Disclaimer
The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.
Liability Disclaimer
In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.
Posting
"If QoS is disabled, there's only one egress per port."
This suggests to me otherwise:
sh platform port-asic stats enqueue gigabitEthernet 0/11
Interface Gi0/11 TxQueue Enqueue Statistics
Queue 0
Weight 0 Frames 2
Weight 1 Frames 0
Weight 2 Frames 0
Queue 1
Weight 0 Frames 0
Weight 1 Frames 159529759
Weight 2 Frames 558551
Queue 2
Weight 0 Frames 0
Weight 1 Frames 0
Weight 2 Frames 0
Queue 3
Weight 0 Frames 0
Weight 1 Frames 0
Weight 2 Frames 695042334
Interesting! I agree it does. So if you do a show mls qos, it shows disabled? If it does show disabled, QoS was never enabled on this device? (I'm wondering whether stats would "survive" a change in QoS mode.) Otherwise, I'm also wondering (especially because of Q1 Wt1) whether WTD is still in effect, recorded per what would be a queue. If none of these, something to go to TAC with. (Might also be a stats recording bug.)
"I.e. you might be able to "borrow" buffer resources reserved for other ports to better deal with microbursts."
If i have only 5 ports enabled on the switch do i have more resources to play with?
Perhaps. What I'm thinking is you much reduce the buffer allocation per port, putting more buffers into the common pool and then hopefully a busy port will draw excess buffers from that pool. (Your original reference has about the best explanation of buffer management I've read.)
What I'm unsure about is whether, for a microburst, a port can draw buffers from the common pool fast enough to not drop packets. If that is an issue, the alternative might be to assign most of the common pool to port reserved buffers. As my understanding is this would apply to all the ports, there would be less buffer resources available per your 5 ports but in a microburst there shouldn't be any delay in obtaining the port reserved buffers.
Also keep in mind, from 10g to gig, some dropping is often normal to indicate to 10g (TCP) flows you don't have 10g of bandwidth. Counter intuitively, increasing buffer (queue) resources can increase the overall drop rate. What you can also do is calculate BDP, divide in half, and take 90% (10g - g) speed difference for the maximum optimal buffer requirement.
