Sierp base 63 - team drive #5

[rogue]

Originally Posted by Puzzle-Peter

S63, thanks for catching that!

I'll have to take a look at the OS. Redhat Enterprise Linux, I'll get the exact data tomorrow.

OK. Once I have that information I might be able to build a version of pfgw that you can run. Another option is for you to build pfgw yourself on one of those boxes that gets this error.

Can you talk someone into upgrading the OS on those boxes?

[Puzzle-Peter]

Originally Posted by rogue

Can you talk someone into upgrading the OS on those boxes?

I don't think so. It would be done by an external IT services company and therefore it would cost money. I'd need a good reason...

Anyway here's the OS info:

box that works:
Linux wumc0519 2.6.18-194.17.4.el5 #1 SMP Mon Oct 25 15:50:53 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux

box that doesn't work:
Linux wumc0214 2.6.9-67.ELsmp #1 SMP Wed Nov 7 13:56:44 EST 2007 x86_64 x86_64 x86_64 GNU/Linux

The error message is simply "FATAL: kernel too old" followed by a segmentation fault.

[rogue]

Originally Posted by Puzzle-Peter

I don't think so. It would be done by an external IT services company and therefore it would cost money. I'd need a good reason...

Anyway here's the OS info:

box that works:
Linux wumc0519 2.6.18-194.17.4.el5 #1 SMP Mon Oct 25 15:50:53 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux

box that doesn't work:
Linux wumc0214 2.6.9-67.ELsmp #1 SMP Wed Nov 7 13:56:44 EST 2007 x86_64 x86_64 x86_64 GNU/Linux

The error message is simply "FATAL: kernel too old" followed by a segmentation fault.

Unfortunately I cannot fix this with my resources. It would require a lot of finagling on Steven's box which could impact his OS's stability. If you are willing to try building it yourself on one of the older boxes, I'll walk you through the process. PM me if you want to do that. I'd hate to see them go unused.

[Puzzle-Peter]

Originally Posted by rogue

Unfortunately I cannot fix this with my resources. It would require a lot of finagling on Steven's box which could impact his OS's stability. If you are willing to try building it yourself on one of the older boxes, I'll walk you through the process. PM me if you want to do that. I'd hate to see them go unused.

The're not unused. The sieves and LLR are running without problems on these boxes. It's only PFGW which I use for the new bases script that doesn't work. It would be nice to be able to run everything on every box though. I'll get back to you in a week when I return from a vacation.

[rogue]

Originally Posted by Puzzle-Peter

The're not unused. The sieves and LLR are running without problems on these boxes. It's only PFGW which I use for the new bases script that doesn't work. It would be nice to be able to run everything on every box though. I'll get back to you in a week when I return from a vacation.

I believe that there is an option with LLR to eliminate k after a prime is found. Gary, can you confirm?

If not, you can use LLR with PRPNet on this base. You would need to set the "usellroverpfgw" to 1 in the server.

[Mini-Geek]

Originally Posted by rogue

I believe that there is an option with LLR to eliminate k after a prime is found.

Yep, put StopOnPrimedK=1 in llr.ini.

[Puzzle-Peter]

Just in case you're wondering: I have stopped sieving at p=81G and started LLR testing. I had reached a removal rate of 18s per factor where I was aiming for ~20. I figured that finding some primes (and thus eliminating part of the tests) would account for the last 2s.

[rogue]

Originally Posted by Puzzle-Peter

Just in case you're wondering: I have stopped sieving at p=81G and started LLR testing. I had reached a removal rate of 18s per factor where I was aiming for ~20. I figured that finding some primes (and thus eliminating part of the tests) would account for the last 2s.

I'm curious about a few things. Did you sieve all remaining k? Did you sieve them to n=25000? Are you using LLR or PFGW to test the output? How many cores are you dedicating to this? When you expect to complete the range you are testing.

BTW, if you want more sieving/PRPing work, R79 is another good candidate. There are 10366 k remaining at n=10000. This would make it about 20% the effort of S63. I considered doing it myself, but I wanted to find a Top 5000 prime or two first, thus my available resources are dedicated to single k conjectures.

[Puzzle-Peter]

Originally Posted by rogue

I'm curious about a few things. Did you sieve all remaining k? Did you sieve them to n=25000? Are you using LLR or PFGW to test the output? How many cores are you dedicating to this? When you expect to complete the range you are testing.

Yes, I sieved them all in one big file 10001<n<25000. I'm using LLR. ATM ~30 cores are working on this, but as other cores finish what they are working on right now, I will move them to S63. So in two or three weeks, I expect to have ~100 cores on S63. It is my goal to finish before the end of the year.

Originally Posted by rogue

BTW, if you want more sieving/PRPing work, R79 is another good candidate. There are 10366 k remaining at n=10000. This would make it about 20% the effort of S63. I considered doing it myself, but I wanted to find a Top 5000 prime or two first, thus my available resources are dedicated to single k conjectures.

When S63 is done I will probably switch to 1k/2k/3k conjectures just to get a change. But well, you can never know...

BTW, LLR is sometimes giving me messages like "5^((N-1)/3)-1 is coprime to N!"
I have never seen this before. Is this a Sierpinski specialty? And what does it mean? sometimes 5^((N-1)/3)-1 is declared composite and the candidate is tested using a different approach / base.

[rogue]

Originally Posted by Puzzle-Peter

Yes, I sieved them all in one big file 10001<n<25000. I'm using LLR. ATM ~30 cores are working on this, but as other cores finish what they are working on right now, I will move them to S63. So in two or three weeks, I expect to have ~100 cores on S63. It is my goal to finish before the end of the year.

When S63 is done I will probably switch to 1k/2k/3k conjectures just to get a change. But well, you can never know...

BTW, LLR is sometimes giving me messages like "5^((N-1)/3)-1 is coprime to N!"
I have never seen this before. Is this a Sierpinski specialty? And what does it mean? sometimes 5^((N-1)/3)-1 is declared composite and the candidate is tested using a different approach / base.

That is a lot of resources. That would be cool to have S63 completed to n=25000. Of course Gary would probably prefer S3 or R3, but if you continue to have those resources available to you, then that would be possible in the future.

The "coprime to N" message is related to a test that LLR does to determine if the base it chose for the PRP test could be used to do the primality test. In this case it couldn't, thus it chose another base. The reason LLR does this is so that residues of composite numbers are always from base 3 PRP tests. If LLR chose a better base up front, then the residues would not be compatible with those produced by other programs which default to base 3, such as PFGW and phrot.

[gd_barnes]

Holy cow. I'd be very impressed if you were able finish that search without dieing of boredom.

I'm curious. How big is the sieve file?