3,563+ factored, NSF Teragrid grant

Message boards : News : 3,563+ factored, NSF Teragrid grant
Message board moderation

AuthorMessage
Greg

Joined: 26 Jun 08
Posts: 640
Credit: 436,677,128
RAC: 191,874
Message 579 - Posted: 17 Sep 2010, 21:33:54 UTC

3,563+ is factored. The composite cofactor was the product of 117-digit and 123-digit prime numbers. Sieving of 6,346+ and 3,607- has begun. In other news, we have received a National Science Foundation Teragrid grant of 300000 CPU-hours to further explore parallelization of the linear algebra step in postprocessing. The linear algebra of 5,409- will be run on Teragrid supercomputers beginning next week.
ID: 579 · Rating: 0 · rate: /
Speedy51

Joined: 16 Oct 09
Posts: 46
Credit: 817,222
RAC: 0
Message 585 - Posted: 21 Sep 2010, 21:56:12 UTC

This is wonderful news. 300000 CPU hours is equivalent to 1.42 years of processing time. Greg would be possible for you to let us know how long the postprocessing of linear algebra of 5,409 takes?
ID: 585 · Rating: 0 · rate: /
Greg

Joined: 26 Jun 08
Posts: 640
Credit: 436,677,128
RAC: 191,874
Message 586 - Posted: 22 Sep 2010, 5:18:13 UTC - in response to Message 585.

Actually it's a bit over 34 years. I think you may have divided by 24 twice. :) The linear algebra for 5,409- has started, but I'm going to have to divide it into multiple runs which may delay its completion somewhat. It looks like it's going to take about 31,000 CPU-hours, or about 3.5 CPU-years, to complete.
ID: 586 · Rating: 0 · rate: /
Speedy51

Joined: 16 Oct 09
Posts: 46
Credit: 817,222
RAC: 0
Message 587 - Posted: 27 Sep 2010, 6:30:08 UTC

Is the linear algebra of 5,409 to big for Teragrid to process in one go? I think I did divide number by 48 (24 twice) instead of one
ID: 587 · Rating: 0 · rate: /
Greg

Joined: 26 Jun 08
Posts: 640
Credit: 436,677,128
RAC: 191,874
Message 588 - Posted: 27 Sep 2010, 6:57:35 UTC - in response to Message 587.

No, I'm just finishing off a bit of time I had remaining in another account, and the calculation runs a bit more efficiently if I use fewer processors. If I were not worried about efficiency, then Teragrid has more than sufficient processors to finish the entire calculation in two days.
ID: 588 · Rating: 0 · rate: /
Speedy51

Joined: 16 Oct 09
Posts: 46
Credit: 817,222
RAC: 0
Message 589 - Posted: 27 Sep 2010, 21:56:35 UTC

Thanks for explaining. I have 1 more question. How dose the information get to the people that are doing the post processing, do you ship the data on disk of is it sent via a high speed data pike?
ID: 589 · Rating: 0 · rate: /
Greg

Joined: 26 Jun 08
Posts: 640
Credit: 436,677,128
RAC: 191,874
Message 590 - Posted: 28 Sep 2010, 1:27:31 UTC - in response to Message 589.

Solving the matrix is the time-consuming part of the post-processing. I process the huge relations file (up to 100 GB for the largest numbers) and generate the matrix locally. The matrix is contained in a 5 - 8 GB file, which I then SFTP to the supercomputer. Transfers from Cal State Fullerton to the supercomputer sites over the internet usually run at 10-15 MB/s, so this usually completes in under 15 minutes. Once the matrix is solved, I transfer the solutions back here and run the square roots (which require the relations file) locally. Through this process, I minimize moving the relations file around.
ID: 590 · Rating: 0 · rate: /
Speedy51

Joined: 16 Oct 09
Posts: 46
Credit: 817,222
RAC: 0
Message 596 - Posted: 29 Sep 2010, 4:51:10 UTC