14.7. Monitoring Active CPU Load

Problem

You want to use MRTG to keep an eye on CPU performance. cfgmaker only sets up graphs for your network interfaces. Now what?

Solution

You'll need to edit mrtg.cfg by hand. Don't use cfgmaker anymore because it will overwrite your changes. This recipe monitors system, user, and nice values, and adds them up to give a snapshot of CPU load.

Add this line to the Global Config Options section:

	# Global Config Options
	LoadMIBs: /usr/share/snmp/mibs/UCD-SNMP-MIB.txt,/usr/share/snmp/mibs/TCP-MIB.txt

Then, add this section at the bottom of the file:

	# monitor CPU load
	#
	Target[xena.cpu]: ssCpuRawUser.0&ssCpuRawUser.0:password@localhost + ssCpuRawSystem.
	0&ssCpuRawSystem.0:password@localhost + ssCpuRawNice.0&ssCpuRawNice.0:
	password@localhost
	Title[xena.cpu]: Xena CPU Load
	RouterUptime[xena.cpu]: password@127.0.0.1
	PageTop[xena.cpu]: <H1>CPU System, User and Nice stats for Xena</H1>
	MaxBytes[xena.cpu]: 100
	ShortLegend[xena.cpu]: %
	YLegend[xena.cpu]: CPU Usage
	Legend1[xena.cpu]: Current CPU percentage load
	LegendI[xena.cpu]: Used
	LegendO[xena.cpu]:
	Options[xena.cpu]: growright,nopercent
	Unscaled[xena.cpu]: ymwd

While you're there, you might as well delete the entry for the loopback device, as you're not going to use it anyway. Now, run mrtg manually three times, or until it stops complaining, and mind your filepaths:

	# env LANG=C mrtg /etc/mrtg.cfg

Then, generate a new index.html page:

	# indexmaker --output=/var/www/mrtg/index.html /etc/mrtg.cfg

Now, you should see something like Figure 14-4 and Figure 14-5.

Figure 14-4. New index page showing CPU load graph

Discussion

These aren't very exciting graphs because they're on a test system with little activity. But, they do show what the entries in mrtg.cfg make the graphs look like. Let's take a tour of the configuration:

LoadMIBs: You have to tell MRTG which MIB files from /usr/share/snmp/ to load. You'll see them cached in /var/www/mrtg/oid-mib-cache.txt. Using the LoadMIBs option means you'll be able to use the symbolic names of the OIDs, rather than having to use their numbers.
Target[xena.cpu]:ssCpuRawUser.0&ssCpuRawUser.0:password@localhost…: The syntax for this line is Keyword[graphname]: value. The Target keyword defines what you want monitored. The value is a list of OID pairs joined by the ampersand, &. The graph name is anything you want, and it must be unique. MRTG only knows how to measure pairs of values, so when you're plotting a single value, just repeat the target definition. Whitespace must surround each target definition. This example contains three target definitions, connected with the plus sign. The plus sign means "add these together."
Title: This is the title of the detailed HTML page.
Figure 14-5. Detail page for the CPU load graph
RouterUptime: Use this to display the system uptime on the detail page.
PageTop: The headline of the detail page.
MaxBytes: This is either a value in bytes, or it behaves as a percentage.
ShortLegend: The units string used for Max, Average, and Current. The default is b/s, bits per second.
YLegend: The Y-axis, or vertical axis, which is for the label of the graph. If you make it too long, it will be silently ignored.
Legend1 (and 2, 3, and 4): Strings for the color legends.
LegendI (and O): Input and output. Because the different values in this graph are added up to a single sum, and each OID pair is the same OID rather than two OIDs, we're only using LegendI.
Options: growright means expand the graph to the right, and nopercent means don't print percentage values. For our CPU graphs printing percentages would be redundant because we already have percentage displays. Without the nopercent option the graph legends it would display like this: Used 65.0 % (65.0%) 35.0 % (35.0%) 6.0 % (6.0%).
Unscaled: The default is scaled, which means MRTG will adjust the graph to make small data more visible. The unscaled option creates a fixed-size graph. ymwd represents year, month, week, day.

man 1 mrtg-reference is your primary configuration reference.

14.7. Monitoring Active CPU Load

Problem

Solution

Discussion

See Also