Cheshire Cat Computing

It's not too clear what you mean here... you are graphing CPU usage from two separate servers? What method are you using to collect the data? Remember that CPU usage is affected by monitoring (Heisenburg in action)

Are you placing both on one graph using the routers.cgi*Graph method, or by using a special script that returns both to the same Target definition? The two CPUs are in different servers, right?

Steve

_________________
Steve Shipway
UNIX Systems, ITSS, University of Auckland, NZ
Woe unto them that rise up early in the morning... -- Isaiah 5:11

From your example: the first example will define two Targets, graphing a single OID from xx.xx.xx.xx (I assume this is different in the second). Then, a 3rd userdefined graph will have both lines on it. There will maybe be a total line giving the sum of the two (although this isnot really relevant in this case, probably you'd prefer an average?)

The second example appears to be identical to one of the first two targets, graphing a single line for the OID on a single server. Can you be more specific as to exactly what you are trying to achieve? I'm not sure what you are saying is more than what you expect... of course the total line will be more than the individuals!

If you want to graph this OID on two different servers, and show an average, the first example would work if you add the 'nototal withaverage' keywords to the Graph definition.

_________________
Steve Shipway
UNIX Systems, ITSS, University of Auckland, NZ
Woe unto them that rise up early in the morning... -- Isaiah 5:11

On problem with monitoring CPU stats are that they are always changing... so, since you are monitoring in two separate targets, it is quite possible that you get different values each time. Depending on how volatile the OID you are monitoring is, it could be changing radically between each sample. This would result in two seemingly identical MRTG definitions collecting different data.

In the first example, you also can see the individual graphs, as well as the userdefined composite. If These are different, then there is a problem (since the data come from the same .rrd file) but I wouldn't expect this to be the case.

You can use 'rrdtool dump' to examine the raw stored numbers in the RRD file, if you want. However, I suspect that you will find both examples are storing different numbers because they are collectnig different numbers -- and if these servers are Windows based, then SNMP monitoring of CPU under Windows is notoriously inaccurate because it doesnt always run in kernel model correctly.

Does anyone else reading have experience of this sort of CPU monitoring?

_________________
Steve Shipway
UNIX Systems, ITSS, University of Auckland, NZ
Woe unto them that rise up early in the morning... -- Isaiah 5:11