Per-flow QoS precedence statistics for IOS-XR

[wblackcenic]

Hello,

We are running QoS on our backbone interfaces to ensure that control-plane and VoIP traffic (Prec 5 / 6) get priority handling in the event of congestion.  In looking at the statistics using the "show policy-map interface [interface #]" command, there appears to be a higher amount of traffic matched on the Prec 5 / 6 class-maps than we would expect. 

My question is whether there is an easy way to determine the details about what specific traffic is being matched within the Prec 5 / 6 class-maps.  I understand that we could apply an inbound policy-map on every interface in our network and look at the statistics from that perspective, but due to the size of our network, that would be quite difficult to do.

Short of actually "sniffing" the traffic, is there another command on the IOS-XR platform that lets you look at traffic statistics per-flow to determine which QoS class-map might get applied?

Thanks for the help.

[xthuijs]

you could apply the span capability on the ingress interface and "capture" the packets and display them on console.

you would apply SPAN to the ingress interface with an ACL matching that traffic you'd be interested in.

Then there is an NP counter that you can use for that : ACL_CAPTURE_NO_SPAN

and you can set an np monitor on that.

the NP monitor is described here:

https://supportforums.cisco.com/document/122386/asr9000xr-how-capture-dropped-or-lost-packets

xander

[wblackcenic]

This explanation makes perfect sense, however, this is a little more intrusive than I was looking for.  I was hoping there was a command similar to the IOS command "show ip accounting" that would give you a brief snapshot of flows traversing a particular router.  It looks like the SPAN method or looking at Netflow flow data is going to be the answer. 

If there is another method you're aware of, I'd love to hear of it.  Otherwise, thank you for your response. 

[xthuijs]

yeah i was going by the requirement of having it all on box, then the only way would be the

acl span with the np monitor counter option described.

alternatively, least impacting will be span acl to a port which connects to a sniffer, but has an offbox requirement and an additional port on a linecard as need. Or it can span to a pw to some remote location.

finally, yeh as you suggest netflow is indeed an option, but that would sample all traffic, regardless of prec, but will record it, and requires then an off box collector found directly attached or routed somewhere in the network.

so yeah you have options :)

cheers!

xander