DTN IP Neighbor Discovery (IPND)

Nodes use DTN IPND beacons, small UDP messages in the IP underlay, to advertise presence. Similarly, IPND beacons received from other nodes serve to detect the availability of DTN neighbors. Nodes SHOULD both send and receive beacons. When the IP underlay is based on the the IPv4 protocol, these beacon messages, detailed in , may be sent as UDP datagrams in either unicast, multicast, or broadcast packets. When the IP underlay uses IPv6, the beacon messages may be sent as either unicast or multicast packets. The beacon message content is agnostic to the underlying transport mode. Broadcast beacons are designed to reach unknown neighbors in neighborhoods within the local network broadcast domain. IPv4 multicast or IPv6 multicast beacons extend the scope of beacon dissemination to potentially include multiple networks across routed boundaries. On broadcast media such as Ethernet or wireless, multicast and broadcast beacons are sent as link-layer broadcast messages. Broadcast and multicast discovery are described in . In contrast, unicast beacons are sent only to explicitly known and enumerated neighbors as described in . Upon discovering a neighbor and its services, IPND can establish a connection to the new neighbor via an IP-based Convergence Layer Adapter (CLA), for example the TCP or Datagram CLA. The CLA then negotiates the connection per its individual specification and installs the appropriate next-hop routing information in the local node.

An IPND node SHOULD send beacons periodically. The time interval between beacons SHOULD be appropriate for the conditions of the network and MAY be configurable. An IPND node MAY make use of the OPTIONAL Beacon Period field in the beacon message to explicitly inform neighbors of the interval on which to expect future beacons. The Beacon Period is not fixed for a given sender and MAY change with each beacon message. If the Beacon Period is included and set to zero, then it SHALL be interpreted as negating any expectation for future beacons. A receiving node SHOULD either know the expected beacon interval of neighbors or extract the interval from the Beacon Period field of arriving messages. The beacon interval along with the existence and receive time of beacons SHOULD be used to determine the state of the sender’s ability to transmit to the receiver (i.e. the up or down state of the sender-to-receiver link). The exact algorithm for determining the link status based on received beacons is implementation-defined.

In the general case, the IP addresses of potential neighbors are not known in advance. To discover unknown neighbors, IPND beacon messages are sent as IP packets with either multicast or broadcast destination addresses. An IPND node MUST support multicast IP destination addresses and multicast IGMP / MLD group membership . A node MAY support IP broadcast destinations. IPv4 multicast addresses for IPND SHOULD be from the IANA assigned local network control block 224.0.0/24 . This block of multicast addresses is intentionally scoped to the local network to prevent dissemination to the wider Internet. Likewise, IPv6 multicast addresses for IPND should have link-local scope . An IPND node MAY also use other multicast addresses as required, such as IPv4 multicast addresses from the IANA assigned Internetwork Control Block or IPv6 multicast addresses with wider scope than link-local. One use case for this would be a mobile ad hoc network (MANET) environment which includes nodes that are not DTN-capable, but do support IP multicast forwarding, e.g. by means of SMF . Those nodes that are DTN-capable would then be able to discover each other over multiple IP hops. In all multicast addressing cases, a node MUST support a configurable IPv4 time-to-live value or IPv6 Hop Limit value for all beacon messages.

An IPND node SHOULD support unicast beacons. Since multicast or broadcast discovery may not always be feasible over internetworks, the IP addresses of potential neighbors reachable only across multiple underlay hops must be explicitly enumerated for discovery. While the neighbor’s address is therefore known, the availability of that neighbor is not known. IPND thus permits DTN nodes to discover available remote neighbors across multiple IP underlay hops when provided with the addresses of those neighbors. In this way, IPND can be used to bridge IP-based DTNs while detecting disconnections among and between the DTNs.

Sending data to an IP address matching a configured beacon destination SHOULD suppress the generation of beacon messages to that destination for a period of time up to but no longer than the beacon sending interval. This suppression SHOULD NOT occur if the parameters of a new beacon message would differ from the preceding beacon including the advertised services () or the Neighborhood Bloom Filter (NBF) (). Upon receiving a data packet from a neighbor where the packets do not represent a beacon, a node SHOULD behave as if a beacon had been received from that neighbor, as follows. If the data packet is addressed to this node via a unicast address, then the behavior SHOULD be as if the implied received beacon contains a Neighborhood Bloom Filter advertisement which indicates the membership of the receiving node in the sender’s 1-hop neighborhood. Otherwise, if the destination address is multicast or broadcast, then the receiving node should presume that the link is bidirectional if and only if its state was bidirectional prior to receiving the data packet (). The sender’s advertised services and beacon period are presumed to be unchanged since the sender’s last received beacon. If no beacons have previously been received from such a neighbor, then it is presumed that there are no services associated with the sender.

Many routing protocols work correctly only when links are bi- directional. In wired IP networks, link bi-directionality can often be presumed. For other types of networks, such as Mobile Ad Hoc Networks (MANETs) this assumption often does not hold. If a link to a neighbor is said to be “up” only because one or more beacon messages have been received from that neighbor over a wireless medium, it is not generally safe to assume that the link is bidirectional. In practice, MANETs often have links that are only unidirectional due to differences in antennae, transmit power, hardware variability, multi-path effects, etc. To discover the bi-directionality of links, an IPND Neighborhood Bloom Filter (NBF) () facility MAY be employed in which each node advertises a Bloom filter representation of the set of neighbors from whom it has received enough recent beacons to be considered “up”. Upon receiving a beacon from an “up” neighbor that advertises an NBF which represents a set containing the receiving node’s ID, the link SHOULD then be considered bi-directional.

depicts the format of beacon messages. Note that IPND follows the DTN convention of using Self-Delimiting Numeric Values (SDNVs) to represent variable length integers. An IPND node MUST use UDP checksums to ensure correctness.

The beacon message is comprised of the following fields: Version: An 8-bit field indicating the version of IPND that constructed this beacon. The present document describes version 0x04 of IPND. This version field is incremented for IPND if either the IPND protocol is modified or the Bundle Protocol version is incremented. In this way the field can also be used to determine the BP version supported by a potential DTN neighbor. Flags: An 8-bit field indicating IPND processing flags. Four flags are currently defined. 0x00 indicates that no special processing should be performed on the beacon. If more than one of the Flags bits is set, then the associated structures will appear in the beacon message according to their bit order (Bit 0 is first). Semantics of bits are described here from least significant (LSb) to most significant (MSb).

Beacon sequence number: A two-octet unsigned integer value incremented once for each beacon transmitted to a particular IP address. EID Length: The byte length of the canonical EID contained in the beacon. The EID length field is an SDNV and is therefore variable length. A two-octet length is shown for convenience of representation. Canonical EID: The canonical end node identifier of the neighbor advertised by the beacon message. The canonical EID is variable length and represented as a Uniform Resource Identifier . Service Block: Optional announced services () in the beacon. Services MAY include CLAs (), routing parameters, a Neighborhood Bloom Filter (), and other implementation-dependent services. Beacon Period: Optional field indicating the sender’s current beacon interval in seconds. A value of zero indicates that the beacon period is undefined. The Beacon Period is an SDNV and is therefore variable length. A two-octet length is shown for convenience of representation.

As described previously, beacon messages may optionally include a block of service availability information. The service block is intended to contain representations of available CLAs, routers, a Neighborhood Bloom Filter, etc., but is sufficiently general to accommodate implementation-specific services provided by the advertising node. For example, the source IP address of a received beacon suffices to identify the remote node at the IP level. However, the IP address alone does not inform other processes via which transport mechanism (e.g. TCP or UDP) or via which transport port the remote node is offering a connection. Similarly, nodes do not know which routers (e.g. PRoPHET ) are running on a remote node in order to inform bundle exchange. Therefore, a beacon MAY contain a service block which serves to notify nodes about the availability of these services. The format of a service block is given in .

A service block is comprised of the following fields: Number of services: The number of services described in the block. The number of services is an SDNV and is therefore variable length. Service Definition(s): A list of service definitions encoded according to the IPND Service Definition TLV encoding (Section 2.6.2) specification.

The IPND Service Definition TLV encoding scheme (IPND-SD-TLV) provides for the standardization of service definitions using a format that focuses on simplicity, flexibility, and efficiency. IPND-SD-TLV borrows many ideas from from the ASN.1 Basic Encoding Rules (BER) specification. Like ASN.1 BER, IPND-SD-TLV structures are generally composed of three distinct parts: Tag: A numeric token which identifies the structure (REQUIRED). Length: A numeric value which specifies the size of the content block (sometimes REQUIRED). Value: The content block, which contains the value(s) described by the tag (REQUIRED). IPND-SD-TLV tags SHALL be 8-bit values, providing IPND a range of 256 possible tag numbers. Tag assignments are designed to provide a basic, standard set of building blocks while remaining flexible enough to allow the implementation of unforeseen specifications. The first 128 tag numbers (i.e. 0-127) SHALL be reserved for standard definitions; the remaining tags (i.e. 128-255) MAY be used for implementation-specific (private) definitions. This design allows a node to inspect the most significant bit (bit-7, zero-indexed) of the tag to determine whether it is a reserved or private value. IPND-SD-TLV defines two classes of data types: Primitive and Constructed (the difference between these data types is discussed below). Reserved tag numbers are designed such that the class of a data type can be determined by examining the second-most significant bit (bit-6, zero-indexed) of the tag. If this bit is not set, the data type is primitive, otherwise it is constructed. As a result of this design, reserved primitive types SHALL be assigned tag numbers 0-63, while reserved constructed types SHALL be assigned tag numbers 64-127. Private tag numbers are always expected to represent constructed data types, therefore private (implementation-specific) constructed types (if in use by IPND) SHALL be assigned tag numbers 128-255. The construction of IPND-SD-TLV tags is depicted in

In order to keep encoded services simple and compact, IPND-SD-TLV SHALL omit the length field in cases where the content’s length is always fixed (e.g. an IP address) or described in-place (e.g. an SDNV value). In the cases where an explicit length field is required (e.g. string content), an IPND node SHALL SDNV encode the length values. Additionally, a length field MUST be included in constructed types immediately following the tag value which describes the length, in bytes, of the structure’s content block. This constraint allows a node to skip constructed types that are unrecognized while reading a received Service Block. An IPND node SHALL SDNV encode these length values. Again, IPND-SD-TLV defines two classes of data types: Primitive and Constructed. Primitive types represent fundamental data types such as integers or strings. An IPND node MUST support the primitive data types specified in . Note that primitive types use one of three distinct length specifiers: Fixed: The content always has a fixed length and SHALL NOT include a length field. Fixed length numeric values (including floating point numbers) SHALL be written in network byte order. Variable: The content is variable length but is encoded as an SDNV, therefore it SHALL NOT include a length field. Explicit: The content is variable and does not describe its own length, therefore it MUST include a length field immediately following the tag value.

Note that a special case exists for representing the empty string and the empty byte array for the “string” and “bytes” data types, respectively. In both cases, “empty” is represented by an explicit length value of 1 and content of a single null byte. Constructed data types represent structures that are composed of other data types. As described earlier, reserved constructed types SHALL be assigned tag numbers 64-127. Additionally, nodes MAY assign tag numbers 128-255 to private constructed types in order to allow for implementation-specific constructed types. An IPND node SHALL use constructed types to specify service definitions as described in Section 2.6.3. It is important to note that the order in which other types are composed within a constructed type need not be explicitly stated. Ordering only becomes an issue in the case where a constructed type (not representing an array structure) contains multiple instances of the same data type. In order to defeat this issue, implementations MUST create data type wrappers in order to differentiate identical types. This design allows IPND to be order-agnostic when it comes to reading data types that compose a constructed type. Appendix B describes an example where data type wrappers are used to differentiate identical fundamental types.

A service is an IPND-SD-TLV structure that represents an advertisement for a DTN-related resource available on the beacon source node. Each service type SHALL have a unique tag number in order to identify it within the service block. Nodes SHALL use the initial set of tag assignments described in (the rationale for tag numbering is described in ).

An IPND node MUST support the service definitions for CLA-TCP-v6 and CLA-UDP-v6; that is, a node MUST support the standard definitions for TCP CLA advertisements and UDP CLA advertisements, respectively (both supporting IPv6 128-bit addresses). An example bitwise representation of the CLA-TCP-v6 service is depicted in {{figure6}. Note that the format of the CLA-UDP-v6 service is identical except for the initial tag number, which would instead be 67 (hex 0x43).

An IPND node MAY support the CLA-TCP-v4, CLA-UDP-v4, CLA-TCP-HN, CLA-UDP-HN, CLA-DCCP-v4, CLA-DCCP-v6 and CLA-DCCP-HN service definitions. Bitwise representations of the CLA-UDP-v4, CLA-TCP-HN and CLA-DCCP-v6 services are depicted in Appendix A. Additionally, a node MAY support the Neighborhood Bloom Filter services (NBF-Hashes and NBF-Bits). These services are described below (). Lastly, a node MAY support any implementation-specific services with tag numbers 128-255. Appendix B describes an example of an implementation-specific service that makes use of private tag number assignments.

In order to efficiently determine link bi-directionality, a node represents the set of its 1-hop neighbors using a Bloom filter referred to as the Neighborhood Bloom Filter (NBF). Upon receiving a beacon from a neighbor that contains NBF service information, a node can quickly determine whether it is in the neighbor’s NBF set, and thereby determine whether the link is bidirectional. Every node that might operate in an environment where discovered links may not be bidirectional SHOULD include NBF service advertisements in its multicast or broadcast beacons which describe the membership of its 1-hop neighbor set. This is especially true if a node’s routing protocol presumes that links are bidirectional. An NBF need not be included within every beacon, but one SHOULD be present within at least one broadcast or multicast beacon following a change in the 1-hop neighborhood of the node. An NBF advertisement MAY be present in every broadcast or multicast beacon. In order to advertise an NBF, an IPND node MUST include two distinct services in the Service Block of some (or all) of its beacons: NBF- Hashes, which describes the hash algorithms used to compute the NBF bit array; and NBF-Bits, which contains the actual bit array of the NBF. The bits set in the NBF-Bits structure MUST be defined by computing hashes on the canonical EID of each 1-hop neighbor considered to be “up”. Each hash algorithm used to compute the NBF bit array MUST be identified in the NBF-Hashes structure (using numerical identifiers; one byte per identifier). Exemplary bitwise formats of fictional NBF-Hashes and NBF-Bits structures are depicted in and , respectively. Note that the NBF bit array in the NBF-Bits structure must be byte-aligned, and SHALL be padded with zero bits at the end of the bit array to achieve byte-alignment.

Different networks naturally have distinct requirements, tolerance for overhead, and node computing resources, so the parameters of the Bloom Filter such as the bit array length, and the number and types of hash algorithms, are not mandated by IPND. However, all nodes participating in such a DTN SHOULD be aware of the same set of hash algorithms and their respective identifiers used in NBF-Hashes structures. NBF services, if present, MAY be ignored by a receiving IPND node if its implementation does not provide for it, or if the parameters of the Bloom filter cannot be determined with certainty (e.g. if the hash function identifiers are not recognized).

IP-based CLAs are generally expected to depend on an IPND implementation module for their discovery service. A CLA MAY opt not to use IPND, either because that CLA does not require discovery or provides its own. Once IPND discovers a new neighbor it MUST inform all CLAs which depend on IPND of the neighbor’s existence and the discovered parameters. The exact means by which IPND communicates with the CLAs is implementation dependent. Similarly, once IPND determines that a link has gone down, it MUST inform all dependent CLAs of the link down event.

Note that an IPND node SHOULD maintain state over all existing neighbors in order to prevent CLAs from needlessly attempting to establish connections between nodes that are already connected. To maintain the current neighbor set, IPND removes stale neighbors after the defined neighbor receive timeout period elapses without receiving any beacon messages from a particular neighbor. Upon detecting a neighbor that is no longer available, IPND MAY provide hints to the CLAs that the neighbor is gone. Note that some CLAs themselves provide keepalive-type functionality and therefore IPND is not necessarily required to detect down neighbors. However, relying on IPND to provide both discovery and availability information provides a single, coherent point in the system design to maintain neighbor information.