Overview:

MSTP, which uses RSTP for rapid convergence, enables VLANs to be grouped into a spanning-tree instance, with each instance having a spanning-tree topology independent of other spanning-tree instances. This architecture provides multiple forwarding paths for data traffic, enables load balancing, and reduces the number of spanning-tree instances required to support a large number of VLANs.

All the details can be found here.

My personal field experience:

It’s funny, I always hear how great MST is (and it does have advantages) but, I only saw it in a production enterprise environment a handful of times. The primary drivers are lots of VLANs (500+) and load balancing traffic (VLAN traffic engineering) across redundant links. Service providers could use this on small metro rings that utilize STP as the ring failover mechanism (RPR may be a better option).

Configuration:

For correct operation, all switches in the MST region must agree on the same CIST regional root. Therefore, any two switches in the region only synchronize their port roles for an MST instance if they converge to a common CIST regional root.

For two or more switches to be in the same MST region, they must have the same VLAN-to-instance map, the same configuration revision number, and the same name.

SW1 (Root for Instance 2 and 3, backup root for Instance 1)

spanning-tree mst configuration

instance 1 vlan 1-100

instance 2 vlan 101-200

instance 3 vlan 201-4094

spanning-tree mst 1 priority 4096

spanning-tree mst 2 priority 0

spanning-tree mst 3 priority 0

Additional information:  
http://en.wikipedia.org/wiki/Spanning_Tree_Protocol#Multiple_Spanning_Tree_Protocol_.28MSTP.29

http://www.ieee802.org/1/pages/802.1s.html