[j-nsp] P-4FE-TX output discards under very light load?

[Pekka Savola]

Hello all,

We're seeing a relatively small number of output discards (usually up 
to 200 a day, usually in the dozens, some days none) on about 5 FE 
interfaces (all of them on different routers on 4xFE PICs) and this 
has been going on for quite some time.  All of the FE interfaces use 
various kinds of copper-to-fiber media converters.

The weird thing here is these interfaces are very low usage -- typical 
traffic is between 0.5-5 Mbit/s.  Even more strange, most of the time 
output drops don't happen during traffic spikes (sometimes up to 20-30 
minutes with 5 minute averages), but apparently during a quiet period.

This leads me to think that that these output drops are not caused by 
congestion of the link but by something else.

Any ideas what that 'something else' could be, or how to go about 
finding it out?

Some pointers/thoughts:
  - the increase is mostly under the "Red dropped Low, TCP" counter.
  - uRPF is applied on all interfaces
  - all interfaces ('interfaces fe-* scheduler-map ...') run a basic
    CoS scheduler/classifier
  - no output filters or rate-limiters
  - the whole PICs are also lightly loaded.  FPCs likewise.
  - some of the interfaces have longer than /30 prefix length, but
    scanning the subnet (trying to trigger ARP resolution drops)
    doesn't seem to increase output drops (are these drops counted
    anywhere?  I guess not..)
  - manual says that flow-control is not supported on 4xFE.  I wonder
    whether this could cause issues.

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings