Help resuming linear algebra step

[lavalamp]

So, my machine had been running linear algebra for six and a half hours, with only 5 minutes to go I think, "It's close enough, I'll play some counter-strike while it finishes." I load the game and ... of course I get a blue screen.

No worries, there are checkpoints right? So I run msieve -s test.dat -l test.log -i test.ini -nf test.fb -t 4 -nc2 again and:

commencing linear algebra
read 3880716 cycles
cycles contain 12919660 unique relations
read 0 relations
error: cannot locate relation 44290431

Oh.

And it wipes the test.dat.mat file for good measure. Luckily I'd made a backup of everything first.

Then I see there are TWO checkpoint files, test.dat.chk and test.dat.bak.chk. So I try using the second checkpoint file and get the same error.

Is there anything I can do to rescue the process at this point without running the entire linear algebra step again?

[EdH]

Originally Posted by lavalamp

So, my machine had been running linear algebra for six and a half hours, with only 5 minutes to go I think, "It's close enough, I'll play some counter-strike while it finishes." I load the game and ... of course I get a blue screen.

No worries, there are checkpoints right? So I run msieve -s test.dat -l test.log -i test.ini -nf test.fb -t 4 -nc2 again and:

commencing linear algebra
read 3880716 cycles
cycles contain 12919660 unique relations
read 0 relations
error: cannot locate relation 44290431

Oh.

And it wipes the test.dat.mat file for good measure. Luckily I'd made a backup of everything first.

Then I see there are TWO checkpoint files, test.dat.chk and test.dat.bak.chk. So I try using the second checkpoint file and get the same error.

Is there anything I can do to rescue the process at this point without running the entire linear algebra step again?

If you have a backup of everything, try running -ncr instead of -nc2. That is the resume command for an interrupted process. -nc2 tells it to start nc2 over from the beginning.

[lavalamp]

Originally Posted by EdH

If you have a backup of everything, try running -ncr instead of -nc2. That is the resume command for an interrupted process. -nc2 tells it to start nc2 over from the beginning.

Ah yes, that did the trick thank-you.

Seems a bit dangerous that nc2 deletes past progress then fails with an error instead of checking first to be honest. I'm very glad I made a manual backup first.

[EdH]

Originally Posted by lavalamp

Ah yes, that did the trick thank-you.

Seems a bit dangerous that nc2 deletes past progress then fails with an error instead of checking first to be honest. I'm very glad I made a manual backup first.

Glad to hear it worked. Yeah, other programs like ggnfs and YAFU refuse to run if they find partial results. I have suffered the loss you (temporarily) experienced. I try to remember the "r" these days.

[jasonp]

The code doesn't know the difference between running in a directory with checkpoint files you don't want versus continuing a job with checkpoint files you do want. It would be a nice feature to encode a fingerprint of the factorization into the temporary files generated.