11-07-2017 02:18 AM - edited 03-08-2019 12:39 PM
Hi, I am puzzled about the QoS statistics on the Cisco Catalyst 3850 (model 3850-12XS), software version 03.07.05E. The CLI command "show policy-map interface" provides class-map statistics for each interface. Below is an output example of this command (real output snippet from my switch).
As you can see it is a stack of two 3850 switches, so the interface IDs are Te1/0/x (Te1/0/1 to Te1/0/12) and Te2/0/x (Te2/0/1 to To2/0/12). FYI in case it should matter: each port is configured in a port-channel across both switches: Te1/0/1 + Te2/0/1, Te1/0/2 + Te2/0/2, Te1/0/3 + Te2/0/3, etc.
But what I don't understand is: the 'packets' counters for each class-map are (as good as) identical for all interfaces on the same stack node. All Te1/0/x interfaces have identical packets counters and all Te2/0/x interfaces have identical packets counters!
This doesn't seem normal behaviour to me, or am I missing something?
Since the output of the command is listed per interface, I assume the counters should also be interface specific and not the same for each interface, no?
hostname#show policy-map interface
TenGigabitEthernet1/0/1
Service-policy input: pm-in-marking
Class-map: class-in-realtime (match-any)
624574798 packets
Class-map: class-in-buscrit (match-any)
554030082 packets
Class-map: class-default (match-any)
26851783296 packets
Match: any
TenGigabitEthernet1/0/2
Service-policy input: pm-in-marking
Class-map: class-in-realtime (match-any)
624574798 packets
Class-map: class-in-buscrit (match-any)
554030082 packets
Class-map: class-default (match-any)
26851783296 packets
Match: any
TenGigabitEthernet1/0/3
Service-policy input: pm-in-marking
Class-map: class-in-realtime (match-any)
624577296 packets
Class-map: class-in-buscrit (match-any)
554031562 packets
Class-map: class-default (match-any)
26851841398 packets
Match: any
TenGigabitEthernet2/0/1
Service-policy input: pm-in-marking
Class-map: class-in-realtime (match-any)
482963561 packets
Class-map: class-in-buscrit (match-any)
784046719 packets
Class-map: class-default (match-any)
13167926772 packets
Match: any
TenGigabitEthernet2/0/2
Service-policy input: pm-in-marking
Class-map: class-in-realtime (match-any)
482963561 packets
Class-map: class-in-buscrit (match-any)
784046719 packets
Class-map: class-default (match-any)
13167926772 packets
Match: any
TenGigabitEthernet2/0/3
Service-policy input: pm-in-marking
Class-map: class-in-realtime (match-any)
482963561 packets
Class-map: class-in-buscrit (match-any)
784046719 packets
Class-map: class-default (match-any)
13167926772 packets
Match: any
PS: the same symptom applies to statistics via SNMP: OID .1.3.6.1.4.1.9.9.166.1.15 = cbQosClassMapStats also provides the same packet counters for each interface (per stack node).
11-07-2017 02:50 AM
Hi. i never noticed this but see the same thing.
I always look at the bytes output / conformed bytes and these counters are different for each interface.
Could be the matched packets are cumulative and the interface specific counters are bytes outpunt and conformed bytes.
11-07-2017 05:56 AM
Hi thanks for the feedback.
You are right, the "bytes output" counters are indeed interface specific.
Unfortunately for ingress qos policies (service-policy input) there is no such "bytes" counter in the output of command "show policy-map interface", there you only get the "packets" counter which has the problem I described earlier.
So this "bytes" counter is only available for egress qos policies (service-policy output).
Nevertheless, egress is nice too.
Do you happen to know what the SNMP OID is for retrieving this "bytes output" counter?
I tried these OIDs but without success:
- cbQosCMPostPolicyByte64 ( .1.3.6.1.4.1.9.9.166.1.15.1.1.10) : non zero values but they are quite different from the "bytes output" values in the CLI.
- cbQosCMPrePolicyByte64 ( .1.3.6.1.4.1.9.9.166.1.15.1.1.6): all zero values.
11-07-2017 06:11 AM
Hi
no, i ve never worked with specific oid s considering the 3850 switch, only with wifi controllers
Maybe just shoot a new question in this forum to where to find those.
There must be someone who knows ;-)
11-07-2017 06:34 AM
OK I will do that!
One more question regarding the "packets" counters:
I just noticed that the described problem only seems to apply to interfaces which are part of a port-channel:
- interfaces Te1/0/1 to Te1/0/8 (each one is part of a different port-channel, as described in initial question) all have the same counters, see below.
- interfaces Te1/0/9, Te1/0/10, Te1/0/12 (not part of a port-channel) do seem to have interface specific counters, see below.
So the problem must be related to the port-channels.
Do you see the same on your switches?
Interfaces in a port-channel:
hostname#show policy-map interface te1/0/1 (snippet)
629119151 packets
558116559 packets
26918031513 packets
hostname#show policy-map interface te1/0/2 (snippet)
629119969 packets
558117874 packets
26918073331 packets
hostname#show policy-map interface te1/0/3 (snippet)
629119969 packets
558117874 packets
26918073331 packets
Interfaces not in a port-channel:
hostname#show policy-map interface te1/0/10 (snippet)
440198 packets
220924 packets
7942384 packets
hostname#show policy-map interface te1/0/12 (snippet)
5263648 packets
4718536 packets
83595286 packets
11-07-2017 07:06 AM - edited 11-07-2017 07:16 AM
Hi
well i have a different qos policy for the uplinks which are in a port-channel.
And getting confused :-)
OK
Port-channel 100 has 2 members: Gi 1/1/1, Gi 2/1/1
show policy-map int Gi 1/1/1
Class-map: CM_VOIP (match-any)
30012737 packets
Match: dscp ef (46)
0 packets, 0 bytes
5 minute rate 0 bps
show policy-map int Gi 2/1/1
Class-map: CM_VOIP (match-any)
21963289 packets
Match: dscp ef (46)
0 packets, 0 bytes
5 minute rate 0 bps
Packet count differs here on the members of the Port-channel. Due to the load balancing algo of a port channel i think.
Now the access-links
sh policy-map int Gi 1/0/1
Class-map: CM_VOIP (match-any)
21170853 packets
Match: dscp ef (46)
0 packets, 0 bytes
sh policy-map int Gi 1/0/2
Class-map: CM_VOIP (match-any)
21170853 packets
Match: dscp ef (46)
0 packets, 0 bytes
5 minute rate 0 bps
Same amount of packets. Overhere it s the other way around, interfaces which are not in a port-channel have the same amount of packets, of which i said earlier i think are cumulative of all 48 interfaces.
Members of port-channels have different counts, i think due to load balancing.
What about switch 2
show policy-map int 2/0/1
Class-map: CM_VOIP (match-any)
15183569 packets
switch 2 (interface 2/0/1) has different counters, so i think its cumulative on a per switch base.
11-07-2017 07:25 AM
Hi
my last post seem to have disappeared here.
i post this again short:
What i see:
interfaces / members of a port channel here have different counts, i think because of load balancing algo on a port-channel
interfaces which are not in a channel all have the same counts on a per switch base in a stack
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide