[FLASH-USERS] Redistribution of block

John ZuHone jzuhone at cfa.harvard.edu
Tue Aug 10 11:04:24 EDT 2010


Seyit,

This is actually not true. FLASH distributes the blocks between the processors using the Morton curve in such a way that each processor is load-balanced, but this is assuming the work is proportional to the cells per block and it has nothing to do with the particles. So in fact you may end up with most of the particles on just a few processors.

In practice this is not a major concern as the particle evolution is one of the computationally "cheapest" parts of the calculation, even if there is no hydro.

There is a way to change this but I have forgotten where it is (and almost certainly you have to go into Paramesh itself to do this).

Best,

John Z

On Aug 10, 2010, at 6:55 AM, Seyit Hocuk wrote:

> Hi Community,
> 
> I assume that when you use the particles routine, FLASH distributes the blocks between processors in such a way that there are approximately equal number of particles in each.
> 
> If there is a routine that handles this, can I change this to some other criteria/parameter, for example to distribute cells with high density in an equal manner (between procs)?
> 
> Kind regards,
> Seyit




More information about the flash-users mailing list