This article was first written in January 2004 for
the BeezNest technical website (http://glasnost.beeznest.org/articles/99)
This has been originally written by Miguel de Icaza (the famous GNOME hacker and creator of Ximian). This was retrieved from:
http://mail.gnome.org/archives/gnome-list/1999-September/msg00036.html.
People usually look at the memory sizes for the processes and misinterpret the information. This is a common mistake.
When talking about memory usage and the reports you get from the operating system, you need to keep in mind the following terms and what they mean:
SIZE
This is the address space seen by the process. If the process maps 65 megs of memory, the address space of the process will be reported as 65 megs of memory. Even if only 100k of memory are actually used. This bit of information is not usually very useful and this is what most people believe is the actual memory usage of a program: it is not.
RSS
This is the "Resident Set Size". This is the amount of memory actually resident on memory. This is a better computation of what is actually loaded in memory. This does not include any piece of code that have been "swapped out". So, for example, if you have a program that uses 100k of memory and the operating system discards or swaps 40k of memory, the RSS will be 60k. The RSS includes both the memory which is unique to this process and the memory shared with other processes. In most modern Unix systems, a large part of this accounts for the memory used by shared libraries. So it usually includes the ammount of memory that is used by other processes for the code that is shared: The libc is usually taken into account. GNOME uses many shared libraries (this makes the applications share more memory).
SHARE
This is the amount of memory from the RSS that is shared with other applications.
So, the actual memory used uniquely by an application is RSS - SHARE. If two programs use library "A", then the process would also be "responsible" for A/2 of the memory, but there are many variations (which pieces of the library each program uses) that make this computation hard.
A good approximation is thus "RSS - SHARE".
The file manager in GNOME for example uses the following libraries:
CORBA libraries:
libgnorba.so.27 => /opt/gnome/lib/libgnorba.so.27 (0x40007000)
libORBitCosNaming.so.0 =>/opt/gnome/lib/libORBitCosNaming.so.0 (0x40011000)
libORBit.so.0 => /opt/gnome/lib/libORBit.so.0 (0x4001a000)
libIIOP.so.0 => /opt/gnome/lib/libIIOP.so.0 (0x4007c000)
libORBitutil.so.0 => /opt/gnome/lib/libORBitutil.so.0(0x40086000)
GNOME graphical libraries:
libgnomeui.so.32 => /opt/gnome/lib/libgnomeui.so.32(0x40089000)
libart_lgpl.so.2 => /opt/gnome/lib/libart_lgpl.so.2(0x4013b000)
libgdk_imlib.so.1 => /opt/gnome/lib/libgdk_imlib.so.1(0x40149000)
Session management libraries:
libSM.so.6 => /usr/X11R6/lib/libSM.so.6 (0x4016a000)
libICE.so.6 => /usr/X11R6/lib/libICE.so.6 (0x40173000)
GTK libraries:
libgtk-1.2.so.0 => /usr/local/lib/libgtk-1.2.so.0 (0x40188000)
libgdk-1.2.so.0 => /usr/local/lib/libgdk-1.2.so.0 (0x402a3000)
Glib libraries:
libgmodule-1.2.so.0 => /usr/local/lib/libgmodule-1.2.so.0(0x402d5000)
libglib-1.2.so.0 => /usr/local/lib/libglib-1.2.so.0(0x403c9000)
X11 libraries:
libXext.so.6 => /usr/X11R6/lib/libXext.so.6 (0x402d9000)
libX11.so.6 => /usr/X11R6/lib/libX11.so.6 (0x402e7000)
GNOME libraries:
libz.so.1 => /usr/lib/libz.so.1 (0x404a4000)
libgnome.so.32 => /opt/gnome/lib/libgnome.so.32 (0x403a3000)
libgnomesupport.so.0 => /opt/gnome/lib/libgnomesupport.so.0(0x403b5000)
libdb.so.2 => /lib/libdb.so.2 (0x403bb000)
Standard Unix libraries:
libdl.so.2 => /lib/libdl.so.2 (0x403ea000)
libm.so.6 => /lib/libm.so.6 (0x4038a000)
libc.so.6 => /lib/libc.so.6 (0x403ff000)
/lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x2aaaa000)
File manager libraries:
libext2fs.so.2 => /lib/libext2fs.so.2 (0x403ed000)
libcom_err.so.2 => /lib/libcom_err.so.2 (0x403fd000)
So, even if it does not use all of the code, the "SIZE" will reflect the maximal size the program would be using if it used all of those libraries.
The more libraries we use the bigger the SIZE goes, even if this does not reflect the actual memory usage.
