[c-nsp] 3850 and output drops

[Peter Rathlev]

On Tue, 2016-09-06 at 18:48 +0200, Sebastian Beutel wrote:
> Everything started when a co-worker tried to troubleshoot a rapidly
> increasing output error counter on a 3850 Switchport.

What kind of interfaces are these? Would it perhaps be downlink
interfaces that connect at a lower rate than the uplink interface on
which traffic arrives? This will exercise the output buffers.

What does "show interface <...> counters errors" say? If the errors are
"OutDiscard" then it's a buffer problem. Otherwise the kind of drops
should indicate where to continue troubleshooting.

Are there any service policies configured on the interfaces? If so,
then "show policy-map interface <...>" might shed some light on why the
drops occur.

You can see hardware counters for an interface with

  show platform qos queue stats Gi 1/0/29

The space between "Gi" and "1/0/29" seems to be necessary at least in
IOS XE 3.3.5.

You might want to take a look at

  "Catalyst 3850: Troubleshooting Output drops"
  http://www.cisco.com/c/en/us/support/docs/switches/catalyst-3850-series-switches/200594-Catalyst-3850-Troubleshooting-Output-dr.html
  (http://tinyurl.com/z6acheu)

There's an example (last "Example 2") policy-map for just assigning
maximum buffers to a single default queue. There are certain caveats
but IMHO something like that should be default on user facing network
interfaces.

-- 
Peter