spanning tree tcn's

carl_townshend · ‎10-31-2007

Hi all, i was told by an expert that spanning tree sometimes slows the network down, he said that TCN's can affect the network each time a pc is switched off and on without portfast, what effect do these tcn's have on the network, ie an access port going up and down ?

rolf.fischer_2 · ‎10-31-2007

As far as I know in this case (access-port, no st portfast) the TNC is forwarded to the Root-Bridge. The Root-Bridge acknowledges the TNC by setting a flag in a standard-BPDU (which are sent all 2 seconds).

As long as the Root-Bridge does like this, the expire-time of the CAM-Tables are set down to (normally) 15 seconds -> now all the MAC-Addresses have to be learned again.

So entries, which are no longer valid, will be cleared.

When the flag is no longer set, the timers are set back to (normally) 300 seconds.

I'm not sure for 100% but I think that's how it works...

HTH

Rolf

glen.grant · ‎10-31-2007

Which is why portfast is used on all access ports . If this is done normally there should be very few TCN's on the network and it should not be a problem at all . We run a very big net and we have never seen this to be a problem if you configure your switches correctly.

carl_townshend · ‎10-31-2007

so, to clariy each time a port goes down, the switch sends a tcn to the root bridge, that then forwards out a bpdu to get all switches to flush there mac address tables in 15 seconds time? does that take that network down when all switches flush there tables ?

rolf.fischer_2 · ‎10-31-2007

Normally it shouldn't. It's just a kind of "cleaning" the CAM-Tables. With every recieved frame the switches learn the MAC-Addresses again.

But I heared actually that in large networks - especially with switches of different vendors and old ones - it can take the network down.

For that reason and for avoiding DHCP-problems you should always use portfast on access-ports.

merritt81 · ‎10-31-2007

The transmission of BPDU's with the TC bit from a root bridge does not take the network down so to speak. Rather, it induces flooding of unicast traffic out all ports in the switched network. Until the receiver of that unicast flow generates traffic the flooding will continue, as the CAM table will not have mac address bound to an interface yet.

This can cause hosts on the network to receive large amounts of input, and if the network is large enough impact performance.

Francois Tallet · ‎10-31-2007

Hi Carl,

Still trying to determine if you should disable STP on the link between your access and distribution? ;-)

If you configure portfast on the edge port, there will be no topology change and no traffic interruption for the host.

Else, there is a slight difference whether you use STP or RSTP/MST.

* STP case: the cam entries are not flushed as a result of a topology change. The network enters a period of "fast aging", meaning that entries that are not revalidated in 20 seconds will be flushed. That should not introduce a lot of flooding in the network, as the big talkers will not have their entries aged out

* RSTP/MST case: here the entries are flushed and you will see much more flooding. Furthermore, during an RSTP/MST convergence, a n edge port (a port connected to a host) that is not configured for portfast might block for 30 seconds if it is synced. For RSTP/MST, it is important that you identify the edge ports in your network and configure portfast on them.

In either case, configuring portfast correctly is the correct solution.

Regards,

Francois