Running GGNFS

[Jeff Gilchrist]

It seems some kind of memory access violation. It has been a while since I have compiled the binaries so I think I should get the latest SVN code and recompile with MPIR. Then we can try those binaries to see if there are still any issues as the bug might have been fixed along the way.

At the same time I will see if I can get an 11e siever going for Windows as well.

[ltd]

I will run the one that crashes for you also. My compile uses the latest SVN.

[ltd]

Again no problem:

N=2442895043335315532241384991110259348565485108389595693855451132244907689766065727669961
  ( 88 digits)
Divisors found:
 r1=13791603738437003037101462071133331419775323 (pp44)
 r2=177129149710631684323603017913556944491924907 (pp45)

So it is either the latest SVN of ggnfs or the SVN version of MPIR that solves the problem.

[siew]

please, advice me

i made patch for factMsieve.pl for computing the pol51m0b.exe and opt in 4 threads.

i receiving 4 file groups, different for each thread.

How can i combine results into one file for msieve? Please, can anyone help.
Do i need to remove some duplicates?

tanx!

[jasonp]

Msieve cannot currently turn the output of pol51m0b into usable GNFS polynomials; it currently has to do the work equivalent to pol51m0b by itself.

You need to combine the files with 'cat' or 'type', then run pol51opt on the result to get a polynomial to use for the run.

[siew]

Yes, i understand, tanx, i

it works like this:

-> Searching leading coefficients from 29844001 to 29845000.
=>  "pol51m0b.exe" -b example.polsel.thread01 -v -v -p 7 -n 4.86E+023 -a 29844 -A 29845 > example.polsel.thread01.log
=>  "pol51opt.exe" -b example.polsel.thread01 -v -v -n 6.13E+021 -N 1.48E+019 -e 2.59E-012 > example.polsel.thread01.log

-> Searching leading coefficients from 29845001 to 29846000.
=>  "pol51m0b.exe" -b example.polsel.thread02 -v -v -p 7 -n 4.86E+023 -a 29845 -A 29846 > example.polsel.thread02.log
=>  "pol51opt.exe" -b example.polsel.thread02 -v -v -n 6.13E+021 -N 1.48E+019 -e 2.59E-012 > example.polsel.thread02.log

-> Searching leading coefficients from 29846001 to 29847000.
=>  "pol51m0b.exe" -b example.polsel.thread03 -v -v -p 7 -n 4.86E+023 -a 29846 -A 29847 > example.polsel.thread03.log
=>  "pol51opt.exe" -b example.polsel.thread03 -v -v -n 6.13E+021 -N 1.48E+019 -e 2.59E-012 > example.polsel.thread03.log

....

at the end i have 4 files which i can append to each other and use as datasource for msieve.exe, am i right?

[jasonp]

The sieving requires you to choose one polynomial (manually) out of the total collection of polynomials you have. You want the polynomial with the largest Murphy E value; the script finds it for you, but I don't know how it does so, or whether you can resume the factorization and have the script start by finishing the poly selection and proceeding to the sieving.

Msieve will not do that by itself, and you really want to use the GGNFS lattice sieve anyway.

[siew]

tanx for the help, seems all ok:

now processing the following:

Selected default factorization parameters for 140 digit level.
Selected lattice siever: gnfs-lasieve4I14e.exe
Creating param file to detect parameter changes...
Q0=9000000, QSTEP=100000.
makeJobFile(): q0=9000000, q1=9100000.
makeJobFile(): Adjusted to q0=9000000, q1=9100000.
Lattice sieving algebraic q-values from q=9000000 to 9100000.
FBsize 603523+0 (deg 5), 1151366+0 (deg 1)

speed is around 0.42 sec/rel

please, how to estimate time for finish this stage?

tanx

[siew]

For my PC it taked 55 minutes for find 40 000 relations, so for

Found 41659 relations, need at least 5532435 to proceed

5 532 432 / 40 000 * 50 min ~ 138 ~ 5.6 days to finnd necessary amount of relations.

Am i right?

[Mini-Geek]

Originally Posted by siew

For my PC it taked 55 minutes for find 40 000 relations, so for

Found 41659 relations, need at least 5532435 to proceed

5 532 432 / 40 000 * 50 min ~ 138 ~ 5.6 days to finnd necessary amount of relations.

Am i right?

Well, not quite.
I'm pretty sure the number of relations found per range will peak a little after the beginning q and slowly drop off, so the average will likely be lower than 40,000.
But let's just say that it does stay at 40,000. So you'll need to have 5,532,432/40,000=138.3108 runs, but since you'll need to finish the 139th run, the total is 139. 139 runs * 50 min per run = 6950 minutes, or about 4.8 days.
However, that's how many relations you'll need to find before the script attempts the first filtering run. If there isn't enough relations, it will return to sieving, then try again when you finish the next set of sieving. I see that it selected default parameters for the 140 digit level. A recent factorization of a 141-digit number of general form took almost 25 million relations. Factorizations this large take quite a while, especially if it's on a single computer, and especially if it's only using one core. With collaboration from several volunteers, that number took 5 days to factor. I wouldn't be surprised if this takes you a month or more. (don't forget that you'll need a machine with plenty of RAM to finish the factorization)

Have you done sufficient ECM and poly selection first so that you aren't just wasting your time by doing an unneeded or inefficient GNFS?

[siew]

skew: 412107.50
# norm 3.64e+020
c5: 8991060
c4: -12127640856442
c3: -4571538298609334349
c2: 1921334488915333355742087
c1: 354841807933800473017446092705
c0: 15303903753718196025579486085196571
# alpha -4.79
Y1: 718317004608979589
Y0: -239659686492979025097250976612

# Murphy_E 2.72e-012, needed was 2.19e-012
type: gnfs
rlim: 18000000
alim: 18000000
lpbr: 28
lpba: 28
mfbr: 52
mfba: 52
rlambda: 2.6
alambda: 2.6

now i'm using 4 cores at 3.5 GHz
Original N is 154 digits

selection of Poly was done at 24 hours in 4 threads too.

i have 16 Gbytes of RAM and 4 TB HDD, so think will enough for this task, also i want to divide task to 3 other Q6600 PC for sieving.