Weird results of Msieve

[VBCurtis]

Originally Posted by chris2be8

What range of leading coefficients to search for a given size of number is another issue. I've not had enough experience to offer advice.
Chris

The rule of thumb for poly select is to spend 3-5% of estimated project-length time in poly select. We don't have any idea whether 3% or 5% (or something else entirely) is correct, but this is what most folks do. For example, a GNFS155 should take a modern quad-core around 10 days to factor, for sieving and linear algebra. That's 40 core-days. 5% of 40 is 2, so spending no more than 2 core-days on poly select for a C155 is wise. GPUs run faster and take more electricity than a CPU core, so perhaps 24-36 hr on a GPU plus whatever CPU time is needed is enough poly select. If you were racing to complete the C155 job by a deadline, I'd aim to begin relations-gathering after 1 day of poly select.

I set the coefficient range quite wide and manually stop the poly select step after enough time has passed, but an automated script using poly coefficients would need a little tweaking of coefficient range for each input.