In the lab I have some 7206VXRs (and 7301s, but lets concentrate on 7206 for now) which are acting as PE routers. I wish to test some basic QoS policies on these for customer facing VoIP services.

On the 7206VXR (with NPE-G1) for example I have two kinds of customers that need QoS'ing; PPP (ADSL) users received over L2TP via a wholesale provider and Ethernet customers.

PPP users are terminated on Virtual-Interfaces, and I can apply QoS policies to them by adding an AV pair to their RADIUS profiles (cisco-avpair = "sub-qos-policy-out=policy-name"). No problems there.

For Ethernet subscribers, they are terminated on separate sub-interfaces as VLANs, and I can apply QoS on the sub-interfaces as follows;

int gi0/0.100
 description Customer 1
 encapsulation dot1q 100
 ip addr
 service-policy output customer-policy

int gi0/0.101
 description Customer 2
 encapsulation dot1q 101
 ip addr
 service-policy output customer-policy

and so on...

I haven't applied any QoS policies anywhere yet, I have simply shown above how and where I intend to enable them. What I want to know before I enable them though, is how can I monitor the 'strain' (computational and resource overhead) placed on the router after applying these policies?

My understanding here is that as we add more subscribers to the router with QoS enabled (not all would need QoS if they don't have VoIP services) more memory resources (buffers) and CPU cycles are going to be required because our ADSL users are often congested (it probably won't be so bad for Ethernet customers as they typically have faster access circuits).

The question is two fold really; how can I monitoring buffer pool usage for QoS policies, and should I? Should I be measuring something else? I am already graphing CPU usage and total memory usage via SNMP (Cacti/Observium/etc), is that all I need, is just watching those figures sufficient? I don't want to pile on subscribers and QoS policies and then exhaust the routers resources and drop packets even though it still has plenty of bandwidth to spare.

  • Did any answer help you? if so, you should accept the answer so that the question doesn't keep popping up forever, looking for an answer. Alternatively, you could provide and accept your own answer. – Ron Maupin Aug 8 '17 at 15:38

as usually, big(modullar) devices and switches have hardware accelerated QOS. So you cant see load of CPU, because qos is implemented in ASIC. Ofcource, it is limited, but limit is in packet per second(1mpps of cef switching npe g1, 2mpps of cef switching - npe g2) , not in bit per second


about your second question, good question ;-) I guess, you shouldnt to monitor qos directly, it is enough to monitor drops, and if you will have them, throubleshoot it - tune buffers\queues\thresholds etc

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.