]> Huawei Tech

Kundalahalli Village, Whitefield, Bangalore Karnataka 560037 India +91-080-49160700 rahul.ietf@gmail.com

Huawei Tech

Kundalahalli Village, Whitefield, Bangalore Karnataka 560037 India +91-080-49160700 rabinarayans@huawei.com

Huawei Tech

W Chang'an Ave Beijing 560037 China zhencao.ietf@gmail.com

General ROLL template This document describes the problems associated with the use of No-Path DAO messaging in RPL and a signaling changes to improve route invalidation efficiency.

RPL specifies a proactive distance-vector based routing scheme. The specification has an optional messaging in the form of DAO messages using which the 6LBR can learn route towards any of the nodes. In storing mode, DAO messages would result in routing entries been created on all intermediate hops from the node's parent all the way towards the 6LBR. RPL allows use of No-Path DAO (NPDAO) messaging to invalidate a routing path corresponding to the given target, thus releasing resources utilized on that path. A No-Path DAO is a DAO message with route lifetime of zero, originates at the target node and always flows upstream towards the 6LBR, signaling route invalidation for the given target. This document explains the problems associated with the current use of NPDAO messaging and also discusses the requirements for an optimized No-Path DAO messaging scheme. Further a new pro-active route invalidation message called as "Destination Cleanup Object (DCO)" is specified which fulfills all mentioned requirements of an optimized route invalidation messaging. 6TiSCH architecture leverages RPL and specifies use of non-storing and storing MOP for its routing operation. Thus an improvement in route invalidation will help optimize 6TiSCH based networks.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119. The document only caters to the RPL's storing mode of operation (MOP). The non-storing MOP does not require use of NPDAO for route invalidation since routing entries are not maintained on 6LRs. Common Ancestor node: 6LR node which is the first common node on the old and new path for the child node. NPDAO: No-Path DAO. A DAO message which has target with lifetime 0. DCO: A new RPL control message type defined by this specification and stands for Destination Cleanup Object. Regular DAO: A DAO message with non-zero lifetime. This document also uses terminology described in .

RPL introduced No-Path DAO messaging in the storing mode so that the node switching its current parent can inform its parents and ancestors to invalidate the existing route. Subsequently parents or ancestors would release any resources (such as the routing entry) it maintains on behalf of target node. The NPDAO message always traverses the RPL tree in upward direction, originating at the target node itself. For the rest of this document consider the following topology:

Node (D) is connected via preferred parent (B). (D) has an alternate path via (C) towards the BR. Node (A) is the common ancestor for (D) for paths through (B)-(G) and (C)-(H). When (D) switches from (B) to (C), [RFC6550] suggests sending No-Path DAO to (B) and regular DAO to (C).

There are following cases in which a node switches its parent and may employ No-Path DAO messaging: Case I: Current parent becomes unavailable because of transient or permanent link or parent node failure. Case II: The node finds a better parent node i.e. the metrics of another parent is better than its current parent. Case III: The node switches to a new parent whom it "thinks" has a better metric but does not in reality. The usual steps of operation when the node switches the parent is that the node sends a No-Path DAO message via its current parent to invalidate its current route and subsequently it tries to establish a new routing path by sending a new DAO via its new parent.

Nodes in LLNs may be resource constrained. There is limited memory available and routing entry records are the one of the primary elements occupying dynamic memory in the nodes. Route invalidation helps 6LR nodes to decide which entries could be discarded to better achieve resource utilization in case of contention. Thus it becomes necessary to have efficient route invalidation mechanism. Also note that a single parent switch may result in a "sub-tree" switching from one parent to another. Thus the route invalidation needs to be done on behalf of the sub-tree and not the switching node alone. In the above example, when Node (D) switches parent, the route invalidation needs to be done for (D), (E) and (F). Thus without efficient route invalidation, a 6LR may have to hold a lot of unwanted route entries.

When a node switches its parent, the NPDAO is to be sent via its previous parent and a regular DAO via its new parent. In cases where the node switches its parent because of transient or permanent parent link/node failure then the NPDAO message is bound to fail. RPL assumes communication link with the previous parent for No-Path DAO messaging. RPL allows use of route lifetime to remove unwanted routes in case the routes could not be refreshed. But route lifetimes in case of LLNs could be substantially high and thus the route entries would be stuck for long.

No-path DAO is sent by the node who has switched the parent but it does not work for the dependent child nodes below it. The specification does not specify how route invalidation will work for sub-childs, resulting in stale routing entries on behalf of the sub-childs on the previous route. The only way for 6LR to invalidate the route entries for dependent nodes would be to use route lifetime expiry which could be substantially high for LLNs. In the example topology, when Node (D) switches its parent, Node (D) generates an NPDAO on its behalf. Post switching, Node (D) transmits a DIO with incremented DTSN so that child nodes, node (E) and (F), generate DAOs to trigger route update on the new path for themselves. There is no NPDAO generated by these child nodes through the previous path resulting in stale entries on nodes (B) and (G) for nodes (E) and (F).

A switching node may generate both an NPDAO and DAO via two different paths at almost the same time. There is a possibility that an NPDAO generated may invalidate the previous route and the regular DAO sent via the new path gets lost on the way. This may result in route downtime thus impacting downward traffic for the switching node. In the example topology, consider Node (D) switches from parent (B) to (C) because the metrics of the path via (C) are better. Note that the previous path via (B) may still be available (albeit at relatively bad metrics). An NPDAO sent from previous route may invalidate the existing route whereas there is no way to determine whether the new DAO has successfully updated the route entries on the new path. An implementation technique to avoid this problem is to further delay the route invalidation by a fixed time interval after receiving an NPDAO, considering the time taken for the new path to be established. Coming up with such a time interval is tricky since the new route may also not be available and it may subsequently require more parent switches to establish a new path.

When the switching node send the NP-DAO message to the previous parent, it is normal that the link to the previous parent is prone to failure. Therefore, it is required that the NP-DAO message MUST be tolerant to the link failure during the switching.

While switching the parent node and sending NP-DAO message, it is required that the routing entries to the dependent nodes of the switching node will be updated accordingly on the previous parents and other relevant upstream nodes.

While sending the NP-DAO and DAO messages, it is possible that the NP-DAO successfully invalidates the previous path, while the newly sent DAO gets lost (new path not set up successfully). This will result into downstream unreachability to the current switching node. Therefore, it is desirable that the NP-DAO is synchronized with the DAO to avoid the risk of route downtime.

As described in , the NPDAO originates at the node switching the parent and traverses upstream towards the root. In order to solve the problems as mentioned in , the draft proposes to add new pro-active route invalidation message called as "Destination Cleanup Object" (DCO) that originates at a common ancestor node between the new and old path. The trigger for the common ancestor node to generate this DCO is the change in the next hop for the target on reception of an update message in the form of regular DAO for the target. In the , when node D decides to switch the path from B to C, it sends a regular DAO to node C with reachability information containing target as address of D and a incremented path sequence number. Node C will update the routing table based on the reachability information in DAO and in turn generate another DAO with the same reachability information and forward it to H. Node H also follows the same procedure as Node C and forwards it to node A. When node A receives the regular DAO, it finds that it already has a routing table entry on behalf of the target address of node D. It finds however that the next hop information for reaching node D has changed i.e. the node D has decided to change the paths. In this case, Node A which is the common ancestor node for node D along the two paths (previous and new), may generate a DCO which traverses downwards in the network. The document in the subsequent section will explain the message format changes to handle this downward flow of NPDAO.

Every RPL message is divided into base message fields and additional Options. The base fields apply to the message as a whole and options are appended to add message/use-case specific attributes. As an example, a DAO message may be attributed by one or more "RPL Target" options which specifies the reachability information for the given targets. Similarly, a Transit Information option may be associated with a set of RPL Target options. The draft proposes a change in DAO message to contain "Invalidate previous route" (I) bit. This I-bit which is carried in regular DAO message, signals the common ancestor node to generate a DCO on behalf of the target node. The I-bit is carried in the transit container option which augments the reachability information for a given set of RPL Target(s).

I (Invalidate previous route) bit: 1 bit flag. The 'I' flag is set by the target node to indicate that it wishes to invalidate the previous route by a common ancestor node between the two paths.

A new ICMPv6 RPL control message type is defined by this specification called as "Destination Cleanup Object" (DCO), which is used for proactive cleanup of state and routing information held on behalf of the target node by 6LRs. The DCO message always traverses downstream and cleans up route information and other state information associated with the given target.

RPLInstanceID: 8-bit field indicating the topology instance associated with the DODAG, as learned from the DIO. K: The 'K' flag indicates that the recipient is expected to send a DCO-ACK back. D: The 'D' flag indicates that the DODAGID field is present. This flag MUST be set when a local RPLInstanceID is used. Flags: The 6 bits remaining unused in the Flags field are reserved for flags. The field MUST be initialized to zero by the sender and MUST be ignored by the receiver. Reserved: 8-bit unused field. The field MUST be initialized to zero by the sender and MUST be ignored by the receiver. DCOSequence: Incremented at each unique DCO message from a node and echoed in the DCO-ACK message. DODAGID (optional): 128-bit unsigned integer set by a DODAG root that uniquely identifies a DODAG. This field is only present when the 'D' flag is set. This field is typically only present when a local RPLInstanceID is in use, in order to identify the DODAGID that is associated with the RPLInstanceID. When a global RPLInstanceID is in use, this field need not be present. Unassigned bits of the DAO Base are reserved. They MUST be set to zero on transmission and MUST be ignored on reception.

The DCO message MAY carry valid options. This specification allows for the DCO message to carry the following options: 0x00 Pad1 0x01 PadN 0x05 RPL Target 0x06 Transit Information 0x09 RPL Target Descriptor The DCO carries a Target option and an associated Transit Information option with a lifetime of 0x00000000 to indicate a loss of reachability to that Target.

A DCO message may contain a Path Sequence in the transit information option to identify the freshness of the DCO message. The Path Sequence in the DCO and should use the same Path Sequence number present in the regular DAO message when the DCO is generated in response to DAO message.

The DCO-ACK message is sent as a unicast packet by a DCO recipient in response to a unicast DCO message.

RPLInstanceID: 8-bit field indicating the topology instance associated with the DODAG, as learned from the DIO. D: The 'D' flag indicates that the DODAGID field is present. This flag MUST be set when a local RPLInstanceID is used. Reserved: 7-bit unused field. The field MUST be initialized to zero by the sender and MUST be ignored by the receiver. DCOSequence: Incremented at each unique DCO message from a node and echoed in the DCO-ACK message. Status: Indicates the completion. Status 0 is defined as unqualified acceptance in this specification. The remaining status values are reserved as rejection codes. DODAGID (optional): 128-bit unsigned integer set by a DODAG root that uniquely identifies a DODAG. This field is only present when the 'D' flag is set. This field is typically only present when a local RPLInstanceID is in use, in order to identify the DODAGID that is associated with the RPLInstanceID. When a global RPLInstanceID is in use, this field need not be present. Unassigned bits of the DAO Base are reserved. They MUST be set to zero on transmission and MUST be ignored on reception.

In , node (D) switches its parent from (B) to (C). The sequence of actions is as follows: Node D switches its parent from node B to node C D sends a regular DAO(tgt=D,pathseq=x+1,I_flag=1) in the updated path to C C checks for routing entry on behalf of D, since it cannot find an entry on behalf of D it creates a new routing entry and forwards the reachability information of the target D to H in a DAO. Similar to C, node H checks for routing entry on behalf of D, cannot find an entry and hence creates a new routing entry and forwards the reachability information of the target D to H in a DAO. Node A receives the DAO, and checks for routing entry on behalf of D. It finds a routing entry but checks that the next hop for target D is now changed. Node A checks the I_flag and generates DCO(tgt=D,pathseq=pathseq(DAO)) to previous next hop for target D which is G. Subsequently, A updates the routing entry and forwards the reachability information of target D upstream DAO(tgt=D,pathseq=x+1,I_flag=x) (the I_flag carries no significance henceforth). Node G receives the DCO and invalidates routing entry of target D and forwards the (un)reachability information downstream to B. Similarly, B processes the DCO by invalidating the routing entry of target D and forwards the (un)reachability information downstream to D. D ignores the DCO since the target is itself. The propagation of the DCO will stop at any node where the node does not have an routing information associated with the target. If the routing information is present and the pathseq associated is not older, then still the DCO is dropped.

Current RPL does not provide a mechanism for route invalidation for dependent nodes. This section describes approaches for invalidating routes of dependent nodes if the implementation chooses to solve this problem. The common ancestor node realizes that the paths for dependent nodes have changed (based on next hop change) when it receives a regular DAO on behalf of the dependent nodes. Thus dependent nodes route invalidation can be handled in the same way as the switching node. Note that there is no way that dependent nodes can set the I_flag in the DAO message selectively since they are unaware that their parent/grand parent node is switching paths. There are two ways to handle dependent node route invalidation: One way to resolve is that the common ancestor does not depend upon the I_flag to generate the reverse NPDAO. The only factor it makes the decision will be based on next_hop change for an existing target to generate the NPDAO. Thus when the switching nodes and all the below dependent nodes advertise a regular DAO, the common ancestor node will detect a change in next hop and generate NPDAO for the same target as in the regular DAO. Another way is that the nodes always set the I_flag whenever they send regular DAO. Thus common ancestor will first check whether I_flag is set and then check whether the next_hop has changed and subsequently trigger DCO if required. This document recommends the approach in point 2. The advantage with I_flag is that the generation of downstream NPDAO is still controlled by the target node and thus is still in control of its own routing state.

We would like to thank Cenk Gundogan, Simon Duquennoy and Pascal Thubert for their review and comments.

IANA is requested to allocate new ICMPv6 RPL control codes in RPL for DCO and DCO-ACK messages. Code Description Reference 0x85 Destination Cleanup Object This document 0x86 Destination Cleanup Object Acknowledgement This document IANA is requested to allocate bit 18 in the Transit Information Option defined in RPL section 6.7.8 for Invalidate route 'I' flag.

The secure versions of DCO and DCO-ACK also have to be considered in the future. The seucrity considerations applicable to DAO, DAO-ACK messaging in RPL is also applicable here.

&RFC6550; &RFC2119; &RFC3552; Thingsquare

This becomes an Appendix.