You can place an RP on a switch when VLANs extend over several switches. However, when you use PIM-SM, ensure that you do not span VLANs on more than two switches.
Use static group-range-to-RP mappings in an SMLT topology as opposed to RP set learning using the Bootstrap Router (BSR) mechanism. Static RP allows for faster convergence in box failure and reset, whereas there are inherent delays in the BSR mechanism as follows:
When a router comes back up after a failover or reset, to accept and propagate (*,g) join requests from surrounding routers (either PIM join messages or local IGMP membership reports) to the RP, a PIM router must determine the address of the RP for each group for which they desire (*,g) state. The PIM router must have the unicast route to the RP address. The route to the RP address is learned by using a unicast routing protocol such as OSPF, and the RP address is either statically configured or dynamically learned using the BSR mechanism.
When a box comes up after a reset, if the RP is not statically configured, it must wait for the BSR to select the RP from candidate RP routers, and then propagate the RP set hop-by-hop to all PIM routers. This must be done before a join message can be processed. If the PIM router receives a join message before it learns the RP set, it drops the join message, and the router waits for another join or prune message to arrive before it creates the multicast route and propagates the join message to the RP. The default Join/Prune timer is 60 seconds, and because of this and the delays inherent in BSR RP-set learning, significant multicast traffic interruptions can occur. If the RP is statically configured, the only delay is in the unicast routing table convergence and the arrival of the Join/Prune messages from surrounding boxes.