[mpich-discuss] Hydra fills in the MPIR_proctable[] incorrectly with multiple processes per node

John DelSignore John.DelSignore at roguewave.com
Mon Mar 18 14:23:21 CDT 2013


Pavan Balaji wrote:
> On 03/18/2013 11:56 AM US Central Time, John DelSignore wrote:
>> But the bigger win is that tool would have to read only 513 separate
>> strings using ptrace() or /proc, which represents a big time
>> savings.
> 
> Doesn't the tool still have to read all the strings?

No.

>  Or are you
> planning to compare pointers to check if they are the same and not read
> the string again if they are the same?

It could be done by comparing pointers, but the way TotalView implements it is through a "target address space data cache". Before TV reads from the target address space for the process (in any target read situation, not just MPIR), it looks in a data cache associated with the target address space, and if the data is already in the cache, there is no reason to reread it from the target address space using ptrace() or /proc.

Cheers, John D.

> 
>  -- Pavan
> 



More information about the discuss mailing list