bloated vs. lightweight WM's

Brian J. Tarricone bjt23 at
Tue May 25 14:36:06 CEST 2004

Olivier wrote:

>>>How do you actually measure the memory used? Your
>>>numbers look fairly
>>>suspicious to me.
>>Try > cat /proc/meminfo
>You know that the OS is caching data, don't you. Better use the "free"
>command and read the "-/+ buffers/cache:" line to get the actual memory
>usage (minus the OS caching which has nothing to do with actual process
>memory usage)
yes - linux, at least, does some very heavy memory caching.  if you 
_really_ want a good measure, reboot your system, go straight into xfce, 
check used memory, reboot, go straight into gnome, check used memory, 
reboot, etc.  even then, it's certainly possible that there might be 
some fluctuations.  ideally, try to turn off as many background services 
as possible (apache, sshd, etc.) before testing.  might not be a bad 
idea to turn off swap as well.  if there's a /proc or sysctl interface 
that deals with memory caching, that would be of use as well.

plus, i'm not sure why everyone is measuring with things like evo, 
firefox, etc. open.  these things have nothing to do with the DE.  the 
only way having extra apps open should effect the DE is in the case of 
the WM, since the WM usually allocates per-window resources to keep 
track of data about them (plus widgets for window decorations).  even 
then, adding applications makes the "benchmark" more unreliable, as apps 
like firefox may initially e.g., allocate bunches of memory which it 
quickly frees on startup, further compounding the caching problem.  if 
you want to get a remotely accurate measurement, it's important to 
_minimise_ the number of other apps open, not the opposite.

these "benchmarks" aren't really that good because they rely on 
reporting facilities that don't tell the whole picture about what's 
going on behind the scenes inside the kernel.  i'd thank the original 
poster to relay this back to gentoo-user, as whoever started this thread 
there is spreading some misinformation based on incorrect assumptions 
and poor methodology.


More information about the Xfce mailing list