20150127, 16:36  #1 
Basketry That Evening!
"Bunslow the Bold"
Jun 2011
40<A<43 89<O<88
3×29×83 Posts 
Ideas for the future beyond justkeepencrunching
If people really wanted a full on challenge.... one possible idea is attempting to try our collective hand at the multiple (S)NFS factory technique.
Considering the amount of data involved (not to mention the massive linear algebra involved), it may not necessarily be a good idea to numbers quite as large as the original paper, but I imagine there's some other Cunningham tables that have several holes with smaller size numbers compared to 2? Or is this a complete pipe dream? I'm not really sure. 
20150127, 17:10  #2  
Basketry That Evening!
"Bunslow the Bold"
Jun 2011
40<A<43 89<O<88
3×29×83 Posts 
Quote:
My original train of thought was, as per other discussion, "the challenge should really be 16e (not that they care of course, but logistically for us, increased sieving for a challenge would only further strain 14e and 15e)... what's a good challenge for 16e?" which led to what I suggested. Edit: It arguably has several other merits as well. For one, it would be a true return to mathematical research, which arguably NFS@Home hasn't really been doing of late (the LA hasn't really been pushing any boundaries). Additionally, it could be publicized thusly "not just another BOINC challenge, actual, specific new research, and we need your help!" Of course, we'd need several forumites with a suitably strong background to lead the effort, not to mention frmky and resources for linear algebra, but it is an idea. Last fiddled with by Dubslow on 20150127 at 17:47 

20150129, 17:55  #3  
Bamboozled!
"πΊππ·π·π"
May 2003
Down not across
2^{5}·331 Posts 
Quote:
I've looked into the idea. Here are some of the things to consider.
So, not impossible, but certainly nontrivial. It's easy to see how some portions could be parallelized: ECM and sieving over BOINC); data collection at multiple sites terminated with a significant sneakernet transfer to a central site; a management team spread over several sites in several time zones. The LA is certainly nontrivial. We could perhaps ask for favours but no guarantees could be made. If a somewhat less trivial challenge is required, why not work out how to run blocked Wiedemann over BOINC? That would definitely be of research interest and of practical use in the future. A substantially less trivial challenge woud be to run polynomial selection for RSA1024 over BOINC. Something like a thousand GPUyears should be enough. 

20150130, 05:10  #4 
Sep 2008
Kansas
2^{2}·821 Posts 

20150130, 09:06  #5 
Bamboozled!
"πΊππ·π·π"
May 2003
Down not across
2^{5}·331 Posts 
So do I, which is why I suggested them ...
To be serious, the BOINC block Wiedemann idea is far better in my view, as long as it is possible to do with decent efficiency, both computational and noncomputational. It is very far from clear how to do that. It would create a tool with longlasting value whereas finding polynomials only allows the factorization of a single integer which, although giving bragging rights within a small community, is unlikely to be of any greater significance. Last fiddled with by xilman on 20150130 at 09:12 Reason: Add proviso 
20150130, 21:01  #6 
Jul 2003
So Cal
2^{2}·11·47 Posts 
I agree that BOINC block Wiedemann is very worthy of serious exploration.

20150130, 21:52  #7 
"Carlos Pinho"
Oct 2011
Milton Keynes, UK
5·7·139 Posts 

20150130, 22:04  #8 
Bamboozled!
"πΊππ·π·π"
May 2003
Down not across
2^{5}×331 Posts 
In that case, I suggest
a) that you read up on BW; b) get hold of the current cadonfs code base and investigate the bwc directory in some detail c) suggest how the BerlekampMassey implementation can be modified so that the other two phases can be split over a significantly large number of systems. 
20150131, 02:59  #9 
Sep 2008
Kansas
2^{2}×821 Posts 
Found this publication interesting (I just read the summary), using CUDA and implementing BW on a GPU cluster.
http://onlinelibrary.wiley.com/doi/1....2896/abstract 
20150131, 03:10  #10 
Tribal Bullet
Oct 2004
2×3×19×31 Posts 
A nice paper; they modified cadonfs :)

20150201, 09:31  #11  
Just call me Henry
"David"
Sep 2007
Cambridge (GMT/BST)
5,821 Posts 
Quote:


Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Some ideas regarding NFS...  paul0  Factoring  3  20150314 19:55 
I broke ECMFermat...or my PC IDEAS?  petrw1  PrimeNet  6  20130427 22:55 
two ideas for NPLB  MiniGeek  No Prime Left Behind  16  20080301 23:32 
GROUP IDEAS  TTn  15k Search  15  20030923 16:28 
Domain name ideas...  Xyzzy  Lounge  17  20030324 16:20 