BOINC NFS sieving - NFS@Home

[jasonp]

I don't understand it either. I used to think the failure happens because the filtering throws away information at a steady rate during the merge phase, and making the target density too large would make the filtering throw away too much. But then Greg tried a huge filtering run with high target density and nothing thrown away, and it still failed.

My current guess is that either it's a bug somewhere that destroys the matrix during the merge phase, or that gauss elimination somehow starts to behave strangely when many matrix columns start to look almost identical.

[Jarod]

Can somebody please advise me what I need to do to continue the LA phase. I get the following lines just before the command prompt exits

10+ relations: 2553021
 heaviest cycle: 28 relations
RelProcTime: 3868
elapsed time 01:04:30

please find log start and resume files attached
I am using target_density = 112 with out the spaces of course. Thanks for any assistance

[frmky]

Originally Posted by Speedy51

Can somebody please advise me what I need to do to continue the LA phase.

Looks like you've finished -nc1, which runs filtering. Now run with
-nc2 -nc3 -t 4 -v
which will run the linear algebra and square roots. Replace the 4 with however many cores your computer has.

[Jarod]

Originally Posted by frmky

Looks like you've finished -nc1, which runs filtering. Now run with
-nc2 -nc3 -t 4 -v
which will run the linear algebra and square roots. Replace the 4 with however many cores your computer has.

Thanks I restarted – nc 2 and all appears well. A quick question will I need to run start command again at some point with nc 3?
The la phase is going to take around 46 hours I should have it completed within the week. It is currently using 6 cores after it has written a checkpoint for the first time I will see if it will work faster on 3 cores

[VBCurtis]

Originally Posted by Speedy51

Thanks I restarted – nc 2 and all appears well. A quick question will I need to run start command again at some point with nc 3?
The la phase is going to take around 46 hours I should have it completed within the week. It is currently using 6 cores after it has written a checkpoint for the first time I will see if it will work faster on 3 cores

Frmky meant for you to run -nc2 -nc3 in the same command. You don't have to run them individually. If you only flagged -nc2, you will have to invoke msieve with -nc3 once the matrix finishes.

If you don't have a reason to run them individually, use -nc next time; that runs all three post-processing phases sequentially.

[fivemack]

Tue Mar 18 17:22:30 2014  prp96 factor: 513771344530953638399207528250780162329591444656481384108792960998927618778379458678020370035651
Tue Mar 18 17:22:30 2014  prp96 factor: 704336210027665042766919139818581008384098536410985081125139978424946084001665660346633709606967

144 hours for 11.9M matrix on three cores of i7/2600K. Log at http://pastebin.com/g3UqPrCi

[fivemack]

I'm afraid my machine won't finish F1893 until after I leave for Mexico; depending on the availability of wifi in Palenque and Merida, the results may not reach the Internet until April 7th or so.

[fivemack]

Originally Posted by VBCurtis

Frmky meant for you to run -nc2 -nc3 in the same command. You don't have to run them individually. If you only flagged -nc2, you will have to invoke msieve with -nc3 once the matrix finishes.

If you don't have a reason to run them individually, use -nc next time; that runs all three post-processing phases sequentially.

There is one oddity: you have to do '-v -nc target_density=112 -t 6' ... the target_density parameter must be right after the -nc on the command line, or it will be ignored.

[RichD]

Originally Posted by RichD

I can start the REAL GW_8_262 download tomorrow.

I've got GW_8_262 running in LA with ETA 38 hrs + sqrt.

Sorry everyone for the confusion.

[swellman]

Originally Posted by fivemack

I tend to use target_density 112, if that doesn't work then 96, if that doesn't work then the default 70. The difference between 112 working and 70 working is often only about 5% of the total relation count.

What I don't quite understand is why I don't get a usable matrix, even with enormous over-sieving, at target densities 128 or over.

Thanks for the guidance. C176_118_93 seems to be heavily over-sieved, or will be. I might try a target density of 128 just to experience the crash if nothing else, falling back to 112, 96, or 70 as needed.

[RichD]

Finally, GW_8_262 splits as:

prp79 factor: 6482638054086919779081042827196458878063715247525998104955700869682305142904453
prp100 factor: 1728418981523664944606126436837534164627159260942767750412042635120454352600710802892574502352523633

40 hrs to solve 7.65M matrix using target_density=116 -t 4 on Core-i5/2500