From: Peter Freddolino (petefred_at_ks.uiuc.edu)
Date: Wed Oct 24 2007 - 21:44:39 CDT
Hi Leandro,
I see what you're after... sorry, I think I missed this initially. Yes,
this is an intrinsic limitation of the tclbc implementation, that
migrating atoms won't carry persistent data with them. I'm not aware of
any way in tclbc to send data along with atoms. I don't think there's
any way to know when an atom is going to migrate without adding some
hooks pretty deep inside namd, which aren't currently present. I know
having some way to synchronize data between processors in tclbc is on
the grand unified namd wish list. Is there any data in particular that
you're wanting to pass along?
Peter
Leandro Martínez wrote:
> HI Peter,
> Thanks for answer. The problem with printing (which is not the most important)
> is that I don't know how to print information outside the calcforces procedure
> for only one processor. But that's not so important, since it limited
> printing only.
>
> I made some tests and the information is not persistant between time-steps, in
> the case that one atom that was assigned to one processor is changed to
> other processor. Something simple as
>
> for { set i 1 } { $i <= $natoms } { incr i } { set test($i) 0 }
> proc calcforces { step unique } {
> global test
> while { [ nextatom ] } { incr test($i) }
> }
>
> should return, at the end, "test = number of steps" for every atom. However,
> if an atom that is being considered in processor 1 passes to processor 2, the
> counter is set to 0 and, therefore, I cannot use the previous
> information for this
> atom to compute anything.
>
> I understand that this is an intrinsic limitation because the tcl
> interpreters are
> really independent, am I wrong? If I knew which atoms were going to be
> exchanged between processors, at which time-step, than I could
> overcome this problem, maybe writing data to a file and reading it
> again, is this information anywhere?
>
> Thanks,
> Leandro.
>
>
>
>
>
>
>
>
>
>
>
>>> Another question is that within the script I update some vectors that will
>>> be used in the next step for the calculations. Fortunatelly there is only
>>> atom-specific information on each vector, in such a way that the
>>> distribution
>>> among several processes is possible. However, I'm not sure how the
>>> scripts handle these updated vectors when the script finishes, I mean,
>>> do all the scripts update the vectors before launching the scripts again
>>> in the next step or this information is lost?
>>> Thanks in advance,
>>> Leandro.
>>>
>> If the information is in variables defined as a global variable inside
>> of the tclBCscript but outside of the calcforces routine, and brought
>> into the calcforces routine as a global, it should be persistent between
>> timesteps. You should probably test this yourself to verify that the
>> information is being passed between timesteps the way you want...
>>
>> Best,
>> Peter
>>
>>>
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:45:25 CST