RSVP Extensions for RMR

A ring node indicates in its IGP updates the ring LSP signaling protocols that it supports. This can be LDP and/or RSVP-TE. Ideally, each node should support both. If the ring is configured with RSVP as the signaling protocol, then once a ring node R_i knows the RID, its ring links and directions, it kicks off ring RSVP LSP signaling automatically.

Ring LSPs differ from regular RSVP-TE LSPs in several ways: 1. Ring LSPs (by construction) form a loop. 2. Ring LSPs are multipoint-to-point. Any ring node can inject traffic into a ring LSP. 3. The bandwidth of a ring LSP can change hop-to-hop. 4. Ring LSPs are protected without the use of bypass or detour LSPs. Ring LSP protection is akin to SONET/SDH ring protection.

After the ring auto-discovery process, each anchor node creates a LSP addressed to itself. This ring LSP contains of a pair of counter-rotating unicast LSPs. So, for a ring containing N nodes, there will be 2N total LSPs signaled. There is no need for ERO object in the Path message. The Path message for ring LSPs has the following format:

<Path Message> ::= <Common Header> [ <INTEGRITY> ] <SESSION> <RSVP_HOP> <TIME_VALUES> <LABEL_REQUEST> [ <SESSION_ATTRIBUTE> ] <sender descriptor list> <sender descriptor list> ::= <sender descriptor>| <sender descriptor list> <sender descriptor> <sender descriptor> ::= <SENDER_TEMPLATE> <SENDER_TSPEC> The anchor node creates 2 Path messages traveling in opposite directions. The SESSION format MUST be as per the description in . The anchor node which creates the LSP will insert it's own address in the "Ring node anchor address" field of the SESSION object. So effectively, the Path messages are addressed to the originating node itself. The SESSION flags of these 2 Path messages are different. The Path message sent to the CW neighbor MUST have the CW flag set in the SESSION object to signal the LSP going in the clockwise direction. The Path message sent to the AC neighbor MUST have the AC flag set to signal the LSP in the anti-clockwise direction. The details for signaling over express links will be given in a future version. When an incoming Path message is received at the ring node R_i, it consults the results of auto-discovery to find the appropriate ring neighbor. If the incoming Path message has CW direction flag set, then R_i sends a Path message to its CW ring neighbor (and vice versa). Thus, there is no need of ERO in the Path message. The Path message is routed locally at each ring based on the ring auto-discovery calculations. The RESV message for ring LSPs also uses the new RING_IPv4 SESSION object. When the Path message originated from the anchor node R_i reaches back to R_i, R_i generates a Resv message. Note that this means that anchor node is both Ingress and Egress for the Path message. The Resv message copies the same ring flags as received in the corresponding Path message. So, a Resv message for a CW LSP goes in the AC direction (unlike the Path message, which goes CW). This is done to correctly match Path and corresponding Resv messages at transit ring nodes. Upon receiving Resv message with CW flag set, the ring node will forward the Resv message to its AC neighbor. Each ring node R_i allocates CW and AC labels for each ring LSP RL_k. As the signaling propagates around the ring, CW and AC labels are exchanged. When R_i receives CW and AC labels for RL_k from its ring neighbors, primary and fast reroute (FRR) paths for RL_k are installed at R_i. Consider the following three nodes of the ring, and their signaling interactions for LSP RL_5 originating from anchor node R5:

P5_CW -> P5_CW -> Q5_CW <- Q5_CW <- ... ------ R7 --------- R8 --------- R9 ------ ... P5_AC <- P5_AC <- Q5_AC -> Q5_AC -> P corresponds to the Path message and Q corresponds to the Resv message. Also, since ring LSPs are MP2P in nature, each ring node SHOULD also signal a Path message towards anchor node. The procedure for that is as follows: When a ring node R5 receives a Path message initiated by anchor node R1(for anchor lsp "lsp1"), R5 SHOULD make a copy of the received Path message for "lsp1". R5 then modifies the sender-template object from the copied Path message for “lsp1”. In the sender-template object, R5 uses the sender address as the loopback address of node R5 and lsp-id = X. R5 then forwards this new Path message to it's ring neighbor. The original anchor Path message has sender address as loopback address of R1 and lsp-id = X. So at this point, there will be 2 different path messages existing for lsp1 First Path message will be for the anchor LSP with sender address = node R1. Second Path message will be for the ring LSP with sender address = node R5. When node R1 receives this modified Path message, it replies with the Resv message containing the same label it advertised for the original anchor lsp "lsp1". The SESSION object of the Resv message will also exactly match with the received Path message. Only the FILTER_SPEC object in the Resv message will have the sender address as loopback of node R5. As this Resv message propagates back towards R5, all the transit nodes also send the same label that they have allocated for the original anchor lsp "lsp1". So no new label routes get installed as part of signaling for this ring lsp. The anchor LSP and all of their associated ring LSPs share label routes. The label actions are described below in .

In the rings, there are no protection LSPs -- no node or link bypass LSPs, no standby LSPs and no detours. Protection is via the "other" direction around the ring, which is why ring LSPs are in counter-rotating pairs. Protection works in the same way for link, node and ring LSP failures. Since each ring LSP is a MP2P LSP, any ring node can inject traffic onto a LSP whose anchor might be a different ring node. To achieve the above, an ingress route will be installed as follows at every ring node J, for a given ring-LSP with anchor Rk (say 1.2.3.4).

1.2.3.4 -> (Push CL_J+1,K, NH: R_J+1) # CW -> (Push AL_J-1,K, NH: R_J-1) # AC CL = Clockwise label AL = Anti-Clockwise label Traffic will either be load balanced in the CW and AC directions or the traffic will be sent on just CW or AC lsp based on parameters such as hop-count, policy etc. Also, 2 transit routes will be installed for the anchor LSP transiting from node Rj as follows:

CL_J,K -> SWAP(CL_J+1,K, NH: R_J+1) #CW -> SWAP(AL_J-1,K , NH: R_J-1) #AC CL = Clockwise label AL = Anti-Clockwise label CW NH has weight 1, AC NH has higher-weight. AL_J,K -> SWAP(AL_J-1,K , NH: R_J-1) #AC -> SWAP(CL_J+1,K, NH: R_J+1) #CW CL = Clockwise label AL = Anti-Clockwise label AC NH has weight 1, CW NH has higher weight. Suppose a packet headed in anti-clockwise direction towards R5 and it arrives at node R8. Lets say that now R8 learns there is a link failure in the AC direction. R8 reroutes this packet back onto the clockwise direction. This reroute action is pre-programmed in the LFIB, to minimize the time between detection of a fault and the corresponding recovery action. At this time, R8 also sends a notification to R7 that the AC direction is not working, so that R7 can similarly switch traffic to the CW direction. These notification SHOULD propagate CW until each traffic source on the ring CW of the failure uses the CW direction.For RSVP-TE, this notification is sent in the form of PathErr message. To provide this notification, the ring node detecting failure SHOULD send a Path Error message with error code of "Notify" and an error value field of ("Tunnel locally repaired"). This Path Error code and value is same as defined in RFC 4090 for the notification of local repair. Note that the failure of a node or a link will not necessarily affect all ring LSPs. Thus, it is important to identify the affected LSPs and only switch the affected LSPs.

A ring node can go down resulting in a smaller ring or a new node can be added to the ring which will increase the ring size. In both of the above cases, the ring auto-discovery process SHOULD kick in and it SHOULD calculate a new ring with the changed ring nodes. When the ring auto-discovery process is complete, IGP will signal RSVP to begin the MBB process for the existing ring LSPs. For this MBB process, the anchor node will create a new Path message with a different "MBB ID" in the SESSION object. All other fields in the SESSION Object will remain same as the existing Path message(before the ring change). This new Path message will then propagate along the ring neighbors in the same way as the original Path message. Each ring neighbor SHOULD forward the Path message to it's appropriate neighbor based on the new auto-discovery calculations. For the ring links which are common between the old and new LSPs, the LSPs will share resources(SE style reservation) on those ring links. Note that here we are using MBB_ID in the SESSION object to share resources instead of the LSP_ID in the SENDER_TEMPLATE Object(which is used in RSVP-TE for sharing resources as described in RFC 3209 ). The LSP_ID use is reserved for a different functionality as described in section .

For RSVP-TE LSPs, bandwidths may be signaled in both directions. However, these are not provisioned either; rather, one does "reverse call admission control". When a service needs to use an LSP, the ring node where the traffic enters the ring attempts to increase the bandwidth on the LSP to the egress. If successful, the service is admitted to the ring.

. R0 . . . R1 . __________|| . . / ________| . R7 / / R2 Anti- | . / / . | Clockwise | . | / . | Clockwise v . | \ . v R6 \ R3 . \ . R5 . . . R4 Let's say that Ring node R5 wants to increase the BW for the LSP whose egress is at node R1. To achieve this BW increase, Ring node R5 has to increase BW along the LSP anchored at node R1(say lsp1). R5 makes a copy of the existing ring Path message for lsp1. R5 then modifies the sender-template object from the copied Path message for “lsp1”. In the sender-template object, R5 uses the sender address as the loopback address of node R5 and lsp-id = X+1. R5 also modifies the TSPEC object which represents the BW increase/decrease in this new Path message. R5 then forwards this new Path message to it's ring neighbor. Note that R5 MUST also continue signaling the original anchor Path message received from ring node R1 for lsp1. The original anchor Path message has sender address as loopback address of R1. Now, let’s say, node 5 wants to increase BW again for lsp1, then R5 adds a new SENDER_TEMPLATE object in the existing Path message for “lsp1” with sender address as loopback of node 5 and lsp-id = X+2. So at this point, there will be 2 different path messages existing for lsp1 First Path message will be for the anchor LSP with sender address = node 1. Second Path message will contain 2 SENDER_TEMPLATE objects as [node5, lsp-id = X+1] and [node5, lsp-id =X+2]. Similarly, if node R6 wants to increase the BW for "lsp1", it SHOULD create a new Path message containing SENDER_TEMPLATE object with sender address = loopback of node 6 and lsp-id = Y+1. Thus, the LSP-ID field is local to each sender node along the ring. If sufficient BW is available all the way towards ring node R1, then this new Path message reaches node R1. R1 generates a Resv message with the correct FILTER_SPEC object corresponding to the received SENDER_TEMPLATE object. This Resv message will also have the correct FLOWSPEC object as per the requested bandwidth. If sufficient BW is not available at some downstream (say node R9), then ring node R9 SHOULD generate a PathErr message with the corresponding Sender Template Object. When node R5 receives this PathErr message, R5 understands that the BW increase was not successful. Note that the existing established bandwidths for lsp1 are not affected by this new PathErr message. When ring node R5 no longer needs the BW reservation, then ring node R5 SHOULD originate a PathTear message with the appropriate Sender Template Object as described above. Every downstream node SHOULD then remove bandwidth allocated on the corresponding link on receipt of this PathTear message. Also, note that as part of this BW increase or decrease process, any ring node does not actually change any label associated with the LSP. So, the label remains same as it was signaled initially when the anchor LSP came up.