Network Working Group R. Ogier Request for Comments: 3684 SRI International Category: Experimental F. Templin Nokia M. Lewis SRI International February 2004
Topology Dissemination Based on Reverse-Path Forwarding (TBRPF)
Status of this Memo
This memo defines an Experimental Protocol for the Internet community. It does not specify an Internet standard of any kind. Discussion and suggestions for improvement are requested. Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2004). All Rights Reserved.
Abstract
Topology Dissemination Based on Reverse-Path Forwarding (TBRPF) is a proactive, link-state routing protocol designed for mobile ad-hoc networks, which provides hop-by-hop routing along shortest paths to each destination. Each node running TBRPF computes a source tree (providing paths to all reachable nodes) based on partial topology information stored in its topology table, using a modification of Dijkstra's algorithm. To minimize overhead, each node reports only *part* of its source tree to neighbors. TBRPF uses a combination of periodic and differential updates to keep all neighbors informed of the reported part of its source tree. Each node also has the option to report additional topology information (up to the full topology), to provide improved robustness in highly mobile networks. TBRPF performs neighbor discovery using "differential" HELLO messages which report only *changes* in the status of neighbors. This results in HELLO messages that are much smaller than those of other link-state routing protocols such as OSPF.
Topology Dissemination Based on Reverse-Path Forwarding (TBRPF) is a proactive, link-state routing protocol designed for mobile ad-hoc networks (MANETs), which provides hop-by-hop routing along shortest paths to each destination. Each node running TBRPF computes a source tree (providing shortest paths to all reachable nodes) based on partial topology information stored in its topology table, using a modification of Dijkstra's algorithm. To minimize overhead, each node reports only *part* of its source tree to neighbors.
TBRPF uses a combination of periodic and differential updates to keep all neighbors informed of the reported part of its source tree. Each node also has the option to report addition topology information (up to the full topology), to provide improved robustness in highly mobile networks.
TBRPF performs neighbor discovery using "differential" HELLO messages which report only *changes* in the status of neighbors. This results in HELLO messages that are much smaller than those of other link- state routing protocols such as OSPF [6].
TBRPF consists of two modules: the neighbor discovery module and the routing module (which performs topology discovery and route computation). An overview of these modules is given in Section 5.
The keywords "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL", when they appear in this document, are to be interpreted as described in BCP 14, RFC 2119 [1].
This document also makes use of internal conceptual variables to describe protocol behavior and external variables that an implementation must allow system administrators to change. The specific variable names, how their values change, and how their settings influence protocol behavior are provided to demonstrate protocol behavior. An implementation is not required to have them in the exact form described here, so long as its external behavior is consistent with that described in this document.
router ID Each node is identified by a unique 32-bit router ID (RID), which for IPv4 is typically equal to the IP address of one of its interfaces. The term "node u" denotes the node whose RID is equal to u.
interface A node's attachment to a communication facility or medium through which it can communicate with other nodes. A node can have multiple interfaces. An interface can be wireless or wired, and can be broadcast (e.g., Ethernet) or point-to-point. Each interface is identified by its IP address. The term "interface I" denotes the interface whose IP address is I.
link A link is an ordered pair of interfaces (I,J) where I and J are on two different nodes, and where interface I has recently received packets sent from interface J. A link (i,j) from node i to node j is said to exist if node i has an interface I and node j has an interface J such that (I,J) is a link. Nodes i and j are called the "tail" and "head" of the link, respectively.
bidirectional link A link (I,J) such that interfaces I and J can both hear each other. Also called a 2-way link.
Ogier, et al. Experimental [Page 4]
RFC 3684 TBRPF February 2004
neighbor node A node j is said to be a neighbor of node i if node i can hear node j on some interface. Node j is said to be a 2-way neighbor if there is a bidirectional link between i and j.
MANET interface Any wireless interface such that two neighbor nodes on the interface need not be neighbors of each other. MANET nodes typically have at least one MANET interface, but this is not a requirement.
topology The topology of the network is described by a graph G = (V, E), where V is the set of nodes u and E is the set of links (u,v) in the network.
source tree The directed tree (denoted T) computed by each node that provides shortest paths to all other reachable nodes.
topology update A message that reports the state of one or more links.
parent The parent of node i for node u is the next node on the computed shortest path from node i to node u.
predecessor The predecessor of a node v on the source tree is the node u such that the link (u,v) is in the source tree.
leaf node A leaf node of the source tree is a node on the source tree that is not the predecessor of any other node on the source tree.
proactive routing protocol A routing protocol in which each node maintains routes to all reachable destinations at all times, whether or not there is currently any need to deliver packets to those destinations. In contrast, an "on-demand" routing protocol discovers and maintains routes only when they are needed.
TBRPF is a proactive routing protocol designed for mobile ad-hoc networks (MANETs). It can support networks with up to a few hundred nodes, and can be combined with hierarchical routing techniques to support much larger networks. Because it employs techniques to
Ogier, et al. Experimental [Page 5]
RFC 3684 TBRPF February 2004
greatly reduce control traffic, TBRPF can support much larger and denser networks than routing protocols based on the classical link- state algorithm (e.g., OSPF).
The number of nodes that can be supported depends on several factors, including the MAC data rate, the rate of topology changes, and the network density (average number of neighbors). Simulations have been reported in which TBRPF has supported as many as 500 nodes. In simulations with 100 nodes and 20 traffic streams (sources), using IEEE 802.11 with a data rate of 2 Mbps, TBRPF was found to generate approximately 80-120 kb/s of routing control traffic for the scenarios considered, which compared favorably with other MANET routing protocols [7][8]. A proof of correctness for TBRPF can be found in references [8] and [9].
The TBRPF Neighbor Discovery (TND) protocol allows each node i to quickly detect the neighbor nodes j such that a bidirectional link (I,J) exists between an interface I of node i and an interface J of node j. The protocol also quickly detects when a bidirectional link breaks or becomes unidirectional.
The key feature of TND is that it uses "differential" HELLO messages which report only *changes* in the status of links. This results in HELLO messages that are much smaller than those of other link-state routing protocols such as OSPF, in which each HELLO message includes the IDs of *all* neighbors. As a result, HELLO messages can be sent more frequently, which allows faster detection of topology changes.
TND is designed to be fully modular and independent of the routing module. TND performs ONLY neighbor sensing, i.e., it determines which nodes are (1-hop) neighbors. In particular, it does not discover 2-hop neighbors (which is handled by the routing module). As a result, TND can be used by other routing protocols, and TBRPF can use another neighbor discovery protocol in place of TND, e.g., one provided by the link layer.
Nodes with multiple interfaces run TND separately on each interface, similar to OSPF. Thus, a neighbor table is maintained for each local interface, and a HELLO sent on a particular interface contains only information regarding neighbors heard on that interface.
Ogier, et al. Experimental [Page 6]
RFC 3684 TBRPF February 2004
We note that, in wireless networks, it is possible for a single interface I to receive packets from multiple interfaces J associated with the same neighbor node. This could happen, for example, if the neighbor uses a directional antenna with different interfaces representing different beams. For this reason, TBRPF includes neighbor interface addresses in HELLO messages, unlike OSPF, which includes only router IDs in HELLO packets.
Each TBRPF node maintains a neighbor table for each local interface I, which stores state information for each neighbor interface J heard on that interface, i.e., for each link (I,J) between interface I and a neighbor interface J. The status of each link can be 1-WAY, 2-WAY, or LOST. The neighbor table for interface I determines the contents of HELLO messages sent on interface I, and is updated based on HELLO messages received on interface I (and possibly on link-layer notifications).
Each TBRPF node sends (on each interface) at least one HELLO message per HELLO_INTERVAL. Each HELLO message contains three (possibly empty) lists of neighbor interface addresses (which are formatted as three message subtypes): NEIGHBOR REQUEST, NEIGHBOR REPLY, and NEIGHBOR LOST. Each HELLO message also contains the current HELLO sequence number (HSEQ), which is incremented with each transmitted HELLO.
In the following overview of the operation of TND, we assume that interface I belongs to node i, and interface J belongs to node j. When a node i changes the status of a link (I,J), it includes the neighbor interface address J in the appropriate list (NEIGHBOR REQUEST/REPLY/LOST) in at most NBR_HOLD_COUNT (typically 3) consecutive HELLOs sent on interface I. This ensures that node j will either receive one of these HELLOs on interface J, or will miss NBR_HOLD_COUNT HELLOs and thus declare the link (J,I) to be LOST. This technique makes it unnecessary for a node to include each 1-WAY or 2-WAY neighbor in HELLOs indefinitely, unlike OSPF.
To avoid establishing a link that is likely to be short lived (i.e., to employ hysteresis), node i must receive (on interface I) at least HELLO_ACQUIRE_COUNT (e.g., 2) of the last HELLO_ACQUIRE_WINDOW (e.g., 3) HELLOs sent from a neighbor interface J, before declaring the link (I,J) to be 1-WAY. When this happens, node i includes J in the NEIGHBOR REQUEST list in each of its next NBR_HOLD_COUNT HELLO messages sent on interface I, or until a NEIGHBOR REPLY message containing I is received on interface I from neighbor interface J.
If node j receives (on interface J) one of the HELLOs sent from interface I that contains J in the NEIGHBOR REQUEST list, then node j declares the link (J,I) to be 2-WAY (unless it is already 2-WAY), and
Ogier, et al. Experimental [Page 7]
RFC 3684 TBRPF February 2004
includes I in the NEIGHBOR REPLY list in each of its next NBR_HOLD_COUNT HELLO messages sent on interface J. Upon receiving one of these HELLOs on interface I, node i declares the link (I,J) to be 2-WAY.
If node i receives a HELLO on interface I, sent from neighbor interface J, whose HSEQ indicates that at least NBR_HOLD_COUNT HELLOs were missed, or if node i receives no HELLO on interface I sent from interface J within NBR_HOLD_TIME seconds, then node i changes the status of link (I,J) to LOST (unless it is already LOST), and includes J in the NEIGHBOR LOST list in each of its next NBR_HOLD_COUNT HELLO messages sent on interface I (unless the link changes status before these transmissions are complete). Node j will either receive one of these HELLOs on interface J or will miss NBR_HOLD_COUNT HELLOs; in either case, node j will declare the link (J,I) to be LOST. In this manner, both nodes will agree that the link between I and J is no longer bidirectional, even if node j can still hear HELLOs from node i.
Each node may maintain and update one or more link metrics for each link (I,J) from a local interface I to a neighbor interface J, representing the quality of the link. Such link metrics can be used as additional conditions for changing the status of a neighbor, based on the link metric going above or below some threshold. TBRPF also allows link metrics to be advertised in topology updates, and to be used for computing shortest paths.
Each node running TBRPF maintains a source tree, denoted T, which provides shortest paths to all reachable nodes. Each node computes and updates its source tree based on partial topology information stored in its topology table, using a modification of Dijkstra's algorithm. To minimize overhead, each node reports only part of its source tree to neighbors. The main idea behind the current version of TBRPF came from PTSP [10], another protocol in which each node reports only part of its source tree. (However, TBRPF differs from PTSP in several ways.) The current version of TBRPF should not be confused with its previous version [11], which is a full-topology routing protocol.
The part of T that a node reports to neighbors is called the "reported subtree" and is denoted RT. Each node reports RT to neighbors in *periodic* topology updates (e.g., every 5 seconds), and reports changes (additions and deletions) to RT in more frequent *differential* updates (e.g., every 1 second). Periodic updates inform new neighbors of RT, and ensure that each neighbor eventually learns RT even if it does not receive all updates. Differential
Ogier, et al. Experimental [Page 8]
RFC 3684 TBRPF February 2004
updates ensure the fast propagation of each topology update to all nodes that are affected by the update. A received topology update is not forwarded, but *may* result in a change to RT, which will be reported in the next differential or periodic update. Whenever possible, topology updates are included in the same packet as a HELLO message, to minimize the number of control packets sent. TBRPF does not require reliable or sequenced delivery of messages, and does not use ACKs or NACKs.
TBRPF supports multiple interfaces, associated hosts, and network prefixes. Information regarding associated interfaces, hosts, and prefixes is disseminated efficiently in periodic and differential updates, similar to the dissemination of topology updates.
The reported subtree RT consists of links (u,v) of T such that u is in the "reported node set" RN, which is computed as follows. Node i includes a neighbor j in RN if and only if node i determines that one of its neighbors may select i to be its next hop on its shortest path to j. To make this determination, node i computes the shortest paths, up to 2 hops, from each neighbor to each other neighbor, using only neighbors (or node i itself) as an intermediate node, and using relay priority (included in HELLO messages) and router ID to break ties. After a node determines which neighbors are in RN, each reachable node u is included in RN if and only if the next hop on the shortest path to u is in RN. A node also includes itself in RN. As a result, the reported subtree RT includes the subtrees of T that are rooted at neighbors in RN, and also includes all local links to neighbors.
We note that neighbors in RN are analogous to multipoint relay (MPR) selectors [12]. Thus, if node i selects neighbor j to be in RN, then node i effectively selects itself to be an MPR of node j. This is quite different from [12], in which a node does not select itself to be an MPR, but selects a subset of its neighbors to be MPRs.
A node with a larger relay priority reports a larger part of its source tree (on average), and is more likely to be selected as a next-hop relay by its neighbors. A node with relay priority equal to 0 is called a non-relay node, and never forwards packets originating from other nodes.
TBRPF does not use sequence numbers for topology updates, thus reducing message overhead and avoiding wraparound problems. Instead, a technique similar to SPTA [13] is used in which, for each link (u,v) reported by one or more neighbors, only the next hop p(u) to u is believed regarding the state of the link. (However, in SPTA each node reports the full topology.) Using this technique, each node maintains a topology graph TG, consisting of links that are believed
Ogier, et al. Experimental [Page 9]
RFC 3684 TBRPF February 2004
to be up, and computes T as the shortest-path tree within TG. To allow immediate rerouting, the restriction that each link (u,v) in TG must be reported by p(u) is relaxed temporarily if p(u) changes to a neighbor that is not reporting the link.
Each node is required to report RT, but may report additional links, e.g., to provide increased robustness in highly mobile networks. More precisely, a node may maintain any subgraph H of TG that contains T, and report the reported subgraph RH, which consists of links (u,v) of H such that u is in RN. For example, H can equal TG, which would provide each node with the full network topology if this is done by all nodes. H can also be a biconnected subgraph that contains T, which would provide each node with two disjoint paths to each other node, if this is done by all nodes.
TBRPF allows the option to include link metrics in topology updates, and to compute paths that are shortest with respect to the metric. This allows packets to be sent along paths that are higher quality than minimum-hop paths.
TBRPF allows path optimality to be traded off in order to reduce the amount of control traffic in networks with a large diameter, where the degree of approximation is determined by the configurable parameter NON_TREE_PENALTY.
Nodes send TBRPF protocol data in contiguous units known as packets. Each packet includes a header, optional header extensions, and a body comprising one or more messages and padding options as needed. To facilitate efficient receiver processing, senders SHOULD insert padding options as necessary to align multi-octet words within the TBRPF packet on natural boundaries (i.e., modulo-8/4/2 addresses for 64/32/16-bit words, respectively). Receivers MUST be capable of processing multi-octet words whether or not aligned on natural boundaries. The following sections specify elements of the TBRPF packet in more detail.
Version (4 bits) The TBRPF version number. This specification documents version 4 of the protocol.
Flags (4 bits) Two bits (L,I) specify which header extensions (if any) follow. Two bits (R) are reserved for future use, and MUST be zero. Any extensions specified by these bits MUST appear in the same order as the bits (i.e., first L, then I) as follows:
L - Length included If the underlying delivery service provides a length field, the sender MAY set L = '0' and omit the length extension. Otherwise, the sender MUST set L = '1' and include a 16-bit unsigned integer length immediately after any previous header field. The length includes all header and data bytes and is written into the length field in network byte order.
Receivers examine the L bit to determine whether the length field is present. If L = '1', the receiver reads the length field to determine the length of the TBRPF packet, including the TBRPF packet header. Receivers discard any TBRPF packet if neither the underlying delivery service nor the TBRPF packet header provide packet length.
I - Router ID (RID) included If the underlying delivery service encodes the sender's RID, the sender MAY set I = '0' and omit the RID field. Otherwise, the sender MUST set I = '1' and include a 4-octet RID in network byte order immediately after any previous header fields. The RID option provides a mechanism for implicit network-level address resolution. A receiver that detects a RID option SHOULD create a binding between the RID and the source address that appears in the network-level header.
The TBRPF packet body consists of the concatenation of one or more TBRPF messages (and padding options where necessary). Messages and padding options within the TBRPF packet body are encoded using the following format:
+-+-+-+-+-+-+-+-+- - - - - |OPTIONS| TYPE | VALUE +-+-+-+-+-+-+-+-+- - - - -
Ogier, et al. Experimental [Page 11]
RFC 3684 TBRPF February 2004
OPTIONS (4 bits) Four option bits that depend on TYPE.
TYPE (4 bits) Identifier for message type or padding option.
VALUE Variable-length field. (Format and length depend on TYPE, as described in the following sections.)
The sequence of elements MUST be processed strictly in the order they appear within the TBRPF packet body; a receiver must not, for example, scan through the packet body looking for a particular type of element prior to processing all preceding elements [2]. TBRPF packet elements include padding options and messages as described below.
Senders MAY insert two types of padding options where necessary, e.g., to satisfy alignment requirements for other elements [2]. Padding options may occur anywhere within the TBRPF packet body. The following two padding options are defined:
Pad1 option (TYPE = 0)
+-+-+-+-+-+-+-+-+ | 0 | 0 | +-+-+-+-+-+-+-+-+
The Pad1 option inserts one octet of padding into the TBRPF packet body; the VALUE field is omitted. If more than one octet of padding is required, the PadN option (described next) should be used, rather than multiple Pad1 options.
The PadN option inserts two or more octets of padding into the TBRPF packet body. The first octet of the VALUE field contains an 8-bit unsigned integer length containing a value between 0 - 253 which specifies the number of zero-valued octets that immediately follow, yielding a maximum total of 255 padding octets.
Additional message types are described as they occur in the following sections. Senders encode messages as specified by the individual message formats. Receivers detect errors in message construction, e.g., messages with unrecognized types, messages with a non-integral number of elements, or with fewer elements than indicated, etc. In all cases, upon detecting an error, the receiver MUST discontinue processing the current TBRPF packet and discard any unprocessed elements.
This section describes the TBRPF Neighbor Discovery (TND) protocol, which allows each node to quickly detect bidirectional links (I,J) between a local interface I and a neighbor interface J, and to quickly detect the loss of such links. The interface between TND and the routing module is defined by the neighbor table maintained by TND and the three procedures Link_Up(I,J), Link_Down(I,J), and Link_Change(I,J), which are called by TND to announce a new link, the loss of a link, and a change in the metric of a link, respectively.
Pri (4 bits) This field indicates the sending node's relay priority, which is an integer between 0 and 15. A node with a higher relay priority is more likely to be selected as the next hop on a route. The value 0 is reserved for non-relay nodes, i.e., nodes that should never forward packets originating from other nodes. A router in normal operation SHOULD have a relay priority equal to 7. A router can change its relay priority dynamically, e.g., when its power supply becomes critical.
n (12 bits) The number of 32-bit neighbor interface addresses in the message.
A HELLO message is the concatenation of a NEIGHBOR REQUEST message, a NEIGHBOR REPLY message, and a NEIGHBOR LOST message, where each of the last two messages is omitted if its list of neighbor interface addresses is empty. Thus, a HELLO message always includes a (possibly empty) NEIGHBOR REQUEST.
Each node maintains, for each of its local interfaces I, a neighbor table, which stores state information for each neighbor interface J from which HELLO messages have recently been received by interface I. The entry for neighbor interface J, in the neighbor table for I, contains the following variables:
nbr_rid(I,J) - The router ID of the node associated with neighbor interface J.
nbr_status(I,J) - The current status of the link (I,J), which can be LOST, 1-WAY, or 2-WAY.
nbr_life(I,J) - The amount of time (in seconds) remaining before nbr_status(I,J) must be changed to LOST if no further HELLO message from interface J is received. Set to NBR_HOLD_TIME whenever a HELLO is received on interface I from interface J.
nbr_hseq(I,J) - The value of HSEQ in the last HELLO message received on interface I from interface J. Used to determine the number of HELLOs that have been missed.
nbr_count(I,J) - The remaining number of times a NEIGHBOR REQUEST/ REPLY/LOST message containing J must be sent on interface I.
hello_history(I,J) - A list of the sequence numbers of the last HELLO_ACQUIRE_WINDOW HELLO messages received on interface I from interface J.
Ogier, et al. Experimental [Page 14]
RFC 3684 TBRPF February 2004
nbr_metric(I,J) - An optional measure of the quality of the link (I,J), represented by an integer between 1 and 255, where smaller values indicate better quality. Defaults to 1 if not used.
nbr_pri(I,J) - The relay priority of the node associated with interface J.
The entry for interface J in the neighbor table for interface I may be deleted if no HELLO has been received on interface I from interface J within the last 2*NBR_HOLD_TIME seconds. (It is kept while NEIGHBOR LOST messages containing J are being transmitted.) The absence of an entry for a given interface J is equivalent to an entry with nbr_status(I,J) = LOST and hello_history(I,J) = NULL.
The three possible values of nbr_status(I,J) have the following informal meanings (the exact meanings are defined by the protocol):
LOST Interface I has not received a sufficient number of HELLO messages recently from Interface J.
1-WAY Interface I has received a sufficient number of HELLO messages recently from Interface J, but the link is not 2-WAY.
2-WAY Interfaces I and J have both received a sufficient number of HELLO messages recently from each other.
Each node MUST send, on each local interface, at least one HELLO message per HELLO_INTERVAL. HELLO messages MAY be sent more frequently than this (e.g., for faster detection of topology changes). However, to avoid the possibility that HSEQ wraps around to the same number before a neighbor that stops receiving HELLO messages changes the status of the link to LOST, the time between two consecutive HELLO messages (sent on a given interface) MUST be greater than NBR_HOLD_TIME/128 second.
To avoid synchronization of control messages, which can result in collisions, HELLO messages SHOULD NOT be transmitted at equal intervals. To achieve this, a node MAY choose the interval between consecutive HELLO messages to be HELLO_INTERVAL - jitter, where jitter is selected randomly from the interval [0, MAX_JITTER].
Ogier, et al. Experimental [Page 15]
RFC 3684 TBRPF February 2004
Each HELLO message always includes a NEIGHBOR REQUEST message, even if its list of neighbor addresses is empty. The NEIGHBOR REQUEST message includes the sequence number HSEQ, which is incremented by 1 (modulo 256) each time a HELLO is sent. The HELLO message also includes a NEIGHBOR REPLY message if its list of neighbor addresses is nonempty, and a NEIGHBOR LOST message if its list of neighbor addresses is nonempty. The contents of these three messages are determined by the following steps at node i for each interface I:
1. For each interface J such that nbr_status(I,J) = LOST and nbr_count(I,J) > 0, include J in the NEIGHBOR LOST message and decrement nbr_count(I,J).
2. For each interface J such that nbr_status(I,J) = 1-WAY and nbr_count(I,J) > 0, include J in the NEIGHBOR REQUEST message and decrement nbr_count(I,J).
3. For each interface J such that nbr_status(I,J) = 2-WAY and nbr_count(I,J) > 0, include J in the NEIGHBOR REPLY message and decrement nbr_count(I,J).
If a node restarts, so that all entries are removed from the neighbor table, then the node MUST ensure that (for each interface) at least one of the following two conditions is satisfied:
1. The difference between the transmission times of the first HELLO sent after restarting and the last HELLO sent before restarting is at least 2*NBR_HOLD_TIME.
2. Letting HSEQ_LAST denote the sequence number of the last HELLO that was sent before restarting, the sequence number of the first HELLO sent after restarting is set to HSEQ_LAST + NBR_HOLD_COUNT + 1 (modulo 256).
Either of these conditions ensures that, if node i with interface I restarts, then each neighbor of node i that has a link (J,I) to interface I will set the status of the link to LOST.
When a node receives a HELLO message, it obtains the IP address of the sending interface from the IP header. If the TBRPF packet header of the received HELLO contains the RID option, then the RID of the sending node is obtained from the TBRPF packet header; otherwise it is equal to the IP address of the sending interface. If node i (with RID equal to i) receives a HELLO message on interface I, sent by node j (with RID equal to j) on interface J, with sequence number HSEQ and relay priority PRI, then node i performs the following steps:
Ogier, et al. Experimental [Page 16]
RFC 3684 TBRPF February 2004
1. If the neighbor table for interface I does not contain an entry for interface J, create one with nbr_rid(I,J) = j, nbr_status(I,J) = LOST (temporarily), nbr_count(I,J) = 0, and nbr_hseq(I,J) = HSEQ.
2. Update hello_history(I,J) to reflect the received HELLO message. If nbr_hseq(I,J) > HSEQ (due to wraparound), set nbr_hseq(I,J) = nbr_hseq(I,J) - 256.
3. If nbr_status(I,J) = LOST and hello_history(I,J) indicates that HELLO_ACQUIRE_COUNT of the last HELLO_ACQUIRE_WINDOW HELLO messages from interface J have been received:
a. If interface I does not appear in the NEIGHBOR REQUEST list or the NEIGHBOR REPLY list, set nbr_status(I,J) = 1-WAY and nbr_count(I,J) = NBR_HOLD_COUNT.
b. Else, set nbr_status(I,J) = 2-WAY and nbr_count(I,J) = NBR_HOLD_COUNT. Call Link_Up(I,J).
4. Else, if nbr_status(I,J) = 1-WAY:
a. If HSEQ - nbr_hseq(I,J) > NBR_HOLD_COUNT, then set nbr_status(I,J) = LOST and nbr_count(I,J) = NBR_HOLD_COUNT.
b. Else, if interface I appears in the NEIGHBOR REQUEST list, set nbr_status(I,J) = 2-WAY and nbr_count(I,J) = NBR_HOLD_COUNT. Call Link_Up(I,J).
c. Else, if interface I appears in the NEIGHBOR REPLY list, set nbr_status(I,J) = 2-WAY and nbr_count(I,J) = 0. Call Link_Up(I,J).
5. Else, if nbr_status(I,J) = 2-WAY:
a. If interface I appears in the NEIGHBOR LOST list, set nbr_status(I,J) = LOST and nbr_count(I,J) = 0. Call Link_Down(I,J).
b. Else, if HSEQ - nbr_hseq(I,J) > NBR_HOLD_COUNT, set nbr_status(I,J) = LOST and nbr_count(I,J) = NBR_HOLD_COUNT. Call Link_Down(I,J).
c. Else, if interface I appears in the NEIGHBOR REQUEST list and nbr_count(I,J) = 0, set nbr_count(I,J) = NBR_HOLD_COUNT.
6. Set nbr_life(I,J) = NBR_HOLD_TIME, nbr_hseq(I,J) = HSEQ, and nbr_pri(I,J) = PRI.
Some link-layer protocols (e.g., IEEE 802.11) provide a notification that the link to a particular neighbor has failed, e.g., after attempting a maximum number of retransmissions. If such an notification is provided by the link layer, then node i SHOULD perform the following step upon receipt of a link-layer failure notification for the link (I,J) from local interface I to neighbor interface J:
If nbr_status(I,J) = 2-WAY, set nbr_status(I,J) = LOST and nbr_count(I,J) = NBR_HOLD_COUNT. Call Link_Down(I,J).
Each node MAY maintain and update one or more link metrics for each link (I,J), representing the quality of the link, e.g., signal strength, number of HELLOs received over some time interval, reliability, stability, bandwidth, etc. Each node MUST declare a neighbor to be LOST if either NBR_HOLD_COUNT HELLOs are missed or if no HELLO is received within NBR_HOLD_TIME seconds; however, a node MAY also declare a neighbor to be LOST based on a link metric being above or below some threshold. Each node MUST receive at least HELLO_ACQUIRE_COUNT of the last HELLO_ACQUIRE_WINDOW HELLOs from a neighbor before declaring the neighbor 1-WAY or 2-WAY; however, a node MAY require an additional condition based on a link metric being above or below some threshold, before declaring the neighbor 1-WAY or 2-WAY. This document does not specify any particular link metric, but an implementation of TBRPF that uses such metrics is considered to be compliant with this specification.
The function Link_Change(I,J) is called to alert the routing module whenever nbr_metric(I,J) changes significantly. If the configurable parameter USE_METRICS is equal to 1, then the metrics nbr_metric(I,J) are used by the routing module for route computation, as described in Section 8.
This section lists the parameters used by the neighbor discovery protocol, and their proposed default values. All nodes MUST be configured to have the same value for all of the following parameters.
Parameter Name Default Value -------------- ------------- HELLO_INTERVAL 1 second MAX_JITTER 0.1 second NBR_HOLD_TIME 3 seconds NBR_HOLD_COUNT 3 HELLO_ACQUIRE_COUNT 2 HELLO_ACQUIRE_WINDOW 3
In addition to the information required by the neighbor discovery protocol, each node running TBRPF maintains a topology table TT, which stores information for each known node and link in the network. Nodes are identified by their RIDs, i.e., node u is the node whose RID is u. The following information is stored in the topology table at node i for each node u and link (u,v):
T(u,v) - Equal to 1 if (u,v) is in node i's source tree T, and 0 otherwise. The previous source tree is also maintained as old_T.
RN(u) - Equal to 1 if u is in node i's reported node set RN, and 0 otherwise. The previous reported node set is also maintained as old_RN.
RT(u,v) - Equal to 1 if (u,v) is in node i's reported subtree RT, and 0 otherwise. Since RT is defined as the set of links (u,v) in T such that u is in RN, this variable need not be maintained explicitly.
TG(u,v) - Equal to 1 if (u,v) is in node i's topology graph TG, and 0 otherwise.
N - The set of 2-way neighbors of node i.
Ogier, et al. Experimental [Page 19]
RFC 3684 TBRPF February 2004
r(u,v) - The list of neighbors that are reporting link (u,v) in their reported subtree RT. The set of links (u,v) reported by neighbor j is denoted RT_j.
r(u) - The list of neighbors that are reporting node u in their reported node set RN.
p(u) - The current parent for node u, equal to the next node on the shortest path to u.
pred(u) - The node that is the predecessor of node u in the source tree T. Equal to NULL if node u is not reachable.
pred(j,u) - The node that is the predecessor of node u in the subtree RT_j reported by neighbor j.
d(u) - The length of the shortest path to node u. If USE_METRICS = 0, d(u) is the number of hops to node u.
reported(u,v) - Equal to 1 if link (u,v) in TG is reported by p(u), and 0 otherwise.
tg_expire(u) - Expiration time for links (u,v) in TG.
rt_expire(j,u) - Expiration time for links (u,v) in RT_j.
nr_expire(u,v) - Expiration time for a link (u,v) in TG such that reported(u,v) = 0. Such non-reported links can be used temporarily during rerouting.
metric(j,u,v) - The metric for link (u,v) reported by neighbor j.
metric(u,v) - The metric for link (u,v) in TG. For a neighbor j, metric(i,j) is the minimum of nbr_metric(I,J) over all 2-WAY links (I,J) from i to j.
cost(u,v) - The cost for link (u,v), equal to metric(u,v) if USE_METRICS = 1, and otherwise equal to 1.
local_if(j) - The address of the preferred local interface for forwarding packets to neighbor j.
nbr_if(j) - The address of the preferred interface of neighbor j.
Ogier, et al. Experimental [Page 20]
RFC 3684 TBRPF February 2004
The routing table consists of a list of tuples of the form (rt_dest, rt_next, rt_dist, rt_if_id), where rt_dest is the destination IP address or prefix, rt_next is the interface address of the next hop of the route, rt_dist is the length of the route, and rt_if_id is the ID of the local interface through which the next hop can be reached.
Each node also maintains three tables that describe associated IP addresses or prefixes: the "interface table", which associates interface IP addresses with router IDs, the "host table", which associates host IP addresses with router IDs, and the "network prefix table", which associates network prefixes with router IDs.
The "interface table" consists of tuples of the form (if_addr, if_rid, if_expire), where if_addr is an interface IP address associated with the router with RID = if_rid, and if_expire is the time at which the tuple expires and MUST be removed. The interface table at a node does NOT contain an entry in which if_addr equals the node's own RID; thus, a node does not advertise its own RID as an associated interface.
The "host table" consists of tuples of the form (h_addr, h_rid, h_expire), where h_addr is a host IP address associated with the router with RID = h_rid, and h_expire is the time at which the tuple expires and MUST be removed.
The "network prefix table" consists of tuples of the form (net_prefix, net_length, net_rid, net_expire), where net_prefix and net_length describe a network prefix associated with the router with RID = net_rid, and net_expire is the time at which the tuple expires and MUST be removed. A MANET may be configured as a "stub" network, in which case one or more gateway routers may announce a default prefix such that net_prefix = net_length = 0. Two copies of each table are kept: an "old" copy that was last reported to neighbors, and the current copy that is updated when association messages are received.
The TOPOLOGY UPDATE message has the two formats, depending on the size of the message. The normal format is as follows, and is used whenever n, NRL, and NRNL all do not exceed 255:
Ogier, et al. Experimental [Page 21]
RFC 3684 TBRPF February 2004
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |M|D|0|0| TYPE | n | NRL | NRNL | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Router ID of u | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Router ID of v_1 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ~ ... ~ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Router ID of v_n | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | metric 1 | metric 2 | ... | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | ... | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The message body contains the n+1 router IDs for nodes u, v_1,...,v_n, which represent the links (u,v_1),..., (u,v_n). The first NRL of the v_k are reported leaf nodes, the next NRNL of the v_k are reported non-leaf nodes, and the last n - (NRL+NRNL) of the v_k are not reported (not in RN).
The M bit indicates whether or not link metrics are included in the message. If M = 1, then a 1-octet metric is included for each of the links (u,v_1),..., (u,v_n), following the last router ID.
The D bit indicates whether or not implicit deletion is used, and must be set to 1 if and only if IMPLICIT_DELETION = 1.
The TOPOLOGY UPDATE message has the following three subtypes:
FULL (TYPE = 5) A FULL update (FULL, n, NRL, NRNL, u, v_1,..., v_n) reports that the links (u,v_1),..., (u,v_n) belong to the sending router's reported subtree RT, and that RT contains no other links with tail u.
ADD (TYPE = 6) An ADD update (ADD, n, NRL, NRNL, u, v_1,..., v_n) reports that the links (u,v_1),..., (u,v_n) have been added to the sending router's reported subtree RT.
DELETE (TYPE = 7) A DELETE update (DELETE, n, NRL, NRNL, u, v_1,..., v_n) reports that the links (u,v_1),..., (u,v_n) have been deleted from the sending router's reported subtree RT.
Ogier, et al. Experimental [Page 22]
RFC 3684 TBRPF February 2004
If n, NRL, or NRNL is larger than 255, then the long format of the TOPOLOGY UPDATE message is used, in which the first 4 octets of the normal format are replaced by the following 8 octets:
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |M|D|1|0| TYPE | 0 | n | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | NRL | NRNL | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
8.3. Interface, Host, and Network Prefix Association Message Formats
The INTERFACE ASSOCIATION (TYPE = 8) and HOST ASSOCIATION (TYPE = 9) messages have the following format:
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |ST | 0 | TYPE | Reserved | n | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Router ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IP Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IP Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | ... | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The message body contains the router ID of the originating node, and n IP addresses of interfaces (TYPE = 8) or hosts (TYPE = 9) that are associated with the router ID. The ST field is defined below.
The NETWORK PREFIX ASSOCIATION message (TYPE = 10) has the following format:
The message body contains the router ID of the originating node, and n network prefixes, each specified by a 1-octet prefix length followed immediately by the prefix, using the minimum number of whole octets required. To minimize overhead, the prefix lengths and prefixes are NOT aligned along word boundaries.
The INTERFACE ASSOCIATION, HOST ASSOCIATION, and NETWORK PREFIX ASSOCIATION messages each have the following three subtypes (similar to those for the TOPOLOGY UPDATE message):
FULL (ST = 0) Indicates that this is a FULL update that includes all interface addresses, host addresses, or network prefixes associated with the given router ID.
ADD (ST = 1) Indicates that the included IP addresses or network prefixes are associated with the router ID, but may not include all such IP addresses or network prefixes.
DELETE (ST = 2) Indicates that the included IP addresses or network prefixes are no longer associated with the router ID.
This section describes the operation of the TBRPF routing module. The operation is divided into the following subsections: periodic processing, updating the source tree and topology graph, updating the routing table, updating the reported node set, generating periodic updates, generating differential updates, processing topology updates, expiring topology information, optional reporting of redundant topology information, local topology changes, generating association messages, processing association messages, and non-relay operation. The operation is described in terms of procedures (e.g., Update_All), which may be executed periodically or in response to some event, and may be called by other procedures. In all procedures, node i is the node executing the procedure.
Each node executes the procedure Update_All() periodically, at least once every DIFF_UPDATE_INTERVAL seconds, which is typically equal to HELLO_INTERVAL. This procedure is defined as follows:
Ogier, et al. Experimental [Page 24]
RFC 3684 TBRPF February 2004
Update_All() 1. For each interface I, create empty message list msg_list(I). 2. For each interface I, generate a HELLO message for interface I and add it to msg_list(I). 3. Expire_Links(). 4. Update_Source_Tree(). 5. Update_Routing_Table(). 6. If REPORT_FULL_TREE = 0, execute Update_RN(); otherwise (the full source tree is reported) Update_RN_Simple(). 7. If current_time >= next_periodic: 7.1. Generate_Periodic_Update(). 7.2. Set next_periodic = current_time + PER_UPDATE_INTERVAL. 8. Else, Generate_Diff_Update(). 9. Generate_Association_Messages(). 10. For each interface I, send the msg_list(I) on interface I. 11. Set old_T = T and old_RN = RN.
8.4.2. Updating the Source Tree and Topology Graph
The procedure Update_Source_Tree() is a variant of Dijkstra's algorithm, which is called periodically and in response to topology changes, to update the source tree T and the topology graph TG. This algorithm computes shortest paths subject to two link cost penalties. The penalty NON_REPORT_PENALTY is added to the cost of links (u,v) that are not currently reported by the parent p(u) so that, whenever possible, a link (u,v) is included in T only if it is currently reported by the parent. To allow immediate rerouting when p(u) changes, it may be necessary to temporarily use a link (u,v) that is not currently reported by the new parent. The penalty NON_TREE_PENALTY is added to the cost of links (u,v) that are not currently in T, to reduce the number of changes to T. When there exist multiple paths of equal cost to a given node, router ID is used to break ties.
The algorithm is defined as follows (where node i is the node executing the procedure):
Update_Source_Tree() 1. For each node v in TT, set d(v) = INFINITY, pred(v) = NULL, old_p(v) = p(v), and p(v) = NULL.
2. Set d(i) = 0, p(i) = i, pred(i) = i.
3. Set S = {i}. (S is the set of labeled nodes.)
4. For each node j in N, set d(j) = c(i,j), pred(j) = i, and p(j) = j. (If USE_METRICS = 0, then all link costs c(i,j) are 1.)
Ogier, et al. Experimental [Page 25]
RFC 3684 TBRPF February 2004
5. While there exists an unlabeled node u in TT such that d(u) < INFINITY: 5.1. Let u be an unlabeled node in TT with minimum d(u). (A heap should be used to find u efficiently.) 5.2. Add u to S (u becomes labeled). 5.3. If p(u) is not equal to old_p(u) (parent has changed): 5.3.1. For each link (u,v) in TG with tail u, if reported(u,v) = 1, set reported(u,v) = 0 and set nr_expire(u,v) = current_time + PER_UPDATE_INTERVAL. 5.3.2. If p(u) is in r(u) (p(u) is reporting u): 5.3.2.1. Set tg_expire(u) = rt_expire(p(u),u). 5.3.2.2. If p(u) = u (u is a neighbor), remove all links (u,v) with tail u from TG. 5.3.2.3. For each link (u,v) with p(u) in r(u,v): 5.3.2.3.1. Add (u,v) to TG and set reported(u,v) = 1. 5.3.2.3.2. Set metric(u,v) = metric(p(u),u,v). If USE_METRICS=1, set c(u,v)=metric(u,v). 5.4. For each node v such that (u,v) is in TG: 5.4.1. If reported(u,v) = 0, set cost = c(u,v) + NON_REPORT_PENALTY. (This penalizes (u,v) if not reported by p(u).) 5.4.2. Else, if p(u) = u AND u is not in r(v), set cost = c(u,v) + NON_REPORT_PENALTY. (This penalizes (u,v) if u is a neighbor and is not reporting v.) 5.4.3. If (u,v) is not in old_T and p(u) != u, set cost = cost + NON_TREE_PENALTY. 5.4.4. If (d(u) + cost, u) is lexicographically less than (d(v), pred(v)), set d(v) = d(u) + c(u,v), pred(v) = u, and p(v) = p(u).
6. Update the source tree T as follows: 6.1. Remove all links from T. 6.2. For each node u other than i such that pred(u) is not NULL, add the link (pred(u), u) to T.
The routing table is updated following any change to the source tree or the association tables (interface table, host table, or network prefix table). The routing table is updated according to procedure Update_Routing_Table(), which is defined as follows:
Update_Routing_Table()
1. Remove all tuples from the routing table.
Ogier, et al. Experimental [Page 26]
RFC 3684 TBRPF February 2004
2. For each node u in TT (other than this node) such that p(u) is not NULL, add the tuple (rt_dest, rt_next, rt_dist, rt_if_id) to the routing table, where: rt_dest = u, rt_if_id = local_if(p(u)), rt_next = nbr_if(p(u)), rt_dist = d(u).
3. For each tuple (if_addr, if_rid, if_expire) in the interface table, if a routing table entry (rt_dest, rt_next, rt_dist, rt_if_id) exists such that rt_dest = if_rid, add the tuple (if_addr, rt_next, rt_dist, rt_if_id) to the routing table.
4. For each tuple (h_addr, h_rid, h_expire) in the host table, if there exists a routing table entry (rt_dest, rt_next, rt_dist, rt_if_id) such that rt_dest = h_rid, add the tuple (h_addr, rt_next, rt_dist, rt_if_id) to the routing table, unless an entry already exists with the same value for h_addr and a lexicographically smaller value for (rt_dist, rt_dest).
5. For each tuple (net_prefix, net_length, net_rid, net_expire) in the network prefix table, if there exists a routing table entry (rt_dest, rt_next, rt_dist, rt_if_id) such that rt_dest = net_rid, add the tuple (net_prefix/net_length, rt_next, rt_dist, rt_if_id) to the routing table, unless an entry already exists with the same value for net_prefix/net_length and a lexicographically smaller value for (rt_dist, rt_dest).
Recall that the reported subtree RT is defined to be the set of links (u,v) in T such that u is in the reported node set RN. Each node updates its RN immediately before generating periodic or differential topology updates.
If REPORT_FULL_TREE = 1 (so that a node reports its entire source tree), then RN simply consists of all reachable nodes, i.e., all nodes u such that pred(u) is not NULL. The procedure that computes RN in this manner is called Update_RN_Simple(). The rest of this section describes how RN is computed assuming REPORT_FULL_TREE = 0.
A node first determines which of its neighbors belong to RN. Node i includes a neighbor j in RN if and only if node i determines that one of its neighbors may select i to be its next hop on its shortest path to j. To make this determination, node i computes the shortest paths, up to 2 hops, from each neighbor to each other neighbor, using only neighbors (or node i itself) as an intermediate node, and using
Ogier, et al. Experimental [Page 27]
RFC 3684 TBRPF February 2004
relay priority and router ID to break ties. If a link metric is used, then shortest paths are computed with respect to the link metric; otherwise min-hop paths are computed.
After a node determines which neighbors are in RN, each node u (other than node i) in the topology table is included in RN if and only if the next hop p(u) to u is in RN. Equivalently, node u is included in RN if and only if u is in the subtree of T rooted at some neighbor j that is in RN. Thus, the reported subtree RT includes the subtrees of T that are rooted at neighbors in RN. Node i also includes itself in RN; thus RT also includes all local links (i,j) to neighbors j.
The precise procedure for updating RN is defined as follows:
Update_RN() 1. Set RN = empty. 2. For each neighbor s in N such that s is in r(s), i.e., such that s is reporting itself: (Initialize to run Dijkstra for source s, for 2 hops.) 2.1. For each node j in N+{i}, set dist(j) = INFINITY and par(j) = NULL. 2.2. Set dist(s) = 0 and par(s) = s. 2.3. For each node j in N+{i} such that (s,j) is in TG: 2.3.1. Set dist(j) = metric(s,j), par(j) = j. 2.3.2. For each node k in N such that (j,k) is in TG: 2.3.2.1. Set cost = metric(j,k). 2.3.2.2. If (dist(j) + cost, nbr_pri(j), j) is lexicographically less than (dist(k), nbr_pri(par(k)), par(k)), set dist(k) = dist(j) + cost and par(k) = j. 2.4. For each neighbor j in N, add j to RN if par(j) = i. 3. Add i to RN. (Node i is always in RN.) 4. For each node u in the topology table, add u to RN if p(u) is in RN.
In some cases it may be desirable to limit the radius (number of hops) that topology information is propagated. Since each TBRPF packet is sent only to immediate (1-hop) neighbors, this cannot be achieved by using a time-to-live field. Instead, the propagation of topology information can be limited to a radius of K hops by limiting RN (at all nodes) to include only nodes that are at most K-1 hops away. Assuming min-hop routing is used, so that d(u) is the number of hops to node u, this can be done by modifying Step 4 of Update_RN() as follows:
4. For each node u in the topology table, add u to RN if p(u) is in RN and d(u) <= K-1.
Every PER_UPDATE_INTERVAL seconds, each node generates and transmits, on all interfaces, a set of FULL TOPOLOGY UPDATE messages (one message for each node in RN that is not a leaf of T), which describes the reported subtree RT. Whenever possible, these messages are included in a single packet, in order to minimize the number of control packets transmitted.
Each topology update message contains the router IDs for n+1 nodes u, v_1,...,v_n, which represent the n links (u,v_1),..., (u,v_n). The n head nodes v_1,..., v_n are divided into three lists in order to convey additional information and thus reduce the number of messages that must be generated. In particular, the first NRL head nodes are leaves of T, thus avoiding the need to generate separate topology update messages for leaf nodes u. Similarly, the last n-(NRL+NRNL) head nodes are not in RN, thus avoiding the need to generate separate topology update messages for nodes u that have been removed from RN.
Periodic update messages are generated according to procedure Generate_Periodic_Update(), defined as follows (where node i is the node executing the procedure):
Generate_Periodic_Update() For each node u in RN (including node i) that is not a leaf of T, add the update (FULL, n, NRL, NRNL, u, v_1,..., v_n) to msg_list(I) for each interface I, where:
(a) v_1,..., v_n are the nodes v such that (u,v) is in T, the first NRL of these are nodes in RN that are leaves of T, the next NRNL of these are nodes in RN that are not leaves of T, and the last n-(NRL+NRNL) of these are not in RN.
(b) If USE_METRICS = 1, then the M (metrics) bit is set to 1 and the link metrics metric(u,v_1),..., metric(u,v_n) are included in the message.
Every DIFF_UPDATE_INTERVAL seconds, if it is not time to generate a periodic update, and if RT has changed since the last time a topology update was generated, a set of TOPOLOGY UPDATE messages describing the changes to RT is generated and transmitted on all interfaces. These messages are constructed according to procedure Generate_Differential_Update(), defined as follows:
Ogier, et al. Experimental [Page 29]
RFC 3684 TBRPF February 2004
Generate_Differential_Update() For each node u in RN: 1. If u is not in old_RN (u was added to RN) and is not a leaf of T, add the update (FULL, n, NRL, NRNL, u, v_1,..., v_n) to msg_list(I) for each I, where: (a) v_1,..., v_n, NRL, and NRNL are defined as above for periodic updates. (b) If USE_METRICS = 1, then the M (metrics) bit is set to 1 and the link metrics metric(u,v_1),..., metric(u,v_n) are included in the message.
2. Else, if u is in old_RN and is not a leaf of T: 2.1. Let v_1,..., v_n be the nodes v such that (u,v) is in T AND at least one of the following 3 conditions holds: (a) (u,v) is not in old_T, or (b) v is in old_RN but not in RN, or (c) v is a leaf and is in RN but not in old_RN. 2.2. If this set of nodes is nonempty, add the update (ADD, n, NRL, NRNL, u, v_1,..., v_n) to msg_list(I) for each interface I, where: (a) NRL and NRNL are defined as above. (b) If USE_METRICS = 1, then the M (metrics) bit is set to 1 and the link metrics metric(u,v_1),..., metric(u,v_n) are included in the message.
3. If u is in old_RN: 3.1. Let v_1,..., v_n be the nodes v such that (u,v) is in old_T but not in TG, and either IMPLICIT_DELETION = 0 or pred(v) is not in RN (or is NULL). (If IMPLICIT_DELETION = 1 and pred(v) is in RN, then the deletion of (u,v) is implied by an ADD update for another link (w,v).) 3.2. If this set of nodes is nonempty, add the update (DELETE, n, u, v_1,..., v_n) to msg_list(I) for each I.
When a packet containing a list (msg_list) of TOPOLOGY UPDATE messages is received from node j, the list is processed according to the procedure Process_Updates(j, msg_list), defined as follows. In particular, this procedure updates TT, TG, and the reporting neighbor lists r(u) and r(u,v). If any link in T has been deleted from TG, then Update_Source_Tree() and Update_Routing_Table() are called to provide immediate rerouting.
Process_Updates(j, msg_list) 1. For each update = (subtype, n, NRL, NRNL, u, v_1,..., v_n) in msg_list:
Ogier, et al. Experimental [Page 30]
RFC 3684 TBRPF February 2004
1.1. Create an entry for u in TT if it does not exist. 1.2. If subtype = FULL, Process_Full_Update(j, update). 1.3. If subtype = ADD, Process_Add_Update(j, update). 1.4. If subtype = DELETE, Process_Delete_Update(j, update). 2. If there exists any link in T that is not in TG: 2.1. Update_Source_Tree(). 2.2. Update_Routing_Table().
Process_Full_Update(j, update) 1. Add j to r(u). 2. Set rt_expire(j,u) = current_time + TOP_HOLD_TIME. 3. For each link (u,v) s.t. j is in r(u,v): 3.1. Remove j from r(u,v). 3.2. If pred(j,v) = u, set pred(j,v) = NULL. 4. If j = p(u) OR p(u) = NULL: 4.1. Set tg_expire(u) = current_time + TOP_HOLD_TIME. 4.2. For each v s.t. (u,v) is in TG, If reported(u,v) = 1, remove (u,v) from TG. 5. Process_Add_Update(j, update).
Process_Add_Update(j, update) For m = 1,..., n: ((u,v_m) is the mth link in update.) 1. Let v = v_m. 2. Create an entry for v in TT if it does not exist. 3. Add j to r(u,v). 4. If j = p(u) OR p(u) = NULL: 4.1. Add (u,v) to TG. 4.2. Set reported(u,v) = 1. 5. If the M (metrics) bit in update is 1: 5.1. Set metric(j,u,v) to the m-th metric in the update. 5.2. If j = p(u) OR p(u) = NULL: 5.2.1. Set metric(u,v) = metric(j,u,v). 5.2.2. If USE_METRICS = 1, set c(u,v) = metric(u,v). 6. If the D (implicit deletion) bit in update is 1: 6.1. Set w = pred(j,v). 6.2. If (w != NULL AND w != u): 6.2.1. Remove j from r(w,v). 6.2.2. If j = p(w), remove (w,v) from TG. 7. Set pred(j,v) = u. (Set new predecessor.) 8. If m <= NRL (v = v_m is a reported leaf): 8.1. Set leaf_update = (FULL, 0, 0, 0, v). 8.2. Process_Full_Update(j, leaf_update). 9. If m > NRL + NRNL (v = v_m is not reported by j): 9.1. Remove j from r(v). 9.2. Set rt_expire(j,v) = 0. 9.3. For each node w s.t. j is in r(v,w), remove j from r(v,w).
Ogier, et al. Experimental [Page 31]
RFC 3684 TBRPF February 2004
9.4. If j = p(v), then for each node w s.t. (v,w) is in TG and reported(v,w) = 1, set reported(v,w) = 0 and set nr_expire(v,w) = current_time + PER_UPDATE_INTERVAL.
Process_Delete_Update(j, update) For m = 1,..., n: ((u,v_m) is the mth link in update.) 1. Let v = v_m. 2. Remove j from r(u,v). 3. If pred(j,v) = u, set pred(j,v) = NULL. 4. If j = p(u), remove (u,v) from TG.
Each node periodically checks for outdated topology information based on the expiration timers tg_expire(u), rt_expire(j,u), and nr_expire(u,v), and removes any expired entries from TG and from the lists r(u) and r(u,v). This is done according to the following procedure Expire_Links(), which is called periodically just before the source tree is updated.
Expire_Links() For each node u in TT other than node i: 1. If tg_expire(u) < current_time, then for each v s.t. (u,v) is in TG, remove (u,v) from TG. 2. Else, for each v s.t. (u,v) is in TG, if reported(u,v) = 0 AND nr_expire(u,v) < current_time, remove (u,v) from TG. 3. For each node j in r(u), if rt_expire(j,u) < current_time: 3.1. Remove j from r(u). 3.2. For each link (u,v) s.t. j is in r(u,v), remove j from r(u,v).
In addition, the following cleanup steps SHOULD be executed periodically to remove unnecessary entries from the topology table TT. A link (u,v) should be removed from TT if it is not in TG and not in old_T. A node u should be removed from TT if all of the following conditions hold: r(u) is empty, r(w,u) is empty for all w, and no link of TG has u as either the head or the tail.
8.4.9. Optional Reporting of Redundant Topology Information
Each node is required to report its reported subtree RT to neighbors. However, each node (independently of the other nodes) MAY report additional links, e.g., to provide increased robustness in highly mobile networks. For example, a node may compute any subgraph H of TG that contains T, and may report the "reported subgraph" RH which consists of links (u,v) of H such that u is in RN. In this case,
Ogier, et al. Experimental [Page 32]
RFC 3684 TBRPF February 2004
each periodic update describes RH instead of RT, and each differential update describes changes to RH. If this option is used, then the parameter IMPLICIT_DELETION MUST be set to 0, since the deletion of a link cannot be implied by the addition of another link if redundant topology information is reported.
This section describes the procedures that are followed when the neighbor discovery module detects a new link, the loss of a link, or a change in the metric for a link.
When a link (I,J) from a local interface I to a neighbor interface J is discovered via the neighbor discovery module, the procedure Link_Up(I,J) is executed, as defined below. Letting j be the neighbor node associated with interface J, Link_Up(I,J) adds j to N (if it does not already belong), updates the preferred local interface local_if(j) and neighbor interface nbr_if(j) so that the link from local_if(j) to nbr_if(j) has the minimum metric among all links from i to j, and updates metric(i,j) to be this minimum metric.
Link_Up(I,J) 1. Let j = nbr_rid(I,J). 2. If j is not in N: 2.1. Add j to N. 2.2. Add (i,j) to TG. 2.3. Set reported(i,j) = 1. 3. If nbr_metric(I,J) < metric(i,j), set local_if(j) = I, nbr_if(j) = J, and metric(i,j) = nbr_metric(I,J). 4. If USE_METRICS = 1, set cost(i,j) = metric(i,j).
When the loss of a link (I,J) from a local interface I to a neighbor interface J is detected via the neighbor discovery module, the procedure Link_Down(I,J) is executed, as defined below. Note that routes are updated immediately when a link is lost, and if the lost link is due to a link-layer failure notification, a differential topology update is sent immediately.
Link_Down(I,J) 1. Let j = nbr_rid(I,J). 2. If there does not exist a link (K,L) from node i to node j with nbr_status(K,L) = 2-WAY: 2.1. Remove j from N. 2.2. Remove (i,j) from TG. 3. If j is in N: 3.1. Let (K,L) be a link from i to j such that nbr_metric(K,L) is the minimum metric among all links from i to j.
Ogier, et al. Experimental [Page 33]
RFC 3684 TBRPF February 2004
3.2. Set local_if(j) = K, nbr_if(j) = L, and metric(i,j) = nbr_metric(K,L). 3.3. If USE_METRICS = 1, set cost(i,j) = metric(i,j). 5. Update_Source_Tree(). 6. Update_Routing_Table(). 7. If j is not in N and lost link is due to link-layer failure notification: 7.1. If (REPORT_FULL_TREE = 0) Update_RN(). 7.2. Else, Update_RN_Simple(). 7.3. Set msg_list = empty. 7.4. Generate_Diff_Update(). 7.5. Send msg_list on all interfaces. 7.6. Set old_T = T and old_RN = RN.
If the metric of a link (I,J) from a local interface I to a neighbor interface J changes via the neighbor discovery module, the following procedure Link_Change(I,J) is executed.
Link_Change(I,J) 1. Let j = nbr_rid(I,J). 2. Let (K,L) be a link from i to j such that nbr_metric(K,L) is the minimum metric among all links from i to j. 3. Set local_if(j) = K, nbr_if(j) = L, and metric(i,j) = nbr_metric(K,L). 4. If USE_METRICS = 1, set cost(i,j) = metric(i,j).
This section describes the procedures used to generate INTERFACE ASSOCIATION, HOST ASSOCIATION, and NETWORK PREFIX ASSOCIATION messages. Addresses or prefixes in the interface table, host table, and network prefix table are reported to neighbors periodically every IA_INTERVAL, HA_INTERVAL, and NPA_INTERVAL seconds, respectively. In addition, differential changes to the tables are reported every DIFF_UPDATE_INTERVAL seconds if it is not time for a periodic update (similar to differential topology updates). Each node reports only addresses or prefixes that are associated with nodes in the reported node set RN; this ensures the efficient broadcast of all associated addresses and prefixes to all nodes in the network.
The generated messages are sent on each interface. Whenever possible, these messages are combined into the same packet, in order to minimize the number of control packets transmitted.
Generate_Interface_Association_Messages() 1. If current_time > next_ia_time: 1.1. Set next_ia_time = current_time + IA_INTERVAL. 1.2. For each node u in RN: 1.2.1. Let addr_1,..., addr_n be the interface IP addresses associated with RID u in the current interface table. 1.2.2. If this list is nonempty, add the INTERFACE ASSOCIATION message (FULL, n, u, addr_1,..., addr_n) to msg_list(I) for each I.
2. Else, for each node u in RN: 2.1. Add the INTERFACE ASSOCIATION message (ADD, n, u, addr_1,..., addr_n) to msg_list(I) for each I, where addr_1,..., addr_n are the interface IP addresses that are associated with RID u in the current interface table but not in the old interface table. 2.2. Add the INTERFACE ASSOCIATION message (DELETE, n, u, addr_1,..., addr_n) to msg_list(I) for each I, where addr_1,..., addr_n are the interface IP addresses that are associated with RID u in the old interface table but not in the current interface table.
Generate_Host_Association_Messages() 1. If current_time > next_ha_time: 1.1. Set next_ha_time = current_time + HA_INTERVAL. 1.2. For each node u in RN: 1.2.1. Let addr_1,..., addr_n be the host IP addresses associated with RID u in the current host table. 1.2.2. If this list is nonempty, add the HOST ASSOCIATION message (FULL, n, u, addr_1,..., addr_n) to msg_list(I) for each I.
2. Else, for each node u in RN: 2.1. Add the HOST ASSOCIATION message (ADD, n, u, addr_1,..., addr_n) to msg_list(I) for each I, where addr_1,..., addr_n are the host IP addresses that are associated with RID u in the current host table but not in the old host table. 2.2. Add the HOST ASSOCIATION message (DELETE, n, u, addr_1,..., addr_n) to msg_list(I) for each I, where addr_1,..., addr_n are the host IP addresses that are associated with RID u in the old host table but not in the current host table.
Ogier, et al. Experimental [Page 35]
RFC 3684 TBRPF February 2004
Generate_Network_Prefix_Association_Messages() 1. If current_time > next_npa_time: 1.1. Set next_npa_time = current_time + NPA_INTERVAL. 1.2. For each node u in RN: 1.2.1. Let length_1, prefix_1,..., length_n, prefix_n be the network prefix lengths and prefixes associated with RID u in the current network prefix table. 1.2.2. If this list is nonempty, add the NETWORK PREFIX ASSOCIATION message (FULL, n, u, length_1, prefix_1, ..., length_n, prefix_n) to msg_list(I) for each I.
2. Else, for each node u in RN: 2.1. Add the NETWORK PREFIX ASSOCIATION message (ADD, n, u, prefix_1,..., prefix_n) to msg_list(I) for each I, where prefix_1,..., prefix_n are the network prefixes that are associated with RID u in the current prefix table but not in the old prefix table.
2.1. Add the NETWORK PREFIX ASSOCIATION message (DELETE, n, u, prefix_1,..., prefix_n) to msg_list(I) for each I, where prefix_1,..., prefix_n are the network prefixes that are associated with RID u in the old prefix table but not in the current prefix table.
When an INTERFACE ASSOCIATION, HOST ASSOCIATION, or NETWORK PREFIX ASSOCIATION message is received from node j, the interface table, host table, or network prefix table, respectively, is updated as described in the following three procedures.
Process_Interface_Association_Messages(j, msg_list) For each message (subtype, n, u, addr_1,..., addr_n) in msg_list such that j = p(u): 1. If subtype = FULL, remove all entries with if_rid = u from the interface table. 2. If subtype = FULL or ADD, then for m = 1,..., n, add the tuple (if_addr, if_rid, if_expire) to the interface table, where: if_addr = addr_m, if_rid = u, if_expire = current_time + IA_HOLD_TIME. 3. If subtype = DELETE, then for m = 1,..., n, remove the tuple (if_addr, if_rid, if_expire) from the interface table, where if_addr = addr_m and if_rid = u.
Ogier, et al. Experimental [Page 36]
RFC 3684 TBRPF February 2004
Process_Host_Association_Messages(j, msg_list) For each message (subtype, n, u, addr_1,..., addr_n) in msg_list such that j = p(u): 1. If subtype = FULL, remove all entries with h_rid = u from the host table. 2. If subtype = FULL or ADD, then for m = 1,..., n, add the tuple (h_addr, h_rid, h_expire) to the host table, where: h_addr = addr_m, h_rid = u, h_expire = current_time + HA_HOLD_TIME. 3. If subtype = DELETE, then for m = 1,..., n, remove the tuple (h_addr, h_rid, h_expire) from the host table, where h_addr = addr_m and h_rid = u.
Process_Network_Prefix_Association_Messages(j, msg_list) For each message (subtype, n, u, length_1, prefix_1, ..., length_n, prefix_n) in msg_list such that j = p(u): 1. If subtype = FULL, remove all entries with net_rid = u from the prefix table. 2. If subtype = FULL or ADD, then for m = 1,..., n, add the tuple (net_prefix, net_length, net_rid, net_expire) to the network prefix table, where: net_prefix = prefix_m, net_length = length_m, net_rid = u, net_expire = current_time + NPA_HOLD_TIME. 3. If subtype = DELETE, then for m = 1,..., n, remove the tuple (net_prefix, net_length, net_rid, net_expire) from the network prefix table, where net_prefix = prefix_m, net_length = length_m, and net_rid = u.
Nodes with relay priority equal to zero are called non-relay nodes, and do not forward packets (of any type) that are received from other nodes. A non-relay node is implemented simply by not generating or transmitting any TOPOLOGY UPDATE messages. A non-relay node may report (in association messages) addresses or prefixes that are associated with itself, but not those associated with other nodes. HELLO messages must be transmitted in order to establish links with neighbor nodes. The following procedures can be omitted in non-relay nodes: Update_RN(), Generate_Periodic_Update(), and Generate_Diff_Update().
This section lists the configurable parameters used by the routing module, and their proposed default values. All nodes MUST have the same value for all of the following parameters except REPORT_FULL_TREE and IMPLICIT_DELETION.
This section describes a mechanism for the efficient best-effort flooding (or network-wide broadcast) of packets to all nodes of a connected ad-hoc network. This mechanism can be considered an optimization of the classical flooding algorithm in which each packet is transmitted by every node of the network. In TBRPF flooding, information provided by TBRPF is used to decide whether a given received flooded packet should be forwarded. As a result, each packet is transmitted by only a relatively small subset of nodes, thus consuming much less bandwidth than classical flooding.
This document specifies that the flooding mechanism use the IPv4 multicast address 224.0.1.20 (currently assigned by IANA for "any private experiment"). Every node maintains a duplicate cache to keep track of which flooded packets have already been received. The duplicate cache contains, for each received flooded packet, the flooded packet identifier (FPI), which for IPv4 is composed of the source IP address, the IP identification, and the fragment offset values obtained from the IP header [14].
When a node receives a packet whose destination IP address is the flooding address (224.0.1.20), it checks its duplicate cache for an entry that matches the packet. If such an entry exists, the node
Ogier, et al. Experimental [Page 38]
RFC 3684 TBRPF February 2004
silently discards the flooded packet since it has already been received. Otherwise, the node retransmits the packet on all interfaces (see the exception below) if and only if the following conditions hold:
1. The TBRPF node associated with the source IP address of the packet belongs to the set RN of reported nodes computed by TBRPF.
2. When decremented, the 'ip_ttl' in the IPv4 packet header (respectively, the 'hop_count' in the IPv6 packet header) is greater than zero.
If the packet is to be retransmitted, it is sent after a small random time interval in order to avoid collisions. If the interface on which the packet was received is not a MANET interface (see the Terminology section), then the packet should not be retransmitted on that interface.
TBRPF is particularly well suited to MANETs consisting of mobile nodes with wireless network interfaces operating in peer-to-peer fashion over a multiple access communications channel. Although applicable across a much broader field of use, TBRPF is particularly well suited for supporting the standard DARPA Internet protocols [3][2]. In the following sections, we discuss practical considerations for the operation of TBRPF on MANETs.
We assume a MANET data link layer that supports broadcast, multicast and unicast addressing with best-effort (not guaranteed) delivery services between neighbors (i.e., a pair of nodes within operational communications range of one another). We further assume that each interface belonging to a node in the MANET is assigned a unicast data link layer address that is unique within the MANET's scope. While such uniqueness is not strictly guaranteed, the assumption of uniqueness is consistent with current practices for deployment of the Internet protocols on specific link layers. Methods for duplicate link layer address detection and deconfliction are beyond the scope of this document.
MANETs are formed as collections of routers and non-routing nodes that use network layer addresses when calculating the MANET topology. We assume that each node has at least one data link layer interface (described above) and that each such interface is assigned a network
Ogier, et al. Experimental [Page 39]
RFC 3684 TBRPF February 2004
layer address that is unique within the MANET. (Methods for network layer address assignment and duplicate address detection are beyond the scope of this document.) We further assume that each node will select a unique Router ID (RID) for use in TBRPF protocol messages, whether or not the node acts as a MANET router. Finally, we assume that each MANET router supports the multi-hop relay paradigm at the network layer; i.e., each router provides an inter-node forwarding service via network layer host routes which reflect the current MANET topology as perceived by TBRPF.
TBRPF employs a proactive neighbor discovery protocol at the network layer that maintains bi-directional link state for neighboring nodes through the periodic transmission of messages. Since TBRPF neighbor discovery messages contain both the data link and network layer address of the sender, implementations MAY perform automatic network-to-data link layer address resolution for the nodes with which they form links. An implementation may use such a mechanism to avoid additional message overhead and potential for packet loss associated with on-demand address resolution mechanisms such as ARP [15] or IPv6 Neighbor Discovery [16]. Implementations MUST respond to on-demand address resolution requests in the normal manner.
10.4. Support for Multiple Interfaces and/or Alias Addresses
MANET nodes may comprise multiple interfaces; each with a unique network layer address. Additionally, MANET nodes may wish to publish alias addresses such as when multiple network layer addresses are assigned to the same interface or when the MANET node is serving as a Mobile IP [17] home agent. Multiple interfaces and alias addresses are advertised in INTERFACE ASSOCIATION messages, which bind each such address to the node's RID.
MANET routers may advertise network prefixes which the router discovered via attached networks, external routes advertised by other protocols, or other means. Network prefixes are advertised in NETWORK PREFIX ASSOCIATION messages, which bind each such prefix to the node's RID.
Non-MANET hosts may establish connections to MANET routers through on-demand mechanisms such as ARP or IPv6 Neighbor Discovery. Such connections do not constitute a MANET link and therefore are not reported in TBRPF topology updates. Non-MANET hosts are advertised
Ogier, et al. Experimental [Page 40]
RFC 3684 TBRPF February 2004
in HOST ASSOCIATION messages, which bind the IP address of each host to the node's RID.
TBRPF packets are communicated using UDP/IP. Port 712 has been assigned by IANA for exclusive use by TBRPF. Implementations in private networks MAY employ alternate data delivery services (i.e., raw IP or local data-link encapsulation). The selection of an alternate data delivery service MUST be consistent among all MANET routers in the private network. In all implementations, the data delivery service MUST provide a checksum facility.
The following sections specify the operation of TBRPF over UDP/IP.
When IPv4 is used, TBRPF nodes obey IPv4 host and router requirements [4][5]. TBRPF packets are sent to the multicast address 224.0.0.2 (All Routers) and thus reach all TBRPF routers within single-hop transmission range of the sender. TBRPF routers MUST NOT forward packets sent to this multicast address.
Since non-negligible packet loss due to link failure, interference, etc. can occur, implementations SHOULD avoid IPv4 fragmentation/ reassembly whenever possible, by splitting large TBRPF protocol packets into multiple smaller packets at the application layer. When fragmentation is unavoidable, senders SHOULD NOT send TBRPF packets that exceed the minimum reassembly buffer size ([4], section 3.3.2) for all receivers in the network.
The specification of TBRPF for IPv6 is the same as for IPv4, except that 32-bit IPv4 addresses are replaced by 128-bit IPv6 addresses. However, to minimize overhead, router IDs remain at 32 bits, similar to OSPF for IPv6 [18].
The TBRPF flooding mechanism specified in this document uses the IPv4 multicast address 224.0.1.20, which is currently assigned by IANA for "any private experiment". In the event that this specification is advanced to standards track, a new multicast address assignment would be requested for this purpose.
Wireless networks are vulnerable to a variety of attacks, including denial-of-service attacks (e.g., flooding and jamming), man-in-the- middle attacks (e.g., interception, insertion, deletion, modification, replaying) and service theft. To counter such attacks, it is important to prevent the spoofing (impersonation) of TBRPF nodes, and to prevent unauthorized nodes from joining the network via neighbor discovery. To achieve this, TBRPF packets can be authenticated using the IP Authentication Header [19][20]. In addition, the Encapsulating Security Payload (ESP) header [21] can be used to provide confidentiality (encryption) of TBRPF packets.
The IETF SEcuring Neighbor Discovery (SEND) Working Group analyzes trust models and threats for ad hoc networks [22]. TBRPF can be extended in a straightforward manner to use SEND mechanisms, e.g., [23].
The authors would like to thank the Army Systems Engineering Office (ASEO) for funding part of this work.
The authors would like to thank several members of the MANET working group for many helpful comments and suggestions, including Thomas Clausen, Philippe Jacquet, and Joe Macker.
The authors would like to thank Bhargav Bellur for major contributions to the original (full-topology) version of TBRPF, Ambatipudi Sastry for his support and advice, and Julie S. Wong for developing a new implementation of TBRPF and suggesting several clarifications to the TBRPF Routing Operation section.
[8] Ogier, R., "Topology Dissemination Based on Reverse-Path Forwarding (TBRPF): Correctness and Simulation Evaluation", Technical Report, SRI International, October 2003.
[10] Ogier, R., "Efficient Routing Protocols for Packet-Radio Networks Based on Tree Sharing", Proc. Sixth IEEE Intl. Workshop on Mobile Multimedia Communications (MOMUC'99), November 1999.
[11] Bellur, B. and R. Ogier, "A Reliable, Efficient Topology Broadcast Protocol for Dynamic Networks", Proc. IEEE INFOCOM '99, New York", March 1999.
[12] Clausen, T. and P. Jacquet, Eds., "Optimized Link State Routing Protocol (OLSR)", RFC 3626, October 2003.
[13] Bertsekas, D. and R. Gallager, "Data Networks", Prentice-Hall, 1987.
[14] Perkins, C., Belding-Royer, E. and S. Das, "IP Flooding in Ad Hoc Mobile Networks", Work in Progress, November 2001.
[15] Plummer, D., "Ethernet Address Resolution Protocol: Or converting network protocol addresses to 48.bit Ethernet address for transmission on Ethernet hardware", STD 37, RFC 826, November 1982.
[16] Narten, T., Nordmark, E. and W. Simpson, "Neighbor Discovery for IP Version 6 (IPv6)", RFC 2461, December 1998.
[17] Perkins, C., Ed., "IP Mobility Support for IPv4", RFC 3344, August 2002.
Ogier, et al. Experimental [Page 43]
RFC 3684 TBRPF February 2004
[18] Coltun, R., Ferguson, D. and J. Moy, "OSPF for IPv6", RFC 2740, December 1999.
[19] Kent, S. and R. Atkinson, "Security Architecture for the Internet Protocol", RFC 2401, November 1998.
[20] Kent, S. and R. Atkinson, "IP Authentication Header", RFC 2402, November 1998.
[21] Kent, S. and R. Atkinson, "IP Encapsulating Security Payload (ESP)", RFC 2406, November 1998.
[22] Nikander, P., "IPv6 Neighbor Discovery Trust Models and Threats", Work in Progress, April 2003.
[23] Arkko, J., "SEcure Neighbor Discovery (SEND)", Work in Progress, June 2003.
Ogier, et al. Experimental [Page 44]
RFC 3684 TBRPF February 2004
Authors' Addresses
Richard G. Ogier SRI International 333 Ravenswood Ave. Menlo Park, CA 94025 USA
Copyright (C) The Internet Society (2004). This document is subject to the rights, licenses and restrictions contained in BCP 78 and except as set forth therein, the authors retain all their rights.
This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Intellectual Property
The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf-ipr@ietf.org.
Acknowledgement
Funding for the RFC Editor function is currently provided by the Internet Society.