MPL Forwarder Select (MPLFS)

The Multicast Protocol for Low-Power and Lossy Networks (MPL) is designed for small devices interconnected by a lossy wireless network such as IEEE 802.15.4. A seed sends a multicast message with a realm-local scope, admin-local scope or higher as specified in . Forwarders forward these messages with an increasing interval size. When the density of forwarders is high, the message may be forwarded by a high number of forwarders that conflict on the link. With extreme forwarder densities and small Trickle intervals, just sending one multicast message may lead to an overload of the communication medium. The number of forwarded messages can be reduced by selecting a minimal set of forwarders. However, for large networks, manually selecting the forwarders is much work, and changing network conditions and configurations make the manual selection an unwanted burden to the network management. This document specifies a protocol that selects the forwarders such that each MPL-enabled device is connected to N_DUPLICATE forwarders, where N_DUPLICATE > 0 can be set. The parameter N_DUPLICATE determines how much path redundancy there is for each MPL message. The value of N_DUPLICATE should be at least 1, because a value of 0 has as result that no forwarder exists in the network during the protocol execution. Moreover, the protocol is distributed and dynamic in nature to face a continuously changing topology. The protocol is inspired by the work described for NeighbourHood Discovery (NHDP) and Simple Multicast Forwarding (SMF) . In contrast to the "HELLO" messages described in , this protocol uses the Trickle protocol to multicast link-local messages, containing a CBOR payload .

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 . Readers of this specification should be familiar with all the terms and concepts discussed in . The following terms are defined in this document: The moment that a node can change its state at messages reception. The following list contains the abbreviations used in this document. TODO, and others to follow.

Nodes participating in MPLFS exchange messages with a format that is described in . A participating node communicates to all its neighbours with link-local multicast messages as described in . Failing links provide a lot of instability. Only messages sent over stable links are accepted. describes a mechanism to refuse messages from unstable links. Each node maintains a set of 1-hop neighbours where each neighbour contains information about its own 1-hop neighbours. On the basis of the contents of the set, the node can decide to become a Forwarder or not, as explained in . The protocol never ends, with a minimum frequency of exchanging maintenance messages specified by an interval size of I_MAX_SELECT. When the set of links is stable, the protocol stabilizes such that there is a path between any two forwarders, and every MPL-enabled node is connected to at least N_DUPLICATE MPL forwarders (when existing), where N_DUPLICATE > 0. N_DUPLICATE can be set dependent on the application requirements. With N_DUPLICATE = 2, it is expected that a multicast message arrives at an intended recipient with very high probability. Nodes have a state that determines whether they are forwarder or not. The state of a node can only be changed by the node itself. To avoid race conditions, (e.g. two nodes simultaneously decide to be no forwarder, while only one is intended) the node with the highest address of all 1-hop neighbours is the only one allowed to change state. Unlike , that considers 3-tuple (Router Priority, MDR Level and Router ID) to allow self state change, this approach only takes into account the node address. Consequently, only k-hop neighbours, with k > 2, can change state simultaneously, and the 1-hop and 2-hop neighbours of a given node can change state one by one.

Each node, n_0, maintains a state with two values: Fixed Forwarder (FF) and No Forwarder (NF). Each node also maintains a set, S1_0, containing information about n_0's 1-hop neighbours and n_0 itself. Each entry, n_i, in S1_0 has the following attributes: the address can be the 64 bit IPv6 address or the short 16 bit address. the average rssi of the messages received by n_0 from n_i. the average rssi of the messages received by n_i from n_0. the number of neighbours, n_ij, of n_i (including n_i) with state = FF. the number of neighbours, n_ij, of n_i with nr_FF < N_DUPLICATE. the number of neighbours, n_ij, of n_i with nr_FF > N_DUPLICATE. the size of S1_i, the set of 1-hop neighbours of n_i. the state of n_i.

A participating node multicasts link-local so-called "neighbour messages" with the Trickle protocol. It uses the multicast address LINK_LOCAL_ALL_NODES as destination. The message sent by n_0 contains the contents of S1_0. The contents of a "neighbour message" from n_i received by n_0 is called M_i. The rssi value associated with the reception of the "neighbour message" is called new_rssi. The message M_i contains information about the set S1_i with the following attributes for all nodes in S1_i: address average-rssi-in nr_FF nr_Under nr_Above size state On reception of M_i from n_i for the first time, the receiving node adds n_i to S1_0, and sets average-rssi-in of n_i in S1_0 to new_rssi. For all following messages from n_i, the average-rssi-in for n_i is calculated in the following way: average-rssi-in := (average-rssi-in*WEIGHT_AVERAGE + new_rssi)/(WEIGHT_AVERAGE+1). The neighbour nodes of M_i are called n_ij. For the n_ij with an address that is equal to the address of n_0: the value of average-rssi-out of n_0 is set equal to the value of average-rssi-in of n_ij. The contents of n_0 is updated with the contents of M_i. Updating includes the following actions: Add n_i to S1_0, if n_i not present in S1_0. Set size of n_i equal to the number of entries in M_i. When n_ij.address = n_j.address, copy the values of nr_Under, nr_Above, nr_FF, and state of n_ij to n_j. When the average-rssi-in and average-rssi-out values of n_i have been averaged over more than WEIGHT_AVERAGE messages, and the averaged RSSI values are smaller than MAXIMUM_RSSI, n_i is called "valid".

The protocol aims at allocating forwarders in the densest part of the network. A dense network is characterized by a high number of neighbours. Therefore, the protocol attempts to assign status FF to the nodes with the highest number of neighbours that have less than N_DUPLICATE neighbours with state = FF (nr_FF < N_DUPLICATE). It is required that a path exists between every two forwarders to prevent network partitioning. Therefore, a node can become forwarder iff one of its neighbours is a forwarder. The consequence of this rule is that one so-called "source-forwarder" must be selected by the network administrator. A likely choice for the "source-forwarder" is the border router. At the start of the selection protocol the node, n_0, sets its state to No Forwarder (NF). It sets the Trickle timer to its minimum interval, I_MIN_SELECT, and starts multicasting M_0 to its neighbours. Every time entries are added to, or removed from, S1_0, the Trickle interval timer is set to I_MIN_SELECT. The executing node, n_0, calculates the following parameter values: max-under is the maximum of the nr_Under attribute of all valid n_i in S1_0. max_address_u is the maximum of the addresses of valid n_i with nr_Under = max-under. max_address_a is the maximum of the addresses of all valid n_i. connected is true iff nr_FF of all neighbouring forwarders is equal to nr_FF of n_0. The information about the state and the nr_Under value of the neighbours comes in asynchronously. Time is needed before the state in a node correctly reflects the state changes of the network. A node can change its state when during the reception of messages of all neighbours, the value of nr_Under has not changed. To calculate its new state, n_0 does the following: set state to FF. set state to NF.

The payload format is /application/cbor . The contents of the message is a CBOR array (Major type 4) of CBOR arrays composed of neighbour address, rssi value, size of S1_i, forwarder state, nr_FF, nr_Under, and nr_Above. Assuming two neighbours, in diagnostic JSON the payload looks like:

The following text recommends default values for the MPLFS protocols. = 0,2; minimum Trickle timer interval. = 10; maximum Trickle timer interval. = 10; number of values to average rssi. = 3; maximum acceptable average rssi value. = 2; requested number of MPL forwarder neighbours for every MPL enabled node.

We are very grateful to

Changes from version 00 to version 01 Definition of S1_0 improved Algorithm changed and simulated Moment of state change specified