New GGNFS problem

[Andi47]

Originally Posted by 10metreh

I am not a GNFS expert. When you make a poly with msieve, it has R0 and R1 values. Which one do you use for a GGNFS "m" value? When using R0 I get a polynomial error. Same with R1.

10metreh

P.S. Henry, I think my joining has induced an increase in your posting rate. Do you agree?

The R0 and R1 are called Y0 and Y1 in GGNFS, the A0, A1, etc. are called c0, c1, etc. in GGNFS. You need not specify an "m" value for GGNFS.

(Feature request to either GGNFS or msieve (or both): input / output polynomials in the same format.)

[henryzz]

Originally Posted by 10metreh

I am not a GNFS expert. When you make a poly with msieve, it has R0 and R1 values. Which one do you use for a GGNFS "m" value? When using R0 I get a polynomial error. Same with R1.

10metreh

P.S. Henry, I think my joining has induced an increase in your posting rate. Do you agree?

yes it has although i have posted more recently in other things as well

if you cant get the msieve polynomials to work in ggnfs dont worry the ggnfs polynomial selection isnt that rubbish
it will help much more for larger jobs
with the size we are doing convinience plays a large part in what we do
i would probably waste the time i save by using the msieve polynomial selection in converting it to the ggnfs format

[10metreh]

I have converted the msieve poly and it runs 4 times faster than the best poly that polyselect has found in the same time!

I haven't tested pol51 but I will stick to msieve for the time being.

P.S. I hope closing the program will save the relations to spairs.out so I can restart - I am trying to save energy.

P.P.S. Currently I am on "total yield: 375000, q=945000" (values not exact). How high do these have to get?

[henryzz]

Originally Posted by 10metreh

I have converted the msieve poly and it runs 4 times faster than the best poly that polyselect has found in the same time!

I haven't tested pol51 but I will stick to msieve for the time being.

P.S. I hope closing the program will save the relations to spairs.out so I can restart - I am trying to save energy.

P.P.S. Currently I am on "total yield: 375000, q=945000" (values not exact). How high do these have to get?

polyselect is rubbish
pol51 is much better and only a little worse than msieve although i dont think msieve is anywhere near as good as it will get

re P.S.
it wont save the relations
especially if they arent using the same polys

re P.P.S.
as a vague estimate q will get to 1.5mil for this factorization
but dont rely on it as it is as vague as my estimate the other day(which was about half my estimate)

[smh]

Originally Posted by 10metreh

... it runs 4 times faster than the best poly that polyselect has found in the same time!

Told you!!

The only advantage of polyselect is that it can find deg. 4 poly's, but a deg5 poly from pol51/msieve will be better than a deg4 poly from polyselect.

Even running msieve's QS will be faster!

[10metreh]

The relations do get saved, but in a file called spairs.add. This suggests that there is a way of resuming work, but I have not found it yet: using factLat.pl as before starts the sieving from the beginning, and saving the spairs.add file as spairs.out is no help. How do I resume?

[henryzz]

i am just doing the linear algebra on a 100 digit gnfs for my aliquot sequence
q went up to 1.4M and that produced 3.6M relations

it will save the relations, at the 100 digit level, every 100k q values
it will also do a postprocessing run to see if you have sieved enough

[schickel]

Originally Posted by 10metreh

The relations do get saved, but in a file called spairs.add. This suggests that there is a way of resuming work, but I have not found it yet: using factLat.pl as before starts the sieving from the beginning, and saving the spairs.add file as spairs.out is no help. How do I resume?

If you are interrupting the siever before the end of the sieve block, yes, it is supposed to resume from where it left off and add in all the relations it found before it was interrupted. I have never had it work correctly for me, so I found the easiest way is to interrupt only after a block has finished and procrels has run. If you interrupt it before procrels runs, the job file is not updated and it will start the block over again.

Probably the best way is to change the range of each block to bring the completion time to a value where you could interrupt it a little easier. What is the digit level of the job and how long does each block take? On my bigger jobs, I usually have the "qstep" set to 100000 which seems to take ~4 hours on my NFS machines. After a couple of blocks you can get a feel for what the relation yield is for each block so that you can tell approximately how far a block is along as the job progresses. On my XP machine the output doesn't scroll while a job is running, so you can tell what the start/end values are, but on my 98 machine, the progress lines scroll the screen, so the only way to tell how far along is know what the yield is for each block.

I run all my machines 24/7, so it doesn't matter how long the blocks are, anyway. (They help keep the bedroom warm this time of year, but do make it hard to keep cool during the summer.....)

Frank

[10metreh]

The procrels error I mentioned in post #1 has recurred. How can I fix it?

[henryzz]

Originally Posted by 10metreh

The procrels error I mentioned in post #1 has recurred. How can I fix it?

could you post your ggnfs.log

[schickel]

Originally Posted by 10metreh

The procrels error I mentioned in post #1 has recurred. How can I fix it?

Are you trying to run procrels manually or is it crashing under the control of the script?

Frank