RPL Routing Pathology In a Network With a Mix of Nodes Operating in Storing and Non-Storing Modes

RPL can operate in storing and non-storing modes. These modes introduce two different ways to perform downward routing. Downward routing is used when a node needs to send a packet to an arbitrary node (e.g., non-DODAG root node) in the network: the packet can go from a node "upward" towards the root and "downwards" to the final destination. The RPL specification allows operating a network with a mix of storing and non-storing modes. RFC 6550 describes special rules to operate such a network: a node that operates with a different Mode of Operation (MOP) than the DODAG root will act as a leaf node in the network. The consensus was that it is unknown if the network would work properly because no one had designed such a network and was left to be explored in the future. In this draft, we document a case in which we allow a mix of nodes operating in storing and non-storing modes to form a single network (e.g, despite having different MOPs) and introduce that RPL's two downwards routing modes, as it is, can cause a routing pathology. This pathology can partition the network, i.e., it can result in scenarios where nodes cannot send packets to the root and the root cannot send packets to the nodes even though these nodes have plenty of physical connectivity in the network. We propose one approach of modifying RPL to prevent this routing pathology. The methodology, introducing a new mode of operation (MOP), has been implemented and tested on an LLN testbed and in process of publication. It is possible there are more elegant approaches to prevent the pathology described.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119. The terminologies used in this document are consistent with the terminologies described in , , and .

Before we describe the routing pathology that arises due to the existence of a mix of nodes operating in storing and non-storing modes, we review the storing and non-storing downwards routing modes that RPL introduces. In Storing mode, a node keeps a (not necessarily) complete list of (nodid, nexthop) for nodes in its subtree. When a node receives a packet, it forwards the packet to the nexthop if the node finds the destination in the list. If it does not find the destination in the list, it forwards the packet to the preferred parent. In Non-storing mode, if a packet does not have routing path in the header, it forwards the packet to the preferred parent. The root in this mode collects and maintains topology information of the network. If the packet makes it to the root, the root computes the path to the destination based on this topology information. The root puts this path in the header and sends it to the next hop. The nodes, upon receiving a packet with a path in the header, forward the packet to the next hop as indicated in the path in the header.

We first examine the effect of this routing pathology for routing collection traffic packets. Lets consider the following network topology. A -> B -> N -> S -> Root (Storing) Note that RPL indicates that, storing mode nodes and non-storing mode nodes use a different mode of operation (MOP) field. Furthermore, if the MOP supported from a DODAG root is not supported at a RPL node, the node can only participate in the RPL network as a leaf node. Say that the Root of the topology is a storing mode node. In this case, S (e.g., a storing mode node) can connect to the Root (a storing mode root) properly as a RPL router node. On the other hand, using the DIOs initiated at the Root, N (e.g., a non-storing mode node) will notice that the Root's MOP and its MOP is different; therefore, will only connect join the RPL network as a leaf node. As a result, A and B, which physically have connections to the root will not be able to join the RPL network. Thus, there is needless network partitioning. While this is an extreme case, other cases where using routes that non-storing mode nodes provide can help optimize the collection routes that RPL nodes form. Next, we examine the routing pathology for downwards traffic packets. Lets consider the same topology as above. Say that we eliminate the rule that RPL introduces of forcing nodes with different MOPs to act as leaf nodes (e.g., no other modifications). In this case, A and B will be able to forward their collection traffic using N. Nevertheless, think of the case where N wants to send a packet to A. Since N is a non-storing mode node, N sends this packet to S because S is the preferred parent. S is operating in storing mode so it looks up node A in its forwarding table and finds that the next hop to reach A is using node N. With the assumption that node N will also know how to reach node A it will forward the packet back to node N. N is operating in non-storing mode so without a source routing header, it will forward the packet back to S. Thus the packet bounces between N and S. Optionally, when using the RPL routing headers, an ICMPv6 error message will be initiated. With the increasing diversity of applications we can envision a network where a part of the network consists of computationally powerful nodes with route storing capabilities and the other part of the network with low-resource nodes that use a non-storing mode and operate together in a single RPL network. Furthermore, on a practical perspective, it is meaningful to use nodes that can contribute in constructing a more efficient DODAG that optimizes the data collection process rather than ignoring a node just because it supports a different MOP. Unfortunately, in such cases, the pathology that we discuss above can arise and cause downwards packets to be dropped and even more, restrict the formation of efficient collection routes.

We describe one way to fix RPL to prevent the pathology described above, while acknowledging that there might be more elegant solutions. In this approach, we acknowledge the fact that non-storing mode nodes are more likely to have strict resource limitations compared to nodes implementing the storing mode. Therefore, we make sure that the most of the required additional capabilities occur at the storing mode nodes rather than the non-storing mode nodes. A new mode of operation (MOP) that allows a node to choose either to implement the storing or non-storing features, or both. The changes below are made compared to the original storing and non-storing modes. Require storing and non-storing nodes to implement source routing header parsing capabilities. RPL DODAG Root nodes supporting this MOP should have the capability to store routes (similar to the non-storing mode option) and also identify storing mode node children nodes. Non-storing nodes send hop-by-hop DAO. Storing nodes keep a table of all the DAO senders and a flag indicating if each of those sender is operating in storing or non-storing mode. This requires allocating one of the bits in the DAO message for a node to indicate if it is operating in storing or non-storing mode. Change the forwarding mechanism in the storing mode node when it receives a downward bound packet: If packet does not have source routing header and the next hop is a storing-mode node, forward as in . If the next hop is a non-storing node, insert the source routing header into the packet and forward, i.e., act like a non-storing root. Using the flag indicating the storing status of nodes in its sub-DODAG, a node constructing a source routing header MAY choose to construct a source routing header only up to the next storing mode node. If the incoming packet has a source routing header, a storing mode node SHOULD obey the route specified in the source routing header to comply with the strict source routing requirements in . If there is a mix of storing and non-storing nodes, we should also be more aggressive about loop detection. More aggressive loop detection will quickly remove the looping packets from the network. Even with the implementation of this suggestion, nodes beyond storing / non-storing nodes will still remain unreachable.

Future work.