has completed 90K-93.4K. (It's actually had some results come in farther up, but I'm only submitting up to the minimum outstanding n--i.e. there's no holes in this lresults file.) lresults is attached.

Note to Gary: The reason why the higher bound of the completed portion of LLRnet's range is lower than the one you marked is because my earlier rough estimate of n-range was based on the leading edge of LLRnet, not what's been completely done with no holes in between.

Ah, OK. So to paraphrase Riese Sieve and SOB, the n-min is now 93.4K and n-max is 94K as of your most recent reporting (more-or-less). I believe those are the terms they use.

I changed the testing limit in the first post here.


Gary

[gd_barnes]

I'll reserve Riesel base 16 n=100K-104K here. Sierp base 16 will be done for that range early Monday and I'll start on it after that. If LLRNet finds a prime in the n=93.4K-100K range, I'll remove the k from the sieved files.


Gary

[gd_barnes]

Don't try this at home...

2 primes...5 k's on different bases knocked out!

Riesel base 4:
16734*4^156852-1 is prime
19464*4^155532-1 is prime

These 2 also take out:
Base 16:
16734*16^78426-1
19464*16^77766-1
-and-
Base 2 odd-n:
8367*2^313705-1

I'm still working on Riesel base 4 k=13854; currently at n=162K base 4. Finding one there would probably knock out k's on 3 different bases.

Who's your daddy?


Gary

[gd_barnes]

Originally Posted by gd_barnes

Don't try this at home...

2 primes...5 k's on different bases knocked out!

Riesel base 4:
16734*4^156852-1 is prime
19464*4^155532-1 is prime

These 2 also take out:
Base 16:
16734*16^78426-1
19464*16^77766-1
-and-
Base 2 odd-n:
8367*2^313705-1

I'm still working on Riesel base 4 k=13854; currently at n=162K base 4. Finding one there would probably knock out k's on 3 different bases.

Who's your daddy?


Gary

I had a sneaking suspicion about these primes this morning after I woke up. When other people have had LLR problems with false primes, the primes usually bunched up. In looking at the results file, I saw that these primes were only 25 tests apart so I ran a double-check:

Original test with 24 composites in between:
19464*2^311064-1 is prime! Time : 200.404 sec.
16734*2^313704-1 is prime! Time : 200.316 sec.

Double-check on same computer with no tests in between:
19464*2^311064-1 is not prime. LLR Res64: 3A387A638BB025BA Time : 200.069 sec.
16734*2^313704-1 is prime! Time : 197.884 sec.



Both tests were run on my Dell core duo work laptop, which has generally been very reliable. I do not overclock any of my machines.

I then ran a triple-check on my main desktop, a 3-Ghz P4. It confirmed the double-check; both the prime and composite residue.

So...the bad news:
1. I have to 'put back' k=19464 has having not found a prime on Riesel bases 4 and 16.
2. I have to rerun the entire batch for k=16734 and k=19464 on a different machine to see if I missed any primes. (not bad; ~2 CPU days)
3. I have to rerun the entire batch for n=100K-104K for Sierp base 16 on a different machine(s) looking for missing primes. (bad bad; ~10-12 CPU days)
4. I will have to unreserve Riesel base 16 n=100K-104K; otherwise it will just sit and wait for as much as 2 weeks. It the LLRNet server has hit n=100K before then, it may as well reserve it.

The only good news is that k=13854 is still knocked out of 3 bases, which includes k=6927 for Base 2 odd-n.

I have relegated my work laptop to sieving for now until I can figure out what is up. My 3 Dell duo laptops (1 work; 2 personal) are the fastest sievers that I have so that's not a bad thing for it.

The only thing I can figure about this is that my work laptop gets far more use on varied tasks and is carried back-and-forth between home and work in all kinds of whether (it has been very cold here last 2 weeks; warmer now) 5 days/week so perhaps one of its components has been slightly compromised. I do shut it down in between home and work.

My 2 personal machines stay at home 99% of the time so I'm comfortable that they don't have the same issue.

I think I jinxed myself when I got a little cocky.


Gary