Thanks, smh.

Unfortunately, msieve came up with a warning saying 'can't find these relations' or something like that (which I anticipated, because I copied some of the original file to speed things up), and went on to the next number in worktodo.ini! This wiped out 85Mb of relations, and half of my time :cry:

Given that the time spent so far has been over 120 hours, I'm not thrilled.

Thanks for helping me out with the files problem. Now if only I wasn't so new to this I could have gotten rid of the second number in worktodo.ini!

I am very grateful to msieve, for all my lack of experience with it, however, and indebted to those on this forum who have helped me.

Thanks everyone!

(Now I'll get back to seiving...) :sad:

[QUOTE=roger;114469]This wiped out 85Mb of relations, and half of my time :cry:

Given that the time spent so far has been over 120 hours, I'm not thrilled.
[/QUOTE]
Better this happened now than later; if you're using NFS, this is less than 10% of the relations that a C114 would need.

Actually, I'm not using NFS.

The 85Mb (about 19k relations) is approximately 6.5%, but with the other .dat file, I had around 14% of the relations needed (not including the ones that would be combined from the 2M partial relations).

So in total, I lost about 60 hours and maybe as much as 12% of the total factorization. But I agree. Better it happen now, then when I have a really big factorization to do.

Roger

[QUOTE]Then what method are you using?[/QUOTE]

The command is :

msieve -nf C114.dat -s C114.dat -v

and it reads form the worktodo.ini file without having to be specified. Myself, I don't know the methods of msieve, only that it works :rolleyes:

I'm not typing -n for the NFS option, so i don't think I'm using it, and the amount of relations needed would I believe be much higher (millions at least, compared to my 300k)

Roger

Yes I did manually copy .dat files to speed up the seiving (I thought if I copied the savefiles, and continued using them, then combined them later, the duplicate relations wouldn't matter much).

Now, I have combined files with hundreds of thousands (!) of duplicate relations.

Should I just start over again, instead using only two savefiles that were never copied, and combine them later? Or continue trying with the duplicates?

Thanks for your time,

Roger

[QUOTE=roger;114573]Yes I did manually copy .dat files to speed up the seiving (I thought if I copied the savefiles, and continued using them, then combined them later, the duplicate relations wouldn't matter much).

Now, I have combined files with hundreds of thousands (!) of duplicate relations.

Should I just start over again, instead using only two savefiles that were never copied, and combine them later? Or continue trying with the duplicates?
[/QUOTE]
For the record, the Readme in the msieve source code has instructions for distributed sieving.

The easiest thing to do is to start from two savefiles that were never copied, then combine them later ([i]only once[/i]). The format of QS relations is stateful, which means you can't just sort the file and delete duplicate lines.

[QUOTE]The easiest thing to do is to start from two savefiles that were never copied, then combine them later (only once). [/QUOTE]

That's what I should have done in the first place.

Thanks for the help, I won't be making these mistakes again!

Is there a 164 digit limit on the number to factor?

I've sieved a 171 digit number (SNFS diff. 193) but can't get msieve to do the post processing.

Is there an easy way around this limitation?

[QUOTE=smh;114939]
I've sieved a 171 digit number (SNFS diff. 193) but can't get msieve to do the post processing.
[/QUOTE]
I'll recompile the application with a larger limit and send via email.

I have also grabbed the binary from the website forgetting that has a default upper limit of 164 digits. Would increasing MAX_MP_WORDS from 17 to 26 (enough for 250 digit inputs) slow down the MPQS runs or significantly increase memory use? If not, it would be convenient not to have to edit the source and recompile. There could still be a stern warning about inputs above 164 digits that we can ignore when completing SNFS runs, or even an error as now with a command line parameter that overrides the check on the size of the inputs.

Thanks,
Greg

[QUOTE=jasonp;115491]Now available. Highlights include:

- added memory and performance optimizations to the linear algebra code. The result should use 15% less memory and should be slightly (~2%) faster[/QUOTE]

I ran the post processing of the factorization of Cullen 641 with both Msieve 1.26 and msieve 1.27. See the attached complete logs.

[CODE]Msieve v. 1.26
Mon Oct 01 20:46:14 2007 commencing linear algebra
Mon Oct 01 20:46:15 2007 factor base loaded:
Mon Oct 01 20:46:15 2007 910077 rational ideals (max prime = 13999981)
Mon Oct 01 20:46:15 2007 909524 algebraic ideals (max prime = 13999969)
Mon Oct 01 20:46:18 2007 read 1954021 cycles
Mon Oct 01 20:46:28 2007 cycles contain 5149943 unique relations
Mon Oct 01 20:49:29 2007 read 5149943 relations
Mon Oct 01 20:49:46 2007 using 32 quadratic characters above 536861292
Mon Oct 01 20:58:11 2007 matrix is 1952700 x 1954021 with weight 173050027 (avg 88.56/col)
Mon Oct 01 20:59:29 2007 filtering completed in 3 passes
Mon Oct 01 20:59:30 2007 matrix is 1940238 x 1940438 with weight 172117384 (avg 88.70/col)
Mon Oct 01 20:59:58 2007 saving the first 48 matrix rows for later
Mon Oct 01 21:00:01 2007 matrix is 1940190 x 1940438 with weight 130122954 (avg 67.06/col)
Mon Oct 01 21:00:01 2007 matrix includes 64 packed rows
Mon Oct 01 21:00:01 2007 using block size 65536 for processor cache size 2048 kB
Mon Oct 01 21:00:14 2007 commencing Lanczos iteration (2 threads)
Tue Oct 02 12:00:36 2007 lanczos halted after 30684 iterations
Tue Oct 02 12:00:47 2007 recovered 48 nontrivial dependencies
Tue Oct 02 12:00:47 2007
Tue Oct 02 12:00:47 2007 commencing square root phase[/CODE]~15:14
[CODE]Msieve v. 1.27
Tue Oct 02 14:42:39 2007 commencing linear algebra
Tue Oct 02 14:42:39 2007 generating factor base
Tue Oct 02 14:42:55 2007 factor base complete:
Tue Oct 02 14:42:55 2007 910077 rational roots (max prime = 13999981)
Tue Oct 02 14:42:55 2007 909524 algebraic roots (max prime = 13999969)
Tue Oct 02 14:43:02 2007 read 1954021 cycles
Tue Oct 02 14:43:11 2007 cycles contain 5149943 unique relations
Tue Oct 02 14:46:25 2007 read 5149943 relations
Tue Oct 02 14:46:44 2007 using 32 quadratic characters above 536861292
Tue Oct 02 14:55:31 2007 matrix is 1952700 x 1954021 with weight 173050027 (avg 88.56/col)
Tue Oct 02 14:56:53 2007 filtering completed in 3 passes
Tue Oct 02 14:56:54 2007 matrix is 1940238 x 1940438 with weight 172117384 (avg 88.70/col)
Tue Oct 02 14:57:24 2007 saving the first 48 matrix rows for later
Tue Oct 02 14:57:27 2007 matrix is 1940190 x 1940438 with weight 130122954 (avg 67.06/col)
Tue Oct 02 14:57:27 2007 matrix includes 64 packed rows
Tue Oct 02 14:57:27 2007 using block size 65536 for processor cache size 2048 kB
Tue Oct 02 14:57:57 2007 commencing Lanczos iteration (2 threads)
Wed Oct 03 05:37:49 2007 lanczos halted after 30680 iterations
Wed Oct 03 05:37:59 2007 recovered 48 nontrivial dependencies
Wed Oct 03 05:38:00 2007
Wed Oct 03 05:38:00 2007 commencing square root phase[/CODE]
~14:56

Exactly 2% faster :-)