Graphing high speed interfaces

[cwinckless]

I thought I had this issue licked, but I'm still seeing some oddities with graphing high speed (gigabit etherchannel and OC48) interfaces from a Cacti installation.

I'm running 0.8.6g and have no access to the underlying machine, just to the software installation. I may be able to get the admin on the system to execute commands if this is recommended, however.

The symptoms seem to vary somewhat, and I have the following:

OC48s: No outbound traffic. Inbound looks close and very low, but I'm expecting to see over 2Gb/s outbound and I'm seeing nothing (NaN, as below)

Port-channel: OUTGOING appears to be correct, inbound reported as NaN.

All these are set up on the same system, using SNMPv2 and 64bit counters as I saw recommended elsewhere when I searched the system. I'm using the default template (Interface -- Traffic bit/sec) for graphing.

All three data sources are also set up identically, using the Interface -- Traffic template, output type ID set as 64bit counters. The only differences are the maximum values, which are adjusted to 4000000000 in the case of the 4GE etherchannel and 2500000000 in the case of the POS/OC48 interfaces.

I have to admit to being stumped. I'd welcome any advice, since I'm fairly sure I'm doing something dumb here.

[cwinckless]

Anyone have any ideas, or even where I could start to look? It's wierd, since it's working with an alternate rrdtool front end...

[gandalf]

....The only differences are the maximum values, which are adjusted to 4000000000 in the case of the 4GE etherchannel and 2500000000 in the case of the POS/OC48 interfaces....

Did you tweak those max values _after_ creating the data sources? Please do a rrdtool info against those files to verify the _actual_ settings.
To check correct readings, switch to DEBUG mode (Settings) for at least one polling cycle. Find the actual snmpget answer in log/cacti.log. Does your findings make sense to you?
Reinhard

[cwinckless]

....The only differences are the maximum values, which are adjusted to 4000000000 in the case of the 4GE etherchannel and 2500000000 in the case of the POS/OC48 interfaces....

Did you tweak those max values _after_ creating the data sources? Please do a rrdtool info against those files to verify the _actual_ settings.
To check correct readings, switch to DEBUG mode (Settings) for at least one polling cycle. Find the actual snmpget answer in log/cacti.log. Does your findings make sense to you?
Reinhard

Thanks. I'll try that; I definitely did tweak it AFTER creating the data sources.

[gandalf]

Yep, that's it. You'll have to run some

Code: Select all

rrdtool tune <file> --maximum <data source>:<value>

to apply your new max values to existing rrd's.
Reinhard

[cwinckless]

Right. I'll get the people with access to the server to do that.

For future configurations, is there a simple step I'm missing in the GUI to do this?

I've attempted it by creating a new data source template with a higher limit (5G), and then creating a new graph template for the OC48s with the traffic_in / traffic_out associated with this data source template.

Then I get stuck on how to actually create a graph associated with this template.

[BSOD2600]

Once you modify the date template with the new min/max values, cacti does not (currently) update the rra files. You must do this manually, as you've found out.

[cwinckless]

Sorry -- wasn't terribly clear.

For future interfaces, what's the process I have to follow to avoid having to do this when I create a graph? I have several cases where I'll be talking about at least 2 and possibly more gig links channeled (which I saw somewhere I can do with a stack graph) but it'd be nice to be able to poll the logical port-channel.

Does that make more sense?

Thanks, by the way, for all the help here. Really appreciate it.

[BSOD2600]

Same logic still applies. If you have the Data Template which your interface(s) use updated with the proper min/max, the next time you go to create a data source/graph, it will have the correct values.

[cwinckless]

Got it. And thanks -- the update the data template for the interface was exactly what I needed to know. I was over-thinking.

[TheWitness]

The MySQL strict patch on the web site for 0.8.6h also has the added benefit of setting the RRD file Max value to "ifSpeed". Therefore, once you have that patch applied, you should no long have to tweak you data sources.

TheWitness

[evilzardoz]

CUrrently running 0.8.6g and ran into a similar problem where interfaces appeared to top out at around 100Mbit which seemed to be the default max value. I suggest changing it to 1000Mbit for future releases.

Would upgrading to 0.8.6h and applying that patch automatically adjust my data sources to ifSpeed or do I still need to do it manually?

Wish there was a way for templates to automatically update the RRAs...

[BSOD2600]

Once the rrd files are created, you must manually adjust them.

[TheWitness]

Please log a feature request. This is a long shot. As a developer, I am quite distracted at the moment. Place the feature request in Mantis (bugs.cacti.net).

Regards,

TheWitness

[bulek]

The MySQL strict patch on the web site for 0.8.6h also has the added benefit of setting the RRD file Max value to "ifSpeed".

This is simply a mistake. Max value should be set to ifSpeed/8. Otherwise you will have it set too high and spikes can occur on such graph.

Besides of this think of subinterfaces like Frame Relay links. Quite often such subinterface speed is set to CIR (ou can set a "bandwidth" statement on Cisco routers which corresponds to ifSpeed). On FR links you can burst much above CIR even up to port speed. If you set Max value to ifSpeed then you will loos all the stats about bursting traffic.

- Piotr