Detailed status of lasievef

Message boards : Questions/Problems/Bugs : Detailed status of lasievef

Author Message
Martin Droppa

Joined: 28 Oct 09
Posts: 3
Credit: 780,178
RAC: 2
Message 1463 - Posted: 30 Dec 2014, 20:32:37 UTC

On main page are detailed status of lasieved; lasievee. Why there isnĀ“t detailed status of lasievef?
Thanks, Martin
____________

Carlos Pinho [TSBTs Pirate]
Volunteer moderator

Joined: 26 Sep 09
Posts: 163
Credit: 7,748,481
RAC: 756
Message 1464 - Posted: 30 Dec 2014, 22:54:04 UTC

Yes you have but not as detailed as those ones. Please check http://escatter11.fullerton.edu/nfs/numbers.html.

Martin Droppa

Joined: 28 Oct 09
Posts: 3
Credit: 780,178
RAC: 2
Message 1465 - Posted: 31 Dec 2014, 7:13:35 UTC - in response to Message 1464.

Thanks. But how many taska must be still done, that 3,697+ will be done? Where could I that find out?
Martin

Carlos Pinho [TSBTs Pirate]
Volunteer moderator

Joined: 26 Sep 09
Posts: 163
Credit: 7,748,481
RAC: 756
Message 1466 - Posted: 31 Dec 2014, 13:35:10 UTC - in response to Message 1465.

Martin,

I think 3,697+ is sieved, Greg must be doing the post-processing on a cluster but it takes a lot of time until the factorization (linear algebra stage)is done. The filtering phase can easily fail if not enough relations were gathered (matrix fail, sieve not enough).

It is hard to predict, as for the other small numbers, how much sieve and relations are needed for these big tasks. Don't forget that NFS@Home is running for records so entering in an unknown field. We can set the sieve region, but we can't determine if enough relations were gathered to build a matrix and keep on the factorization.

What Greg is doing, and he can correct if I am wrong, is to set first the Q range and then keep increasing it as progress goes on and filtering phase is tested on a cluster. If it fails more relations are needed and therefore more wu's are created. Each wu manages 2k Q values.

Best regards,

Carlos

Martin Droppa

Joined: 28 Oct 09
Posts: 3
Credit: 780,178
RAC: 2
Message 1467 - Posted: 1 Jan 2015, 0:39:09 UTC - in response to Message 1466.

Carlos,

Best regards,
Martin

Carlos Pinho [TSBTs Pirate]
Volunteer moderator

Joined: 26 Sep 09
Posts: 163
Credit: 7,748,481
RAC: 756
Message 1468 - Posted: 1 Jan 2015, 11:36:20 UTC

Martin,

Just a few numbers so you can think and be more clear about sieving.
Example, 3,766+ was sieved from q=230M to q=2600M, meaning it was needed 1.185M wu's. M means million.
Log file of the post-processing can be found here:

http://www.mersenneforum.org/showpost.php?p=370115&postcount=128

So filtering took 51 hours and can only be run on a single core. 660727075 unique relations were found by sieving. Then LA phase was run with "MPI process 0 of 256" and "initialized process (0,0) of 16 x 16 grid" with "commencing Lanczos iteration (8 threads)". This means it was run on a big cluster with 256 nodes each one having 8 cores, totalizing 2048 cores. Each node ran for ~100-120 hours.

Square root phase took 186 hours and can only be run on a single core.

Cheers,

Carlos

Ken_g6

Joined: 6 Jan 10
Posts: 1
Credit: 1,352,988
RAC: 2,485
Message 1536 - Posted: 26 May 2015, 19:30:28 UTC

That page also doesn't tell me what my WU is. In this case, what's a G10m359? You're not trying to factor M359, are you? (I assume not.)

Carlos Pinho [TSBTs Pirate]
Volunteer moderator

Joined: 26 Sep 09
Posts: 163
Credit: 7,748,481
RAC: 756
Message 1537 - Posted: 27 May 2015, 9:42:10 UTC - in response to Message 1536.