This is a script I stumbled upon written in Python that shows memory usage per process in Linux.
Credit for the site where I found the script here: http://unixlive.editboard.com/
# Try to determine how much RAM is currently being used per program.
# Note the per program, not per process. So for example this script
# will report mem used by all httpd process together. In detail it reports:
# sum(all RSS for process instances) + max(shared mem for any process instance)
# The shared calculation below will factor out shared text and
# libs etc. within a program, but not between programs. So there
# will always be some overestimation. This will be the same for
# all processes that just use libc for e.g. but more for others
# that use larger shared libs like gnome, kde etc.
# Author: P@draigBrady.com
# V1.0 06 Jul 2005 Initial release
# V1.1 11 Aug 2006 root permission required for accuracy
# V1.2 08 Nov 2006 Add total to output
# Use KiB,MiB,... for units rather than K,M,...
# V1.3 22 Nov 2006 Ignore shared col from /proc/$pid/statm for
# 2.6 kernels up to and including 2.6.9.
# There it represented the total file backed extent
# V1.4 23 Nov 2006 Remove total from output as it's meaningless
# (the shared values overlap with other programs).
# Display the shared column. This extra info is
# useful, especially as it overlaps between programs.
# V1.5 26 Mar 2007 Remove redundant recursion from human()
# V1.6 05 Jun 2007 Also report number of processes with a given name.
# Patch from email@example.com
# All interpreted programs where the interpreter is started
# by the shell or with env, will be merged to the interpreter
# (as that's what's given to exec). For e.g. all python programs
# starting with "#!/usr/bin/env python" will be grouped under python.
# You can change this by changing comm= to args= below but that will
# have the undesirable affect of splitting up programs started with
# differing parameters (for e.g. mingetty tty[1-6]).
# For 2.6 kernels up to and including 2.6.13 and later 2.4 redhat kernels
# (rmap vm without smaps) it can not be accurately determined how many pages
# are shared between processes in general or within a program in our case:
# A warning is printed if overestimation is possible.
# In addition for 2.6 kernels up to 2.6.9 inclusive, the shared
# value in /proc/$pid/statm is the total file-backed extent of a process.
# We ignore that, introducing more overestimation, again printing a warning.
# I don't take account of memory allocated for a program
# by other programs. For e.g. memory used in the X server for
# a program could be determined, but is not.
# This script assumes threads are already merged by ps
# use ps just to enumerate the pids and names
# so as to remove the race between reading rss and shared values
import sys, os, string
if os.geteuid() != 0:
sys.stderr.write("Sorry, root permission required.\n");
for char in "-_":
return (int(kv), int(kv), int(kv))
for line in open("/proc/"+str(pid)+"/smaps").readlines()
return sum([int(line.split()) for line in shared_lines])
elif (2,6,1) <= kv <= (2,6,9):
return 0 #lots of overestimation, but what can we do?
for line in os.popen("ps -e -o rss=,pid=,comm=").readlines():
size, pid, cmd = map(string.strip,line.strip().split(None,2))
if int(pid) == our_pid:
continue #no point counting this process
continue #ps gone away
if shareds[cmd] < shared:
#Note shared is always a subset of rss (trs is not always)
count[cmd] += 1
count[cmd] = 1
#Add max shared mem for each program
for cmd in cmds.keys():
sort_list = cmds.items()
sort_list=filter(lambda x:x,sort_list) #get rid of zero sized processes (kernel threads)
#The following matches "du -h" output
#see also human.py
def human(num, power="Ki"):
while num >= 1000: #4 digits
num /= 1024.0
return "%.1f %s" % (num,power)
def cmd_with_count(cmd, count):
return "%s (%u)" % (cmd, count)
print " Private + Shared = RAM used\tProgram \n"
for cmd in sort_list:
print "%8sB + %8sB = %8sB\t%s" % (human(cmd-shareds[cmd]), human(shareds[cmd]), human(cmd),
print "\n Private + Shared = RAM used\tProgram \n"
#Warn of possible inaccuracies
#1 = accurate
#0 = some shared mem not reported
#-1= all shared mem not reported
if kv[:2] == (2,4):
if open("/proc/meminfo").read().find("Inact_") == -1:
elif kv[:2] == (2,6):
if (2,6,1) <= kv <= (2,6,9):
vm_accuracy = shared_val_accurate()
if vm_accuracy == -1:
sys.stderr.write("Warning: Shared memory is not reported by this system.\n")
sys.stderr.write("Values reported will be too large.\n")
elif vm_accuracy == 0:
sys.stderr.write("Warning: Shared memory is not reported accurately by this system.\n")
sys.stderr.write("Values reported could be too large.\n")