question about actual n tested
 

hello,

actually we are at n around 76000

as i can see, we are doing the 75000-100000 range

from the stat, we have this:

upper_bound amount
50000 55555
75000 10659
100000 30999
125000 29048

we were working in the 125000 upper bound, now in the 100000 upper bound.

wouldn't it be more productive to first do the 55555 test from the 50000_upper_bound and after the last 10659 test from the 75000_upper_bound ?

The thing is that the real first pass test are n>100000.
All the lower test have been done before but we have no residues
to prove that all test have really been done.

So we decided to recover all the residues in small steps.
The chances that we missed a prime are not high enough to force us to
run all recover tests before runing the normal first pass tests.

So every week or two i activate another 5000-10000 n range for recovery.
At the moment the active recovery windows are:
65000-75000 for sierpinski
75000-85000 for riesel


Lars

ok

now i understand.

thanks

As far as I understand, residues from prp version 2.x (which, I think, is faster than version 3.x on Pentium III processors) have been rejected. Prp version 3.x prints out both the new-style residues and the old ones. Would it not be better to just have someone run prp 3.x on these numbers and then the old residues could be checked (so that we have a double-check) and we would then have a new residue too? I don't know if LLR gives both types of residue or not.
Thoughts?