Re: [Apbs-users] meaning of dime with mg-para

[David Gohara <sdg0919@gm...>]

>

Sorry... mail problems today...

  dime IS tied to the processor. So each processor will have nx, ny,  
nz grid points (based on dime). If the processor count is 1 (pdime 1 1  
1) then the number of total grid points is equal to dime. When you  
split the calculation up into multiple processors each processor gets  
a grid dimensioned at the value of dime.

   This is true of APBS 0.4.0.

Dave

On Oct 30, 2007, at 10:07 AM, David Gohara wrote:

>
> On Oct 30, 2007, at 8:24 AM, Mark Abraham wrote:
>
>> David Gohara wrote:
>>> Hi Mark,
>>>
>>> dime specifies the global dimension and pdime simply specifies the
>>> breakup in each direction (x,y,z).  In the case of MPI it's the  
>>> number
>>> of MPI tasks that will be generated in each dimension. For  
>>> asynchronous
>>> mode (async) it will break the job into N x M x O tasks, but run  
>>> them
>>> serially.
>>
>> Hi David,
>>
>> Thanks for the reply, but I don't think that's quite what I asked.  
>> Your
>> statement that "dime specifies the global dimension" doesn't tell me
>> whether "setting dime constrains the number of grid points per  
>> processor
>> or the number of grid points over the whole processor array".
>> Re-phrasing, when pdime != 1 1 1, I'm asking whether "global  
>> dimension"
>> is intensive or extensive with respect to the component calculations?
>>
>> I now see in the 0.5.1 documentation that dime is the number of grid
>> points *per processor*. If that also applies to 0.4.0 then I'm  
>> happy. If
>> not, how does the break-up occur and where does it get output?
>>
>> Regards,
>>
>> Mark
>>
>> -------------------------------------------------------------------------
>> This SF.net email is sponsored by: Splunk Inc.
>> Still grepping through log files to find problems?  Stop.
>> Now Search log events and configuration files using AJAX and a  
>> browser.
>> Download your FREE copy of Splunk now >> http://get.splunk.com/
>> _______________________________________________
>> apbs-users mailing list
>> apbs-users@...
>> https://lists.sourceforge.net/lists/listinfo/apbs-users
>

Re: [Apbs-users] meaning of dime with mg-para

[Nathan Baker <baker@cc...>]

Hi Mark --

I've updated the APBS user manual a bit to clarify this.  This update  
will be available with the next APBS release but I've attached it here  
for reference.

Sorry for the confusion.

-- Nathan

Re: [Apbs-users] meaning of dime with mg-para

[David Gohara <sdg0919@gm...>]

Hi Mark,

   Sorry I misspoke.

   dime IS tied to the processor. So each processor will have nx, ny,  
nz grid points (based on dime). If the processor count is 1 (pdime 1 1  
1) then the number of total grid points is equal to dime.
On Oct 30, 2007, at 8:24 AM, Mark Abraham wrote:

> David Gohara wrote:
>> Hi Mark,
>>
>>  dime specifies the global dimension and pdime simply specifies the
>> breakup in each direction (x,y,z).  In the case of MPI it's the  
>> number
>> of MPI tasks that will be generated in each dimension. For  
>> asynchronous
>> mode (async) it will break the job into N x M x O tasks, but run them
>> serially.
>
> Hi David,
>
> Thanks for the reply, but I don't think that's quite what I asked.  
> Your
> statement that "dime specifies the global dimension" doesn't tell me
> whether "setting dime constrains the number of grid points per  
> processor
> or the number of grid points over the whole processor array".
> Re-phrasing, when pdime != 1 1 1, I'm asking whether "global  
> dimension"
> is intensive or extensive with respect to the component calculations?
>
> I now see in the 0.5.1 documentation that dime is the number of grid
> points *per processor*. If that also applies to 0.4.0 then I'm  
> happy. If
> not, how does the break-up occur and where does it get output?
>
> Regards,
>
> Mark
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc.
> Still grepping through log files to find problems?  Stop.
> Now Search log events and configuration files using AJAX and a  
> browser.
> Download your FREE copy of Splunk now >> http://get.splunk.com/
> _______________________________________________
> apbs-users mailing list
> apbs-users@...
> https://lists.sourceforge.net/lists/listinfo/apbs-users

Re: [Apbs-users] meaning of dime with mg-para

[Gernot Kieseritzky <gernotf@ch...>]

Hi!

Mark Abraham schrieb:
> Hi,
> 
> When using mg-para with (for example) "pdime 2 2 2", does setting dime 
> constrain the number of grid points on each processor, or the number of 
> grid points over the whole processor array? The User Guide doesn't seem 
> to imply either of them.

In the context of mg-para the "dime" parameter gives the number of grid
points per processor. It is explained on
http://apbs.sourceforge.net/doc/user-guide/index.html#mg-para :

"Please note that some of the parameters change in meaning a bit for
this type of calculation. In particular, dime  should be interpreted as
the number of grid points per processor. This interpretation helps
manage the amount of memory per-processor -- generally the limiting
resource for most calculations."

You can use the "tools/manip/psize.py" tool to find a reasonable
combination of pdime, dime and grid resolution.

Greetings,
Gernot Kieseritzky

Re: [Apbs-users] meaning of dime with mg-para

[Mark Abraham <Mark.Abraham@an...>]

David Gohara wrote:
> Hi Mark,
> 
>   dime specifies the global dimension and pdime simply specifies the 
> breakup in each direction (x,y,z).  In the case of MPI it's the number 
> of MPI tasks that will be generated in each dimension. For asynchronous 
> mode (async) it will break the job into N x M x O tasks, but run them 
> serially.

Hi David,

Thanks for the reply, but I don't think that's quite what I asked. Your 
statement that "dime specifies the global dimension" doesn't tell me 
whether "setting dime constrains the number of grid points per processor 
or the number of grid points over the whole processor array". 
Re-phrasing, when pdime != 1 1 1, I'm asking whether "global dimension" 
is intensive or extensive with respect to the component calculations?

I now see in the 0.5.1 documentation that dime is the number of grid 
points *per processor*. If that also applies to 0.4.0 then I'm happy. If 
not, how does the break-up occur and where does it get output?

Regards,

Mark

Re: [Apbs-users] meaning of dime with mg-para

[David Gohara <sdg0919@gm...>]

Hi Mark,

   dime specifies the global dimension and pdime simply specifies the  
breakup in each direction (x,y,z).  In the case of MPI it's the number  
of MPI tasks that will be generated in each dimension. For  
asynchronous mode (async) it will break the job into N x M x O tasks,  
but run them serially.

   Hope that helps,

Dave

On Oct 30, 2007, at 12:14 AM, Mark Abraham wrote:

> Hi,
>
> When using mg-para with (for example) "pdime 2 2 2", does setting dime
> constrain the number of grid points on each processor, or the number  
> of
> grid points over the whole processor array? The User Guide doesn't  
> seem
> to imply either of them.
>
> Thanks,
>
> Mark
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc.
> Still grepping through log files to find problems?  Stop.
> Now Search log events and configuration files using AJAX and a  
> browser.
> Download your FREE copy of Splunk now >> http://get.splunk.com/
> _______________________________________________
> apbs-users mailing list
> apbs-users@...
> https://lists.sourceforge.net/lists/listinfo/apbs-users

[Apbs-users] meaning of dime with mg-para

[Mark Abraham <Mark.Abraham@an...>]

Hi,

When using mg-para with (for example) "pdime 2 2 2", does setting dime 
constrain the number of grid points on each processor, or the number of 
grid points over the whole processor array? The User Guide doesn't seem 
to imply either of them.

Thanks,

Mark