Discussion related to cado-nfs

[Emmanuel Thomé <Emmanuel.Thome@inria.fr>]

Hi,

On Fri, Aug 23, 2019 at 12:04:47PM +0100, Paul Leyland wrote:
> I'm sure you are all tired of hearing of bugs so I thought I'd share this
> logging output from the siever:
> 
> 
> Info:Lattice Sieving: Marking workunit GCW11244-_sieving_65930000-65940000
> as ok (0.0% => ETA Wed Feb 15 18:46:52 2023)
> Info:Lattice Sieving: Marking workunit GCW11244-_sieving_65920000-65930000
> as ok (0.0% => ETA Wed Jun  2 23:18:51 2021)
> Info:Lattice Sieving: Marking workunit GCW11244-_sieving_65910000-65920000
> as ok (0.0% => ETA Wed Nov 11 18:48:35 2020)
> 
> 
> That's what I like to see.  At the rate that the ETA is decreasing, the
> computation should be finished some time in the third millennium BC!

Well, why not ?

You seem to surmise that the ETA would decrease linearly. That is not
true. We want the ETA to converge to the real finish time, and I see
above no indication that it won't be the case. Having an ETA that is way
way off at the beginning is totall expected. Indeed, it is computed based
on a linear fit of the computation progress *thus far*, e.g. with the
formula:

    ETA = now + (work left to do) / (work already done) * (time spent thus 
far).

E.