[firedrake] parallel weirdness

[Hiroe Yamazaki]

Hi Lawrence,

you can reproduce the problem by runing test_linear.py at
bitbucket.org/colinjcotter/slicemodels/branch/periodic_parallel. With
1 and 2 cores, it gives different results.

The code Colin is describing with line numbers is an old-version of
slicemodels.py, which is found here:

https://gist.github.com/anonymous/c308131558ff54fec9c4

In this version I added this term

- (-dt*div(w)*pbar)*dx

in line 270, which was previously added to the RHS in line 139.
With this change, the parallelization problem seemed to be fixed and I
got the same results with 1 and 2 cores, but but it makes no sense to
us.

All the best,
Hiroe


2014-11-18 20:46 GMT+00:00 Mitchell, Lawrence
<lawrence.mitchell at imperial.ac.uk>:
> Hello,
>
> On 18 Nov 2014, at 18:57, Colin Cotter <colin.cotter at imperial.ac.uk> wrote:
>
>> Dear all,
>> We have a rather weird parallel bug, and Hiroe and I would appreciate some suggestions. The relevant code is in bitbucket.org/colinjcotter/slicemodels/branch/periodic_parallel revision 622fb24.
>> We were observing different results on 1 and 2 cores, which Hiroe fixed by moving some code, but it makes no sense to us!
>> In line 270, you can see an extra term,
>> - (-dt*div(w)*pbar)*dx,
>> which was previously added to the RHS in line 139
>> so, this changes implementation but not the algorithm
>> either, this extra term is solved into a contribution of self.ures in L142, in which case it pops up as part of self.ures in l270, or we write it explicitly in the form there in general, we don't wish to write it explicitly in the form though, and I want to understand what is going on!
>>
>> Some further investigation showed that the problem seems to occur in the pressure solve on line 393, resulting in a nonsense value of self.Deltap occurring. This is just inverting a DG mass system so it ought to be straightforward, and the inputs are (approximately) the same.
>>
>> Do let me know if you have any ideas,
>
>
> Can you describe what I need to do to reproduce the problem?  I have looked at the repository, but don't know where to begin.
>
> Lawrence