top - display and update information about the top cpu processes
top [ -SbiInqTu ] [ -dcount ] [ -stime ] [ -ofield ] [ -Uusername ] [
Top displays the top processes on the system and periodically updates
this information. If standard output is an intelligent terminal (see
below) then as many processes as will fit on the terminal screen are
displayed by default. Otherwise, a good number of them are shown (around
20). Raw cpu percentage is used to rank the processes. If number is
given, then the top number processes will be displayed instead of the
Top makes a distinction between terminals that support advanced
capabilities and those that do not. This distinction affects the choice
of defaults for certain options. In the remainder of this document, an
"intelligent" terminal is one that supports cursor addressing, clear
screen, and clear to end of line. Conversely, a "dumb" terminal is one
that does not support such features. If the output of top is redirected
to a file, it acts as if it were being run on a dumb terminal.
-S Show system processes in the display. The definition of system
processes is system dependent (See below).
-b Use "batch" mode. In this mode, all input from the terminal is
ignored. Interrupt characters (such as ^C and ^\) still have an
effect. This is the default on a dumb terminal, or when the output
is not a terminal.
-i Use "interactive" mode. In this mode, any input is immediately read
for processing. See the section on "Interactive Mode" for an
explanation of which keys perform what functions. After the command
is processed, the screen will immediately be updated, even if the
command was not understood. This mode is the default when standard
output is an intelligent terminal.
-I Do (or do not) display idle processes (toggle). By default, top
displays only active (ready to run or running on a processor)
-n Use "non-interactive" mode. This is identical to "batch" mode.
-q Renice top to -20 so that it will run faster. This can be used when
the system is being very sluggish to improve the possibility of
discovering the problem. This option can only be used by root.
-T List data for individual kernel threads. Normally the information
presented is a summation across all threads of the process. This is
of use when examining POSIX threaded processes.
-u Do not take the time to map uid numbers to usernames. Normally, top
will read as much of the file "/etc/passwd" as is necessary to map
all the user id numbers it encounters into login names. This option
disables all that, while possibly decreasing execution time. The
uid numbers are displayed instead of the names.
Show only count displays, then exit. A display is considered to be
one update of the screen. This option allows the user to select the
number of displays he wants to see before top automatically exits.
For intelligent terminals, no upper limit is set. The default is 1
for "batch" mode and dumb terminals.
Set the delay between screen updates to time seconds. The default
delay between updates is 4 seconds.
Sort the process display area on the specified field. The field
name is the name of the column as seen in the output, but in lower
case. Likely values are "cpu", "size", "res", and "time", but may
vary on different operating systems. Note that not all operating
systems support this option.
Show only those processes owned by username. This option currently
only accepts usernames and will not understand uid numbers.
Both count and number fields can be specified as "infinite", indicating
that they can stretch as far as possible. This is accomplished by using
any proper prefix of the keywords "infinity", "maximum", or "all". The
default for count on an intelligent terminal is, in fact, infinity.
The environment variable TOP is examined for options before the command
line is scanned. This enables a user to set his or her own defaults.
The number of processes to display can also be specified in the
environment variable TOP. The options -I, -S, and -u are actually
toggles. A second specification of any of these options will negate the
first. Thus a user who has the environment variable TOP set to "-I" may
use the command "top -I" to see only active processes.
When top is running in "interactive mode", it reads commands from the
terminal and acts upon them accordingly. In this mode, the terminal is
put in "CBREAK", so that a character will be processed as soon as it is
typed. Almost always, a key will be pressed when top is between
displays; that is, while it is waiting for time seconds to elapse. If
this is the case, the command will be processed and the display will be
updated immediately thereafter (reflecting any changes that the command
may have specified). This happens even if the command was incorrect. If
a key is pressed while top is in the middle of updating the display, it
will finish the update and then process the command. Some commands
require additional information, and the user will be prompted
accordingly. While typing this information in, the user's erase and kill
keys (as set up by the command stty) are recognized, and a newline
terminates the input.
These commands are currently recognized (^L refers to control-L):
^L Redraw the screen.
h or ?
Display a summary of the commands (help screen).
q Quit top.
d Change the number of displays to show (prompt for new number).
Remember that the next display counts as one, so typing d1 will make
top show one final display and then immediately exit.
n or #
Change the number of processes to display (prompt for new number).
s Change the number of seconds to delay between displays (prompt for
k Send a signal ("kill" by default) to a list of processes. This acts
similarly to the command kill(1)).
r Change the priority (the "nice") of a list of processes. This acts
similarly to the command renice(8)).
u Display only processes owned by a specific username (prompt for
username). If the username specified is simply "+", then processes
belonging to all users will be displayed.
e Display a list of system errors (if any) generated by the last kill
or renice command.
i (or I) Toggle the display of idle processes.
In "batch" mode, all input from the terminal is ignored. Interrupt
characters (such as ^C and ^\) still have an effect. This is the default
on a dumb terminal, or when the output is not a terminal.
By default in "batch" mode, only one screen is displayed before top
It should be noted that much of the information displayed by top is based
on calculations across the interval between the previous screen and the
current screen. If only one screen is displayed, as is the the default
for "batch" mode, then there is no previous screen to use as a base, so
some of the information will not be available on the first screen. For
example, it will only display information on processes that are currently
running, and some of the percentages may be zero.
To get more useful information using "batch" mode, it is recommended that
more than one screen be requested. For example, the command "top -b -d2"
will display two screens of information. The second screen will contain
more useful information than the first.
The actual display varies depending on the specific variant of Unix that
the machine is running. This description may not exactly match what is
seen by top running on this particular machine. Differences are listed
at the end of this manual entry.
The top few lines of the display show general information about the state
of the system, including the last process id assigned to a process (on
most systems), the three load averages, the current time, the number of
existing processes, the number of processes in each state (sleeping,
running, starting, zombies, and stopped), and a percentage of time spent
in each of the processor states (executing in user mode, executing in
kernel mode, idle, etc). It also includes information about physical and
virtual memory allocation.
The remainder of the screen displays information about individual
processes. This display is similar in spirit to ps(1) but it is not
exactly the same. PID is the process id, PGRP is the process group id,
USERNAME is the name of the process's owner (if -u is specified, a UID
column will be substituted for USERNAME), PRI is the current priority of
the process, SIZE is the total size of the process (text, data, and
stack), RES is the current amount of resident memory (both SIZE and RES
are given in kilobytes), STATE is the current state (one of "sleep",
"WAIT", "run", "idl", "zomb", or "stop"), TIME is the number of system
and user cpu seconds that the process has used, WCPU, when displayed, is
the weighted cpu percentage (this is the same value that ps(1) displays
as CPU), CPU is the raw percentage and is the field that is sorted to
determine the order of the processes, and COMMAND is the name of the
command that the process is currently running (if the process is swapped
out, this column is marked "<swapped>").
The "ABANDONED" state (known in the kernel as "SWAIT") was abandoned,
thus the name. A process should never end up in this state.
William LeFebvre, EECS Department, Northwestern University
TOP user-configurable defaults for options.
/dev/kmem kernel memory
/dev/mem physical memory
/etc/passwd used to map uid numbers to user names
/vmunix system image
The default for -I is "do not show idle processes". This may be
confusing if you're expecting top to display a screenful of processes by
default. To appease folks who can't stand that behavior, I have added
the ability to set "default" options in the environment variable TOP (see
the OPTIONS section). Those who want idle processes displayed need only
set the environment variable TOP to "-I".
The command name for swapped processes should be tracked down, but this
would make the program run slower.
As with ps(1), things can change while top is collecting information for
an update. The picture it gives is only a close approximation to
kill(1), ps(1), stty(1), mem(4), renice(8)
Multiprocessor support was added. Both number of CPUs and the CPU-ID for
running processes are shown. On MP systems the sum of %CPU (or %WCPU)
numbers may be higher than 100% (one full CPU utilization) but never
higher than 100 * the number of CPUs.
The IRIX implementation doesn't nlist(2) the kernel to get addresses of
symbols which makes it less system dependent. All values are retrieved
using the sysmp(2), syssgi(2), and proc(4) interfaces. The same
executable should run on any SGI hardware (Uniprocessor, SMP, and S2MP)
running IRIX 6.2 and up.
"last pid" is not supported.
A process is considered to be a "system" process if it has uid=0 and
ppid=1. This is a compromise for the sake of backward compatibility. In
new IRIX systems many of the traditional "system processes" are
implemented either within the kernel as threads, or as real-time
System processes display can be toggled interactively using the 'S' key.
Idle processes display can be toggled interactively using the 'i' key.
Memory sizes are followed by their units: K (Kilobytes), M (Megabytes), G
(Gigabytes) or T (Terabytes).
The CPU states are sampled every clock tick and provide a "best estimate"
of the system CPU activity summed across all processors. On IRIX they
idle the cpu was idle and not waiting for anything
usr the cpu was executing some user mode code
ker the cpu was executing some kernel mode code
wait the cpu was idle with some i/o outstanding
xbrk the cpu was executing code in a fault path
intr the cpu was handling an interrupt
For example: a high "wait" number means that the system is I/O bound. A
close to 100% "usr" number indicates CPU bound processes with little
system activity. "xbrk" and "intr" should be close to 0.0% under normal
circumstances. Larger numbers may indicate pathologically high low-level
driver activity, memory shortage (high VM activity) etc., in which case
osview(1) should give a better picture of the system state.
The meanings of the idle and wait values changed in the IRIX 6.5.13
release. For details, see the sar(1) man page.
The PRI (priority) field supports advanced IRIX real-time and scheduling
schemes using the following formats where <pri> is a decimal priority
<pri> Normal (timeshared)
g<pri> Gang scheduled
bc Batch Critical
On pre 6.4 systems the supported formats are:
<pri> Normal (timeshared)
For more details on new IRIX scheduling and priorities, see npri(1),
nice(2), realtime(5), and all the sched_XXX(2) services.
%WCPU (weighted CPU) is a decaying average of %CPU over a longer period
than just the last sampling period. For short running processes %WCPU
should be close to %CPU as the two values only diverge over time.
This version of top supports sorting by various fields via the command
line -o (order) option or the 'o' interactive option. The aliases 'mem'
(equivalent to 'size'), and 'rss' (equivalent to 'res') are accepted for
friendliness. In addition, the IRIX version of top understands prefixes
of field names, like '-os', instead of '-osize', to be used. In case of
prefix ambiguities, the first one to match will be used. Use 'h' (help)
to list the supported orders.
The old IRIX gr_top is now implemented as a perl script on top of the new
top to reduce bloat. In addition to the -Bcolor option to set the
background color, you may now control the foreground color with -Fcolor,
and the title bar color with -Tcolor. Colors are either RGB values or
Xwindow color names, e.g. purple. (See showrgb for legal color names.)
SGI specific "See Also" should include: osview(1), npri(1), realtime(5),
Several more sophisticated performance tools are optionally provided by
SGI notably PCP (the Performance CoPilot).
Unlike other top versions, the IRIX version should support thousands of
processors and an even larger number of processes. Thus it avoids preallocating
a maximum size for process data. In case the number of
processes grows too far a new initialization of data structures is
performed by the program restarting itself. This restarting forgets any
interactive settings and may cause a change in the Weighted CPU%
averages. This should be an extremely rare occurrence but should be
Original porting to SGI/IRIX 5.x was done by Sandeep Cariapa. Later
enhancements and further IRIX ports by Larry McVoy, John Schimmel, and
PPPPaaaaggggeeee 7777 [ Back ]