HI Juan,
This is not a bug. We need to dig this issue further.
The counter "Total output drops:" is incremented each time the output queue
was overloaded, and therefore some packets had to be dropped due to a lack
of
output buffers. If this counter for output drops was incrementing while you
are on the
router, and you did a "show interface", the output queue would most
definitely have been 40/40,
which means it was full. If the output queue is 0/40 like in the capture you
provided, the
output drops should not be incrementing at that particular time.
The output queue can become full during periods where the interface is
sending close to full capacity of traffic, or if there are short bursts of
high traffic rates that needs to go out on that interface.
1) If you clear the counters, how is the rate of the output drop
accumulation?
2) Verify that the cable is good by using a cable tester or replace it with
a known good cable.
3) Verify that the Network Interface Card (NIC) is compatible and working
properly. Refer to
< http://www.cisco.com/en/US/customer/products/hw/switches/ps700/products_tec
h_note09186a00800a7af0.shtml>
http://www.cisco.com/en/US/customer/products/hw/switches/ps700/products_tech
_note09186a00800a7af0.shtml
4) Verify that the flow control settings are the same on both sides of a
Ethernet link.
5)If you have transient congestion, or microbursts, you might be able to reduce drops by increasing buffer resources. On a 3750, believe that's only possible by enabling QoS, which often introduces even more drops because by default it evenly distributes resources to four egress queues rather than one. (This is why Paolo suggested disabling QoS, if enabled.)
Tuning 3750 QoS is possible, but it's somewhat complex. There's a good document on theses forums, https://supportforums.cisco.com/docs/DOC-8093, discussing 3750 QoS buffer tuning.
When tuning for drop avoidance, most try increasing queuing resources, but sometimes decreasing queuing resources is the "better" solution. The latter, though, is not well understood by most. Also a LAN 3750's advanced QoS capabilities are few.
(BTW, huge buffers/queues may avoid drops but then you may have transient high latencies.)
The simplest solution might be to obtain increased bandwidth from your provider, e.g. another 100 Mbps link or a (fractional) gig link.
If you want to minimize drops using QoS features, you'll want to replace the 3750, on this link, with an ISR. Again, though, drop reduction using QoS isn't trivial and you're probably going to have some drops even with "ideal" QoS. QoS also is the "better" solution if you have traffic with different service requirements.
Action plan:
kindly apply the following:
- Change the load interval from 5 minute to 30 seconds >> load-interval 30
"under the interface mode "
- Clear the counters.
- disable qos( conf t #no mls qos) but this will disable the qos .
- After that if you would still see the discards increasing , please capture
the following commands 2-3 times (1 minute interval between each output)
#sh controllers
#sh interfaces
#show interface counter error
#show platform port-asic stats drop
#sh interface interface-ID | i output drop
#show mls qos int stat ////////this for the interfaces seeing the
output drops
Regards
Inayath
*Plz rate the usefull posts.
