[c-nsp] Input Queue 7500

[Anton Kapela]

> I've noticed the input queue interface ethernet from my 7513 
> (drops and flushes is increasing)

That counter doesn't mean what Cisco lets you think it means.... 

>  Encapsulation ARPA, loopback not set
>   Keepalive set (10 sec)
>   Full-duplex, 100Mb/s, 100BaseTX/FX
>   ARP type: ARPA, ARP Timeout 04:00:00
>   Last input 00:00:00, output 00:00:00, output hang never
>   Last clearing of "show interface" counters 00:35:20
>   Input queue: 5/75/1053/6148 (size/max/drops/flushes); Total 
> output drops:
> 0   <<<<< SEE HERE!

> This behavior is normal, for that traffic (About 45Mbps in 
> some cases) ???

That should not be observed unless you're: running cef-less, being
directly attacked, or attached to noisy-ass layer-2 that faces, oh, say,
a few thousand collocation customers with 50 or 60 ip address
secondary's configured. Any of those situations could be causing what
you're seeing, and let me be clear: none of them would be characterized
as normal.

> How can I decrease that counters? some suggestions for set up 
> in Ethernet Interface???

First of all, 'sh int' input queue is a depiction of the state of
packets which were placed into queue for the "ip_input" process; i.e.
these are packets waiting to be "process switched." This is generally
bad. Before you go any further in troubleshooting this, check to ensure
that cef is enabled on your router. 

If it is, then start debugging and checking your switching path. You can
do this by becoming more familiar with the 7500 switching architecture.
I'd suggest the following URLs to get yourself started:

1)	http://tinyurl.com/7fw4p

Be sure to focus on this part of the page:

2)	http://tinyurl.com/wg8qz


Also, use the following commands on interfaces which you're seeing the
input queue drops on:

#sh int FastEthernet3/0/0 switching

#sh int FastEthernet3/0/0 stats

#sh int FastEthernet3/0/0 accounting

In a nutshell, when you see actual ip_input queue drops (which is what
the 'sh int' is displaying), things are "not good." 

Any protocol which requires RSP/RP attention must be punted to the host
processor. This includes stuff like certain layer-2 ethertype
broadcasts, ARP, BFD, OSPF Hellos, certain other multicast, DHCP, and
CDP, in addition to packets directly addressed to the routers interfaces
or control plane. If cef is disabled entirely on your box, then this
counter will be including buffered packets about to be switched, which
it normally (when cef is running) would not. 

You can get a feel for what protocol 'groups' are listening (read:
configuring cef to punt packets up to the host rsp/rp) on an interface
by looking at the routers hardware & software IDB allocation. To see it,
use:

#sh idb

[snip]

HWIDB#1   Serial1/0/0:1 (HW IFINDEX(10001), HW SB CDP(4), Serial(3), MAC
ADDR(1))
HWIDB#2   Serial1/0/0:2 (HW IFINDEX(10001), HW SB CDP(4), Serial(3), MAC
ADDR(1))
HWIDB#3   Serial1/0/0:3 (HW IFINDEX(10001), HW SB CDP(4), Serial(3), MAC
ADDR(1))
[snip]

SWIDB#1   Serial1/0/0:1 (SW CDP(4), NetBIOS(2), DSS(3), KEEPALIVE(1))
SWIDB#2   Serial1/0/0:2 (SW CDP(4), NetBIOS(2), DSS(3), KEEPALIVE(1))
SWIDB#3   Serial1/0/0:3 (SW CDP(4), NetBIOS(2), DSS(3), KEEPALIVE(1))
[snip]

Good luck,

-Tk