Broken Switching Fabric

[splazo]

How can something "Break" the switching fabric? I have a Linux server that is uplinked to a Intel 510T switch that is uplinked to another Intel 510T switch which then goes gigE to a cisco 3500XL and then gigE to a cisco 6506. If I dump the packets on the server and exclude all ip addresses bound to the machine, I still see lots of traffic. Most of it is HSRP and ARPS. But I also see full TCP sessions that are going to and from other servers that are on different switches. It's not like I see all the traffic comeing from other hosts, just every second or so I'll see a speratic TCP sessions (port 80 most of the time - not broadcasts) from other hosts. Any clue on where I should start looking?

[dklem]

A couple things to check. First, check to be sure that the destination mac address is in the cam table. If not, we are going to flood the packet.

Also, please make sure that you have all your end user stations set with portfast enabled. Without it enabled, these ports can generate TCNs, which cause the cam table to go into fast aging (15 secs, vs. 300secs). Check the following document for more details and troubleshooting steps.

http://www.cisco.com/warp/customer/473/17.html

Dave