Let's pu{sh to fi}nish base 2- to 1000 bits!

[R.D. Silverman]

Originally Posted by fivemack

Only three machines, but one of them is a four-socket 48-CPU 64GB Opteron that I bought when a special offer made it about the most cost-effective, and certainly the most convenient, way to acquire that many computrons (the others are a 16GB i7/4770 and a 32GB i7/4930K; the post-processing is done on that last one)

I basically just run jobs with make

G=$(shell seq 0 12999)
S=/home/nfsworld/gnfs-batalov/gnfs-lasieve4I16e

all: $(patsubst %,%.t1,$G)

%.t1:
	$(S) snfs -r -f $(shell echo $*\*10000+20000000 | bc) -c 10000 2> $*.t
	wc -c $*.t > $*.t1

and manually write makefiles that run for about a month on however-many CPUs.

Very well done.

Now let's (collectively perhaps?) tackle the next first hole in the table.
Let's push to finish base 2 to 1000 bits.

Perhaps the IGG will stop doing easy factorizations and help out.

Or perhaps not......

[jyb]

Originally Posted by R.D. Silverman

Very well done.

Now let's (collectively perhaps?) tackle the next first hole in the table.
Let's push to finish base 2 to 1000 bits.

Perhaps the IGG will stop doing easy factorizations and help out.

Or perhaps not......

Perhaps continually insulting people by referring to them as the IGG is not the best way to enlist their aid.

And perhaps the barrier to performing such factorizations is not the lack of willing helpers, but rather the need for coordination. Perhaps you can volunteer in that regard. Or perhaps not....

[R.D. Silverman]

Originally Posted by jyb

Perhaps continually insulting people by referring to them as the IGG is not the best way to enlist their aid.

And perhaps the barrier to performing such factorizations is not the lack of willing helpers, but rather the need for coordination. Perhaps you can volunteer in that regard. Or perhaps not....

I call a spade a spade.

Actually it has been stated repeatedly by members of the IGG (within this forum!) that they
do not want to work on difficult factorizations because it
is "boring". One sieves for a long period and nothing happens. They want
more immediate results; hence "instant gratification".

Or maybe they just have ADD?

We have coordination: NFS@Home.

[jyb]

Originally Posted by R.D. Silverman

We have coordination: NFS@Home.

Well if you're satisfied with the factorizations that NFS@Home is doing, then what's the problem? And if you're not (because you'd rather work on different numbers), then I guess we don't really have coordination, do we?

Again I ask whether you'd care to volunteer your services toward coordinating the factorization of a number you care about? If not, then I scarcely see how you have cause to complain about others' indolence.

[Xyzzy]

Again I ask whether you'd care to volunteer your services toward coordinating the factorization of a number you care about?

That is a good question!

We would be willing to allocate one (Linux) computer to your effort, if you decide to coordinate something.

We bet others would as well.

[wombatman]

I've got 10 to 18 (effective) cores to add with some guidance.

[R.D. Silverman]

Originally Posted by jyb

Well if you're satisfied with the factorizations that NFS@Home is doing, then what's the problem? And if you're not (because you'd rather work on different numbers), then I guess we don't really have coordination, do we?

Again I ask whether you'd care to volunteer your services toward coordinating the factorization of a number you care about? If not, then I scarcely see how you have cause to complain about others' indolence.

I have done so in the past. The first instance of NFSNET was organized and run by me, then later hosted by Richard Wackerbarth. (who did a nice job)

I can coordinate, but I can not send data nor receive data owing to a very strict firewall and corporate policy.
I can not give out assignments except
at a level of course granularity. And while I have CPU resources internally
at work they are strictly devoted to projects. The only machine on which I
could place data that I received from an outside source is my home PC. It is
inadequate for large post processing.

"The problem" is that NFS@Home has limited resources. It is not doing
large/first hole Cunningham numbers since it is focused at the moment on
a C221 via GNFS. More resources would allow work on the former.
Actually, I do not understand the criteria by which they have been selecting
their numbers over the past 1-2 years. At one time they were doing
wanted Cunningham numbers.

I lack the resources to perform leading edge factorizations. I do numbers
that are within reach of my resources; currently these are Homogeneous
Cunningham first holes.

[fivemack]

The Cunningham tables appear to be looking after themselves, thanks mostly to Ryan Propper who clearly has access to some very substantial resources.

On the whole I would rather run large jobs than complain that other people aren't using their superior resources to run the particular jobs that I'm interested in; convincing people to do what you want is a diplomatic rather than a mathematical problem. NFS@home's criterion seems fairly simple: on the large-jobs side they do GNFS at the edge of the practical, on the small-jobs side they do whatever people express interest in: I've a few times sent them a decent polynomial for a C17x and then done the linear algebra myself. Cunningham first-holes are bigger than small jobs, so don't get done.

I'm also unsure, in this world where computers do clock themselves down and save significant power when idle, that using computers other than the personally-owned ones of explicit volunteers is necessarily the right way to go.

[jyb]

Originally Posted by R.D. Silverman

I have done so in the past. The first instance of NFSNET was organized and run by me, then later hosted by Richard Wackerbarth. (who did a nice job)

I can coordinate, but I can not send data nor receive data owing to a very strict firewall and corporate policy.
I can not give out assignments except
at a level of course granularity. And while I have CPU resources internally
at work they are strictly devoted to projects. The only machine on which I
could place data that I received from an outside source is my home PC. It is
inadequate for large post processing.

"The problem" is that NFS@Home has limited resources. It is not doing
large/first hole Cunningham numbers since it is focused at the moment on
a C221 via GNFS. More resources would allow work on the former.
Actually, I do not understand the criteria by which they have been selecting
their numbers over the past 1-2 years. At one time they were doing
wanted Cunningham numbers.

I lack the resources to perform leading edge factorizations. I do numbers
that are within reach of my resources; currently these are Homogeneous
Cunningham first holes.

Well then allow me to make a suggestion: take a look at the factorizations which Fivemack has previously coordinated on this forum. You'll note that:

1) Assignments are given out (actually, just taken) at a level of coarse granularity.
2) The sieving work need not be done at all by the coordinator (though Tom did in fact do quite a bit of the sieving).

That should address your concerns, at least up to the point of sieving. I suspect you can find willing volunteers to store the data and do the post-processing, though I concede I might be wrong about that. It would probably be worth checking with Tom to get some idea of how much iron he thinks is actually required for post-processing job on e.g. 2,983+.

The real value that Tom added was:
1) Coming up with a good polynomial (in the case of GNFS jobs).
2) Doing a bunch of test-sieving in order to come up with good parameters.
3) Maintaining the thread which contains reservations.
4) Doing periodic relation maintenance (e.g. duplicate elimination, post-processing test runs).
5) Maintaining a server to act as a relation dump.
6) Doing the post-processing.
7) Maybe some other stuff I don't know about, but he would most likely share.

Do any of those, other than 6, seem too onerous for you? Again, I would be surprised if you didn't find someone willing to take on 6.

I think you'll find plenty of people here will jump at the chance to join in on a factorization if someone (e.g. you) can perform these steps. What do you say?

[fivemack]

The linear algebra for a kilobit SNFS job takes somewhere between a month and a season on an i7-4930K or similar (six cores, four memory controllers, you probably do want 32G of memory); the relations take low-three-digit gigabytes of disc space, maybe a week of download over slow ADSL.

RDS is entirely aware of what resources are required - he's coordinated some enormous projects - but is not in a position to buy them. His perfectly reasonable point is that individuals are unwilling to commit to six-month runs, but he considers this a problem to be ranted about rather than a political truth to be worked around.

nfs@home works around it by using spare cycles from people who couldn't care less about factorisation, but there's only space for one nfs@home, someone else has done the tedious work to set it up, and the reward for that tedious work is that he gets to set the priorities. I'm happy to run linear algebra for as many 1000ish-bit SNFS jobs as nfs@home is prepared to run, but Greg has perfectly reasonably chosen to run limit-of-the-possible GNFS jobs instead.

[R.D. Silverman]

Originally Posted by jyb

Well then allow me to make a suggestion: take a look at the factorizations which Fivemack has previously coordinated on this forum. You'll note that:

1) Assignments are given out (actually, just taken) at a level of coarse granularity.

I need everyone's email address to mail assignments.
Or I could post them here.

2) The sieving work need not be done at all by the coordinator (though Tom did in fact do quite a bit of the sieving).

This I can do, if the post-processing software can handle CWI relation
format.

The real value that Tom added was:
1) Coming up with a good polynomial (in the case of GNFS jobs).

Not needed for Cunningham numbers.

2) Doing a bunch of test-sieving in order to come up with good parameters.

Not needed. Read my paper: Optimal Parameterization of SNFS.

However: Who would create the input files? I do not have Msieve/GGNFS.
(and am not allowed to install them)

3) Maintaining the thread which contains reservations.

Maintained through this forum?

4) Doing periodic relation maintenance (e.g. duplicate elimination, post processing test runs).

Impossible. My home PC can't handle a data set that large (it is 5 years old)
I am behind a very strict firewall/restrictions at work. I can't accept data from an external source.

5) Maintaining a server to act as a relation dump.
6) Doing the post-processing.
7) Maybe some other stuff I don't know about, but he would most likely share.

Also impossible.

I think you'll find plenty of people here will jump at the chance to join in on a factorization if someone (e.g. you) can perform these steps. What do you say?