BOINC NFS sieving - NFS@Home

[pinhodecarlos]

Just a note. Keep monitoring the less payable sievers since at least one of the big players is hitting hard on the small F, Greg too(for his post processing). Not sure if the French team is running or not for the team badges.

[swellman]

C303_150_119 from the XYYXF project is a SNFS 311. It is one of ten remaining C3XX left in the XYYXF project, all of which have been run through ECM to t66+ by Yoyo and others, some to even higher levels.

It is near (perhaps beyond?) the practical limit of the 16e_small siever. Best yield comes from a 34/34 job sieved on the rational side. Greg has agreed to let this job run, with a target of 2-2.1B relations.

n: 643864214334634268895374202342247423078904934597133752583128069256047803365128938205024001859004848051848833245470196507782923791701338997148666967666245375097001045160291064832511719269478532381478970673816809653020664057169299457025413360035912745471402021110752895305461548827923164133406738190535207
skew: 2.3051
type: snfs
size: 311
c6: 1
c0: 150
Y1: -7738807338473109582614865884818187755242394882082999
Y0: 33252567300796508789062500000000000000000000
rlim: 225000000
alim: 225000000
lpbr: 34
lpba: 34
mfbr: 100
mfba: 68
rlambda: 3.8
alambda: 3.1

Results of test sieving on the rational side with Q in blocks of 1000:

MQ        Norm_yield      Speed (sec/rel)
40           3987              0.479
60           3299              0.544
80           3046              0.576
100          2685              0.663
140          2381              0.733
180          2262              0.780
240          1881              0.902
300          1794              0.865
400          1533              0.986
500          1425              1.148
600          1343              1.155
800          1142              1.283
1000         1078              1.317
1200         1108              1.272
1400          950              1.450
1600          865              1.610

Suggesting a sieving range for Q of 40M-1610M to generate 2.1B raw relations. Which is a LOT of sieving.

Any suggested improvements are welcome. I do not plan to post/enqueue this job until 16e_small digests its current tasks, as well as give other projects a chance to post. 2,2694M anyone?

If something goes wrong with this job, e.g. high dup rate, we can always add relations by sieving over Q of say 40-400M on the -a side but that should be considered an emergency measure IMO.

As this SNFS 311 job is roughly equivalent to a GNFS 203, I am surprised at how difficult it is to sieve. More to it than pure SNFS I suppose, the escore/awkwardness of the coefficients have an effect too.

[charybdis]

Originally Posted by swellman

Any suggested improvements are welcome. I do not plan to post/enqueue this job until 16e_small digests its current tasks, as well as give other projects a chance to post. 2,2694M anyone?

If something goes wrong with this job, e.g. high dup rate, we can always add relations by sieving over Q of say 40-400M on the -a side but that should be considered an emergency measure IMO.

As this SNFS 311 job is roughly equivalent to a GNFS 203, I am surprised at how difficult it is to sieve. More to it than pure SNFS I suppose, the escore/awkwardness of the coefficients have an effect too.

Firstly, I don't know where you got the idea that SNFS 311 is similar difficulty to GNFS 203; that would be a ratio of 0.65 rather than the more normal 0.68-0.69. Secondly, this isn't SNFS 311, it's SNFS 314 (did you forget that you multiplied up by 150?). That suggests a GNFS difficulty of 213-216, so no wonder it's probably too big for 16e-small. I'd suggest asking Greg if he could put this on the main 16e queue as it will sieve much faster there; it would hold up 16e-small for a couple of months at least.

I'm hoping to test-sieve 2,2694M in the coming days.

[swellman]

Originally Posted by charybdis

Firstly, I don't know where you got the idea that SNFS 311 is similar difficulty to GNFS 203; that would be a ratio of 0.65 rather than the more normal 0.68-0.69.

I and others use SNFS = (GNFS-30)*1.8

It’s an old rule of thumb, though I can’t speak to its accuracy. Maybe it breaks down above GNFS 200?

Secondly, this isn't SNFS 311, it's SNFS 314 (did you forget that you multiplied up by 150?). That suggests a GNFS difficulty of 213-216, so no wonder it's probably too big for 16e-small.

I was using the standard convention with xyyx numbers SNFS = x*log(y) for difficulty, plus 311 is what Yafu spit out for this particular polynomial (there’s an error in Yafu’s SNFS diff calculation method when Cn > 1 though that’s not the case here). Y0 is -150^20 and Y1 is 119^25. But even a SNFS 314 is ~GNFS 204 again using the above rule of thumb.

But I concede that this would be a brutally difficult job for 16e_small.

I'd suggest asking Greg if he could put this on the main 16e queue as it will sieve much faster there; it would hold up 16e-small for a couple of months at least.

No doubt the main 16e siever can run this considerably faster. The point of the exercise is to define the upper limit for 16e_small. But you’ve pointed out the ugly part of a job this size - it will definitely hold up things for 8-10 weeks.

I'm hoping to test-sieve 2,2694M in the coming days.

Excellent.

[swellman]

I’m not trying to finesse this past anyone here - if folks voice their discontent about enqueuing a job this difficult then I won’t do it. @charybdis does not seem in favor of the idea.

Greg has agreed to let 16e_small sieve it but perhaps it is a bridge too far, sucking up too many community resources for a single outlier data point.

We could sneak up on the upper SNFS limit for 16e_small with say SNFS 295, 300, 395 etc jobs.

[charybdis]

Originally Posted by swellman

I and others use SNFS = (GNFS-30)*1.8

It’s an old rule of thumb, though I can’t speak to its accuracy. Maybe it breaks down above GNFS 200?

I think that must have been intended for smaller, personal-size jobs. It breaks down well below GNFS-200; for example, it suggests GNFS-185 = SNFS-279, and having run both, I can tell you that SNFS-279 is close to twice as hard. NFS@Home data will tell a similar story. A ratio of 0.685 is about right for the numbers I've run in the GNFS 175-190 and SNFS 260-280 range.

For SNFS-310+ you can start directly comparing e-scores for sextics against degree 6 GNFS polynomial scores, which again suggests your XYYX number is comparable to roughly GNFS-216.

I was using the standard convention with xyyx numbers SNFS = x*log(y) for difficulty, plus 311 is what Yafu spit out for this particular polynomial (there’s an error in Yafu’s SNFS diff calculation method when Cn > 1 though that’s not the case here). Y0 is -150^20 and Y1 is 119^25.

SNFS difficulty = log(Y1^d * f(Y0/Y1)) where f is the algebraic poly and d is its degree. Your poly isn't really for 150^119 + 119^150, it's for 150^120 + 150*119^150, which gives the correct difficulty.

[VBCurtis]

I don't see a hard limit as a concept for any of the sievers- before the era of f-small, we showed that 15e can handle GNFS-198 and similar-sized SNFS. That doesn't mean it's a good idea, because there are faster ways to handle those jobs.

The same seems true here- f-small *can* handle this job, but why choose a slower queue when a faster one exists? By the time we're below yield 2.0 on 34LP, there's no doubt that larger lim's are desirable.

That said, my disposition (beyond nfs@home or this forum) is to prioritize efficiency over "can I do this?", so maybe that's why I think this job isn't a great idea on f-small.

[frmky]

The queue for 16e is long. In terms of calendar time, it would be completed more quickly in 16e_small.

[charybdis]

Originally Posted by frmky

The queue for 16e is long. In terms of calendar time, it would be completed more quickly in 16e_small.

...but many other numbers would be completed 10 weeks later in calendar time as a result. If this was some "high-priority" number - for example, if an original base-2 Cunningham had been knocked down to c215 after an ECM hit - then there would be a case for getting it done quickly, but as it stands I don't see why this number should be given special treatment.

[chris2be8]

Is something wrong with escatter11.fullerton.edu?

$ ping escatter11.fullerton.edu
ping: escatter11.fullerton.edu: Temporary failure in name resolution

I can connect to other sites such as here so I think it's not my internet connection.

[swellman]

Originally Posted by chris2be8

Is something wrong with escatter11.fullerton.edu?

$ ping escatter11.fullerton.edu
ping: escatter11.fullerton.edu: Temporary failure in name resolution

I can connect to other sites such as here so I think it's not my internet connection.

It’s offline for me as well.