OddPerfect enthusiasts & others: ECM help needed!

[frmky]

I concur that ECM can now wind down. I was shooting for about 0.5*t60, and we're about there. But we still have a while before sieving starts. 5,409- still has about 185,000 tasks to go, and at the current rate of about 6,000/day it'll take another month.

[R.D. Silverman]

Originally Posted by frmky

I concur that ECM can now wind down. I was shooting for about 0.5*t60, and we're about there. But we still have a while before sieving starts. 5,409- still has about 185,000 tasks to go, and at the current rate of about 6,000/day it'll take another month.

The web site shows a whole bunch of numbers between 5,409- and 3,607-
that are waiting to be done. Are you skipping those?

[debrouxl]

Based on http://escatter11.fullerton.edu/nfs/...ead.php?id=211 , I think that NFS@home clients are going to sieve some of these integers and 3,607- in a concurrent manner.

[jasonp]

Many of the NFS@Home participants do not have the gigabyte per core that the 16e lattice siever requires, so the project queues up smaller jobs in between the bigger ones, that are suitable for the 15e siever. The smaller jobs proceed in parallel.

(Not to put words in Greg's mouth, but he's on west coast time and I'm awake now)

[bdodson]

Originally Posted by R.D. Silverman

The web site shows a whole bunch of numbers between 5,409- and 3,607-
that are waiting to be done. Are you skipping those?

There are two queues among the numbers on the "Status" page; and
3,607- is next on the queue for the 16e siever. That's "next" as in
starting in c. one month, as the 5,409- tasks finish. The numbers
in between will be done with the 15e siever (<= snfs difficulty 270,
and what were the "smallest available" gnfs). Plenty of time for more
ECM (towards the rest of t60); most of the remainder of my curves
will run on 32-bit xeons.

-Bruce

(off topic Postscript: the Batalov+Dodson number M919 finished
last night, C261 = p126*p135; one digit below the Childers/Dodson
2nd place record.)

[R.D. Silverman]

Originally Posted by bdodson

There are two queues among the numbers on the "Status" page; and
3,607- is next on the queue for the 16e siever. That's "next" as in
starting in c. one month, as the 5,409- tasks finish. The numbers
in between will be done with the 15e siever (<= snfs difficulty 270,
and what were the "smallest available" gnfs). Plenty of time for more
ECM (towards the rest of t60); most of the remainder of my curves
will run on 32-bit xeons.

-Bruce

(off topic Postscript: the Batalov+Dodson number M919 finished
last night, C261 = p126*p135; one digit below the Childers/Dodson
2nd place record.)

I would have thought that some of the other numbers (e.g. the 180+ digit
GNFS jobs) required the 16e siever as well. Is this not the case?

The status page does not show the separate queues.

[frmky]

Originally Posted by R.D. Silverman

The web site shows a whole bunch of numbers between 5,409- and 3,607-
that are waiting to be done. Are you skipping those?

See what those east of me said...

[frmky]

Originally Posted by R.D. Silverman

I would have thought that some of the other numbers (e.g. the 180+ digit
GNFS jobs) required the 16e siever as well. Is this not the case?

The status page does not show the separate queues.

GNFS-180 jobs run just fine with 15e. I have yet to see how the GNFS-184 job does, though. I've kept explicit mention of the separate 15e/16e queues off the status page for simplicity, but as a general rule SNFS < 271 will be done with 15e and larger with 16e.

[jrk]

Originally Posted by jasonp

Many of the NFS@Home participants do not have the gigabyte per core that the 16e lattice siever requires,

Originally Posted by frmky

GNFS-180 jobs run just fine with 15e. I have yet to see how the GNFS-184 job does, though. I've kept explicit mention of the separate 15e/16e queues off the status page for simplicity, but as a general rule SNFS < 271 will be done with 15e and larger with 16e.

Greg, if you are interested, you may want to do some trials on those 16e jobs to see if they can benefit from using 3 large primes on the side you are sieving the special-Q on. Doing that increases the yield somewhat for the smaller special-Q in the sieving range (not as much for larger Q), and allows for the factor base limit to be reduced as well without destroying yield. With the lower memory requirement of a smaller fb you may be able to fit the 16e tasks onto more PCs.

Note that the GGNFS siever has a limit of 96 bits for mpqs, though (and you'll get a lot of "mpqs failed" at that size). 90bit algebraic cofactors were used for 6,353+, with large primes up to 31 bits. Using exactly 3*lpb was worse, probably because it is harder to find good 3-way splits near that limit without one prime being too big. So if you have 33bit large primes, 95 or 96bit mpqs should suffice.

It's hard to tell exactly how much 6,353+ benefited from this (the limited sieving trial suggested a marginal 8% or so speed gain overall, most of that coming from the small Q end). A SNFS >270 will likely benefit more clearly.

Remember to set the lambda value to something larger than log(2^mfb)/log(fblim), if you decide to try this.

[Jeff Gilchrist]

883 curves done @ 26e7, default B2.

[frmky]

Originally Posted by jrk

Greg, if you are interested, you may want to do some trials on those 16e jobs to see if they can benefit from using 3 large primes on the side you are sieving the special-Q on.

Do you still have the filtering log from 6,353+? I'd like to see how that progressed.