BARMON

Section: Maintenance Commands (8)
Updated: November 1991
Index Return to Main Contents

NAME

barmon - display performance bar graphs for several hosts

SYNOPSIS

barmon host1 host2 ... [ interval ]

DESCRIPTION

barmon uses the rstatd(8) to gather system statistics from the specified hosts and then displays that information interactively on a tty using curses(3). The display is a series of bar graphs divided by host on the Y axis and activity type on the X axis. It is typically used to watch a number of processors running distributed applications. Other uses include monitoring a network of machines for overloading, balance, etc.

Upon start up, barmon will be a little sluggish because it is pinging all the hosts for statistics. After the information is retrieved, the screen is cleared and a graph like so will be shown (this graph shows one machine paging, all others are idle):

          [% USER CPU][% SYS CPU ][Pkts   100][Disk    40][Page    60]
sbox0     [          ][          ][          ][          ][          ]
sbox1     [UUU       ][          ][          ][DDDDDDDDDD][PPPPPPPPPP]
sbox4     [          ][          ][          ][          ][          ]

The graph is updated every interval seconds (default: 5 seconds).

All of the fields of barmon are averaged over interval. The fields are:

% USER CPU: Percentage of time spent on user CPU in the last interval.
% SYS CPU: Percentage of time spent on system (kernel) CPU in the last interval.
PKTS: Number of network packets (both input and output) seen in the last interval.
DISK: Number of disk operations (both input and output) seen in the last interval.
PAGE: Number of paging operations (both input and output) seen in the last interval.

The fields with numbers in the title are auto scaled each interval. If a field is marked 200, that means a full bar indicates 200 events, a half bar indicates 100, and so on. All the fields are auto scaled across all the hosts, so if the standard deviation is high, you start losing information.

BUGS

The rstat()ing hosts effect the timing enough that sometimes the numbers come out wrong. They are usually too large. I tried to address this somewhat by fudging the numbers down and by using interval timers instead of sleep(3). Don't use short intervals.

Silently ignores unreachable hosts. However, a dead host tends to slow down startup quite a bit.

A host that crashes causes the program to exit.

Rstatd(8) does not dig out all the info I'd like. It would be cool if we could implement vmstat(8) and xperfmon(8) with rstatd(8).

Barmon(8) does not display all the information it could. The I/O operations could be displayed as "IIIOOOO" to indicate "in" or "out" direction since rstat(8) does tell you this information.

(I'd refer you to rstat(3) to figure out the fields but that seems to be undocumented. Sigh. I figured it out by looking at the rstat.h include file.)

Rstatd(8) seems to be a Sun only thing - lobby BSD to include it.

AUTHOR

Larry McVoy, Stanford University, lm@sunburn.stanford.edu

This document was created by man2html, using the manual pages.
Time: 21:51:34 GMT, February 02, 2023