L1193
 

L1193 splits as:

[CODE]prp101 factor: 10952406243948994025559441584939174982528513885797898600315900849747185346663920239218585542502976901
prp128 factor: 38423525351878925840834786618336821458014668017212382604241848415388493701424326251927639927990070796852858081758597718401115501[/CODE]

Updated the NFS@Home infrastructure with L1193.

jasonp: I'm torture-testing msieve's filtering code (trunk version) with the full set of ~222M unique relations for L1354 :smile:
I set target density to 110; msieve had to filter five times before deciding the matrix was dense enough, at only 59.96 (previous rounds: 33.80, 38.85, 43.96, 50.01). Tomorrow morning, we'll see whether the matrix was dense enough for nontrivial dependencies to be found...
At least, msieve r901 has improved the run times in such a way that the ~7.1M matrix can be chewed through in an estimated ~26h on that Sandy Bridge Xeon E3-1230 on 4 threads, so it won't be too much of a problem if I have to start from scratch. It's trivial to cut relations from the dataset with modern versions of msieve, through filter_maxrels.

BTW, I'm looking for a direct way to force the initial ideals maximum weight. filter_lpbound doesn't seem to be it:
[code] filter_lpbound=X have filtering start by only looking
at ideals of size X or larger[/code]
The ideals weight started at 20; if I had forced it to 45 or 50, filtering could have succeeded on the first round, and produced a matrix of density higher than 60.

Assuming it doesn't exist yet, adding this knob definitely makes less sense than working on msieve-db or msieve-lacuda, and possibly makes no sense whatsoever, so I'm not demanding that you add this to msieve, even though it's certainly much easier :smile:

~26h20 of LA for L1354, and the factors are in after ~50' of square root (and an odd dependency 0 with many cycles ?).

[code]commencing square root phase
reading relations for dependency 0
read 7083479 cycles
cycles contain 17853616 unique relations
read 17853616 relations
algebraic side is not a square!
reading relations for dependency 1
read 3541703 cycles
cycles contain 8921408 unique relations
read 8921408 relations
multiplying 8921408 relations
multiply complete, coefficients have about 494.44 million bits
initial square root is modulo 744462673
sqrtTime: 3139
prp75 factor: 101814304324334006347720247956531431054730671614952768502158077738161907507
prp89 factor: 50221617492347410835001189762529438377673186530427303161494399851009900695146062023971609[/code]

Adding the extra knob is easy.

Your first dependency behaved badly because dependencies are numbered from 1 onward; asking for dependency 0 really means asking the matrix reading code for the entire matrix! (Looks like I need to check for that :)

Edit: lots of problems here are fixed in SVN904

Well, I hadn't explicitly asked for dependency 0 :smile:
Just plain old msieve -i ... -s ... -nf ... -nc "target_density=110" -t 4 -v.

EDIT: OK, I'll update to HEAD next time.

I see from the original version that passing any NFS configuration string started the square root with dependency 0, so I messed up the patch to fix problems that Carlos reported. Grrr.

I'll take F1953 next.

L1213 progress report
 

[QUOTE=Speedy51;344844]It is currently at 35.3% give or take 12.6 days of the LA process to go.I upgraded to version 1.52 this is the reason why the speed increase has happened I'm still running it on three cores[/QUOTE]
There is another 5.95 days to go on the LA process. Currently 69.2% complete

ETA two days for L1821, assuming the Haswell machine doesn't mysteriously turn into a penguin again

Finally got L1953.dat file downloaded and two copies made. Filtering done and
ETA is less than 48 hours!

I might grab another before I go on holiday the following week.
No promises.

Grabbing L1694