Try not to use (M)STP, but if you have to:
1) Limit the scope of STP-like domain
2) Enable BPDU filter on "customer" facing interfaces (otherwise prepare
to almost constant root bridge negotiation)
3) If you want to use multivendor environment look very carefully on
vlan to MSTP instance mapping
For example on Foundry/Brocade FWS switches some vlan id's are
restricted so you're not able to map/bind it to MSTP instance, on Cisco
you can map any vlan. Becasue you have to use same vlan to MSTP mapping
(configuration digest) in the same region - this can split your network
into two regions.
4) Make sure you aways know where root node is for each instance (just
configure it manually to be 100% sure)
5) 2 instances + default seems to work quite stable on 10-20 switches
(just my experience) and gives a kind of resource usage balancing.
My experiences with STP-like technologies are that you have to be very
restrictive - don't enable it to customer and be sure he will not create
a loop over your network. The worst thing (nightmare) is to debug
STP/MSTP when something goes wrong. If you want to apply it to big LAN,
you have a chance to win this game if you'll be careful. But if you have
a plan to build metro network (or something more) just forget about this
(if you know Brocade devices try to switch to MLX/CES platform and run
MPLS :-)
Cheers
Tomek
Brashear, Jonathan pisze:
| Article list | Name | Date |
| Click Here | Brashear, Jonathan |
||
| Currently Viewing : This Article | Tomasz Szewczyk | 2010-02-02 08:28:21 |