Alarming on bandwidth changes

[Stuart Weenig]

I want to be able to generate an alarm when the bandwidth of an interface changes. The use case is that we sometimes have clients who are set to auto/auto but the negotiation gets redone and drops from 1Gbps to 100Mbps or 10Mbps. We'd like to detect that. I can do this in NV today simply by creating a baseline for the ifstats.speed expression and setting a threshold where ifstats.speed < ifstats.speed.baseline - 1. However, we're looking at moving to CA PM and don't want to build anything else in NV that we'll be taking away when we move to CAPM. Is something like this possible in CAPM? 

[Michael Poller]

Hi Stuart,

 

Against the OOTB Interface Metric Family there are a few speed related metrics we can alarm against.

 

There is the Historical Interface Speed:

 

Speed In and Out:

 

I haven't tested such a rule, but it appears you can configure one like this:

That can be set for comparison to a Fixed Value or a Standard Deviation.

 

Any chance any of that would meet your needs?

 

Thanks,

Mike

[Stuart Weenig]

I'm not sure that rule would quite get what i'm going for. Definitely not the "fixed value" one because that would assume that all my interfaces are running at the same speed (which obviously they are not). 

I guess i'm seeking clarity about which metrics have the standard deviation calculated. Is it calculated on each metric for each element? If that's the case, then a rule looking for a standard deviation on speed should do it. However, if i were developing this, i would have taken the shortcut and not calculated a standard deviation on a metric that isn't likely to change. Can anyone confirm this works?

[Michael Poller]

There is no config to set up standard deviations. As far as I can tell we're gathering it for all metrics.

 

Baseline Calculations - CA Performance Management - 3.1 - CA Technologies Documentation 

[Tim Johnson]

Mike,

   Regarding this post. Lemme ask you this.  Would you consider this method, using STD Deviation in the following context??

For Example: if the interface for a circuit is say at 60% on average then suddenly drops to 5% in 4 polling cycles, to trigger it an alert

[Bob_Keville]

Hi Stuart - did Mike's responses answer your question? 

 

Bob

[Stuart Weenig]

The linked article doesn't state it outright and Michael Poller's statement of "as far as i can tell" isn't very reassuring. 

 

However, we've gone ahead and rolled out this feature in NetVoyant with the assumption that we will, as far as we can tell, be able to deliver this same functionality when we move to PM.