Noob Question: What to use

[jux]

My experience is that msieve always takes too long in poly select (using the 3-4% figure): I just did a 120 digit factorization as a benchmark and msieve spent 3.75 hours on polynomial selection while only spending 6 hours on nfs.
Can I tell msieve to reduce this time? I already ran tune() once.

[VBCurtis]

I run the poly-select step manually on its own, followed by the factmsieve.py python script for NFS. As such, I just cancel (ctrl-c) msieve when I think it has run long enough, and fire up factmsieve.py.

Also, poly select is single-threaded while NFS is multi-threaded; so 3 hrs of one thread compared to 6 hrs quad-threaded is high, but not terrible; an hour for poly select is probably enough, while the other two hours of poly select might save a thread-hour of NFS. The time isn't totally wasted, merely less efficient.

I don't know how to tell yafu/msieve a priori to shorten the deadline for poly select.

[bsquared]

You can shorten it, but you need to know about how many coefficients you want to search a priori. Launch yafu with a "-np X,Y" argument, where X and Y are the start and stop leading coefficients. Msieve is similar I think.

Also, yafu poly select is multi-threaded.

[jux]

If yafu poly select was running with one thread and NFS was running with eight, then the select time is quite reasonable.
If poly select is multi-threaded, it's still using only 20% of my CPU. Since I leave factoring to be done overnight, I might only bring out the GPU version of msieve for large numbers.

Just for reference, it took me around 10 full days to factor a 150 digit number.