[Dune] restarting parallel code

[Dedner, Andreas]

The backup/restore method in in the dune alugrid bindings and the method simple take
an iostream (dune/grid/alugrid/3d/grid.hh).
The way to move to a higher number of processores (at least using alu) is to backup your
grid and data on n processors (files grid_1,grid_2,...grid_n and data_0,data_1,...data_n)
Restart your program with m>n processors. Proc 1 through n restore the grid from the data files,
Proc n+1,..,m setup an empty grid. When you call your loadBalance method, everything should
work. 
Andreas
________________________________________
From: Marco Cisternino [marco.cisternino at optimad.it]
Sent: 04 February 2013 14:50
To: Dedner, Andreas; dune at dune-project.org
Subject: Re: [Dune] restarting parallel code

Il 04/02/2013 15:14, Dedner, Andreas ha scritto:
> It might be a good idea to switch to a newer dune version.
> The file I was referring to is from 2011...
You're right and I'm planning to do this.
> If you really do not want to do that then alugrid has  methods backup/restore
> you could use directly.
I really don't know how. Alu lacks in documentation, I can't imagine how
to understand the way these methods work.
>
> The backup/restore method are only for the grid, you have to backup user data
> yourself. But if you are storing your data in vector like structures, then you can just dump
> them to disk since the indexsets are restored.
I think this is true for restoring the calculations with the same number
of processes, doesn't it?
I can't figure out how to use indexsets with a new number of processes.
Aren't they local (in the sense of the process)?

Marco
>
> Best Andreas
> ________________________________________
> From: Marco Cisternino [marco.cisternino at optimad.it]
> Sent: 04 February 2013 13:14
> To: Dedner, Andreas
> Cc: dune at dune-project.org
> Subject: Re: [Dune] restarting parallel code
>
> Hi Andreas, thank you for your quick reply. I'm using dune 2.1.1 and I
> cannot find that file.
> I found this:
> dune/grid/geometrygrid/backuprestore.hh
> Is it the same?
> I had a look to it. But what about user data?
> Thanks again.
>
> Marco
>
>
>
> Il 04/02/2013 13:54, Dedner, Andreas ha scritto:
>> Hi Marco.
>>
>> Have a look at dune/grid/common/backuprestore.hh
>> I do not know if how many grids implement this class but alugrid and alberta do.
>>
>> Best
>> Andreas
>> ________________________________________
>> From: dune-bounces+a.s.dedner=warwick.ac.uk at dune-project.org [dune-bounces+a.s.dedner=warwick.ac.uk at dune-project.org] on behalf of Marco Cisternino [marco.cisternino at optimad.it]
>> Sent: 04 February 2013 12:42
>> To: dune at dune-project.org
>> Subject: [Dune] restarting parallel code
>>
>> Hi duners,
>> I wonder if anyone of you has already think about a stop and restart
>> system for his parallel code.
>> I would like to stop my code, save the results and restart the code with
>> a new number of process and the possibility of coarsening the grid.
>> Any suggestion is more than appreciated.
>> Thanks.
>>
>> Marco
>>
>> --
>> Marco Cisternino, Ph.D.
>> Optimad Engineering s.r.l.
>> www.optimad.it
>> marco.cisternino at optimad.it
>> +3901119719782
>>
>>
>> _______________________________________________
>> Dune mailing list
>> Dune at dune-project.org
>> http://lists.dune-project.org/mailman/listinfo/dune
>>
>>
> --
> Marco Cisternino, Ph.D.
> Optimad Engineering s.r.l.
> www.optimad.it
> marco.cisternino at optimad.it
> +3901119719782
>
>
>

--
Marco Cisternino, Ph.D.
Optimad Engineering s.r.l.
www.optimad.it
marco.cisternino at optimad.it
+3901119719782