Stateless SRv6 Point-to-Multipoint Path

The Segment Routing (SR) for unicast or Point-to-Point (P2P) path is described in . For SR multicast or Point-to-Multipoint (P2MP) path/tree, it may be implemented through using multiple SR P2P paths. The function of a SR P2MP path/tree from an ingress node to multiple (say n) egress/leaf nodes is implemented by n SR P2P paths. These n P2P paths are from the ingress to those n egress/leaf nodes of the P2MP path/tree. This solution may waste some network resources such as link bandwidth. An alternative solution proposed in uses a number of P2MP chain tunnels to implement a P2MP path/tree from an ingress to n egress/leaf nodes. Each P2MP chain tunnel is a tunnel from the ingress to a leaf node as its tail end and may have some leaf nodes as its bud nodes along the tunnel. This alternative solution improves the usage of network resources over the solution above using pure P2P paths. However, these two solutions are based on SR P2P paths. A solution for a SR P2MP path/tree using a P2MP multicast tree is proposed in . For a SR P2MP path/tree from an ingress/root to multiple egress/leaf nodes, a multicast P2MP tree is created to deliver the traffic from the ingress/root to the egress/leaf nodes. The state of the tree is instantiated in the forwarding plane by a controller such as PCE at Root node, intermediate Replication nodes and Leaf nodes of the tree. This is not consistent with the SR principles in which no state is stored at the core of the network. This document describes a new solution for a SRv6 Point-to-Multipoint (P2MP) Path/Tree to deliver the traffic from the ingress of the path to the multiple egresses/leaves of the path in a SR domain. This solution uses a P2MP multicast tree without storing its state in the core of the network for a SR P2MP path/tree like a SR P2P path. For distinguishing a SRv6 P2MP path/tree used in the other solutions with storing some states in the core, a new name, called stateless SRv6 P2MP path/tree, is used in the solution in this document. Even though SRv6 P2MP path/tree and stateless SRv6 P2MP path/tree are used interchangeably in the document, they both mean stateless SRv6 P2MP path/tree.

For a SR P2P path from its ingress to its egress, a segment list for the path is provided to the ingress. The ingress pushes the list into a packet, and the packet is delivered to the egress according to the segment list without any state in the core of the network. For a SR P2MP path from its ingress to multiple egress/leaf nodes, a segment list for the P2MP path is provided to the ingress. The ingress pushes the list into a packet, and the packet is delivered to the multiple egress/leaf nodes according to the segment list without any state in the core of the network. shows a SR P2MP path from ingress/root R to four egress/leaf nodes L1, L2, L3 and L4. Nodes P1, P2, P3 and P4 are the transit nodes of the P2MP path. Suppose that X-m is the segment identifier (SID) of node X. X-m is an adjacent SID or node SID. For simplicity, we assume X-m is a node SID in the illustrations below. R-m, P1-m, P2-m, P3-m, P4-m, L1-m, L2-m, L3-m and L4-m are the SIDs of the nodes on the SR P2MP path. They are multicast SIDs or replication SIDs in general. A multicast SID is a SID from a multicast SID block. In a SR domain supporting SR multicast, each node has a multicast node SID, which is globally significant. A multicast SID of a node on a SR P2MP path is associated with the SIDs of its next hop (or say downstream) nodes. When the node receives a packet with its multicast SID, it duplicates and sends the packet to each of its next hop nodes according to their SIDs. If node P on a SR P2MP path has B (B > 1) next hop nodes along the path, the SID of node P, P-m, MUST be a multicast SID when it is in the segment list for the P2MP path. The SIDs of the B next hop nodes just follow P-m in the segment list. When node P receives the packet with P-m as destination address (DA), it duplicates and sends the packet to each of the B next hop nodes along the P2MP path.

<P1-m, P2-m, P3-m, L1-m, L2-m, P4-m, L3-m, L4-m> is a segment list for the SR P2MP path in to be pushed into a packet at ingress/root R. Node P1 has 2 next hop nodes P2 and P3 along the P2MP path. The next hop nodes' SIDs P2-m and P3-m follow P1-m, which is P1's multicast SID. When P1 receives a packet with DA = P1-m transported by the P2MP path, it duplicates and sends the packet to its next hop nodes P2 and P3 according to P1-m, P2-m and P3-m. The number of branches or next hops from node P1 is a value of one argument in P1-m, called N-Branches. The value of N-Branches in P1-m is 2. With this information, node P1 duplicates and sends the packet to 2 next hop nodes P2 and P3, which are indicated by the 2 SIDs P2-m and P3-m following P1-m. The number of SIDs under node P1 is a value of another argument in P1-m, called N-SIDs. It is the number of the SIDs encoding the sub-trees from P1 and the SIDs following. The sub-trees are encoded by 7 SIDs following P1-m in the segment list. The value of N-SIDs in P1-m is 7. Since there are 2 branches or next hops (i.e., L1 and L2) from node P2, the value of N-Branches in P2-m is 2. The two sub-trees from P2 are encoded by 2 SIDs (i.e., L1-m and L2-m) and there are 3 SIDs (i.e., P4-m, L3-m, L4-m) following them. The value of N-SIDs in P2-m is 5 (2 + 3). With this information, before sending the packet to node P2, node P1 sets DA to P2-m, SL in SRH to 5 (the N-SIDs in DA = P2-m), and sends the packet to DA (i.e., P2). Since there are 1 branch or next hop (i.e., P4) from node P3, the value of N-Branches in P3-m is 1. The sub-tree from P3 is encoded by 3 SIDs (i.e., P4-m, L3-m and L4-m) and no SIDs following them. The value of N-SIDs in P3-m is 3. With this information, before sending the packet to node P3, node P1 sets DA to P3-m, SL in SRH to 3 (the N-SIDs in DA = P3-m), and sends the packet to DA (i.e., P3). Each node on the SR P2MP path sends the packet to its next hop nodes according to the segment list and no state is stored in any transit node (i.e., the core of the network). The packet is delivered to the egress/leaf nodes from the ingress.

For a sub-tree ST of a SR P2MP path from the ingress node of the P2MP path, suppose that the multicast SID of the next hop node NH is mSID; there are B branches (i.e., outgoing interfaces) to the next hop node BNH-j (j = 1, ..., B) from node NH along the sub-tree, the multicast SID of BNH-j is mSID-j; SidSeq-j (j = 1, ..., B) is the SID sequence in the segment list encoding the sub-trees from node BNH-j. Sub-tree ST is encoded as segment list

\___/ \____________________/ \______/ \________/ SIDs of NH B branches/next-hops sub-trees sub-trees BNH-j of node NH from BNH-1 from BNH-B ]]> where mSID contains the number of branches in its N-Branches field, which is B, and the number of SIDs in its N-SIDs field, which is the number of the SIDs encoding the sub-trees from NH and the SIDs following (No SID following in this case). The SIDs following mSID encode the sub-trees. The value of N-SIDs field in mSID is B plus the number of the SIDs in SidSeq-1, ..., SidSeq-B. mSID-j (j = 1, ..., B) contains the number of branches in its N-Branches field, which is the number of branches from node BNH-j, and the number of SIDs in its N-SIDs field, which is the number of the SIDs in SidSeq-j to SidSeq-B. For the P2MP path in Figure 1 from ingress node R to egress nodes L1, L2, L3 and L4, there is one sub-tree from R. Suppose that the multicast SIDs of P1, P2, P3, P4, L1, L2, L3 and L4 are P1-m, P2-m, P3-m, P4-m, L1-m, L2-m, L3-m and L4-m respectively. The sub-tree is encoded as segment list

\__/ \___________/ \________/ \______________/ SIDs of P1 2 branches/next-hops sub-trees sub-tree P2 and P3 of node P1 from P2 from P3 where ]]> L1-m, L2-m is the SID sequence (SidSeq-1) in the segment list encoding the sub-trees from P2. P4-m, L3-m, L4-m is the SID sequence (SidSeq-2) in the segment list encoding the sub-tree from P3. P1-m's N-Branches field is set to 2 since there are 2 branches from P1 and its N-SIDs field to 7 since there are 7 SIDs following P1-m, which "points" to the sub-tree from P1. P2-m's N-Branches field is set to 2 since there are 2 branches from P2 and its N-SIDs field to 5 since there are 5 SIDs in SidSeq-1 and SidSeq-2. The N-SIDs = 5 acts as a pointer to the sub-tree from P2. P3-m's N-Branches field is set to 1 since there is 1 branch from P3 and its N-SIDs field to 3 since there are 3 SIDs in SidSeq-2. The SIDs = 3 acts as a pointer to the sub-tree from P3. P4-m's N-Branches field is set to 2 and its N-SIDs field to 2. shows in details the segment list, which is an encoding of the sub-tree of the SR P2MP path from R via P1 to L1, L2, L3 and L4.

A bud node is considered as a loopback leaf of itself. The bud node will have one more branch for this loopback leaf. For example, suppose that L4 is a bud node and connected to a leaf L5 (not shown in ). The N-Branches in L4-m as multicast SID of bud L4 is 2 since there are 2 branches from L4: one to L5 and the other to L4 itself as a leaf. shows in details the segment list, which is an encoding of the sub-tree of the SR P2MP path from R via P1 to L1, L2, L3, L4 and L5. For L4-m as multicast SID of bud L4, its N-Branches = 2, N-SIDs = 2. The N-SIDs = 2 acts as a pointer to the sub-tree from L4. This sub-tree has 2 branches: one from L4 to L5, and the other from L4 (loopback) to L4 itself. The others in are the same as or similar to those in .

This section describes the procedures or behaviors on the ingress, transit and egress/leaf node of a SR P2MP path to deliver a packet received from the path to its destinations.

For a packet to be transported by a SR P2MP Path, the ingress of the P2MP path duplicates the packet for each sub-tree of the SR P2MP path branching from the ingress, pushes the segment list encoding the sub-tree into the packet by executing H.Encaps and sends the packet to the next hop node along the sub-tree. Regarding to the finite size of the segment list, a sub-tree can be "split" into multiple sub-trees such that each of the sub-trees can be encoded in the segment list of the finite size. For example, there is one sub-tree from the ingress R of the SR P2MP path in via next hop node P1 towards egress/leaf nodes L1, L2, L3 and L4. For this sub-tree, the ingress R duplicates the packet, set the destination address (DA) to P1-m (i.e., multicast SID of node P1), pushes the segment list without P1-m (i.e., <P2-m, P3-m, L1-m, L2-m, P4-m, L3-m, L4-m>) encoding the sub-tree into a Segment Routing Header (SRH) of the packet by executing H.Encaps and sends the packet to DA (i.e., node P1). The contents of the multicast SIDs P1-m, P2-m, P3-m, L1-m, L2-m, P4-m, L3-m, L4-m are shown in . Suppose that the duplicated packet is Pkt0 for the sub-tree. The execution of H.Encaps pushes an IPv6 header (i.e., SRH) to Pkt0 and sets some fields in the header to produce an encapsulated packet Pkt'. Pkt' is represented in the following:

]]> where DA=P1-m means that the destination address (DA) is set to P1-m; SA=R means that the source address (SA) is set to R; SL=7 means that the number of Segments Left (SL) is 7.

When a transit node of a SR P2MP path receives a packet transported by the P2MP path, the DA of the packet is a multicast SID of the node and the packet contains a segment list for the next hops and the sub-trees of the transit node. The DA and the segment list comprise the information for encoding the sub-trees. For example, when node P1 receives a packet transported by the SR P2MP path in , the packet's DA is P1-m (which is a multicast SID of node P1) and the segment list in the packet is <P2-m, P3-m, L1-m, L2-m, P4-m, L3-m, L4-m>. The N-Branches field (which has value of B) of the DA indicates that there are B branches or next hops from the transit node. The N-SIDs field of the DA indicates the number of SIDs for the B sub-trees from the transit node. The multicast SIDs of the B next hop nodes are the first B multicast SIDs of the segment list in the packet. For example, the N-Branches field (which has value of 2) of DA = P1-m indicates that there are 2 branches or next hops from node P1. The N-SIDs field (which has value of 7) of the DA = P1-m indicates that there are 7 SIDs for the 2 sub-trees from node P1. The first multicast SID (P2-m) of the segment list is the SID of the first next hop node (P2); The second multicast SID (P3-m) of the segment list is the SID of the second next hop node (P3). After the multicast SIDs of the next hop nodes, there are B SidSeqs (SIDs sequences) for the B sub-trees. The N-SIDs field (which has value of S1) of the first multicast SID of the next hop nodes indicates that there are S1 SIDs from SidSeq-1 to SidSeq-B; the N-SIDs field (which has value of S2) of the second multicast SID of the next hop nodes indicates that there are S2 SIDs from SidSeq-2 to SidSeq-B; and so on. For example, there are 2 SidSeqs for the 2 sub-trees from node P1 after the multicast SIDs P2-m and P3-m of the next hop nodes P2 and P3. The N-SIDs field of P2-m (the first multicast SID of the next hop nodes) has value of 5, indicating that there are 5 SIDs from SidSeq-1 to SidSeq-2. The N-SIDs field of P3-m (the second multicast SID of the next hop nodes) has value of 3, indicating that there are 3 SIDs from SidSeq-2. The transit node duplicates the packet for each next hop under it, sets the DA of the duplicated packet to the multicast SID of the next hop, SL in SRH to the N-SIDs in the DA, and sends the packet to the DA (i.e., the next hop). For example, node P1 duplicates the packet for the first next hop P2, sets DA to P2-m (multicast SID of P2), SL in SRH to 5 (N-SIDs in P2-m), and sends the packet Pkt' to DA (i.e., P2).

]]> Node P1 duplicates the packet for the second next hop P3, sets DA to P3-m (multicast SID of P3), SL in SRH to 3 (N-SIDs in P3-m), and sends the packet Pkt' to DA (i.e., P3).

]]> The behavior of Multicast SID is executed by node N when the DA of the packet received by N is N's Multicast SID. It is a variant of the Endpoint behavior in Section 4.1 of with the change from S13 - S15 to S13a - S15b below.

This change duplicates the packet for each of B branches or sub-trees from N, sends the duplicated packet to the next hop node along the branch through setting the DA of the duplicated packet to the multicast SID of the next hop node, SL in SRH to the N-SIDs in DA to pop SIDs and have the SIDs sequence encoding the sub-trees from the next hop at the top of the segment list in SRH, and submitting the duplicated packet to the egress IPv6 FIB lookup for transmission to the new destination DA (i.e., the next hop).

When an egress node of a SR P2MP path receives a packet transported by the P2MP path, the DA of the packet is the Multicast SID of the egress node and SL = 0. The egress node proceeds to process the next header in the packet (refer to S03 in Section 4.1 of ).

A controller such as PCE can compute a stateless SRv6 P2MP path and send it to its ingress. For a packet to be transported by the path, the ingress encapsulates the packet with the path and the packet will be delivered to the egresses of the path without any states in the network core. An example architecture using PCE as a controller is illustrated in . There is a connection (i.e., PCE session) between the PCE and (the PCC running on) each of the PEs, which are possible ingress nodes in the network domain. Note that some of connections between the PCE and PEs are not shown in the figure.

The PCE has the information about the network domain from the IGP or BGP (BGP-LS). The information includes link bandwidth, link colors, node SIDs, and so on. A separate multicast SID could be provisioned on every replication node and the PCE gets the SID on the node from IGP or BGP. The PCE maintains the current status of the network resource usage in its local TED (Traffic Engineering Database), and the status of every stateless SRv6 P2MP path in its local LSP-DB (Label Switch Path Database). Upon receiving a request for a stateless SRv6 P2MP path from a user or application, the PCE computes a path based on the network resource availability stored in the TED. After a path satisfying the given constraints is found, the PCE constructs a stateless SRv6 P2MP path using the multicast SIDs of the nodes on the path and encodes the structure of the P2MP path/tree into the parameters of the SIDs. In fact, the stateless SRv6 P2MP path is a segment list consisting of multicast SIDs with parameter values. And then the PCE sends the segment list representing the path to the ingress node of the path in a PCEP message such as PCInitiate. After receiving the path from the PCE, the ingress node establishes the path by creating a forwarding entry in its FIB. For every multicast packet to be transported by the path, the forwarding entry encapsulates the packet with the segment list and the packet will be delivered to the egress nodes of the path along the path without any state in the core of the network.

Protections for a SR P2MP path can be classified into two types: global protection and local protection.

For a primary SR P2MP path from an ingress node R1 to multiple egress nodes Li (i = 1, ..., n), a backup SR P2MP path from an ingress node R1' to multiple egress nodes Li' (i = 1, ..., n) is set up to provide global protection for the primary SR P2MP path. If R1' is the same as R1, the failure of the ingress node R1 of the primary SR P2MP path is not protected; otherwise (i.e., R1' and R1 are different and connected to the same traffic source), the failure of the ingress node R1 is protected. If Li' is the same as Li (i = 1, ..., n), the failure of the egress nodes Li (i = 1, ..., n) of the primary SR P2MP path is not protected; otherwise (i.e., Li' and Li are different and connected to the same destination), the failure of the egress nodes Li is protected. When a failure happens on the primary SR P2MP path and is detected by the source of the traffic or other entity, the traffic to be transported by the primary SR P2MP path is switched to the backup SR P2MP path, which sends the traffic from its ingress node R1' to its egress nodes Li' (i = 1, ..., n).

Local protection or say Fast Reroute (FRR) of a SR P2P path is proposed in and . It can be applied to FRR of a SR P2MP path in a similar way. But FRR for SR P2MP path is more complicated. More details will be added later.

TBD

The authors would like to thank Acee Lindem, Jeffrey Zhang, Rishabh Parekh, Arvind Venkateswaran and Daniel Voyer for their valuable comments and suggestions on this draft.

For simplicity, 64 bits for Common Prefix, 16 bits for Node ID, 8 bits for the number of branches (N-Branches) and 8 bits for the number of SIDs (N-SIDs) are used when G-SRv6 compression method is applied for <P1-m, P2-m, P3-m, L1-m, L2-m, P4-m, L3-m, L4-m> at ingress node R in . The Destination Address (DA) is illustrated below in . It contains the Common Prefix of 64 bits, node P1's ID of 16 bits, the value 2 for the number of branches (N-Branches) of 8 bits, and the value 7 for the number of SIDs (N-SIDs) of 8 bits.

The IPv6 header is shown in . Ingress node R sends a packet with the IPv6 header to the DA.