Question

I am trying to measure the memory consumption of a (MPI) Fortran project that I'm working on. I measured the memory consumption with GNU's time(1) by using /usr/bin/time -v, which gives the maximum resident set size in kbytes. Because I've seen that some older versions of time had troubles with memory measurement, I wanted to make sure that it was working fine, so I wrote the following small program to test it:

program timingcheck

  implicit none
  integer :: i, s, Mb,arrlen
  integer,dimension(:), allocatable :: memarr

  i = 1
  s = sizeof(i) !number of bytes integer i occupies
  Mb = 1024*1024

  !Allocate array of size 100 Mb
  arrlen = 100 * Mb / s
  allocate(memarr(1:arrlen)) 

  memarr = 1

  call sleep(3)

  deallocate(memarr)

end program timingcheck

So far, so good. I compile the program with gfortran, without any special flags. By running /usr/bin/time -v ./timingcheck.out I get:

Command being timed: "./timingcheck.out"
User time (seconds): 0.02
System time (seconds): 0.00
Percent of CPU this job got: 1%
Elapsed (wall clock) time (h:mm:ss or m:ss): 0:03.03
Average shared text size (kbytes): 0
Average unshared data size (kbytes): 0
Average stack size (kbytes): 0
Average total size (kbytes): 0
Maximum resident set size (kbytes): 104144
Average resident set size (kbytes): 0
Major (requiring I/O) page faults: 0
Minor (reclaiming a frame) page faults: 655
Voluntary context switches: 2
Involuntary context switches: 3
Swaps: 0
File system inputs: 0
File system outputs: 0
Socket messages sent: 0
Socket messages received: 0
Signals delivered: 0
Page size (bytes): 4096
Exit status: 0

So /usr/bin/time gives me the maximal memory consumption as 104144 kb, which translates to 101.7 Mb.

Here's the question: where do those 1.7 Mb extra memory usage come from? I see multiple possibilities:

I made a mistake in my program that I'm not aware of
There is some extra memory needed in Fortran's way of storing the array that I'm not aware of
it's because of an imprecision of /usr/bin/time's memory measurement

Is it one (or multiple) of those? Or something completely different?

Edit: ldd timingcheck.out gives:

linux-vdso.so.1 =>  (0x00007ffde1b53000)
libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007f421485d000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f4214494000)
libquadmath.so.0 => /usr/lib/x86_64-linux-gnu/libquadmath.so.0 (0x00007f4214254000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f4213f4b000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f4213d35000)
/lib64/ld-linux-x86-64.so.2 (0x000055c3d15ec000)

Edit 2: I've run the measurement for different array sizes. These are the results:

Intended array size:    Measured Peak Memory Usage:
10  Mb                  12.07  Mb
50  Mb                  51.98  Mb
100 Mb                  101.83 Mb   
200 Mb                  201.87 Mb       
500 Mb                  502.03 Mb

Interestingly enough, the peak memory consumption seems to vary a little each time I run the measurement. What might be the reason behind this?

Unexpectedly high peak memory consumption of Fortran program as measured by GNU time(1)

0 个答案: