[c-nsp] Tx Queue for 7609 WS67xx cards

Fri Aug 30 11:12:57 EDT 2013

On 30/08/13 15:49, Victor Lyapunov wrote:
> Hello all
>
> Got a question about queuing functionality in WS67xx cards. We have a
> typical 7609 setup with TenG uplink and GigE downlink interfaces.
>
> During the day we see an increase in packet drops on the GigE downlink
> interfaces. The average bandwidth is kept around 750Mbps but I suppose
> short bursts may result in dropped packets. Is there any way to improve
> this behavior? (e.g increase output buffer?)

Since you've got QoS enabled, you need to pay attention to:

  1. The division of the buffer space between queues
  2. The RED drop threshold(s) within those queues
  3. What queues your QoS config has traffic being put into

Why do you have QoS enabled? Is it just for CoPP? If so, all your 
traffic will be going into queue 1, and the default config only gives it 
a fraction of the total buffer space, and 70% drop threshold. You can 
see this in your "sh queueing" output:

> lab7609#sh mls qos queuing interface g1/22
>   Weighted Round-Robin
>    Port QoS is enabled
>    Trust state: trust DSCP
>    Extend trust state: not trusted [COS = 0]
>    Default COS is 0
>      Queueing Mode In Tx direction: mode-cos
>      Transmit queues [type = 1p3q8t]:
>      Queue Id    Scheduling  Num of thresholds
>      -----------------------------------------
>         01         WRR                 08
>         02         WRR                 08
>         03         WRR                 08
>         04         Priority            01
>
>      WRR bandwidth ratios:  100[queue 1] 150[queue 2] 200[queue 3]
>      queue-limit ratios:     50[queue 1]  20[queue 2]  15[queue 3]  15[Pri
> Queue]

^^^ pay close attention to the above values

>
>      queue tail-drop-thresholds
>      --------------------------
>      1     70[1] 100[2] 100[3] 100[4] 100[5] 100[6] 100[7] 100[8]
>      2     70[1] 100[2] 100[3] 100[4] 100[5] 100[6] 100[7] 100[8]
>      3     100[1] 100[2] 100[3] 100[4] 100[5] 100[6] 100[7] 100[8]

...and these ^^^

> Is the output queue (even when we enable mls qos) 40 packets long (have
> read that Tx buffer length is around 1Mbyte)?

I don't know where you got "40" from, but the thresholds are in % and 
the weights are unitless.

> In mls mode can we configure its length using "hold-queue <len>" ?

No.

> The traffic in our network is mostly best effort (not care much about
> additional delay). Is there a way to decrease output packet drops
> for WS67xx cards operating in mls qos?

Yes, assign more queue to the BE queue.

> (If this has already been addressed in the past please provide reference to
> previous thread)

This has been discussed lots of times, and I'm not about to spend time 
googling - you can do that!