[QUOTE=akruppa]
can you let the ECMNET server hand out only 19^193-1 assignments
[/QUOTE]
It's been done. This is sooner than I was expecting!

William

I think it would be best if Alex could coordinate the SNFS sieving with the people who were trying to factor 5^349-1, in order to make this stage faster.

>Are you going to attempt the 5 roadblock?

Yes.

>If so, I'll stop my ECM work (which is not assigned by William).

That would be nice.

Thanks,
Alex

>I think it would be best if Alex could coordinate the SNFS sieving with the people who were
>trying to factor 5^349-1, in order to make this stage faster.

Part of the reason I'm taking on this numbers is that I wanted something that lets me run the sievers unattended for a while. Sieving is progressing nicely already and won't take terribly long - there are a couple of Athlons at my University I can use. I have little personal time at the moment and could not coordinate distributed sieving - if I may, I would like to attempt this one on my own. If there is interest (and no ECM factor), we could do the 19 roadblock with voluteer sievers later on.

Alex

What resources do you estimate you will need to do the linear algebra?

The size of the number is very close to 2,811- that NFSNET did (very slightly smaller) so I expect I will get a matrix of similar size. In [URL=http://www.mersenneforum.org/showthread.php?t=2364]this thread[/URL], xilman states the size of the (properly merged) matrix as (9.5M)^2 with 670M non-zero entries. This means the matrix could fit into 2.5GB of memory and could be done on a single-cpu 32-bit machine, if need be - it would take forever, though.

I plan to oversieve a lot more than NFSNET did, so I should get a smaller and less heavy matrix. I'm not yet entirely sure where I'll solve it - maybe I could use a few Athlon XP at the chair for Efficient Algorithms (the same ones that are doing the sieving right now) with 100MB Ethernet, but there may be a much better alternative.

I've often heard about a Opteron cluster with dual-cpu nodes and Infiniband interconnects at our department, which would be ideal for the matrix jobs. I'll need to talk to the responsible ones and ask if I can use that, but supposedly the cluster is not utilised too much, so I'm having high hopes there. I'd then still have to get the parallel version of CWI's block lanczos, and compile and test it on the cluster.

My current order of preference for the linear algebra is: 1. that Opteron cluster, 2. the Efficient Algorithms Athlons, 3. go begging CWI or Jens Franke for help, 4. buy ram, stick it in a single machine and wait for it.

Alex

Progress
 

Sieving is about 50% done now. Note, I had started sieving quite a while before officially reserving the number.

Alex

Have you sorted out where you're going to be doing the algebra?

No, not yet. I have an exam on October 2nd, studying for that is what I focus on most of the time now. Sieving will take until well after that date so I'll go looking for suitable hardware sometime in October.

Alex

Alex,

How was your exam?

Hope it went well,
Pace

I, ummm, kinda misremembered the date :alex: it is on the 10th in fact - monday.

Alex