Number of relations from ggnfs

[Greebley]

I was examining the number of relations needed by choosing random ggnfs runs with a low min_rels (default 0.2). This is with the current ggnfs and msieve (1.49). (Note I used aliqueit to call msieve/ggnfs automatically)

My thought was to try to improve on the value of MIN_RELS.

One interesting thing is the jump in relations needed between 97 and 98. This is when the LPBA/R values switches.

To explain this graph, the Min value is the value where we didn't have enough relations, and the Max value is the value where we did have enough. It is more than one run combined, so the Min can be higher than the Max when one run didn't have enough for a higher value than another one did.

http://i322.photobucket.com/albums/n...fs_Min_Max.jpg

What I think the jump means is that the value for LPBA/R switches too soon. Below are the values from def-par.txt used for ggnfs, and the key value is the 25,25 at 95 (the second entry) becoming 26,26 at 100. However, I am not positive on this. What do you think? Could we lessen the number of relations in the 98-101? range by lowering the 26 to 25 for the 100 entry (try the 103)? I am unsure what would happen if the LPBA/R which this is based on was too small. Will it break things or will the number of relations needed just get big? My intuition is the latter so that if we picked the values perfectly there would be no jumps.

Any thoughts? Note that from the graph, I don't yet know if there is a problem between 110 and 112 when we go to 27. I will hopefully have more values past 112 as I am increasing the size of tests (doing smaller first).

This code is from the def-par.txt for ggnfs

gnfs,90,4,58,2000,2.5e-4,0.30,220,15,10000,600,700000,700000,25,25,44,44,2.4,2.4,40000,2000000,200
gnfs,95,4,60,2000,1.0e-4,0.30,220,15,10000,600,1200000,1200000,25,25,45,45,2.4,2.4,60000,2000000,200
gnfs,100,5,58,1500,3.0e-3,0.4,220,15,10000,2000,1800000,1800000,26,26,48,48,2.5,2.5,100000,4000000,300
gnfs,103,5,59,2000,9.0e-4,0.35,200,15,15000,2000,2300000,2300000,26,26,49,49,2.6,2.6,100000,4000000,300
gnfs,106,5,59,2000,6.0e-4,0.25,200,15,15000,2000,2500000,2500000,26,26,49,49,2.6,2.6,150000,4000000,300
gnfs,110,5,61,2000,1.5e-4,0.3,250,15,50000,2400,3200000,3200000,27,27,50,50,2.6,2.6,100000,4000000,300
gnfs,112,5,61,2000,1.6e-4,0.25,250,15,50000,2800,3500000,3500000,27,27,50,50,2.6,2.6,100000,4000000,300
gnfs,118,5,63,2000,2.6e-5,0.28,250,20,50000,3600,4500000,4500000,27,27,50,50,2.4,2.4,60000,4000000,300

[Greebley]

I can also post the values when I am done if anyone is interested.

Given the number of digits, I am figuring out the smallest value such that there was enough relations and the largest value such that it wasn't enough relations.

I am hoping with 1.49 I won't get the broken case where the number of relations is too small so the linear algebra fails and so ggnfs fails.

[science_man_88]

Originally Posted by Greebley

I can also post the values when I am done if anyone is interested.

Given the number of digits, I am figuring out the smallest value such that there was enough relations and the largest value such that it wasn't enough relations.

I am hoping with 1.49 I won't get the broken case where the number of relations is too small so the linear algebra fails and so ggnfs fails.

there's a thread estimating minimum relation I currently have 34 test.n files in my folder now (admittedly no log files but I didn't change anything I'm not smart enough to)