[petsc-users] Memory logging in fortran

[Colin McAuliffe]

Removing the have task info stuff leads to getrusage being called.  
Though there must be a bug in the os x getrusage because the max  
resident set size it returns is too large to be right. I realize this  
is not related to petsc, but if you are interested the python script:

import resource
import sys

t = range(0,10000)

print "Size of t is ",sys.getsizeof(t),"bytes"
print "Max resident set size is  
",resource.getrusage(resource.RUSAGE_SELF)[2]*1024,"bytes"

gives the output:

Size of t is  80072 bytes
Max resident set size is  4521459712 bytes

I would guess the two values should be in the same ballpark, and  
actually the computed resident set size is larger than the amount of  
memory I have to begin with! Hopefully the apple developers would be  
able to help.

Thanks
Colin

Quoting Barry Smith <bsmith at mcs.anl.gov>:

>
> On May 22, 2013, at 8:32 PM, Colin McAuliffe <cjm2176 at columbia.edu> wrote:
>
>> According to gdb the program goes here
>>
>> 110: #elif defined(PETSC_HAVE_TASK_INFO)
>> 111:   *mem = 0;
>>
>> The configuration flags from my petsc compilation indicate that   
>> both PETSC_HAVE_TASK_INFO and PETSC_HAVE_GETRUSAGE are 1. I'm not   
>> sure what have task info is supposed to do, is it safe to set it to  
>>  0 to see if this fixes the problem?
>
>     Remove the PETSC_HAVE_TASK_INFO  stuff completely and then rerun  
>  make and see if it solves the problem.
>
>      At some point the TASK_INFO stuff worked on the Apple some   
> years ago, perhaps they turned if off.
>
>    Barry
>
>>
>> Quoting Barry Smith <bsmith at mcs.anl.gov>:
>>
>>>
>>> On May 22, 2013, at 7:06 PM, Colin McAuliffe <cjm2176 at columbia.edu> wrote:
>>>
>>>> Hi Barry, thanks for the quick response.
>>>>
>>>> I am using os x and so I would expect that petsc would be able to  
>>>>   use getrusage in the following lines of mem.c:
>>>>
>>>> 115: #elif defined(PETSC_HAVE_GETRUSAGE)
>>>> 116:   getrusage(RUSAGE_SELF,&temp);
>>>>
>>>> Is this not the case?
>>>
>>>   You can run in the debugger, put a break point at that line to    
>>> verify it goes there and if it does check the values of temp after  
>>>   the call. (Sometimes Apple's docs are more optimistic than   
>>> reality.)
>>>
>>>   Barry
>>>
>>>>
>>>> Here is the link to apple's getrusage manual page
>>>> http://developer.apple.com/library/mac/#documentation/Darwin/Reference/ManPages/10.7/man2/getrusage.2.html
>>>>
>>>> Quoting Barry Smith <bsmith at mcs.anl.gov>:
>>>>
>>>>>
>>>>>  Colin,
>>>>>
>>>>>   PetscMemoryGetCurrentUsage() and MaximumUsage() us calls to   
>>>>> the   underlying operating system to get how much memory the   
>>>>> process is   using, as such they will also include external   
>>>>> package memory.   Unfortunately it depends on the operating   
>>>>> system providing this   information and often it does not, this   
>>>>> is why you get 0. If you  are  a hacker you could look at the   
>>>>> source for   PetscMemoryGetCurrentUsage() and see how to get the  
>>>>>  information  for  your OS and add it to this routine.
>>>>>
>>>>>  Barry
>>>>>
>>>>>
>>>>>
>>>>> On May 22, 2013, at 6:31 PM, Colin McAuliffe   
>>>>> <cjm2176 at columbia.edu> wrote:
>>>>>
>>>>>> Hi all,
>>>>>>
>>>>>> When calling PetscMemoryGetCurrentUsage or     
>>>>>> PetscMemoryGetMaximumUsage in a fortran code the memory usage    
>>>>>>  returned is always zero. Adding command line options such as    
>>>>>>  -malloc_log and -memory_info doesn't change this result. Is   
>>>>>> there   something else I'm missing? Also, will use of these two  
>>>>>>   functions  give memory used by external packaged called by   
>>>>>> petsc?
>>>>>>
>>>>>> All the best,
>>>>>> Colin
>>>>>>
>>>>>> --
>>>>>> Colin McAuliffe
>>>>>> PhD Candidate
>>>>>> Columbia University
>>>>>> Department of Civil Engineering and Engineering Mechanics
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Colin McAuliffe
>>>> PhD Candidate
>>>> Columbia University
>>>> Department of Civil Engineering and Engineering Mechanics
>>>
>>>
>>>
>>
>>
>>
>> --
>> Colin McAuliffe
>> PhD Candidate
>> Columbia University
>> Department of Civil Engineering and Engineering Mechanics
>
>
>



-- 
Colin McAuliffe
PhD Candidate
Columbia University
Department of Civil Engineering and Engineering Mechanics