[c-nsp] Portchannel, ttl 1 packets filling input queue.

Drew Weaver drew.weaver at thenap.com
Tue Nov 17 13:49:12 EST 2009


Hey all,

I had been suffering from some input/output queue drops on the Catalyst side of a connection between a [Cat6500 (Sup 720-3BXL) WS-6724-SFP] and a GSR 12810 /w SIP-601 & SPA10x1GE-V2.

Since this link was tremendously busy I thought perhaps it was simply a matter of micro bursts exceeding the maximum bandwidth of the interface, and instead of upgrading to 10GE for a microburst, I decided to create a port-channel.

So I created the port channel using two ports on the 6724-SFP and two ports on the SPA10x1GE-V2.

Since the GSR doesn't support anything but etherchannel (for what reason I can't tell you) I used etherchannel.

I noticed as soon as this port-channel interface came up that the input queue was immediately getting drops/flushes so I did some:

sh buffers input-interface port-channel 1 dump

several times in there I saw this:

source: x.x.x.x, destination: y.y.y.y, id: 0x0000, ttl: 1,
TOS: 0 prot: 17, source port 32136, destination port 9810

where x.x.x.x is a host on my network and y.y.y.y is a host on the Internet.

pretty much every time I ran it I saw several packets like this (all with TTL 1).

This continued until I broke the port-channel and put everything back to how it was.

I ran that same command:

sh buffers input-interface g4/19 1 dump

On the physical interface connecting the two (without the port channel) and I didn't get the same results.

Does anyone know of any bugs or anything with port-channel, or any caveats that might explain what I am running into?

thanks,
-Drew



More information about the cisco-nsp mailing list