[c-nsp] Queue Drops on lightly loaded interfaces on 6500/SUP2/MFSC2

[Sam Stickland]

Hi,

I've got consistent queue drops on a couple of lightly loaded interfaces on a 
cisco 6500/SUP2/MFSC2.

Over a five minute period our monitoring system reports:

FastEthernet4/1 : (iE 0/ oE 0), (iQ 30/ oQ 4850)
FastEthernet4/5 : (iE 0/ oE 0), (iQ 0/ oQ 12)
FastEthernet4/3 : (iE 0/ oE 0), (iQ 0/ oQ 3)

(iE is input errors, oQ is output queue drops etc.).

At various points all of these interfaces have input or output queue drops of 
some description, pretty much constantly (although Fa4/1 features more heavily 
than most).

The load on these interfaces is pretty low, here's the five minute averages:

Fa4/1 : In:  27.3 Mb/s (2.7%)		Out:  29.3 Mb/s (2.9%)
Fa4/3 : In:  6619.5 kb/s (0.7%)		Out:  9576.9 kb/s (1.0%) Fa4/5 : In: 
3376.8 kb/s (0.3%)  	Out:  1750.0 kb/s (0.2%)

All of these seem to low to cause major output queue drops, unless the flows 
were really really bursty, which I doubt.

All of these ports are dot1q trunks. For ports Fa4/1 and Fa4/3 the input and 
output hold-queues have been increased to 4096.

sh int fa4/1 stats shows zero's for all counters, while fa4/3 records a tiny 
number of processed switched packets (<5 pkts/min).

> From time to time the input queue drops also show up on the SVIs, 
presumably in sudden bursts. Here's another five minute snapshot just taken:

Vlan1105 : (iE 0/ oE 0), (iQ 430/ oQ 0)
Vlan803  : (iE 0/ oE 0), (iQ 414/ oQ 0)
Vlan221  : (iE 0/ oE 0), (iQ 111/ oQ 0)

None of these interfaces seem to be excessively process switching:

#sh int vlan 221 stats
Vlan221
           Switching path    Pkts In   Chars In   Pkts Out  Chars Out
                Processor       5882     565362       1287     135592
              Route cache      41003   36676785        696      67067
        Distributed cache   10983892 2497519934   10969818 2130831460
                    Total   11030777 2534762081   10971801 2131034119

#sh int vlan 1105 stats
Vlan1105
           Switching path    Pkts In   Chars In   Pkts Out  Chars Out
                Processor        400      31708       2025     129822
              Route cache          0          0          0          0
        Distributed cache    4596762  292278134    3891159  239773059
                    Total    4597162  292309842    3893184  239902881

MFSC CPU utilisation on this box is

five seconds 17%/13%; one minute: 22%; five minutes: 23%

and for the Sup2

five seconds: 14%/5%; one minute: 23%; five minutes:  23%

So it doesn't look to be excessively overloaded.

The input queue drops are bursty enough that I haven't been able to see any of 
the packets ever fill up the input queues, and get a chance to look at "show 
buffers input-interface xxxx"

The output queue drops really don't appear to be the result of congested 
ethernet medium (although I accept they could be).

How do I go about debugging this?

I've read all of http://www.cisco.com/warp/public/63/queue_drops.html, but I 
haven't got any further.

Sam