A P2MP policy represents a multicast tree from the root node to a set of leaf nodes, and is a single provider tunnel. A P2MP policy can contain redundant trees from the root to leaf nodes, each with its own preference. This redundancy is implemented using Candidate Paths (CPs). Each CP represents a P2MP tree with its own Traffic Engineering (TE) constraints. The CPs can be optimized based on link failures or IGP optimizations. Each CP can contain multiple P2MP LSPs represented by path instance IDs. The CP can perform make-before-break between these path instances (P2MP LSPs).
A P2MP policy is relevant only on the root node where the P2MP tree is instantiated. The P2MP policy, which is identified by the tuple <root ID, tree ID>, does not include forwarding information for the P2MP LSP. The policy only contains information about the root and leaf nodes and the TE, which is required to set up the tree from the root to the leaf nodes. The forwarding information is part of the replication segment. The root, transit, and leaf nodes contain replication segments.