Riesel base 6 - team drive #4 - EIGHT OR BUST!

[MyDogBuster]

Ian, Max, or Mathew, can you copy the "PRPnet servers for NPLB" first post from the NPLB project over to here and obviously call it "PRPnet servers for CRUS"? Then tweak it for this project and sticky it. Port 1300 will be for S9 (misc.) efforts and port 1400 will be for the future base 2/4 even-odd-k-n effort.

Done. The URL's will have to be tweaked by whoever (Max) sets up the stats page similar to NPLB's. Send me the links and i will edit them in.

I say load away Max.

[Puzzle-Peter]

I started a sieve for R6, n=2M to 5M. Not sure if I will reserve it for LLR testing when sieving is done or maybe people will be interested in continuing this team drive? We'll see...

[gd_barnes]

Originally Posted by Puzzle-Peter

I started a sieve for R6, n=2M to 5M. Not sure if I will reserve it for LLR testing when sieving is done or maybe people will be interested in continuing this team drive? We'll see...

All of the base 6 and 16 team drives will be continued at some point. We needed to reduce the number of stickies so I unstickied them while they are in a lull. I'm sure that there will be some people who will be willing to continue testing the team drives when we promote them in the future after finishing sieving.

I am currently sieving S6 for n=1M-2M. I should be at P=72T in about a week.

[henryzz]

Originally Posted by Puzzle-Peter

I started a sieve for R6, n=2M to 5M. Not sure if I will reserve it for LLR testing when sieving is done or maybe people will be interested in continuing this team drive? We'll see...

Some of those numbers will want P-1 factoring as well as sieving. If a doublecheck is planned for this file at some point then most I think will want P-1.

[Puzzle-Peter]

Originally Posted by henryzz

Some of those numbers will want P-1 factoring as well as sieving. If a doublecheck is planned for this file at some point then most I think will want P-1.

I have reached the point where finding a factor is slower than LLRing the smallest candidates in the range. I never really got familiar with factoring so a few hints on where to start and what to do would be welcome. I have downloaded ecm and it's working on my system but what is the best way to tackle a list of candidates which has ~16K entries?

[henryzz]

With numbers this size you need to use Prime95. srfile has an option for generating pfactor lines -p. It then outputs worktodo.txt lines for Prime95.
Is this file going to be double checked at some point? If so you would want -p 2 otherwise -p 1.
How deep have you sieved?

[Puzzle-Peter]

Double check will probably happen in a very far future, if ever. I am now at p=500T.

[henryzz]

Originally Posted by Puzzle-Peter

Double check will probably happen in a very far future, if ever. I am now at p=500T.

With the same file?

I suspect you are safe just going for 1 test saved. It will raise the barrier on where you start P-1 though.

[VBCurtis]

Originally Posted by Puzzle-Peter

I have reached the point where finding a factor is slower than LLRing the smallest candidates in the range. I never really got familiar with factoring so a few hints on where to start and what to do would be welcome. I have downloaded ecm and it's working on my system but what is the best way to tackle a list of candidates which has ~16K entries?

P-1 factoring is similar to ECM, but ECM is a randomized-seed method while P-1 uses a fixed seed (so to speak). Imagine ECM as a bit like throwing darts at a dartboard, hoping to hit the spot where a factor is hiding; P-1 would be like "is there a factor hiding on the double-1 space?" There is no point in asking that question multiple times, so P-1 is run just once for a set of parameters.

P-1 runs faster then ECM, which is why it is sometimes done even when ECM is deemed inefficient. A rule of thumb is that P-1 is usually done with para meters 10 times as big as ECM parameters deemed efficient. You have trial-sieved to 500T = 5e14 = roughly half of 15-digit factors (and all smaller ones), so ECM at the 15-digit level would be redundant. ECM is traditionally (per RDS' paper on optimal ECM efforts) done in 5-digit levels, so any ECM curves would be run at 20-digit level, which is B1= 11e3. If P-1 is done at 10x the B1 of ECM, that's around 1e5 for the B1 bound on P-1.

If you get Prime95 working on this file, please post what B1 and B2 it picks, and where in the file it deems efficient to start.

[Puzzle-Peter]

I tried running P95 just to see how it works on a small number of candidates. It keeps telling me that [candidate] does not need P-1, even if I raise the trial factoring level from 48bit to 95 bit by manually editing worktodo.txt. Number of tests saved=1 or =2 does not change a thing. What's going wrong here?

[Mini-Geek]

Originally Posted by Puzzle-Peter

I tried running P95 just to see how it works on a small number of candidates. It keeps telling me that [candidate] does not need P-1, even if I raise the trial factoring level from 48bit to 95 bit by manually editing worktodo.txt. Number of tests saved=1 or =2 does not change a thing. What's going wrong here?

Raising the TF level would lower the need for P-1, not raise it (since you're reducing the range in which P-1 can find a factor to factors >95 bits, instead of all >48). Having a larger candidate would also raise the need to P-1, I believe.

Since Mersennes have the property that a factor must be of the form 2kp+1, P-1 for Mersennes has an advantage (over general form numbers) by including p in the set of primes for B1. Hence, Mersennes can benefit from P-1 much more readily than most numbers. Is it possible that the candidates/TF levels here really do mean that P-1 is not a useful thing?

You could try raising the number of tests saved to 10 or 100 (and keep the TF at a realistic level, or set it lower) to see if that makes Prime95 think P-1 is useful. If we're right near the break-even point, maybe we'll want to override Prime95's assumptions and run P-1, but it sounds entirely possible that it's made the best choice already.