- Hi,

Everytime i try a general number, getdepts is doing some ecm the first time

it runs.

- On what number is it doing the ecm?

- 50* B1= 2000 and 150* B1= 5000 won't give you many large factors. Why so

many curves with such small bounds?

- Am i right gmp-ecm 4c is used? v5.03 might give some extra performance .

Regards,

Sander - --- In ggnfs@yahoogroups.com, "Sander Hoogendoorn" <sander@s...> wrote:
> Everytime i try a general number, getdepts is doing some ecm the

first time

> it runs.

It appears to be the discriminant of the polynomial c_d^(d-1)*f(x/c_d).

>

> - On what number is it doing the ecm?

> - 50* B1= 2000 and 150* B1= 5000 won't give you many large factors.

Why so

> many curves with such small bounds?

I'm not familiar with the theory, so this is a wild guess, but perhaps

the factorization is optional (it can help, but it's not required)?

Alternately, it might be sufficient that we only know the small

factors of the discriminant.

--

Sam - --- In ggnfs@yahoogroups.com, "Sander Hoogendoorn" <sander@s...> wrote:
> Hi,

first time

>

> Everytime i try a general number, getdepts is doing some ecm the

> it runs.

Why so

>

> - On what number is it doing the ecm?

> - 50* B1= 2000 and 150* B1= 5000 won't give you many large factors.

> many curves with such small bounds?

performance .

> - Am i right gmp-ecm 4c is used? v5.03 might give some extra

>

It would be easy to externalize GMP-ECM as is done with the ECMNet

> Regards,

> Sander

Client. Granted, this adds a little overhead since it would not be

linked into GGNFS, but so little time is spent with GMP-ECM as opposed

to sieving, it shouldn't be a hardship on anyone. This would allow it

to use whatever version of GMP-ECM the user wants. If Chris is

interested, I can share with him the current ECMNet Client source so

he can see how it is done. - Hi Mark,

On the first run, it is using ECM to try to factor the discriminant

of the polynomial. This is needed to compute an integral basis for the

number field. Having said that, it is not really necessary to have a

full integral basis - something reasonably close will suffice. And so,

it is not really necessary to have a full factorization of the

discriminant. However, it is important to be fairly certain that we

have all the small factors - thus the ECM.

It is annoying that it takes so long on the first run, but this was

really just overkill on my part. The newest version (still very

unstable, tho I did just u/l it to the devel area) won't try as many

curves and so it won't take quite as long. So I don't think the extra

speed increase of linking in a newer version is worth the penalty of

added difficulty compiling. It's not terribly bad, really, since it

only happens on the first run (and it is much quicker now since it

won't try as many curves).

Cheers,

Chris

--- In ggnfs@yahoogroups.com, "Mark Rodenkirch" <mgrogue@w...> wrote:

>

>

> --- In ggnfs@yahoogroups.com, "Sander Hoogendoorn" <sander@s...> wrote:

> > Hi,

> >

> > Everytime i try a general number, getdepts is doing some ecm the

> first time

> > it runs.

> >

> > - On what number is it doing the ecm?

> > - 50* B1= 2000 and 150* B1= 5000 won't give you many large factors.

> Why so

> > many curves with such small bounds?

> > - Am i right gmp-ecm 4c is used? v5.03 might give some extra

> performance .

> >

> > Regards,

> > Sander

>

> It would be easy to externalize GMP-ECM as is done with the ECMNet

> Client. Granted, this adds a little overhead since it would not be

> linked into GGNFS, but so little time is spent with GMP-ECM as opposed

> to sieving, it shouldn't be a hardship on anyone. This would allow it

> to use whatever version of GMP-ECM the user wants. If Chris is

> interested, I can share with him the current ECMNet Client source so

> he can see how it is done. > It's not terribly bad, really, since it only happens on the

Ehhh, it did this again for _every_ dependency during the square root fase.

> first run (and it is much quicker now since it

> won't try as many curves).

What i did, is let getdepts run a few curves with B1=20K and B1=50K and

factor the remainder with gmp-ecm and ppsiqs. The factors found have to be

added to the log file if i understand things correctely.

Sander