mls qos output drops

[dan.letkeman]

Hello,

I have some 3560 switches where I have enabled the global qos command "mls qos".  When I have this enabled I notice a lot of output drops on some of the slower links.  What exactly does the mls qos command do to cause output drops?  With it on, I see about a 50 percent decrease in speed over some of our wireless links.

I'm not running using QOS as of right now, but when I want to in the future, do I need the "mls qos" command?  If I do, (and I know there is more too it than just "mls qos"), but won't that defeat the purpose of qos, if the global command is just going to slow my links down?

Thanks,

Dan.

[Edison Ortiz]

Can you post the output from the drops with and without mls qos and the switch config?

Regards

Edison

[dan.letkeman]

No sure what you mean by post the output from the drops...

Input queue: 0/75/0/0 (size/max/drops/flushes); Total output drops: 4857894

The switch config is very basic.  This is the port connect to another building with a wireless bridge.  This is the port i'm seeing the output drops on when i enable mls qos

interface GigabitEthernet0/24
switchport trunk encapsulation dot1q
switchport trunk native vlan 2056
switchport trunk allowed vlan 40,56,700,2056
switchport mode trunk

[Edison Ortiz]

And if you clear the counter on the interface and disable 'mls qos' the output drops do not increment?

That's odd.

When you enable mls qos, you activate the SRR scheduler in hardware where queues share bandwidth in a round-robin fashion, more info can be found at:

http://www.cisco.com/en/US/docs/solutions/Enterprise/WAN_and_MAN/QoS_SRND/QoSDesign.html#wp999395

When you have 'mls qos' disabled, the packets will enter and leave the switch and receive equal treatment.

I never seen SRR starving a queue unless the egress queue needs to buffer this traffic due to the slow partner link.

Also, perhaps the traffic was being dropped silently with mls qos disabled and not reported by the interface.

Are you suffering from latency issue due to these drops or the latency is the same with and without 'mls qos'?

Regards

Edison.

[otn-cco]

Was there a fix for your output drops ??

We are seeing a similar pattern on a 3750 (turn on mls qos, output drops increase, turn it of, all stops - no QoS, no SRR policies)...

Seems there is a bug ID for mls qos trust dcsp and packet drops, but not just turning on mls qos.

Cheers,

-P

[dan.letkeman]

I have not found a fix for it yet.  I'm running (C3560-IPSERVICES-M), Version 12.2(50)SE3.

[toec · Accepted solution]

Sorry for digging up an old post, but we just had this same "problem" on one of our blade switches (CBS3130 to be exact).

Also, please correct me if i seem wrong.

Since the traffic is subject to the QoS SRR scheduler and its queues, wouldnt it be subject to the configured thresholds of those queues aswell?

If i recall from a QoS course, the thresholds for the queues are made at the time when there was nothing faster than T1 internet connections. Our output queue for QoS is 100packets large on each threshold/queue, we are currently pushing ~8000pps.

This output is from our problem switch with 'mls qos' enabled, but not used in class-maps or policy-maps or anything such. Just simply enabled.

#sh mls qos interface gi1/0/1 statistics

GigabitEthernet1/0/1 (All statistics are in packets)

......

  output queues dropped:

queue:    threshold1   threshold2   threshold3

-----------------------------------------------

queue 0:            0            0            0

queue 1:     20821053            0            0

queue 2:            0            0            0

queue 3:            0            0            0

......

These dropped packets are the same amount as OutDiscards for the same interface.

I have now disabled 'mls qos' with 'no mls qos' (how else? ) and cleared my interface counters aswell as the mls qos counters, but not had it running for long enough to notice anything except that packets are still counted in the 'output queues enqueued' and its thresholds as this output shows from the same command as above:

......

  output queues enqueued:

queue:    threshold1   threshold2   threshold3

-----------------------------------------------

queue 0:           0           0           0

queue 1:    11299434           0        7956

queue 2:           0           0           0

queue 3:           0           0        6520

......

I think that if you must have 'mls qos' enabled, and if you are experiencing OutDiscards that you should tune your QoS buffers and/or thresholds.

[dan.letkeman]

I believe you are correct.  I actually found out that if I enable auto qos on a port so the switch install's all of the default qos settings, the problem is almost completely gone because I actually have the buffers and thresholds set.

Thanks,

Dan.