Re: shared memory in computational chemistry

[Peter Schüller <ps@brumma.com>]

Hi Francesco,

Francesco Pietra wrote:
> Hi Peter:
>
> Thanks. I set 
>
> echo "60000000" >/proc/sys/kernel/shmmax
>
> because the segment to allocate was ca 39mb. In
> between the rows reporting the steady approach to scf
> energy, a row appeared:
>
> 1 warning: armci set_mem_offset: offset changed 0 to
> 8192

I am not sure if this means trouble, but as it is only a warning and as 
you said "completed successfully" I assume the results also made sense ;)

> and the calculation of partial charges completed
> successfully.
>
> As to a permanent solution, is any drawback in setting
> shmmax as above (or even to a higher value)
> permanently? Probably this issue of allocating big
> fragments will become routine, though I can't predict
> their size.

I am sorry but I don't really know about the internals of shared memory 
in clusters.

I know that it is necessary to increase shmmax considerably to run 
Oracle (If there is need for a large block cache and for speed).

I also know that if the shared memory is not explicitly freed by the 
using process, it will not be freed automatically and has to be freed by 
the user by "ipcrm" or similar commands in order to free memory for 
other applications (or in order to restart the same application).

As long as your program frees the memory and you don't need the rest of 
the memory while your program is running you should be fine in my 
experience (from non cluster environments).

Best Regards,

Peter