[Ecm-discuss] GMP-ECM finds input N

Alexander.Kruppa at loria.fr Alexander.Kruppa at loria.fr
Lun 30 Mar 17:14:48 CEST 2009


Paul Zimmermann wrote:
>        Dear Sam,
> 
>> Date: Mon, 30 Mar 2009 08:31:32 -0400 (EDT)
>> From: Samuel S Wagstaff <ssw at cerias.purdue.edu>
>>
>> Paul,
>>
>> Sometimes the GMP-ECM program finds the input number N, but it is
>> printed incorrectly.  Some examples follow.  Some days it does this
>> for 10% of the runs; other days it never does it.
>>
>> Have you any idea what is wrong?  I suspect a hardware or system
>> problem, but I don't know what to tell the systems people.
>>
>> Thanks,
>>
>> Sam
> > [...]
> I suspect a defect memory. I have started identical runs on a Core 2 (with the
> same version of GMP-ECM and GMP) and they do not stop immediately like yours
> runs. To convince the systems people you could redo those runs with the same
> parameters (e.g., echo 278...531 | ecm -sigma 4289042661 250000000). If sigma
> is given, GMP-ECM should be deterministic. If you can produce two runs that
> give different results, clearly it is a systems problem.
> 
> Do other users of the computers also report similar problems?
> 
> Hope this helps,
> Paul

You could also do runs with "residue verboseness," enabled by adding "-v 
-v" to the command line. That way GMP-ECM prints the residue after stage 
1, and the product of the polynomials values at the end of stage 2. If 
these differ from one run to another with the same N, B1, B2, and sigma 
values, then there's very likely something wrong with the hardware. If 
you wish to compare residues between different machines, check that all 
the stage 2 parameters agree. They are printed with at least one "-v", e.g.,

  Using B1=50000, B2=12746592, polynomial x^2, sigma=3108370201
  dF=512, k=5, d=4620, d2=13, i0=-2


What type of hardware is the system that shows this erratic behaviour? 
Are the errors more likely when it is hot in the room?

Best regards,
Alex




Plus d'informations sur la liste de diffusion Ecm-discuss