Frame storm on Catalyst 3500

[CHM1]

I have a backbone which is composed of 6 Catalyst C3500 in a loop (Fiber Gigabit and 100 TX); the loop is regulated by STP; these switches are powered via UPS.
There are also peripheral switches which have no UPS.
When there is an AC main power fail (every wednesday morning) and the power comes again, the peripheral switches reboot, the backbone not (which is normal), but there is a frame storm on the network and nobody can work anymore.
To resolve, we must unplug one of the redundant link in the loop.
The STP config of the switches seems good and works very well in case of breakdown of a link.
What's happening ?
Thanks

 

[wybnormal]

I think I can point you to where the problem is but I can not tell exactly what to do, pehaps someone else can. It sounds like a spanning tree issue with the redundant link. There are config issues with Spanning tree and the use of backup links like this. It may be as simple as settting up the root bridge as a hard code or blocking BPUs on the redundant link. Hopefully someone else can provide more hard details for *us*. If I read below correctly, then you do need to block the redundant port.

Mike S
------------ Cisco Notes --------------

Here is something that Cisco published:
<<Layer 2 switches run Spanning Tree Protocol (STP) to discover the Layer 2 topology and to prevent network loops. Spanning Tree enables LAN switches to be connected in redundant manner without creating loops. One drawback of Spanning Tree is that it is recomputed every time there is a change in topology or there is a link or switch failure resulting in loss of connectivity. In typical networks the minimum convergence time for the Spanning Tree is about 50 seconds based on the default values of Spanning Tree timers. Cisco has incorporated several enhancements to the STP to achieve network convergence in the order of 2-5 seconds thus increasing the network availability. Examples of such enhancements include UplinkFast, PortFast and BackboneFast.

UplinkFast helps achieve convergence in the order of 2 seconds upon failure of an active redundant uplink to the LAN backbone. With UplinkFast, the switch instantaneously cuts over to the new Spanning Tree root because the standby redundant link goes directly to the forwarding state from the blocked state. For each downstream MAC address the switch knows how to reach, it sends a proxy multicast frame on the new root port at a configured update rate to rapidly restore reachability information. The upstream switches update their tables to use the new path.

>>>

 

[bjhagroo]

Wybnormal has pointed you in the right direction. I support a site with more than 40 -5500 and 6500 series. I will give you our config. Our user ports are configured with port fast enabled, uplinkfast is disabled on all connections including inter-switch links. Portfast is disabled on all trunked switch links and most core to distribution switches are configured with fast etherchannel. We run about 40 Vlans with 150-400 users per Vlan. I have seen STP loops caused by user devices acting like routers, some servers and/or workstations running portfast enable while others are not, long STP connvergence because ROOT BRIDGE is not in the logical middle, Cisco's version of STP not running on all switches(PVST). PVST requires that Cisco ISL run on trunk links between switches. PVST runs a seperate instance of spanning tree for each Vlan(this is recommended). I use Cisco PVST with connections to a Synoptics switched subnet(not recommended by Cisco), but i just set the uplink to the CISCO VLAN on the CISCO side and on the Synoptics side to VLAN 1, no portfast and no trunking on this port. There are 3 versions of STP, make sure you are only running one. IEEE CST,PVST and PVST+. I suggest just PVST.