Hi,
First time here at LQ, so if the post is in the wrong place, this is a bit programming related. I've been trying to find the maximum resident memory of a process in a memory constraint situation.

Instead of using top, or ps, which gives me real-time snapshot at the memory usage of a process, is there a way to determine the maximum resident memory used by a process?
/usr/bin/time seem to provide this functionality within the format string

time -f "%M" ./foo
gives
0

That is 0KB of resident memory, but I know the process will take more than 0KB of memory

foo is a compiled c++ code of the following

#include <map>

int main(){
std::map<int,double> tmp;
for(int i = 0; i < 100000; i++){
tmp[i] = i * 42;
}
sleep(5);
return 0;
}

a run with `time ./foo` gives

1.05user 0.02system 0:01.10elapsed 98%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+10030minor)pagefaults 0swaps

The minor page-fault is suspicious, does that mean memory is being used.. but not recorded? if so, how can I know how much memory is being mapped?

Hi,

Welcome to LQ!

Which version of Linux and which compiler are you using?
What's your shell?

There are several 'odd' things about your example.

1. It won't compile w/o atting '#include <unistd.h>'
2. your output for time gives the binaries output even though, as you call it w/o the full path, it should be overridden by the bash built-in
3. compiled and run on my system (with full path, /usr/bin/time -f "%M" ./foo does produce sensible output
   
   Code:

   $ /usr/bin/time -f "%M"  ./foo
   28496

Cheers,
Tink

I had thought "time" was untrustworthy - but I can't remember why I thought that.
Looks like it uses wait4, which uses getrusage, so it should be o.k. Be aware that wait3/4 count children as well if there are any.

Minor page faults are faults that don't result in physical I/O/ - i.e. they can be resolved from the page cache (usually). These are good page faults - the mm system is working as designed.

Trying to ascertain memory usage under Linux is a can of worms.

Hi Tink,
I'm compiling with g++ version 4.4.1 on Ubuntu 9.10 i686 2.6.31-19.
I've also tried the same thing on my Debian unstable 2.6.18.8, which has g++ 4.4.3.
Using gnome-terminal.

My program compiles fine without the #include <unistd.h>, maybe that is included by default with g++.

But yes, I have been using the full path /usr/bin/time but that does not seem to produce any sensible result like you have.

@syg00
Thanks for clearing up the minor pagefault.