capture commands from each PID and calculate the memory consuming of command in GIGA ( with sort way )

Question

I am running this approach in order to verify the commands that consuming memory from high to low on linux redhat machine

I wrote this approach as 1024/1024 , in order to get the values in GIGA

ps -eo size,pid,user,command --sort -size |
awk '{ hr=$1/1024/1024 ; printf("%13.2f GIGA ",hr) }
     { for ( x=4 ; x<=NF ; x++ ) { printf("%s ",$x) } print "" }' |
cut -d "" -f2 | cut -d "-" -f1 | more

        58.13 GIGA java
        38.39 GIGA /home/mc/lok_agent/mc.pl
        36.00 GIGA /usr/lib/jvm/java
        23.99 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
        19.42 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
        19.39 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
        19.37 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
        17.42 GIGA java
        15.43 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
        15.42 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
        15.23 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
        10.18 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
         7.22 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
         7.02 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
         5.71 GIGA /usr/jdk64/jdk1.8.0_112/bin/java
         5.57 GIGA /usr/bin/gnome
.
.
.

Until now the sum of above commands is much more then 200G

But when I run free -g I get used 168

free -g
              total        used        free      shared  buff/cache   available
Mem:            251         168           2           0          81          82
Swap:            15          11           4

So something with my approach isn’t right

ps -eo size,pid,user,command --sort -size |     awk '{ hr=$1/1024/1024 ; printf("%13.2f GIGA ",hr) } { for ( x=4 ; x<=NF ; x++ ) { printf("%s ",$x) } print "" }' |    cut -d "" -f2 | cut -d "-" -f1 |more

any advice what is wrong with my syntax ?

Answer 1

There are different ways of measuring memory used in processes. The problem is processes share memory (think of two processes using the same shared library, it's in-memory once) and you can have resident (on the RAM chips), swapped (on disk) or virtual (either/both).

Using rss instead of size will eliminate the swapped/virtual issue but not the shared.

The (probably) next release of procps will have ps and top making available PSS and USS fields, for proportional/unique segment size. These either allocate (e.g. 3 processes using a shared library get 1/3rd the size of that library) or ignore the shared part.