Linear algebra reservations, progress and results

[swellman]

Originally Posted by Xyzzy

We hesitate to attempt other work until someone else tries this job. Either our method/computer/whatever is wrong or the data is bad.

Odd that. Did you get a plethora of errors during nc1? The fact that you got into nc2 implies a feasible matrix was built, but maybe something went horribly wrong.

What was the command line you used to start this job after the dataset finished downloading? Did you have to make several attempts using various values of target_density to get into nc2? Posting the log file might be useful, assuming it’s not too huge to post without editing out a lot of error messages.

[Xyzzy]

Originally Posted by swellman

Odd that. Did you get a plethora of errors during nc1?

No errors other than it rejecting a few bad relations.

Originally Posted by swellman

What was the command line you used to start this job after the dataset finished downloading?

nice ./msieve -v -nc target_density=120 -t 2

Originally Posted by swellman

Did you have to make several attempts using various values of target_density to get into nc2?

No, 120 worked. We did try 110 as well.

Originally Posted by swellman

Posting the log file might be useful, assuming it’s not too huge to post without editing out a lot of error messages.

One of the log files is here: http://www.mersenneforum.org/showthread.php?p=476269

[swellman]

I’ve begun downloading this job on an attempt to duplicate Xyzzy’s issues, or punch through the factorization with a 32Gb machine. Will also first attempt to build the matrix @TD=130.

[VBCurtis]

I think the part of xyxxy's log in an above post indicates the matrix was not at the density requested; this is likely oversieved. Reducing input relations a bit should help with the matrix.

[Xyzzy]

Originally Posted by VBCurtis

I think the part of xyxxy's log in an above post indicates the matrix was not at the density requested; this is likely oversieved. Reducing input relations a bit should help with the matrix.

If this is the case, can you just chop out a bunch of relations or do you have to remove certain ones?

What line from the log file gave you this clue?

[swellman]

I’ve built a matrix @TD=134, though it’s too early to know if it’s stable. 9.2 Mb matrix, which is currently 0.7% complete and has an ETA of 342 hours using 4 cores +4HT on an old Asus gaming machine with 32 Gb memory.

[VBCurtis]

Originally Posted by Xyzzy

If this is the case, can you just chop out a bunch of relations or do you have to remove certain ones?

What line from the log file gave you this clue?

You said you set TD to 120, but the snippet of the log you posted (in the other thread, about errors) showed a density of 72 or 74. msieve has a flag to truncate the number of relations, called filter_maxrels.

[fivemack]

Mon Jan  8 03:52:05 2018  p75 factor: 801555151960799518693380147420555646755304231564827417719955137744350336133
Mon Jan  8 03:52:05 2018  p162 factor: 122488940752089154227459526241667811037638237196160215174532126697824168900400497894701744177033066245529267216221847139479978147962889684942895545571155822482249

About 432 hours on 7 cores E5-2650v2 for a 27.9M density-136 (138 didn't work) matrix. Log attached and at https://pastebin.com/6xiuSfNg

[richs]

Reserving phi_13(phi_5(phi_7(13))/151*151841) cofactor in 14d, please.

fivemack: the directory name is C164_32607907713428723311_13

[VBCurtis]

Reserving 13_2_836m when ready.

[swellman]

Reserving C231_143_54 for postprocessing, thank you.