[c-nsp] excessive buffer overrun/ignored packets on GEIP+ interface

[Anton Kapela]

> 7507 w/GEIP+ --------> Cat6509 (MM fibre)

First, we need more info about your config. I can guess what vip you
have on the geip+, but let us know. Also, cef or dcef? Tell us what
other hardware you're running, i.e. provide info on the types of pa,
vip, and int processors in your router.

> I have plenty of headroom on the link, so it isnt a BW 
> availability problem. 
> 
> Anyone have any clues as to what's causing this? 

Often this is how rx-side buffering manifests in dcef mode. On an
example router here, vip 4 holds a pa-fe-tx and a pa-t3. Vip 4's
accumulator will hold packets in buffer whenever the final output pa &
vip that a packet is destined for has no free memd buffer space
available. Example:

#sh controllers vip 4 accumulator

[snip]
Serial1/0/0:21:
 MEMD txacc 0x0122: 2065268 in, 10958 drops (0 paks, 0/113/375 bufs)
1536kbps
   No MEMD acc: 2065268 in
      Limit drops  : 10958 normal pak drops, 0 high prec pak drops
      Buffer drops : 0 normal pak drops, 0 high prec pak drops

Checking the int, we see:

  Last clearing of "show interface" counters 15w5d
  Input queue: 0/75/0/0 (size/max/drops/flushes); Total output drops:
41463
  Queueing strategy: VIP-based fair queuing
  5 minute input rate 0 bits/sec, 0 packets/sec
  5 minute output rate 0 bits/sec, 0 packets/sec

..so, this is expected. Total pa/vip output drops exceed the rx-side vip
accumulator drops, i.e. ~40k egress drops and ~10k ingress drops;
downloadin' fools with bittorrent, perhaps. 

Now, the ds3/pa-tx:

FastEthernet4/0/0:
 local txacc 0x1A0A: 650 in, 0 drops (0 paks, 0/24414/24414 bufs)
100000kbps
Serial4/1/0:
 local txacc 0x1A12: 5604 in, 0 drops (0 paks, 0/10793/10793 bufs)
44210kbps

..as we expect, no input drops on are shown, but we can see at least one
t1 on vip 1 had enough output queue abuse to have rx size drops in the
input vip. We don't see any pa-level input drops or overruns, so this
means our vip was able to throw away packets fast enough, otherwise the
pa would be causing the drops.

Without more info on your setup it's hard to get very specific, but it
could be a sign that your vip cannot rx discard fast enough and at
times, the geip+ pa is doing so instead. Also possible that your cat6 is
spewing non-cef-switchable packets at your router at a high rate thus
being discarded. Check the #sh cef not-cef-switched statistics, see if
these counters correlate with your input discards at all. 

Finally, read over these urls:

http://www.cisco.com/en/US/products/hw/modules/ps3088/products_tech_note
09186a0080094345.shtml

http://www.frameip.com/dos-cisco/queue_drops.pdf

-Tk