Spanning-tree, portfast and network loops

[clsmns]

I have a medium-sized network with approx. 20 3500-XL access switches, each with 1 GBIC 5484 (SX) down to a central Catalyst 6509 with gigabit boards.

Each first gigabit interface in each access switch is running spanning-tree normally. Each fastethernet interface in each access switch is configured with the portfast feature, with the exception of the last one which is used as a backup link with a UTP cable down to the central switch.

The problem is that a couple of time some other people have move cables and they disconnected the UTP backup link. After that they reconnected this backup link to another interface which is configured with portfast and this have generated a network loop and everything crashed.

So, I would like to receive suggestions in order to avoid the full crash of the entire network when this happens again. I suppose that when a loop of this nature occurs there is a kind of a broadcast storm, bpdus of the stp trying to converge. But, how can I limit them? How can I instruct the central switch to shutdown only the giga interface to the switch that is originating the loop?

Thanks in advance,

Hector Miranda

IBM de Chile

hmiranda@cl.ibm.com

[efrahim]

Nothing much can be done here- except put each switch in its own vlans that would help you to reduce the problem to one vlan or mark the cable so that no one move.

i think the reason is - the switch when in portfast , skip learning and listenining, n go to forwarding , but actually it suppose to go in the blocking..

There is a feature bpdu guard but i think it is not available on the 3500/2900Xl switches but it is available in 2950 n 3550 that can help you here..

Here is the URL about the bpdu guard

http://www.cisco.com/warp/customer/473/65.html

hope this helps

[clsmns]

Thanks. Bpdu guard is not available on 3500XL.

But, when a network loop occurs, what is traveling over the wire? A lot of bpdus? A broadcast storm? Because if it is a broadcast storm I think I can control the level of broadcasts in each gigabit backbone interface and after a threshold is reached, simply shutdown the interface. In this case I would like to know what level (percentage) of broadcasts is acceptable in a single gigabit ethernet interface.

Can anybody tell me that?

Thanks in advance,

Hector Miranda

IBM de Chile

CCNA/CCDA/CCNP

[aybra]

Hi

not necessairly brodcast only can cause loop, any paket can create loop.

physical loop cause trouble in mac table,

the switche find that Mac address is connected to tow different ports so il send the packet to port and the other siwtch resend it from the second port and that create loop.

thanks

[milan.kulik · Accepted solution]

Hi,

BPDU guard IS AVAILABLE on 3500XL.

Just upgrade to 12.0(5)WC5a.

HTH,

Milan

[clsmns]

Thanx, I already did it.

[dh8a]

You might want to explore Uni DIrectional Link Dectection options as well. It was developed to aid in detecting the status and misconfigured connections. It should disable only the port that it detects the connection problem on.

Regards