[QE-users] phonons calculations with -ni parallelization freezing

[Pavel Ondračka]

Hi,

so a small update, it seems like the calculation was really stuck
because it did not end with max_seconds but instead was just killed by
the scheduler later. 

I would like to recover from this somehow as I have already 80% of the
q-points calculated, so I'm thinking about using the GRID
parallelization to run the remaining q-points by hand. 

The question is would I be able to later combine the partially finished
-ni run with the GRID run and be able to run EPW on top of this?

Best regards
Pavel

On Mon, 2022-02-07 at 13:18 +0100, Pavel Ondračka wrote:
> Hi,
> 
> I've been running some ph.x calculations with qe-6.8 using the -ni +
> -
> nk parallelization. It started OK, but now it seems like some of the
> q
> points are stuck for unknow reasons. Some of the -ni process groups
> completed all of the work but some of them stop when they finish one
> q
> point and shoudl start to work on another. 
> 
> It just prints the new q-point that it should run, for example:
>      Calculation of q =    0.1000000   0.2886751   0.0000000
> and then no more output.
> 
> This is already a second run of recover=.true. calculation (the
> initial
> run was stopped cleanly with max_seconds).
> 
> One output from both a process group where it finishes cleanly
> "out.1_0" and from the currently stuck one (stuck for a day already)
> "out.2_0" is attached.
> 
> My in file:
> &inputph
> prefix='Ti3AlC2',
> ldisp=.true.,
> fildyn='dyn.xml',
> nq1=10,
> nq2=10,
> nq3=2,
> tr2_ph=1.0d-16,
> max_seconds=414000,
> nmix_ph=5,
> alpha_mix(1)=0.3,
> fildvscf='dvscf',
> recover=.true.,
> /
> 
> Any ideas how I can salvage this calculation? I want to do EPW run on
> top of it. 
> 
> Best regards
> Pavel Ondračka