L2605A results
 

[code]
Tue Dec 25 12:03:22 2012 prp59 factor: 31555613270751818364644721374370821508605167782878068098001
Tue Dec 25 12:03:22 2012 prp116 factor: 11035950277565764539587430093466269078344279095605067665849536916334366226728444521184440488703339995545570590923051
[/code]

Nine hours to download 15648MB of gzipped relations, 7.5 hours for cycle formulation, 70 mins for matrix preparation, 30 hours x 24 CPUs for Lanczos, 70 mins x 8 CPUs for square root. Matrix was produced with target_density=110 and was

7762611 x 7762788 (3336.4 MB) with weight 959463017 (123.60/col).

I'll do a target_density=70 filtration overnight and see how the Lanczos ETA compares; this was a thoroughly oversieved example.

[code]prp53 factor: 59122173114692957881233579305551458987234232111538323
prp108 factor: 821940190331419058367581972388250175903231973889848170934645125119484232268565543468934179860153660230245341[/code]

[B]853_83_minus1[/B] is currently too short on relations. After removing 45M duplicates, only 169169085 unique relations remained (that's after +1M free rels; quartics have many more free rels than sextics; this is a sextic). Filtering didn't produce cycles.

Need another +15% relations. Preferably in a separate directory(853_83_minus1_extra/), so that the old relations wouldn't have to be ftp'd again.
________________

P.S. Retried with density 70: a 14.4M cycle set was built. It is silly to solve such a huge matrix when a 8-9M matrix is easily expected with some minimal additional sieiving.

977_67_minus1 ends in
[code]prp92 factor: 34260142867603184002466442104359991495022313591783050473340438216525379973625039331088283587
prp99 factor: 756401130000036565972471720653891809354065981664584170942564984975043549470339665458114100127601073[/code]

Batalov: expanded the range for 853_83_minus1 from 30M-220M to 30M-250M. If you kept a copy of the original .dat.gz file, you can resume the download :smile:

Thanks for expanding, that's good!

As for resume, this data is served over http, not [URL="ftp://escatter11.fullerton.edu/nfs_data"]ftp[/URL]. Unsure how one would resume.

I use `wget -c`; curl can certainly do something equivalent :smile:

I've moved the old relations to a different file so the new relations will be in their own file.

Thanks!

L2605A target density 70 results
 

With a target density of 70, the cycle formulation was very slightly faster (25760 vs 26801 seconds), but the matrix was

8958606 x 8958788 (2673.1 MB) with weight 777424561 (86.78/col)

and the Lanczos ETA was 34h22m rather than ETA=29h51m (actual-time=30h8m) with td=110.

Was this over-sieved enough for td=130? Find out in about 28000 seconds, in our next tedious episode of The Joys of Relation Filtering.

With a target density of 130, 26743 seconds of -nc1 produce 7410788 cycles; the matrix is 7410611 x 7410788 (3646.8 MB) with weight 1043563923 (140.82/col)

and ETA is 28h28m

(jasonp, would it be possible to have some ability to do the full-merge pass multiple times with different target densities without doing the clique removal N times; the full-merge takes 10-20 minutes and the clique removal four or five hours in this case)

I'll claim L2605B too, unless frmky's cluster is starting to feel cold and unloved. Looks like about 36 hours to go on the sieving.