Actually I think Greg's filtering jobs for NFSNET problems regularly got a max of 12 or 13 relations per cycle, and the vast majority of them finished without incident.

Is this log better?
 

It's still only

Tue Mar 15 15:44:39 2011 heaviest cycle: 13 relations

after using batalov's modification

It doesn't look much different, though the average relation-set is a little more dense. <shrug> I don't see a reason to avoid moving forward.

Indeed, just a tiny bit denser (and comparing filtering runs side-by-side, same amount of iterations of in-memory singleton removal; one would hope to lure the process into more of these by feeding it with more starting ideals but still fitting in memory? maybe the same thing with <=70 or 90 but less than 200 which is already known not to fit... or maybe gain access to a 32-48Gb memory machine for filtering only?).

The distribution of cycle lengths looks a bit better (and taller just by one). Well, it may very well work. "It's hard to make predictions, especially about the future."

Look what the Easter Bunny brought
 

[code]
Tue Mar 15 17:01:13 2011 matrix is 22118025 x 22118273 (6497.9 MB) with weight 1668868718 (75.45/col)
Tue Mar 15 17:01:13 2011 sparse part has weight 1482192101 (67.01/col)
Tue Mar 15 17:02:46 2011 commencing Lanczos iteration (4 threads)
Tue Mar 15 17:07:29 2011 checkpointing every 30000 dimensions
Sat Apr 23 18:57:31 2011 error: corrupt state, please restart from checkpoint
Sat Apr 23 19:35:10 2011 restarting at iteration 328285 (dim = 20760019)
Mon Apr 25 23:10:32 2011 lanczos halted after 349762 iterations (dim = 22118023)
Mon Apr 25 23:11:19 2011 recovered 27 nontrivial dependencies
Tue Apr 26 05:10:06 2011 sqrtTime: 21224
Tue Apr 26 05:10:06 2011 prp86 factor: 13631026947184719408017710782061183922818222029213157259955839900234789678553530335889
Tue Apr 26 05:10:06 2011 prp101 factor: 97733201303936765730020949309738330435443943197369644808690186562410926043389032719036397111868080737
[/code]

Thanks again to jrk for finding the polynomial, to bdodson for doing 70% of the sieving, to all the other sieving participants, and to jasonp and batalov for helping me out after the first matrix-job failed. Prof Wagstaff has been informed.

I did run four sqrt jobs in parallel for this one (what's the point in having a 32G machine if you don't hammer it from time to time); but by Murphy's law three of them gave factors on the first dependency.

Third time is a charm!
Congratulations! Sic itur ad astra!

So, which star next?
 

12^283+1 C194 ?
10^284+1 C200 ?

There's something appealingly round about C200, and I'll have a machine capable of doing the matrix by the time that the sieving is anywhere close to done. Will ask about ECM status and think how much I can devote to polynomial search.

[QUOTE=fivemack;236865]... but we could probably get the ... factors by Easter.[/QUOTE][QUOTE=fivemack;259628][code]Tue Apr 26 05:10:06 2011 prp86 factor: 13631026947184719408017710782061183922818222029213157259955839900234789678553530335889
Tue Apr 26 05:10:06 2011 prp101 factor: 97733201303936765730020949309738330435443943197369644808690186562410926043389032719036397111868080737
[/code][/QUOTE]Good prediction skillz. :tu: