Asymmetric Extended Route Optimization (AERO)

The following sections specify the operation of IP over Asymmetric Extended Route Optimization (AERO) links:

S1; X2->S2)| | MSP M1 | +-+---------+--+-+ +--------------+ | Secured | | +--------------+ |AERO Server S1| | tunnels | | |AERO Server S2| | Nbr: C1, R1 +-----+ | +-----+ Nbr: C2, R1 | | default->R1 | | | default->R1 | | X1->C1 | | | X2->C2 | +-------+------+ | +------+-------+ | AERO Link | | X===+===+===================+==)===============+===+===X | | | | +-----+--------+ +--------+--+-----+ +--------+-----+ |AERO Client C1| | AERO Proxy P1 | |AERO Client C2| | Nbr: S1 | |(Proxy Nbr Cache)| | Nbr: S2 | | default->S1 | +--------+--------+ | default->S2 | | MNP X1 | | | MNP X2 | +------+-------+ .--------+------. +-----+--------+ | (- Proxyed Clients -) | .-. `---------------' .-. ,-( _)-. ,-( _)-. .-(_ IP )-. +-------+ +-------+ .-(_ IP )-. (__ EUN )--|Host H1| |Host H2|--(__ EUN ) `-(______)-' +-------+ +-------+ `-(______)-' ]]> presents the AERO link reference model. In this model: the AERO link is an overlay network service configured over one or more underlying INET partitions which may be managed by different administrative authorities and have incompatible protocols and/or addressing plans. AERO Relay R1 aggregates Mobility Service Prefix (MSP) M1, discovers Mobile Network Prefixes (MNPs) X* and advertises the MSP via BGP peerings over secured tunnels to Servers (S1, S2). Relays use the SPAN service to bridge disjoint segments of a partitioned AERO link. AERO Servers S1 and S2 configure secured tunnels with Relay R1 and also act as Mobility Anchor Points (MAPs) and default routers for their associated Clients C1 and C2. AERO Clients C1 and C2 associate with Servers S1 and S2, respectively. They receive Mobile Network Prefix (MNP) delegations X1 and X2, and also act as default routers for their associated physical or internal virtual EUNs. Simple hosts H1 and H2 attach to the EUNs served by Clients C1 and C2, respectively. AERO Proxy P1 configures a secured tunnel with Relay R1 and provides proxy services for AERO Clients in secured enclaves that cannot associate directly with other AERO link neighbors. Each node on the AERO link maintains an AERO interface neighbor cache and an IP forwarding table the same as for any link. Although the figure shows a limited deployment, in common operational practice there will normally be many additional Relays, Servers, Clients and Proxys.

AERO Relays provide hybrid routing/bridging services (as well as a security trust anchor) for nodes on an AERO link. Relays use standard IPv6 routing to forward packets both within the same INET partitions and between disjoint INET partitions based on a mid-layer IPv6 encapsulation known as the SPAN header. The inner IP layer experiences a virtual bridging service since the inner IP TTL/Hop Limit is not decremented during forwarding. Each Relay also peers with Servers and other Relays in a dynamic routing protocol instance to provide a Distributed Mobility Management (DMM) service for the list of active MNPs (see ). Relays present the AERO link as a set of one or more Mobility Service Prefixes (MSPs) but as link-layer devices need not connect directly to the AERO link themselves unless an administrative interface is desired. Relays configure secured tunnels with Servers, Proxys and other Relays; they further maintain IP forwarding table entries for each Mobile Network Prefix (MNP) and any other reachable non-MNP prefixes. AERO Servers provide default forwarding services and a Mobility Anchor Point (MAP) for AERO Client Mobile Nodes (MNs). Each Server also peers with Relays in a dynamic routing protocol instance to advertise its list of associated MNPs (see ). Servers facilitate PD exchanges with Clients, where each delegated prefix becomes an MNP taken from an MSP. Servers forward packets between AERO interface neighbors and track each Client's mobility profiles. AERO Clients register their MNPs through PD exchanges with AERO Servers over the AERO link, and distribute the MNPs to nodes on EUNs. A Client may also be co-resident on the same physical or virtual platform as a Server; in that case, the Client and Server behave as a single functional unit. AERO Proxys provide a conduit for ANET AERO Clients to associate with AERO Servers in external INETs. Client and Servers exchange control plane messages via the Proxy acting as a bridge between the ANET/INET boundary. The Proxy forwards data packets between Clients and the AERO link according to forwarding information in the neighbor cache. The Proxy function is specified in . AERO Gateways are Servers that provide forwarding services between the AERO interface and INET/EUN interfaces. Gateways are provisioned with MNPs the same as for an AERO Client, and also run a dynamic routing protocol to discover any non-MNP IP routes. The Gateway advertises the MSP(s) to INETs, and distributes all of its associated MNPs and non-MNP IP routes via BGP peerings with Relays. AERO Relays, Servers, Proxys and Gateways are critical infrastructure elements in fixed (i.e., non-mobile) INET deployments and hence have permanent and unchanging INET addresses. AERO Clients are MNs that connect via ANET interfaces, i.e., their ANET addresses may change when the Client moves to a new ANET connection.

The AERO routing system comprises a private instance of the Border Gateway Protocol (BGP) that is coordinated between Relays and Servers and does not interact with either the public Internet BGP routing system or any underlying INET routing systems. In a reference deployment, each Server is configured as an Autonomous System Border Router (ASBR) for a stub Autonomous System (AS) using an AS Number (ASN) that is unique within the BGP instance, and each Server further uses eBGP to peer with one or more Relays but does not peer with other Servers. Each INET of a multi-segment AERO link must include one or more Relays, which peer with the Servers and Proxys within that INET. All Relays within the same INET are members of the same hub AS using a common ASN, and use iBGP to maintain a consistent view of all active MNPs currently in service. The Relays of different INETs peer with one another using eBGP. Relays advertise the AERO link's MSPs and any non-MNP routes to each of their Servers. This means that any aggregated non-MNPs (including "default") are advertised to all Servers. Each Relay configures a black-hole route for each of its MSPs. By black-holing the MSPs, the Relay will maintain forwarding table entries only for the MNPs that are currently active, and packets destined to all other MNPs will correctly incur Destination Unreachable messages due to the black-hole route. In this way, Servers have only partial topology knowledge (i.e., they know only about the MNPs of their directly associated Clients) and they forward all other packets to Relays which have full topology knowledge. Servers maintain a working set of associated MNPs, and dynamically announce new MNPs and withdraw departed MNPs in eBGP updates to Relays. Servers that are configured as Gateways also redistribute non-MNP routes learned from non-AERO interfaces via their eBGP Relay peerings. Clients are expected to remain associated with their current Servers for extended timeframes, however Servers SHOULD selectively suppress updates for impatient Clients that repeatedly associate and disassociate with them in order to dampen routing churn. Servers that are configured as Gateways advertise the MSPs via INET/EUN interfaces, and forward packets between INET/EUN interfaces and the AERO interface using standard IP forwarding. Scaling properties of the AERO routing system are limited by the number of BGP routes that can be carried by Relays. As of 2015, the global public Internet BGP routing system manages more than 500K routes with linear growth and no signs of router resource exhaustion . More recent network emulation studies have also shown that a single Relay can accommodate at least 1M dynamically changing BGP routes even on a lightweight virtual machine, i.e., and without requiring high-end dedicated router hardware. Therefore, assuming each Relay can carry 1M or more routes, this means that at least 1M Clients can be serviced by a single set of Relays. A means of increasing scaling would be to assign a different set of Relays for each set of MSPs. In that case, each Server still peers with one or more Relays, but institutes route filters so that BGP updates are only sent to the specific set of Relays that aggregate the MSP. For example, if the MSP for the AERO link is 2001:db8::/32, a first set of Relays could service the MSP 2001:db8::/40, a second set of Relays could service 2001:db8:0100::/40, a third set could service 2001:db8:0200::/40, etc. Assuming up to 1K sets of Relays, the AERO routing system can then accommodate 1B or more MNPs with no additional overhead (for example, it should be possible to service 1B /64 MNPs taken from a /34 MSP and even more for shorter prefixes). In this way, each set of Relays services a specific set of MSPs that they advertise to the native Internetwork routing system, and each Server configures MSP-specific routes that list the correct set of Relays as next hops. This arrangement also allows for natural incremental deployment, and can support small scale initial deployments followed by dynamic deployment of additional Clients, Servers and Relays without disturbing the already-deployed base. Server and Relays can use the Bidirectional Forwarding Detection (BFD) protocol to quickly detect link failures that don’t result in interface state changes, BGP peer failures, and administrative state changes. BFD is important in environments where rapid response to failures is required for routing reconvergence and, hence, communications continuity. A full discussion of the BGP-based routing system used by AERO is found in . The system provides for Distributed Mobility Management (DMM) per the distributed mobility anchoring architecture .

For IPv6 MNPs, the AERO routing system includes ordinary IPv6 routes. For IPv4 MNPs, the AERO routing system includes IPv6 routes based on an IPv4-embedded IPv6 address format discussed in .

A Client's AERO address is an IPv6 link-local address with an interface identifier based on the Client's delegated MNP. Relay, Server and Proxy AERO addresses are assigned from the range fe80::/96 and include an administratively-provisioned value in the lower 32 bits. For IPv6, Client AERO addresses begin with the prefix fe80::/64 and include in the interface identifier (i.e., the lower 64 bits) a 64-bit prefix taken from one of the Client's IPv6 MNPs. For example, if the AERO Client receives the IPv6 MNP: 2001:db8:1000:2000::/56 it constructs its corresponding AERO addresses as: fe80::2001:db8:1000:2000 fe80::2001:db8:1000:2001 fe80::2001:db8:1000:2002 ... etc. ... fe80::2001:db8:1000:20ff For IPv4, Client AERO addresses are based on an IPv4-mapped IPv6 address formed from an IPv4 MNP and with a prefix length of 96 plus the MNP prefix length. For example, for the IPv4 MNP 192.0.2.32/28 the IPv4-mapped IPv6 MNP is: 0:0:0:0:0:FFFF:192.0.2.16/124 (also written as 0:0:0:0:0:FFFF:c000:0210/124) The Client then constructs its AERO addresses with the prefix fe80::/64 and with the lower 64 bits of the IPv4-mapped IPv6 address in the interface identifier as: fe80::FFFF:192.0.2.16 fe80::FFFF:192.0.2.17 fe80::FFFF:192.0.2.18 ... etc. ... fe80:FFFF:192.0.2.31 Relay, Server and Proxy AERO addresses are allocated from the range fe80::/96, and MUST be managed for uniqueness. The lower 32 bits of the AERO address includes a unique integer value between 1 and 0xfffffffe (e.g., fe80::1, fe80::2, fe80::3, etc., fe80::ffff:fffe) as assigned by the administrative authority for the link. If the link spans multiple SPAN segments, the AERO addresses are assigned to each segment in 1x1 correspondence with SPAN addresses (see: ). The address fe80:: is the IPv6 link-local Subnet-Router anycast address, and the address fe80::ffff:ffff is reserved as the unspecified AERO address. The lowest-numbered AERO address from a Client's MNP delegation serves as the "base" AERO address (for example, for the MNP 2001:db8:1000:2000::/56 the base AERO address is fe80::2001:db8:1000:2000). The Client then assigns the base AERO address to the AERO interface and uses it for the purpose of maintaining the neighbor cache entry. The Server likewise uses the AERO address as its index into the neighbor cache for this Client. If the Client has multiple AERO addresses (i.e., when there are multiple MNPs and/or MNPs with prefix lengths shorter than /64), the Client originates ND messages using the base AERO address as the source address and accepts and responds to ND messages destined to any of its AERO addresses as equivalent to the base AERO address. In this way, the Client maintains a single neighbor cache entry that may be indexed by multiple AERO addresses. The Client's Subnet-Router anycast address can be statelessly determined from its AERO address by simply transposing the AERO address into the upper N bits of the Anycast address followed by 128-N bits of zeroes. For example, for the AERO address fe80::2001:db8:1:2 the Subnet-Router anycast address is 2001:db8:1:2::. AERO addresses for mobile node Clients embed a MNP as discussed above, while AERO addresses for non-MNP destinations are constructed in exactly the same way. A Client AERO address therefore encodes either an MNP if the prefix is reached via the SPAN or a non-MNP if the prefix is reached via a Gateway.

An AERO link configured over a single INET appears as a single unified link with a consistent underlying network addressing plan. In that case, all nodes on the link can exchange packets via simple INET encapsulation, since the underlying INET is connected. In common practice, however, an AERO link may be partitioned into multiple "segments", where each segment is a distinct INET potentially managed under a different administrative authority (e.g., as for worldwide aviation service providers such as ARINC, SITA, Inmarsat, etc.). Individual INETs may also themselves be partitioned internally, in which case each internal partition is seen as a separate segment. The addressing plan of each segment is consistent internally but will often bear no relation to the addressing plans of other segments. Each segment is also likely to be separated from others by network security devices (e.g., firewalls, proxies, packet filtering gateways, etc.), and in many cases disjoint segments may not even have any common physical link connections at all. Therefore, nodes can only be assured of exchanging packets directly with correspondents in the same segment, and not with those in other segments. The only means for joining the segments therefore is through inter-domain peerings between AERO Relays. The same as for traditional campus LANs, multiple AERO link segments can be joined into a single unified link via a virtual bridging service termed the "SPAN". The SPAN performs link-layer packet forwarding between segments (i.e., bridging) without decrementing the network-layer TTL/Hop Limit. The SPAN model is depicted in :

. . . . . . . . . . . . . . .. . . . . . . . . ]]> To support the SPAN, AERO links require a reserved /64 IPv6 "SPAN Service Prefix (SSP)". Although any routable IPv6 prefix can be used, a Unique Local Address (ULA) prefix (e.g., fd00::/64) is recommended since border routers are commonly configured to prevent packets with ULAs from being injected into the AERO link by an external IPv6 node and from leaking out of the AERO link to the outside world. Each segment in the SPAN assigns a unique sub-prefix of SSP::/96 termed a "SPAN Partition Prefix (SPP)". For example, a first segment could assign fd00::1000/116, a second could assign fd00::2000/116, a third could assign fd00::3000/116, etc. The administrative authorities for each segment must therefore coordinate to assure mutually-exclusive SPP assignments, but internal provisioning of the SPP is an independent local consideration for each administrative authority. A "SPAN address" is an address taken from a SPP and assigned to a Relay, Server, Gateway or Proxy interface. SPAN addresses are formed by simply replacing the upper portion of an administratively-assigned AERO address with the SPP. For example, if the SPP is fd00::1000/116, the SPAN address formed from the AERO address fe80::1001 is simply fd00::1001. An "INET address" is an address of a node's interface connection to an INET. AERO/SPAN/INET address mappings are maintained as permanent neighbor cache entires as discussed in . AERO Relays serve as bridges to join multiple segments into a unified AERO link over multiple diverse administrative domains. They support the bridging function by first establishing forwarding table entries for their SPPs either via standard BGP routing or static routes. For example, if three Relays ('A', 'B' and 'C') from different segments serviced the SPPs fd00::1000/116, fd00::2000/116 and fd00::3000/116 respectively, then the forwarding tables in each Relay are as follows: fd00::1000/116->local, fd00::2000/116->B, fd00::3000/116->C fd00::1000/116->A, fd00::2000/116->local, fd00::3000/116->C fd00::1000/116->A, fd00::2000/116->B, fd00::3000/116->local These forwarding table entries are permanent and never change, since they correspond to fixed infrastructure elements in their respective segments. This provides the basis for a link-layer forwarding service that cannot be disrupted by routing updates due to node mobility. With the SPPs in place in each Relay's forwarding table, control and data packets sent between AERO nodes in different segments can therefore be carried over the SPAN via encapsulation. For example, when a source AERO node in segment A forwards a packet with IPv6 address 2001:db8:1:2::1 to a target AERO node in segment C with IPv6 address 2001:db8:1000:2000::1, it first encapsulates the packet in a SPAN header with source SPAN address taken from fd00::1000/116 (e.g., fd00::1001) and destination SPAN address taken from fd00::3000/116 (e.g., fd00::3001). Next, it encapsulates the SPAN message in an INET header with source address set to its own INET address (e.g., 192.0.2.100) and destination set to the INET address of a Relay (e.g., 192.0.2.1). SPAN encapsulation is based on Generic Packet Tunneling in IPv6 ; the encapsulation format in the above example is shown in :

In this format, the inner IP header and packet body are the original IP packet, the SPAN header is an IPv6 header prepared according to , and the INET header is prepared according to . A packet is said to be "forwarded/sent into the SPAN" when it is encapsulated as described above then forwarded via a secured tunnel to a neighboring Relay. This gives rise to a routing system that contains both MNP routes that may change dynamically due to regional node mobility and SPAN routes that never change. The Relays can therefore provide link-layer bridging by sending packets into the SPAN instead of network-layer routing according to MNP routes. As a result, opportunities for packet loss due to node mobility between different segments are mitigated. With reference to , for a Client's AERO address the SPAN address is simply set to the Subnet-Router anycast address. For non-link-local addresses, the destination SPAN address may not be known in advance for the first few packets of a flow sent via the SPAN. In that case, the SPAN destination address is set to the original packet's destination, and the SPAN routing system will direct the packet to the correct SPAN egress node. (In the above example, the SPAN destination address is simply 2001:db8:1000:2000::1.)

For IPv4 MNPs, Servers inject a "SPAN Compatibility Prefix (SCP)" that embeds the MNP into the BGP routing system. The SCP begins with the upper 64 bits of the SSP, followed by the constant string "0000:FFFF" followed by the IPv4 MNP. For example, if the SSP is fd00::/64 and the MNP is 192.0.2.0/24 then the SCP is fd00::FFFF:192.0.2.0/120. This allows for encapsulation of IPv4 packets in IPv6 headers with "SPAN Compatibility Addresses (SCAs)". In this example, the SCA corresponding to the SCP is simply fd00::FFFF:192.0.2.0, and can be used as the SPAN destination address for packets forwarded via the SPAN. This allows for forwarding of initial IPv4 packets over IPv6 SPAN routes, followed by route optimization for direct communications.

AERO interfaces are virtual interfaces configured over one or more underlying interfaces classified as follows: Native interfaces have global IP addresses that are reachable from any INET correspondent. All Server, Gateway and Relay interfaces are native interfaces, as are INET-facing interfaces of Proxys. NATed interfaces connect to a private network behind a Network Address Translator (NAT). The NAT does not participate in any AERO control message signaling, but the Server can issue control messages on behalf of the Client. Clients that are behind a NAT are required to send periodic keepalive messages to keep NAT state alive when there are no data packets flowing. If no other periodic messaging service is available, the Client can send RS messages to receive RA replies from its Server(s). VPNed interfaces use security encapsulation to a Virtual Private Network (VPN) server that also acts as an AERO Server. As with NATed links, the Server can issue control messages on behalf of the Client, but the Client need not send periodic keepalives in addition to those already used to maintain the VPN connection. Proxyed interfaces connect to an ANET that is separated from the open INET by an AERO Proxy. Unlike NATed and VPNed interfaces, the Proxy can actively issue control messages on behalf of the Client. Direct interfaces connect a Client directly to a neighbor without crossing any ANET/INET paths. An example is a line-of-sight link between a remote pilot and an unmanned aircraft. AERO interfaces use encapsulation (see: ) to exchange packets with AERO link neighbors over Native, NATed or VPNed interfaces. AERO interfaces do not use encapsulation over Proxyed and Direct underlying interfaces. AERO interfaces maintain a neighbor cache for tracking per-neighbor state the same as for any interface. AERO interfaces use ND messages including Router Solicitation (RS), Router Advertisement (RA), Neighbor Solicitation (NS) and Neighbor Advertisement (NA) for neighbor cache management. AERO interfaces send ND messages with an Overlay Multilink Network Interface (OMNI) option formatted as specified in . The OMNI option includes prefix registration information and "ifIndex-tuples" containing link quality information for the AERO interface's underlying interfaces. When encapsulation is used, AERO interface ND messages MAY also include an AERO Source/Target Link-Layer Address Option (S/TLLAO) formatted as shown in :

In this format, Type and Length are set the same as specified for S/TLLAOs in , with trailing zero padding octets added as necessary to produce an integral number of 8 octet blocks. The S/TLLAO includes N ifIndex-tuples in correspondence to ifIndex-tuples that appear in the OMNI option. Each ifIndex-tuple includes the folllowing information: ifIndex[i] - the same value as in the corresponding ifIndex-tuple included in the OMNI option. V[i] - a bit that identifies the IP protocol version of the address found in the Link Layer Address [i] field. The bit is set to 0 for IPv4 or 1 for IPv6. Reserved[i] - MUST encode the value 0 on transmission, and ignored on reception. Link Layer Address [i] - the IPv4 or IPv6 address used as the encapsulation source address. The field is 4 bytes in length for IPv4 or 16 bytes in length for IPv6. Port Number [i] - the upper layer protocol port number used as the encapsulation source port, or 0 when no upper layer protocol encapsulation is used. The field is 2 bytes in length. If an S/TLLAO is included, the first S/TLLAO ifIndex-tuple MUST correspond to the first OMNI option ifIndex-tuple, and any additional S/TLLAO ifIndex-tuples MUST correspond to a proper subset of the remaining OMNI option ifIndex-tuples. Any S/TLLAO ifIndex-tuple having an ifIndex value that does not appear in an OMNI option ifindex-tuple is ignored. If the same ifIndex value appears in multiple ifIndex-tuples, the first tuple is processed and the remaining tuples are ignored. Any S/TLLAO ifIndex-tuples can therefore be viewed as inter-dependent extensions of their corresponidng OMNI option ifIndex-tuples, i.e., the OMNI option and S/TLLAO are companion options that are interpreted in conjunction with each other. A Client's AERO interface may be configured over multiple underlying interface connections. For example, common mobile handheld devices have both wireless local area network ("WLAN") and cellular wireless links. These links are typically used "one at a time" with low-cost WLAN preferred and highly-available cellular wireless as a standby. In a more complex example, aircraft frequently have many wireless data link types (e.g. satellite-based, cellular, terrestrial, air-to-air directional, etc.) with diverse performance and cost properties. If a Client's multiple underlying interfaces are used "one at a time" (i.e., all other interfaces are in standby mode while one interface is active), then ND message OMNI options include only a single ifIndex-tuple and set to a constant value. In that case, the Client would appear to have a single interface but with a dynamically changing link-layer address. If the Client has multiple active underlying interfaces, then from the perspective of ND it would appear to have multiple link-layer addresses. In that case, ND message OMNI options MAY include multiple ifIndex-tuples - each with a value that corresponds to a specific interface. The OMNI option MUST include a first ifIndex-tuple that corresponds to the interface over which the ND message is sent. Every ND message need not include all OMNI and/or S/TLLAO ifIndex-tuples; for any ifIndex-tuple not included, the neighbor considers the status as unchanged. Relay, Server and Proxy AERO interfaces may be configured over one or more secured tunnel interfaces. The AERO interface configures both an AERO address and its corresponding SPAN address, while the underlying secured tunnel interfaces are either unnumbered or configure the same SPAN address. The AERO interface encapsulates each IP packet in a SPAN header and presents the packet to the underlying secured tunnel interface. For Relays that do not configure an AERO interface, the secured tunnel interfaces themselves are exposed to the IP layer with each interface configuring the Relay's SPAN address. Routing protocols such as BGP therefore run directly over the Relay's secured tunnel interfaces. For nodes that configure an AERO interface, routing protocols such as BGP run over the AERO interface but do not employ SPAN encapsulation. Instead, the AERO interface presents the routing protocol messages directly to the underlying secured tunnels without applying encapsulation and while using the SPAN address as the source address. This distinction must be honored consistently according to each node's configuration so that the IP forwarding table will associate discovered IP routes with the correct interface.

AERO Servers, Proxys and Clients configure AERO interfaces as their point of attachment to the AERO link. AERO nodes assign the MSPs for the link to their AERO interfaces (i.e., as a "route-to-interface") to ensure that packets with destination addresses covered by an MNP not explicitly assigned to a non-AERO interface are directed to the AERO interface. AERO interface initialization procedures for Servers, Proxys, Clients and Relays are discussed in the following sections.

When a Server enables an AERO interface, it assigns AERO/SPAN addresses and configures permanent neighbor cache entries for neighbors in the same SPAN segment by consulting the ROS list for the segment. The Server also configures secured tunnels with one or more neighboring Relays and engages in a BGP routing protocol session with each Relay. The AERO interface provides a single interface abstraction to the IP layer, but internally comprises multiple secured tunnels as well as an NBMA nexus for sending encapsulated data packets to AERO interface neighbors. The Server further configures a service to facilitate ND/PD exchanges with AERO Clients and manages per-Client neighbor cache entries and IP forwarding table entries based on control message exchanges. Gateways are simply Servers that run a dynamic routing protocol between the AERO interface and INET/EUN interfaces (see: ). The Gateway provisions MNPs to networks on the INET/EUN interfaces (i.e., the same as a Client would do) and advertises the MSP(s) for the AERO link over the INET/EUN interfaces. The Gateway further provides an attachment point of the AERO link to the non-MNP-based global topology.

When a Proxy enables an AERO interface, it assigns AERO/SPAN addresses and configures permanent neighbor cache entries the same as for Servers. The Proxy also configures secured tunnels with one or more neighboring Relays and maintains per-Client neighbor cache entries based on control message exchanges.

When a Client enables an AERO interface, it sends an RS message with ND/PD parameters over an ANET interface to a Server in the MAP list, which returns an RA message with corresponding parameters. (The RS/RA messages may pass through a Proxy in the case of a Client's Proxyed interface.) After the initial ND/PD message exchange, the Client assigns AERO addresses to the AERO interface based on the delegated prefix(es). The Client can then register additional ANET interfaces with the Server by sending an RS message over each ANET interface.

AERO Relays need not connect directly to the AERO link, since they operate as link-layer forwarding devices instead of network layer routers. Configuration of AERO interfaces on Relays is therefore OPTIONAL, e.g., if an administrative interface is needed. Relays configure secured tunnels with Servers, Proxys and other Relays; they also configure AERO/SPAN addresses and permanent neighbor cache entries the same as Servers. Relays engage in a BGP routing protocol session with a subset of the Servers on the local SPAN segment, and with other Relays on the SPAN (see: ).

Each AERO interface maintains a conceptual neighbor cache that includes an entry for each neighbor it communicates with on the AERO link per . AERO interface neighbor cache entries are said to be one of "permanent", "symmetric", "asymmetric" or "proxy". Permanent neighbor cache entries are created through explicit administrative action; they have no timeout values and remain in place until explicitly deleted. AERO Servers and Proxys maintain permanent neighbor cache entries for all other Servers and Proxys within the same SPAN segment. Each entry maintains the mapping between the neighbor's network-layer AERO address and corresponding INET address. The list of all permanent neighbor cache entries for the SPAN segment is maintained in the segment's ROS list. Symmetric neighbor cache entries are created and maintained through RS/RA exchanges as specified in , and remain in place for durations bounded by ND/PD lifetimes. AERO Servers maintain symmetric neighbor cache entries for each of their associated Clients, and AERO Clients maintain symmetric neighbor cache entries for each of their associated Servers. The list of all Servers on the AERO link is maintained in the link's MAP list. Asymmetric neighbor cache entries are created or updated based on route optimization messaging as specified in , and are garbage-collected when keepalive timers expire. AERO route optimization sources (ROSs) maintain asymmetric neighbor cache entries for active targets with lifetimes based on ND messaging constants. Asymmetric neighbor cache entries are unidirectional since only the ROS and not the target (e.g., a Client's MAP) creates an entry. Proxy neighbor cache entries are created and maintained by AERO Proxys when they process Client/Server ND/PD exchanges, and remain in place for durations bounded by ND/PD lifetimes. AERO Proxys maintain proxy neighbor cache entries for each of their associated Clients. Proxy neighbor cache entries track the Client state and the address of the Client's associated Server. To the list of neighbor cache entry states in Section 7.3.2 of , Proxy and Server AERO interfaces add an additional state DEPARTED that applies to symmetric and proxy neighbor cache entries for Clients that have recently departed. The interface sets a "DepartTime" variable for the neighbor cache entry to "DEPARTTIME" seconds. DepartTime is decremented unless a new ND message causes the state to return to REACHABLE. While a neighbor cache entry is in the DEPARTED state, packets destined to the target Client are forwarded to the Client's new location instead of being dropped. When DepartTime decrements to 0, the neighbor cache entry is deleted. It is RECOMMENDED that DEPARTTIME be set to the default constant value REACHABLETIME plus 10 seconds (40 seconds by default) to allow a window for packets in flight to be delivered while stale route optimization state may be present. When a target Server (acting as a MAP) receives a valid NS message used for route optimization, it searches for a symmetric neighbor cache entry for the target Client. The MAP then returns a solicited NA message without creating a neighbor cache entry for the ROS, but creates or updates a target Client "Report List" entry for the ROS and sets a "ReportTime" variable for the entry to REPORTTIME seconds. The MAP resets ReportTime when it receives a new authentic NS message, and otherwise decrements ReportTime while no NS messages have been received. It is RECOMMENDED that REPORTTIME be set to the default constant value REACHABLETIME plus 10 seconds (40 seconds by default) to allow a window for route optimization to converge before ReportTime decrements below REACHABLETIME. When the ROS receives a solicited NA message response to its NS message, it creates or updates an asymmetric neighbor cache entry for the target network-layer and link-layer addresses. The ROS then (re)sets ReachableTime for the neighbor cache entry to REACHABLETIME seconds and uses this value to determine whether packets can be forwarded directly to the target, i.e., instead of via a default route. The ROS otherwise decrements ReachableTime while no further solicited NA messages arrive. It is RECOMMENDED that REACHABLETIME be set to the default constant value 30 seconds as specified in . The ROS also uses the value MAX_UNICAST_SOLICIT to limit the number of NS keepalives sent when a correspondent may have gone unreachable, the value MAX_RTR_SOLICITATIONS to limit the number of RS messages sent without receiving an RA and the value MAX_NEIGHBOR_ADVERTISEMENT to limit the number of unsolicited NAs that can be sent based on a single event. It is RECOMMENDED that MAX_UNICAST_SOLICIT, MAX_RTR_SOLICITATIONS and MAX_NEIGHBOR_ADVERTISEMENT be set to 3 the same as specified in . Different values for DEPARTTIME, REPORTTIME, REACHABLETIME, MAX_UNICAST_SOLICIT, MAX_RTR_SOLCITATIONS and MAX_NEIGHBOR_ADVERTISEMENT MAY be administratively set; however, if different values are chosen, all nodes on the link MUST consistently configure the same values. Most importantly, DEPARTTIME and REPORTTIME SHOULD be set to a value that is sufficiently longer than REACHABLETIME to avoid packet loss due to stale route optimization state.

Client AERO interfaces avoid encapsulation over Direct underlying interfaces and Proxyed underlying interfaces for which the first-hop access router is AERO-aware. Other AERO interfaces encapsulate packets according to whether they are entering the AERO interface from the network layer or if they are being re-admitted into the same AERO link they arrived on. This latter form of encapsulation is known as "re-encapsulation". For packets entering the AERO interface from the network layer, the AERO interface copies the "TTL/Hop Limit", "Type of Service/Traffic Class" , "Flow Label" (for IPv6) and "Congestion Experienced" values in the packet's IP header into the corresponding fields in the encapsulation header(s). For packets undergoing re-encapsulation, the AERO interface instead copies these values from the original encapsulation header into the new encapsulation header, i.e., the values are transferred between encapsulation headers and *not* copied from the encapsulated packet's network-layer header. (Note especially that by copying the TTL/Hop Limit between encapsulation headers the value will eventually decrement to 0 if there is a (temporary) routing loop.) For IPv4 encapsulation/re-encapsulation, the AERO interface sets the DF bit as discussed in . AERO interfaces configured over INET underlying interfaces encapsulate each packet in a SPAN header, then encapsulate the resulting SPAN packet in an INET header according to the next hop determined in the forwarding algorithm in . If the next hop is reached via a secured tunnel, the AERO interface uses an INET encapsulation format specific to the secured tunnel type (see: ). If the next hop is reached via an unsecured underlying interface, the AERO interface instead uses Generic UDP Encapsulation (GUE) or an alternate minimal encapsulation format . When GUE encapsulation is used, the AERO interface next sets the UDP source port to a constant value that it will use in each successive packet it sends, and sets the UDP length field to the length of the SPAN packet plus 8 bytes for the UDP header itself plus the length of the GUE header (or 0 if GUE direct IP encapsulation is used). For packets sent to a Server or Relay, the AERO interface sets the UDP destination port to 8060, i.e., the IANA-registered port number for AERO. For packets sent to a Client, the AERO interface sets the UDP destination port to the port value stored in the neighbor cache entry for this Client. The AERO interface then either includes or omits the UDP checksum according to the GUE specification. AERO interfaces observes the packet sizing and fragmentation considerations found in .

AERO interfaces decapsulate packets destined either to the AERO node itself or to a destination reached via an interface other than the AERO interface the packet was received on. When the encapsulated packet arrives in multiple fragments, the AERO interface reassembles as discussed in . Further decapsulation steps are performed according to the appropriate encapsulation format specification.

AERO nodes employ simple data origin authentication procedures. In particular: AERO Relays, Servers and Proxys accept encapsulated data packets and control messages received from secured tunnels. AERO Servers and Proxys accept encapsulated data packets and NS messages used for Neighbor Unreachability Detection (NUD) received from a source found in the ROS list. AERO Proxys and Clients accept packets that originate from within the same secured ANET. AERO Clients and Gateways accept packets from downstream network correspondents based on ingress filtering. AERO nodes silently drop any packets that do not satisfy the above data origin authentication procedures. Further security considerations are discussed .

IP packets enter a node's AERO interface either from the network layer (i.e., from a local application or the IP forwarding system) or from the link layer (i.e., from an AERO interface neighbor). All packets entering a node's AERO interface first undergo data origin authentication as discussed in . Packets that satisfy data origin authentication are processed further, while all others are dropped silently. Packets that enter the AERO interface from the network layer are forwarded to an AERO interface neighbor. Packets that enter the AERO interface from the link layer are either re-admitted into the AERO link or forwarded to the network layer where they are subject to either local delivery or IP forwarding. In all cases, the AERO interface itself MUST NOT decrement the network layer TTL/Hop-count since its forwarding actions occur below the network layer. AERO interfaces may have multiple underlying interfaces and/or neighbor cache entries for neighbors with multiple ifIndex-tuple registrations (see ). The AERO interface uses each packet's DSCP value (and/or other traffic discriminators such as port number) to select an outgoing underlying interface based on the node's own QoS preferences, and also to select a destination link-layer address based on the neighbor's underlying interface with the highest preference. AERO implementations SHOULD allow for QoS preference values to be modified at runtime through network management. If multiple outgoing interfaces and/or neighbor interfaces have a preference of "high", the AERO node replicates the packet and sends one copy via each of the (outgoing / neighbor) interface pairs; otherwise, the node sends a single copy of the packet via an interface with the highest preference. AERO nodes keep track of which underlying interfaces are currently "reachable" or "unreachable", and only use "reachable" interfaces for forwarding purposes. The following sections discuss the AERO interface forwarding algorithms for Clients, Proxys, Servers and Relays. In the following discussion, a packet's destination address is said to "match" if it is the same as a cached address, or if it is covered by a cached prefix (which may be encoded in an AERO address).

When an IP packet enters a Client's AERO interface from the network layer the Client searches for an asymmetric neighbor cache entry that matches the destination. If there is a match, the Client uses one or more "reachable" neighbor interfaces in the entry for packet forwarding. If there is no asymmetric neighbor cache entry, the Client instead forwards the packet toward a Server (the packet is intercepted by a Proxy if there is a Proxy on the path). When an IP packet enters a Client's AERO interface from the link-layer, if the destination matches one of the Client's MNPs or link-local addresses the Client decapsulates the packet (if necessary) and delivers it to the network layer. Otherwise, the Client drops the packet and MAY return a network-layer ICMP Destination Unreachable message subject to rate limiting (see: ).

For control messages originating from or destined to a Client, the Proxy intercepts the message and updates its proxy neighbor cache entry for the Client. The Proxy then forwards a (proxyed) copy of the control message. (For example, the Proxy forwards a proxied version of a Client's NS/RS message to the target neighbor, and forwards a proxied version of the NA/RA reply to the Client.) When the Proxy receives a data packet from a Client within the ANET, the Proxy searches for an asymmetric neighbor cache entry that matches the destination and forwards the packet as follows: if the destination matches an asymmetric neighbor cache entry, the Proxy uses one or more "reachable" neighbor interfaces in the entry for packet forwarding via encapsulation. If the neighbor interface is in the same SPAN segment, the Proxy forwards the packet directly to the neighbor; otherwise, it forwards the packet to a Relay. else, the Proxy encapsulates and forwards the packet to a Relay while using the packet's destination address as the SPAN destination address. (If the destination is an AERO address, the Proxy instead uses the corresponding Subnet-Router anycast address for Client AERO addresses and the SPAN address for administratively-provisioned AERO addresses.). When the Proxy receives an encapsulated data packet from an INET neighbor or from a secured tunnel, it accepts the packet only if data origin authentication succeeds and the SPAN destination address is its own address. If the packet is a SPAN fragment, the Proxy then adds the fragment to the reassembly buffer and returns if the reassembly is still incomplete. Otherwise, the Proxy reassembles the packet (if necessary) and continues processing. Next, the Proxy searches for a proxy neighbor cache entry that matches the destination. If there is a proxy neighbor cache entry in the REACHABLE state, the Proxy decapsulates and forwards the packet to the Client. If the neighbor cache entry is in the DEPARTED state, the Proxy instead re-encapsulates the packet and forwards it to a Relay. If there is no neighbor cache entry, the Proxy instead discards the packet.

For control messages destined to a target Client's AERO address that are received from a secured tunnel, the Server (acting as a MAP) intercepts the message and sends an appropriate response on behalf of the Client. (For example, the Server sends an NA message reply in response to an NS message directed to one of its associated Clients.) If the Client's neighbor cache entry is in the DEPARTED state, however, the Server instead forwards the packet to the Client's new Server as discussed in . When the Server receives an encapsulated data packet from an INET neighbor or from a secured tunnel, it accepts the packet only if data origin authentication succeeds. If the SPAN destination address is its own address, the Server reassembles if necessary and discards the SPAN header (if the reassembly is incomplete, the Server instead adds the fragment to the reassembly buffer and returns). The Server then continues processing as follows: if the destination matches a symmetric neighbor cache entry in the REACHABLE state the Server prepares the packet for forwarding to the destination Client. If the current header is a SPAN header, the Server reassembles if necessary and discards the SPAN header. The Server then forwards the packet according to the cached link-layer information, while using SPAN encapsulation for the Client's Proxyed/Native interfaces, simple INET encapsulation for NATed/VPNed interfaces, or no encapsulation for Direct interfaces. else, if the destination matches a symmetric neighbor cache entry in the DEPARETED state the Server re-encapsulates the packet and forwards it using the SPAN address of the Client's new Server as the destination. else, if the destination matches an asymmetric neighbor cache entry, the Server uses one or more "reachable" neighbor interfaces in the entry for packet forwarding via the local INET if the neighbor is in the same SPAN segment or via a Relay otherwise. else, if the destination is an AERO address that is not assigned on the AERO interface the Server drops the packet. else, the Server (acting as a Gateway) releases the packet to the network layer for local delivery or IP forwarding. Based on the information in the forwarding table, the network layer may return the packet to the same AERO interface in which case further processing occurs as below. (Note that this arrangement accommodates common implementations in which the IP forwarding table is not accessible from within the AERO interface. If the AERO interface can directly access the IP forwarding table, the forwarding table lookup can instead be performed internally from within the AERO interface itself.) When the Server's AERO interface receives a data packet from the network layer or from a NATed/VPNed/Direct Client, it processes the packet according to the network-layer destination address as follows: if the destination matches a symmetric or asymmetric neighbor cache entry the Server processes the packet as above. else, the Server encapsulates the packet and forwards it to a Relay. For administratively-assigned AERO address destinations, the Server uses the SPAN address corresponding to the destination as the SPAN destination address. For Client AERO address destinations, the Server uses the Subnet-Router anycast address corresponding to the destination as the SPAN destination address. For all others, the Server uses the packet's destination IP address as the SPAN destination address.

Relays forward packets over secured tunnels the same as any IP router. When the Relay receives an encapsulated packet via a secured tunnel, it removes the INET header and searches for a forwarding table entry that matches the destination address in the next header. The Relay then processes the packet as follows: if the destination matches one of the Relay's own addresses, the Relay submits the packet for local delivery. else, if the destination matches a forwarding table entry the Relay forwards the packet via a secured tunnel to the next hop. If the destination matches an MSP without matching an MNP, however, the Relay instead drops the packet and returns an ICMP Destination Unreachable message subject to rate limiting (see: ). else, the Relay drops the packet and returns an ICMP Destination Unreachable as above. As for any IP router, the Relay decrements the TTL/Hop Limit when it forwards the packet. If the packet is encapsulated in a SPAN header, only the Hop Limit in the SPAN header is decremented, and not the TTL/Hop Limit in the inner packet header.

The AERO interface is the node's attachment to the AERO link. For AERO link neighbor underlying interface paths that do not require encapsulation, the AERO interface sends unencapsulated IP packets. For other paths, the AERO interface acts as a tunnel ingress when it sends packets to the neighbor and as a tunnel egress when it receives packets from the neighbor. AERO interfaces configure an MTU the same as for any IP interface, however the MTU does not reflect the physical size of any links in the path but rather determines the maximum size for reassembly. AERO interfaces observe the packet sizing considerations for tunnels discussed in and as specified below. The Internet Protocol expects that IP packets will either be delivered to the destination or a suitable Packet Too Big (PTB) message returned to support the process known as IP Path MTU Discovery (PMTUD) . However, PTB messages may be crafted for malicious purposes or lost in the network . This can be especially problematic for tunnels, where a condition known as a PMTUD "black hole" can result. For these reasons, AERO interfaces employ operational procedures that avoid interactions with PMTUD, including the use of fragmentation when necessary. AERO interfaces observe three different types of fragmentation. Source fragmentation occurs when the AERO interface (acting as a tunnel ingress) fragments the encapsulated packet into multiple fragments before admitting each fragment into the tunnel. Network fragmentation occurs when an encapsulated packet admitted into the tunnel by the ingress is fragmented by an IPv4 router on the path to the egress. Finally, link-layer fragmentation (aka link adaptation) occurs at a layer below IP and is coordinated between underlying data link endpoints. IPv6 specifies a minimum link Maximum Transmission Unit (MTU) of 1280 bytes . Although IPv4 specifies a smaller minimum link MTU of 68 bytes , AERO interfaces also observe the IPv6 minimum for IPv4 even if encapsulated packets may incur network fragmentation. IPv6 specifies a minimum Maximum Reassembly Unit (MRU) of 1500 bytes , while the minimum MRU for IPv4 is only 576 bytes (but, note that many standard IPv6 over IPv4 tunnel types already assume a larger MRU than the IPv4 minimum). AERO interfaces therefore configure an MTU that MUST NOT be smaller than 1280 bytes, MUST NOT be larger than the minimum MRU among all nodes on the AERO link minus the encapsulation overhead ("ENCAPS"), and SHOULD NOT be smaller than 1500 bytes. AERO interfaces also configure a Maximum Segment Unit (MSU) as the maximum-sized encapsulated packet that the ingress can inject into the tunnel without source fragmentation. The MSU value MUST NOT be larger than 1280 bytes if there is no operational assurance that a larger size can traverse the link along all paths. The network layer proceeds as follows when it forwards an IP packet to the AERO interface. For each IPv4 packet that is larger than the AERO interface MTU and with DF set to 0, the network layer uses IPv4 fragmentation to break the packet into a minimum number of non-overlapping fragments where the first fragment is no larger than the MTU and the remaining fragments are no larger than the first. For all other IP packets, if the packet is larger than the AERO interface MTU, the network layer drops the packet and returns a PTB message to the original source. Otherwise, the network layer admits each IP packet or fragment into the AERO interface. For each IP packet admitted into AERO interface, if the neighbor is reached via an underlying interface that does not require encapsulation the AERO interface proceeds according to the underlying interface MTU. If the packet is no larger than the underlying interface MTU, the AERO interface presents the packet to the underlying interface. Otherwise, for IPv4 packets with DF set to 0 the AERO interface uses IPv4 fragmentation to break the packet into fragments no larger than the underlying interface MTU. For other packets, the AERO interface either performs link adaptation or drops the packet and returns a PTB message to the original source. (If the original source corresponds to a local application, the PTB would appear to have originated from a router on the path when in fact it was locally generated from within the AERO interface.) In the same way, when a packet that has been admitted into the AERO link reaches a target neighbor but is too large to be delivered over the final-hop underlying interface, the target either performs link adaptation or drops the packet and returns a PTB. Link adaptation is preferred in both cases when possible to avoid packet loss. For underlying interfaces that require encapsulation, the AERO interface (acting as a tunnel ingress) instead encapsulates the packet and performs path MTU procedures according to the specific encapsulation format. For INET interfaces, the ingress encapsulates the packet in a SPAN header. If the SPAN packet is larger than the MSU, the ingress source fragments the SPAN packet into a minimum number of non-overlapping fragments where the first fragment is no larger than the MSU and the remaining fragments are no larger than the first. The ingress then encapsulates each SPAN packet/fragment in an INET header and admits them into the tunnel. For IPv4, the ingress sets the DF bit to 0 in the INET header in case any network fragmentation is necessary. The encapsulated packets will be delivered to the egress, which reassembles them into a whole packet if necessary. By fragmenting at the SPAN layer instead of lower layers, standard IPv6 fragmentation and reassembly ensures that IPv4 issues such as data corruption due to reassembly misassociations will not occur . The ingress sends each fragment with minimal delay (i.e., in a multi-fragment burst) so that individual fragments are unlikely to be diverted to different destinations due to routing fluctuations. Since the SPAN header and IPv6 fragment extension header reduces the room available for packet data, but the original source has no way to control its insertion, the ingress MUST include their lengths in ENCAPS even for packets in which the header is absent.

In light of the above considerations, AERO interfaces MUST configure an MTU of 9180 bytes (i.e., the same as specified in ). This means that the AERO interface MUST be capable of reassembling original IP packets up to 9180 bytes in length. When an IP packet is admitted into an AERO interface, the interface encapsulates the packet using SPAN encapsulation and fragments the encapsulated packet into fragments that are no larger than 1280 bytes. The fragments will be reassembled by the tunnel egress that services the final destination. The AERO interfaces of Clients behind Proxys MAY see underlying interfaces with MTUs smaller than 9180 (but no smaller than the IP minimum link MTU). If a Client's underlying interfaces configure a diversity of MTUs (e.g., 1280, 1500, 4KB, 8KB, etc.) then neighbors on the link would appear to have multiple MTUs. IPv6 Path MTU Discovery accounts for this possibility since MTU discovery must be performed even between nodes that appear to be connected to the same link. Applications that cannot tolerate loss in the network due to MTU restrictions should restrict themselves to sending packets no larger than the IP minimum link MTU, i.e., even if the current path MTU would appear to support a larger size. This is due to the fact that routing changes could cause the path to traverse links with smaller MTUs at any given point in time.

When an AERO node admits a packet into the AERO interface, it may receive link-layer or network-layer error indications. A link-layer error indication is an ICMP error message generated by a router in the INET on the path to the neighbor or by the neighbor itself. The message includes an IP header with the address of the node that generated the error as the source address and with the link-layer address of the AERO node as the destination address. The IP header is followed by an ICMP header that includes an error Type, Code and Checksum. Valid type values include "Destination Unreachable", "Time Exceeded" and "Parameter Problem" . (AERO interfaces ignore all link-layer IPv4 "Fragmentation Needed" and IPv6 "Packet Too Big" messages since they only emit packets that are guaranteed to be no larger than the IP minimum link MTU as discussed in .) The ICMP header is followed by the leading portion of the packet that generated the error, also known as the "packet-in-error". For ICMPv6, specifies that the packet-in-error includes: "As much of invoking packet as possible without the ICMPv6 packet exceeding the minimum IPv6 MTU" (i.e., no more than 1280 bytes). For ICMPv4, specifies that the packet-in-error includes: "Internet Header + 64 bits of Original Data Datagram", however Section 4.3.2.3 updates this specification by stating: "the ICMP datagram SHOULD contain as much of the original datagram as possible without the length of the ICMP datagram exceeding 576 bytes". The link-layer error message format is shown in (where, "L2" and "L3" refer to link-layer and network-layer, respectively):

The AERO node rules for processing these link-layer error messages are as follows: When an AERO node receives a link-layer Parameter Problem message, it processes the message the same as described as for ordinary ICMP errors in the normative references . When an AERO node receives persistent link-layer Time Exceeded messages, the IP ID field may be wrapping before earlier fragments awaiting reassembly have been processed. In that case, the node should begin including integrity checks and/or institute rate limits for subsequent packets. When an AERO node receives persistent link-layer Destination Unreachable messages in response to encapsulated packets that it sends to one of its asymmetric neighbor correspondents, the node should process the message as an indication that a path may be failing, and optionally initiate NUD over that path. If it receives Destination Unreachable messages over multiple paths, the node should allow future packets destined to the correspondent to flow through a default route and re-initiate route optimization. When an AERO Client receives persistent link-layer Destination Unreachable messages in response to encapsulated packets that it sends to one of its symmetric neighbor Servers, the Client should mark the path as unusable and use another path. If it receives Destination Unreachable messages on many or all paths, the Client should associate with a new Server and release its association with the old Server as specified in . When an AERO Server receives persistent link-layer Destination Unreachable messages in response to encapsulated packets that it sends to one of its symmetric neighbor Clients, the Server should mark the underlying path as unusable and use another underlying path. When an AERO Server or Proxy receives link-layer Destination Unreachable messages in response to an encapsulated packet that it sends to one of its permanent neighbors, it treats the messages as an indication that the path to the neighbor may be failing. However, the dynamic routing protocol should soon reconverge and correct the temporary outage. When an AERO Relay receives a packet for which the network-layer destination address is covered by an MSP, if there is no more-specific routing information for the destination the Relay drops the packet and returns a network-layer Destination Unreachable message subject to rate limiting. The Relay writes the network-layer source address of the original packet as the destination address and uses one of its non link-local addresses as the source address of the message. When an AERO node receives an encapsulated packet for which the reassembly buffer it too small, it drops the packet and returns a network-layer Packet Too Big (PTB) message. The node first writes the MRU value into the PTB message MTU field, writes the network-layer source address of the original packet as the destination address and writes one of its non link-local addresses as the source address.

AERO Router Discovery, Prefix Delegation and Autoconfiguration are coordinated as discussed in the following Sections.

Each AERO Server on the link configures a PD service to facilitate Client requests. Each Server is provisioned with a database of MNP-to-Client ID mappings for all Clients enrolled in the AERO service, as well as any information necessary to authenticate each Client. The Client database is maintained by a central administrative authority for the AERO link and securely distributed to all Servers, e.g., via the Lightweight Directory Access Protocol (LDAP) , via static configuration, etc. Clients receive the same service regardless of the Servers they select. AERO Clients and Servers use ND messages to maintain neighbor cache entries. AERO Servers configure their AERO interfaces as advertising interfaces, and therefore send unicast RA messages with a short Router Lifetime value (e.g., REACHABLETIME seconds) in response to a Client's RS message. Thereafter, Clients send additional RS messages to keep Server state alive. AERO Clients and Servers include PD parameters in RS/RA messages (see for ND/PD alternatives). The unified ND/PD messages are exchanged between Client and Server according to the prefix management schedule required by the PD service. If the Client knows its MNP in advance, it can instead employ prefix registration by including its AERO address as the source address of an RS message and with an OMNI option with valid prefix registration information for the MNP. If the Server (and Proxy) accept the Client's MNP assertion, they inject the prefix into the routing system and establish the necessary neighbor cache state. The following sections specify the Client and Server behavior.

AERO Clients discover the addresses of Servers in a similar manner as described in . Discovery methods include static configuration (e.g., from a flat-file map of Server addresses and locations), or through an automated means such as Domain Name System (DNS) name resolution . Alternatively, the Client can discover Server addresses through a layer 2 data link login exchange, or through a unicast RA response to a multicast/anycast RS as described below. In the absence of other information, the Client can resolve the DNS Fully-Qualified Domain Name (FQDN) "linkupnetworks.[domainname]" where "linkupnetworks" is a constant text string and "[domainname]" is a DNS suffix for the AERO link (e.g., "example.com"). To associate with a Server, the Client acts as a requesting router to request MNPs. The Client prepares an RS message with PD parameters and includes a Nonce and Timestamp option if the Client needs to correlate RA replies. If the Client already knows the Server's AERO address, it includes the AERO address as the network-layer destination address; otherwise, it includes the link-scoped All-Routers multicast (ff02::2) or Subnet-Router anycast (fe80::) address as the network-layer destination. If the Client already knows its own AERO address, it uses the AERO address as the network-layer source address; otherwise, it uses the unspecified AERO address (fe80::ffff:ffff) as the network-layer source address. The Client next includes an OMNI option in the RS message to register its link-layer information with the Server. The Client sets the OMNI option prefix registration information according to the MNP, and includes a first ifIndex-tuple corresponding to the underlying interface over which the Client will send the RS message. The Client MAY include additional ifIndex-tuples specific to other underlying interfaces. The Client MAY also include an SLLAO with a single link-layer address corresponding to the first OMNI option ifIndex-tuple. The Client sets a "primary" flag in the OMNI option if it wishes to enable proxy keepalives on this underlying interface. The Client then sends the RS message (either directly via Direct interfaces, via INET encapsulation for NATed interfaces, via a VPN for VPNed interfaces, via a Proxy for proxyed interfaces or via a Relay for native interfaces) and waits for an RA message reply (see ). The Client retries up to MAX_RTR_SOLICITATIONS times until an RA is received. If the Client receives no RAs, or if it receives an RA with Router Lifetime set to 0, the Client SHOULD abandon this Server and try another Server. Otherwise, the Client processes the PD information found in the RA message. Next, the Client creates a symmetric neighbor cache entry with the Server's AERO address as the network-layer address and the Server's encapsulation and/or link-layer addresses as the link-layer address. The Client records the RA Router Lifetime field value in the neighbor cache entry as the time for which the Server has committed to maintaining the MNP in the routing system, and caches the other RA configuration information including Cur Hop Limit, M and O flags, Reachable Time and Retrans Timer. The Client then autoconfigures AERO addresses for each of the delegated MNPs and assigns them to the AERO interface. The Client also caches any MSPs included in Route Information Options (RIOs) as MSPs to associate with the AERO link, and assigns the MTU value in the MTU option to the underlying interface. The Client then registers additional underlying interfaces with the Server by sending RS messages via each additional interface. The RS messages include the same parameters as for the initial RS/RA exchange, but with destination address set to the Server's AERO address and with the initial OMNI option ifIndex-tuple corresponding to the underlying interface. The Client sets a "primary" flag in the OMNI option if it wishes to enable proxy keepalives on this underlying interface. Following autoconfiguration, the Client sub-delegates the MNPs to its attached EUNs and/or the Client's own internal virtual interfaces as described in to support the Client's downstream attached "Internet of Things (IoT)". The Client subsequently sends additional RS messages over each underlying interface before the Router Lifetime received for that interface expires. After the Client registers its underlying interfaces, it may wish to change one or more registrations, e.g., if an interface changes address or becomes unavailable, if QoS preferences change, etc. To do so, the Client prepares an RS message to send over any available underlying interface. The RS includes an OMNI option with prefix registration information specific to its MNP, with a first ifIndex-tuple specific to the selected underlying interface, and with any additional ifIndex-tuples specific to other underlying interfaces. The Client includes fresh ifIndex-tuple values to update the Server's neighbor cache entry. When the Client receives the Server's RA response, it has assurance that the Server has been updated with the new information. If the Client wishes to discontinue use of a Server it issues an RS message over any underlying interface with an OMNI option with a prefix release indication. When the Server processes the message, it releases the MNP, sets the symmetric neighbor cache entry state for the Client to DEPARTED and returns an RA reply with Router Lifetime set to 0. After a short delay (e.g., 2 seconds), the Server withdraws the MNP from the routing system.

AERO Servers act as IP routers and support a PD service for Clients. Servers arrange to add their AERO addresses to a static map of Server addresses for the link and/or the DNS resource records for the FQDN "linkupnetworks.[domainname]" before entering service. Server addresses should be geographically and/or topologically referenced, and made available for discovery by Clients on the AERO link. When a Server receives a prospective Client's RS message on its AERO interface, it SHOULD return an immediate RA reply with Router Lifetime set to 0 if it is currently too busy or otherwise unable to service the Client. Otherwise, the Server authenticates the RS message and processes the PD parameters. The Server first determines the correct MNPs to delegate to the Client by searching the Client database. When the Server delegates the MNPs, it also creates a forwarding table entry for each MNP so that the MNPs are propagated into the routing system (see: ). For IPv6, the Server creates an IPv6 forwarding table entry for each MNP. For IPv4, the Server creates an IPv6 forwarding table entry with the SPAN Compatibility Prefix (SCP) corresponding to the IPv4 address. The Server next creates a symmetric neighbor cache entry for the Client using the base AERO address as the network-layer address and with lifetime set to no more than the smallest PD lifetime. Next, the Server updates the neighbor cache entry by recording the information in each ifIndex-tuple in the RS OMNI option. The Server also records the actual SPAN/INET addresses in the neighbor cache entry. If an SLLAO was present, the Server also compares the SLLAO address information for the first ifIndex-tuple with the SPAN/INET information to determine if there is a NAT on the path. Next, the Server prepares an RA message using its AERO address as the network-layer source address and the network-layer source address of the RS message as the network-layer destination address. The Server sets the Router Lifetime to the time for which it will maintain the symmetric neighbor cache entry, and sets Cur Hop Limit, M and O flags, Reachable Time and Retrans Timer to values appropriate for the AERO link. The Server includes the delegated MNPs, any other PD parameters and an OMNI option with an ifIndex-tuple with ifIndex set to 0. The Server then includes one or more RIOs that encode the MSPs for the AERO link, plus an MTU option (see ). The Server finally forwards the message to the Client using SPAN/INET, INET, or NULL encapsulation as necessary. After the initial RS/RA exchange, the Server maintains a ReachableTime timer for the Client's symmetric neighbor cache entry set to expire after Router Lifetime seconds. If the Client (or Proxy) issues additional RS messages, the Server sends an RA response and resets ReachableTime. If the Server receives an ND message with PD release indication it sets the Client's symmetric neighbor cache entry to the DEPARTED state and withdraws the MNP from the routing system after a short delay (e.g., 2 seconds). If ReachableTime expires before a new RS is received, the Server deletes the neighbor cache entry and withdraws the MNP without delay. The Server processes any ND/PD messages pertaining to the Client and returns an NA/RA reply in response to solicitations. The Server may also issue unsolicited RA messages, e.g., with PD reconfigure parameters to cause the Client to renegotiate its PDs, with Router Lifetime set to 0 if it can no longer service this Client, etc. Finally, If the symmetric neighbor cache entry is in the DEPARTED state, the Server deletes the entry after DepartTime expires. Note: Clients SHOULD notify former Servers of their departures, but Servers are responsible for expiring neighbor cache entries and withdrawing routes even if no departure notification is received (e.g., if the Client leaves the network unexpectedly). Servers SHOULD therefore set Router Lifetime to REACHABLETIME seconds in solicited RA messages to minimize persistent stale cache information in the absence of Client departure notifications. A short Router Lifetime also ensures that proactive Client/Server RS/RA messaging will keep any NAT state alive (see above). Note: All Servers on an AERO link MUST advertise consistent values in the RA Cur Hop Limit, M and O flags, Reachable Time and Retrans Timer fields the same as for any link, since unpredictable behavior could result if different Servers on the same link advertised different values.

When DHCPv6 is used as the ND/PD service back end, AERO Clients and Servers are always on the same link (i.e., the AERO link) from the perspective of DHCPv6. However, in some implementations the DHCPv6 server and ND function may be located in separate modules. In that case, the Server's AERO interface module can act as a Lightweight DHCPv6 Relay Agent (LDRA) to relay PD messages to and from the DHCPv6 server module. When the LDRA receives an authentic RS message, it extracts the PD message parameters and uses them to construct an IPv6/UDP/DHCPv6 message. It sets the IPv6 source address to the source address of the RS message, sets the IPv6 destination address to 'All_DHCP_Relay_Agents_and_Servers' and sets the UDP fields to values that will be understood by the DHCPv6 server. The LDRA then wraps the message in a DHCPv6 'Relay-Forward' message header and includes an 'Interface-Id' option that includes enough information to allow the LDRA to forward the resulting Reply message back to the Client (e.g., the Client's link-layer addresses, a security association identifier, etc.). The LDRA also wraps the OMNI option and SLLAO into the Interface-Id option, then forwards the message to the DHCPv6 server. When the DHCPv6 server prepares a Reply message, it wraps the message in a 'Relay-Reply' message and echoes the Interface-Id option. The DHCPv6 server then delivers the Relay-Reply message to the LDRA, which discards the Relay-Reply wrapper and IPv6/UDP headers, then uses the DHCPv6 message to construct an RA response to the Client. The Server uses the information in the Interface-Id option to prepare the RA message and to cache the link-layer addresses taken from the OMNI option and SLLAO echoed in the Interface-Id option.

Clients may connect to ANETs that require a perimeter security gateway to enable communications to Servers in outside INETs. In that case, the ANET can employ an AERO Proxy. The Proxy is located at the ANET/INET border and listens for RS messages originating from or RA messages destined to ANET Clients. The Proxy acts on these control messages as follows: when the Proxy receives an RS message from a new ANET Client, it first authenticates the message then examines the network-layer destination address. If the destination address is a Server's AERO address, the Proxy proceeds to the next step. Otherwise, if the destination is All-Routers multicast or Subnet-Router anycast, the Proxy selects a "nearby" Server that is likely to be a good candidate to serve the Client and replaces the destination address with the Server's AERO address. Next, the Proxy creates a proxy neighbor cache entry and caches the Client and Server link-layer addresses along with the OMNI option information and any other identifying information including Transaction IDs, Client Identifiers, Nonce values, etc. The Proxy then replaces the SLLAO in the RS message (if present) with a new SLLAO with a single ifIndex-tuple matching the first ifIndex-tuple in the OMNI option and with the Link Layer Address and Port Number fields set to the Proxy's SPAN address. The Proxy finally encapsulates the (proxyed) RS message in a SPAN header with destination set to the Server's SPAN address then forwards the message into the SPAN. when the Server receives the RS, it authenticates the message then creates or updates a symmetric neighbor cache entry for the Client with the Proxy's SPAN address as the link-layer address. The Server then sends an RA message back to the Proxy via the SPAN. when the Proxy receives the RA, it authenticates the message and matches it with the proxy neighbor cache entry created by the RS. The Proxy then caches the PD route information as a mapping from the Client's MNPs to the Client's ANET address, caches the Server's advertised Router Lifetime and sets the neighbor cache entry state to REACHABLE. The Proxy then replaces the RA SLLAO with an SLLAO with its own ANET address, sets the P bit in the RA flags field, sets the OMNI option "primary" flag according to the cached value from the RS, optionally rewrites the Router Lifetime and forwards the (proxyed) message to the Client. If the RA included an MTU option, the Proxy rewrites the MTU value (if necessary) to the minimum of the received MTU value and the MTU of the underlying ANET interface. After the initial RS/RA exchange, the Proxy forwards any Client data packets for which there is no matching asymmetric neighbor cache entry to a Relay via the SPAN. The Proxy instead forwards any Client data destined to an asymmetric neighbor cache target directly to the target according to the link-layer information - the process of establishing asymmetric neighbor cache entries is specified in . While the Client is still attached to the ANET, the Proxy sends NS, RS and/or unsolicited NA messages to update the Server's symmetric neighbor cache entries on behalf of the Client and/or to convey QoS updates. If the "primary" flag was set, the Proxy performs periodic RS/RA exchanges on the Client's behalf according to the cached Server lifetime. This allows for higher-frequency Proxy-initiated RS/RA messaging over well-connected INET infrastructure supplemented by lower-frequency Client-initiated RS/RA messaging over constrained ANET data links. If the Server ceases to send solicited advertisements, the Proxy deletes the neighbor cache entry and sends unsolicited RAs on the ANET interface with destination set to All-Nodes multicast (ff02::1) and with Router Lifetime set to zero to inform Clients that the Server has failed. Although the Proxy engages in ND exchanges on behalf of the Client, the Client can also send ND messages on its own behalf, e.g., if it is in a better position than the Proxy to convey QoS changes, etc. For this reason, the Proxy marks any Client-originated solicitation messages (e.g. by inserting a Nonce option) so that it can return the solicited advertisement to the Client instead of processsing it locally. If the Client becomes unreachable, the Proxy sets the neighbor cache entry state to DEPARTED and retains the entry for DEPARTTIME seconds. While the state is DEPARTED, the Proxy forwards any packets destined to the Client to a Relay. The Relay in turn forwards the packets to the Client's current Server. When DepartTime expires, the Proxy deletes the neighbor cache entry and discards any further packets destined to this (now forgotten) Client. When a neighbor cache entry transitions to DEPARTED, some of the fragments of a multiple fragment packet may have already arrived at the Proxy while others are en route to the Client's new location, however no special attention in the reassembly algorithm is necessary when re-routed packets are simply treated as loss. Since the fragments of a multiple-fragment packet are sent in minimal inter-packet delay bursts, such occasions will be rare. In some ANETs that employ a Proxy, the Client's MNP can be injected into the ANET routing system. In that case, the Client can send data messages without encapsulation so that the ANET native routing system transports the unencapsulated packets to the Proxy. This can be very beneficial, e.g., if the Client connects to the ANET via low-end data links such as some aviation wireless links. If the first-hop ANET access router is AERO-aware, the Client can avoid encapsulation for both its control and data messages. When the Client connects to the link, it can send an unencapsulated RS message with source address set to its AERO address and with destination address set to the AERO address of the Client's selected Server or to All-Routers multicast or Subnet-Router anycast. The Client includes an OMNI option formatted as specified in . The Client then sends the unencapsulated RS message, which will be intercepted by the AERO-Aware access router. The access router then encapsulates the RS message in an ANET header with its own address as the source address and the address of a Proxy as the destination address. The access router further remembers the address of the Proxy so that it can encapsulate future data packets from the Client via the same Proxy. If the access router needs to change to a new Proxy, it simply sends another RS message toward the Server via the new Proxy on behalf of the Client. In some cases, the access router and Proxy may be one and the same node. In that case, the node would be located on the same physical link as the Client, but its message exchanges with the Server would need to pass through a security gateway at the ANET/INET border. The method for deploying access routers and Proxys (i.e. as a single node or multiple nodes) is an ANET-local administrative consideration.

In environments where fast recovery from Server failure is required, Proxys SHOULD use proactive Neighbor Unreachability Detection (NUD) to track Server reachability in a similar fashion as for Bidirectional Forwarding Detection (BFD) . Proxys can then quickly detect and react to failures so that cached information is re-established through alternate paths. The NUD control messaging is carried only over well-connected ground domain networks (i.e., and not low-end aeronautical radio links) and can therefore be tuned for rapid response. Proxys perform proactive NUD with Servers for which there are currently active ANET Clients by sending continuous NS messages in rapid succession, e.g., one message per second. The Proxy sends the NS message via the SPAN with the Proxy's AERO address as the source and the AERO address of the Server as the destination. When the Proxy is also sending RS messages to the Server on behalf of ANET Clients, the resulting RA responses can be considered as equivalent hints of forward progress. This means that the Proxy need not also send a periodic NS if it has already sent an RS within the same period. If the Server fails (i.e., if the Proxy ceases to receive advertisements), the Proxy can quickly inform Clients by sending multicast RA messages on the ANET interface. The Proxy sends RA messages on the ANET interface with source address set to the Server's address, destination address set to All-Nodes multicast, and Router Lifetime set to 0. The Proxy SHOULD send MAX_FINAL_RTR_ADVERTISEMENTS RA messages separated by small delays . Any Clients on the ANET that had been using the failed Server will receive the RA messages and associate with a new Server.

While data packets are flowing between a source and target node, route optimization SHOULD be used. Route optimization is initiated by the first eligible Route Optimization Source (ROS) closest to the source as follows: For Clients on VPNed, NATed and Direct interfaces, the Server is the ROS. For Clients on Proxyed interfaces, the Proxy is the ROS. For Clients on native interfaces, the Client itself is the ROS. For correspondent nodes on INET/EUN interfaces serviced by a Gateway, the Gateway is the ROS. The route optimization procedure is conducted between the ROS and the target Server/Gateway acting as a Route Optimization Responder (ROR) in the same manner as for IPv6 ND Address Resolution and using the same NS/NA messaging. The target may either be a MNP Client serviced by a Server, or a non-MNP correspondent reachable via a Gateway. The procedures are specified in the following sections.

While data packets are flowing from the source node toward a target node, the ROS performs address resolution by sending an NS message to receive a solicited NA message from the ROR. When the ROS sends an NS, it includes the AERO address of the ROS as the source address (e.g., fe80::1) and the AERO address corresponding to the data packet's destination address as the destination address (e.g., if the destination address is 2001:db8:1:2::1 then the corresponding AERO address is fe80::2001:db8:1:2). The NS message includes an OMNI option with a single ifIndex-tuple with ifIndex set to 0, and an SLLAO with the SPAN address of the ROS. The message also includes a Nonce and Timestamp option if the ROS needs to correlate NA replies. The ROS then encapsulates the NS message in a SPAN header with source set to its own SPAN address and destination set to the data packet's destination address, then sends the message into the SPAN without decrementing the network-layer TTL/Hop Limit field.

When the Relay receives the NS message from the ROS, it discards the INET header and determines that the ROR is the next hop by consulting its standard IPv6 forwarding table for the SPAN header destination address. The Relay then forwards the SPAN message toward the ROR the same as for any IPv6 router. The final-hop Relay in the SPAN will deliver the message via a secured tunnel to the ROR.

When the ROR receives the NS message, it examines the AERO destination address to determine whether it has a neighbor cache entry and/or route that matches the target. If there is no match, the ROR drops the NS message. Otherwise, the ROR continues processing as follows: if the target belongs to an MNP Client neighbor in the DEPARTED state the ROR changes the NS message SPAN destination address to the SPAN address of the Client's new Server, forwards the message into the SPAN and returns from processing. If the target belongs to an MNP Client neighbor in the REACHABLE state, the ROR instead adds the AERO source address to the target Client's Report List with time set to ReportTime. If the target belongs to a non-MNP route, the ROR continues processing without adding an entry to the Report List. The ROR then prepares a solicited NA message to send back to the ROS but does not create a neighbor cache entry. The ROR sets the NA source address to the destination AERO address of the NS, and includes the Nonce value received in the NS plus the current Timestamp. If the target belongs to an MNP Client, the ROR then includes an OMNI option with prefix registration length set to the length of the MNP; otherwise, set to the maximum of the non-MNP prefix length and 64. (Note that a /64 limit is imposed to avoid causing the ROS to set short prefixes (e.g., "default") that would match destinations for which the routing system includes more-specific prefixes.) The ROR next includes a first ifIndex-tuple in the OMNI option with ifIndex set to 0. If the target belongs to an MNP Client, the ROR next includes additional ifIndex-tuples in the OMNI option for each of the target Client's underlying interfaces with current information for each interface The ROR then includes a TLLAO option with ifIndex-tuples in one-to-one correspondence with the tuples that appear in the OMNI option. For NATed, VPNed and Direct interfaces, the link layer addresses are the SPAN address of the ROR. For Proxyed interfaces, the link-layer addresses are the SPAN addresses of the Proxy's INET interfaces. For native interfaces, the link-layer addresses are the SPAN addesses of the Client's native interfaces. The ROR finally encapsulates the NA message in a SPAN header with source set to its own SPAN address and destination set to the source SPAN address of the NS message, then forwards the message into the SPAN without decrementing the network-layer TTL/Hop Limit field.

When the Relay receives the NA message from the ROR, it discards the INET header and determines that the ROS is the next hop by consulting its standard IPv6 forwarding table for the SPAN header destination address. The Relay then forwards the SPAN-encapsulated NA message toward the ROS the same as for any IPv6 router. The final-hop Relay in the SPAN will deliver the message via a secured tunnel to the ROS.

When the ROS receives the solicited NA message, it caches the source SPAN address then discards the INET and SPAN headers. The ROS next verifies the Nonce and Timestamp values (if present), then creates an asymmetric neighbor cache entry for the ROR and caches all information found in the solicited NA OMNI and TLLAO options. The ROS finally sets the asymmetric neighbor cache entry lifetime to REACHABLETIME seconds.

Following route optimization, the ROS forwards future data packets destined to the target via the addresses found in the cached link-layer information. The route optimization is shared by all sources that send packets to the target via the ROS, i.e., and not just the source on behalf of which the route optimization was initiated. While new data packets destined to the target are flowing through the ROS, it sends additional NS messages to the ROR before ReachableTime expires to receive a fresh solicited NA message the same as described in the previous sections (route optimization refreshment strategies are an implementation matter, with a non-normative example given in ). The ROS uses the cached SPAN address of the ROR as the NS SPAN destination address, and sends up to MAX_UNICAST_SOLICIT NS messages separated by 1 second until an NA is received. If no NA is received, the ROS assumes that the current ROR has become unreachable and deletes the neighbor cache entry. Subsequent data packets will trigger a new route optimization per to discover a new ROR while initial data packets travel over a suboptimal route. If an NA is received, the ROS then updates the asymmetric neighbor cache entry to refresh ReachableTime, while (for MNP destinations) the ROR adds or updates the ROS address to the target Client's Report List and with time set to ReportTime. While no data packets are flowing, the ROS instead allows ReachableTime for the asymmetric neighbor cache entry to expire. When ReachableTime expires, the ROS deletes the asymmetric neighbor cache entry. Any future data packets flowing through the ROS will again trigger a new route optimization. The ROS may also receive unsolicited NA messages from the ROR at any time (see: ). If there is an asymmetric neighbor cache entry for the target, the ROS updates the link-layer information but does not update ReachableTime since the receipt of an unsolicited NA does not confirm that the forward path is still working. If there is no asymmetric neighbor cache entry, the ROS simply discards the unsolicited NA. In this arrangement, the ROS holds an asymmetric neighbor cache entry for the ROR, but the ROR does not hold an asymmetric neighbor cache entry for the ROS. The route optimization neighbor relationship is therefore asymmetric and unidirectional. If the target node also has packets to send back to the source node, then a separate route optimization procedure is performed in the reverse direction. But, there is no requirement that the forward and reverse paths be symmetric.

AERO nodes perform Neighbor Unreachability Detection (NUD) per . NUD is performed either reactively in response to persistent link-layer errors (see ) or proactively to confirm reachability. The NUD algorithm is based on periodic authentic NS/NA message exchanges. The algorithm may further be seeded by ND hints of forward progress, but care must be taken to avoid inferring reachability based on spoofed information. For example, authentic RS/RA exchanges may be considered as acceptable hints of forward progress, while spurious data packets should not be. When an ROR directs an ROS to a neighbor with one or more target link-layer addresses, the ROS can proactively test each direct path by sending an initial NS message to elicit a solicited NA response. While testing the paths, the ROS can optionally continue sending packets via the SPAN, maintain a small queue of packets until target reachability is confirmed, or (optimistically) allow packets to flow via the direct paths. In any case, the ROS should only consider the neighbor unreachable if NUD fails over multiple target link-layer address paths. When a ROS sends an NS message used for NUD, it uses its AERO addresses as the IPv6 source address and the AERO address corresponding to a target link-layer address as the destination. For each target link-layer address, the source node encapsulates the NS message in SPAN/INET headers with its own SPAN address as the source and the SPAN address of the target as the destination, If the target is located within the same SPAN segment, the source sets the INET address of the target as the destination; otherwise, it sets the INET address of a Relay as the destination. The source then forwards the message into the SPAN. Paths that pass NUD tests are marked as "reachable", while those that do not are marked as "unreachable". These markings inform the AERO interface forwarding algorithm specified in . Proxys can perform NUD to verify Server reachability on behalf of their proxyed Clients to reduce Client-initated control messaging overhead.

AERO is a Distributed Mobility Management (DMM) service. Each Server is responsible for only a subset of the Clients on the AERO link, as opposed to a Centralized Mobility Management (CMM) service where there is a single network mobility collective entity for all Clients. Clients coordinate with their associated Servers via RS/RA exchanges to maintain the DMM profile, and the AERO routing system tracks all current Client/Server peering relationships. Servers provide a Mobility Anchor Point (MAP) for their dependent Clients. Clients are responsible for maintaining neighbor relationships with their Servers through periodic RS/RA exchanges, which also serves to confirm neighbor reachability. When a Client's underlying interface address and/or QoS information changes, the Client is responsible for updating the Server with this new information. Note that for Proxyed interfaces, however, the Proxy can perform the RS/RA exchanges on the Client's behalf. Mobility management considerations are specified in the following sections.

Servers acting as MAPs accommodate Client mobility and/or QoS change events by sending unsolicited NA messages to each ROS in the target Client's Report List. When a MAP sends an unsolicited NA message, it sets the IPv6 source address to the Client's AERO address and sets the IPv6 destination address to All-Nodes multicast. The MAP also includes an OMNI option with prefix registration information, with a first ifIndex-tuple with ifIndex set to 0, and with additional ifIndex-tuples for the target Client's remaining interfaces. The MAP then includes a TLLAO with corresponding ifIndex-tuples, with the link layer address of the first tuple set to the MAP's SPAN address and with link layer addresses of the remaining tuples set to the corresponding target SPAN addresses. The MAP finally encapsulates the message in a SPAN header with source set to its own SPAN address and destination set to the SPAN address of the ROS, then sends the message to a Relay which in turn forwards it to the ROS. As discussed in Section 7.2.6 of , the transmission and reception of unsolicited NA messages is unreliable but provides a useful optimization. In well-connected Internetworks with robust data links unsolicited NA messages will be delivered with high probability, but in any case the MAP can optionally send up to MAX_NEIGHBOR_ADVERTISEMENT unsolicited NAs to each ROS to increase the likelihood that at least one will be received. When the ROS receives an unsolicited NA message, it ignores the message if there is no existing neighbor cache entry for the Client. Otherwise, it uses the included OMNI option and TLLAO information to update the neighbor cache entry, but does not reset ReachableTime since the receipt of an unsolicited NA message from the target Server does not provide confirmation that any forward paths to the target Client are working. If unsolicited NA messages are lost, the ROS may be left with stale address and/or QoS information for the Client for up to REACHABLETIME seconds. During this time, the ROS can continue sending packets according to its stale neighbor cache information. When ReachableTime is close to expiring, the ROS will re-initiate route optimization and receive fresh state information. In addition to sending unsolicited NA messages to the current set of ROSs for the Client, the MAP also sends unsolicited NAs to the former link-layer address for any ifIndex-tuple for which the link-layer address has changed. The NA messages update Proxys or Servers that cannot easily detect (e.g., without active probing) when a formerly-active Client has departed.

When a Client needs to change its ANET addresses and/or QoS preferences (e.g., due to a mobility event), either the Client or its Proxys send RS messages to the Server via the SPAN with an OMNI option and SLLAO that include an ifIndex-tuple with the new link quality and address information. Up to MAX_RTR_SOLICITATION RS messages MAY be sent in parallel with sending actual data packets in case one or more RAs are lost. If all RAs are lost, the Client SHOULD re-associate with a new Server. When the Server receives the Client's changes, it sends unsolicited NA messages to all nodes in the Report List the same as described in the previous section.

When a Client needs to bring new underlying interfaces into service (e.g., when it activates a new data link), it sends an RS message to the Server via the underlying interface with an OMNI option with appropriate link quality values and with an SLLAO (if necessary) with link-layer address information for the new link..

When a Client needs to remove existing underlying interfaces from service (e.g., when it de-activates an existing data link), it sends an RS message to its Server with an OMNI option with appropriate link quality values. If the Client needs to send RS messages over an underlying interface other than the one being removed from service, it MUST include an ifIndex-tuple for the sending interface as the first tuple and include additional ifIndex-tuples with appropriate link quality values for any underlying interfaces being removed from service.

When a Client associates with a new Server, it performs the Client procedures specified in . The Client also includes a notification identifier in the RS message OMNI option per if it wants the new Server to notify the old Server. When the new Server receives the Client's RS message, it responds by returning an RA as specified in . If the Client's RS includes a notification identifier, the new Server also prepares an RS or unsolicited NA message to send to the old Server. The RS/NA message includes the Client's AERO address as the source address, the old Server's AERO address as the destination address, and an OMNI option and S/TLLAO with an ifIndex-tuple with ifIndex set to 0. The OMNI option includes a prefix release indication, and the S/TLLAO includes the SPAN address of the new Server. For RS messages, the new Server retries up to MAX_RTR_SOLICITATIONS attempts until an RA is received. (Note that the Client can alternatively send RS/NA messages with a release indication to the old Server on its own behalf, however, this additional Client messaging may be undesirable in some environments. Note also that the choice of using RS or unsolicited NA is based on the need for a reliable acknowledgement; in environments where Router Lifetimes can be expected to be short, sending up to MAX_NEIGHBOR_ADVERTISEMENT unsolicited NAs may be sufficient.) When the old Server processes the RS/NA, it changes the symmetric neighbor cache entry state to DEPARTED, sets the link-layer address of the Client to the address found in the S/TLLAO, and sets DepartTime to DEPARTTIME seconds. For RS messages, the old Server then returns an immediate RA message with Router Lifetime set to 0. After a short delay (e.g., 2 seconds) the old Server withdraws the Client's MNP from the routing system. After DepartTime expires, the old Server deletes the symmetric neighbor cache entry. The old Server also sends unsolicited NA messages to all ROSs in the Client's Report List with an OMNI option with prefix release indication, with a single ifIndex-tuple with ifIndex set to 0 and with the SPAN address of the new Server in a companion TLLAO. When the ROS receives the NA, it caches the address of the new Server in the existing asymmetric neighbor cache entry and marks the entry as STALE. Subsequent data packets will then flow according to any existing cached link-layer information and trigger a new NS/NA exchange via the new Server. Clients SHOULD NOT move rapidly between Servers in order to avoid causing excessive oscillations in the AERO routing system. Examples of when a Client might wish to change to a different Server include a Server that has gone unreachable, topological movements of significant distance, movement to a new geographic region, movement to a new SPAN segment, etc. When a Client moves to a new Server, some of the fragments of a multiple fragment packet may have already arrived at the old Server while others are en route to the new Server, however no special attention in the reassembly algorithm is necessary when re-routed fragments are simply treated as loss. Since the fragments of a multiple-fragment packet are sent in a minimal inter-packet delay burst, such occasions will be rare.

The AERO Client provides an IGMP (IPv4) or MLD (IPv6) proxy service for its EUNs and/or hosted applications . The Client forwards IGMP/MLD messages over any of its underlying interfaces for which group membership is required. The IGMP/MLD messages may be further forwarded by a first-hop ANET access router acting as an IGMP/MLD-snooping switch , then ultimately delivered to an AERO Proxy/Server acting as a Protocol Independent Multicast - Sparse-Mode (PIM-SM, or simply "PIM") Designated Router (DR) . AERO Gateways also act as PIM routers (i.e., the same as AERO Proxys/Servers) on behalf of nodes on INET/EUN networks. The behaviors identified in the following sections correspond to Source-Specific Multicast (SSM) and Any-Source Multicast (ASM) operational modes.

When an ROS (i.e., an AERO Proxy/Server/Gateway) "X" acting as PIM router receives a Join/Prune message from a node on its downstream interfaces containing one or more ((S)ource, (G)roup) pairs, it updates its Multicast Routing Information Base (MRIB) accordingly. For each S belonging to a prefix reachable via X's non-AERO interfaces, X then forwards the (S, G) Join/Prune to any PIM routers on those interfaces per . For each S belonging to a prefix reachable via X's AERO interface, X originates a separate copy of the Join/Prune for each (S,G) in the message using its own AERO address as the source address and ALL-PIM-ROUTERS as the destination address. X then encapsulates each message in a SPAN header with source address set to the SPAN address of X and destination address set to S then forwards the message into the SPAN. The SPAN in turn forwards the message to AERO Server/Gateway "Y" that services S. At the same time, if the message was a Join, X sends a route-optimization NS message toward each S the same as discussed in . The resulting NAs will return the AERO address for the prefix that matches S as the network-layer source address and TLLAOs with the SPAN addresses corresponding to any ifIndex-tuples that are currently servicing S. When Y processes the Join/Prune message, if S located behind any Native, Direct, VPNed or NATed interfaces Y acts as a PIM router and updates its MRIB to list X as the next hop in the reverse path. If S is located behind any Proxys "Z"*, Y also forwards the message to each Z* over the SPAN while continuing to use the AERO address of X as the source address. Each Z* then updates its MRIB accordingly and maintains the AERO address of X as the next hop in the reverse path. Since the Relays in the SPAN do not examine network layer control messages, this means that the (reverse) multicast tree path is simply from each Z* (and/or Y) to X with no other multicast-aware routers in the path. If any Z* (and/or Y) is located on the same SPAN segment as X, the multicast data traffic sent to X directly using SPAN/INET encapsulation instead of via a Relay. Following the initial Join/Prune and NS/NA messaging, X maintains an asymmetric neighbor cache entry for each S the same as if X was sending unicast data traffic to S. In particular, X performs additional NS/NA exchanges to keep the neighbor cache entry alive for up to t_periodic seconds . If no new Joins are received within t_periodic seconds, X allows the neighbor cache entry to expire. Finally, if X receives any additional Join/Prune messages for (S,G) it forwards the messages to each Y and Z* in the neighbor cache entry over the SPAN. At some later time, Client C that holds an MNP for source S may depart from a first Proxy Z1 and/or connect via a new Proxy Z2. In that case, Y sends an unsolicited NA message to X the same as specified for unicast mobility in . When X receives the unsolicited NA message, it updates its asymmetric neighbor cache entry for the AERO address for source S and sends new Join messages to any new Proxys Z2. There is no requirement to send any Prune messages to old Proxys Z1 since source S will no longer source any multicast data traffic via Z1. Instead, the multicast state for (S,G) in Proxy Z1 will soon time out since no new Joins will arrive. After some later time, C may move to a new Server Y2 and depart from old Sever Y1. In that case, Y1 sends Join messages for any of C's active (S,G) groups to Y2 while including its own AERO address as the source address. This causes Y2 to include Y1 in the multicast forwarding tree during the interim time that Y1's symmetric neighbor cache entry for C is in the DEPARTED state. At the same time, Y1 sends an unsolicited NA message to X with an OMNI option and TLLAO with ifIndex-tuple set to 0 and a release indication to cause X to release its asymmetric neighbor cache entry. X then sends a new Join message to S via the SPAN and re-initiates route optimization the same as if it were receiving a fresh Join message from a node on a downstream link.

When an ROS X acting as a PIM router receives a Join/Prune from a node on its downstream interfaces containing one or more (*,G) pairs, it updates its Multicast Routing Information Base (MRIB) accordingly. X then forwards a copy of the message to the Rendezvous Point (RP) R for each G over the SPAN. X uses its own AERO address as the source address and ALL-PIM-ROUTERS as the destination address, then encapsulates each message in a SPAN header with source address set to the SPAN address of X and destination address set to R, then sends the message into the SPAN. At the same time, if the message was a Join X initiates NS/NA route optimization the same as for the SSM case discussed in . For each source S that sends multicast traffic to group G via R, the Proxy/Server Z* for the Client that aggregates S encapsulates the packets in PIM Register messages and forwards them to R via the SPAN. R may then elect to send a PIM Join to Z* over the SPAN. This will result in an (S,G) tree rooted at Z* with R as the next hop so that R will begin to receive two copies of the packet; one native copy from the (S, G) tree and a second copy from the pre-existing (*, G) tree that still uses PIM Register encapsulation. R can then issue a PIM Register-stop message to suppress the Register-encapsulated stream. At some later time, if C moves to a new Proxy/Server Z*, it resumes sending packets via PIM Register encapsulation via the new Z*. At the same time, as multicast listeners discover individual S's for a given G, they can initiate an (S,G) Join for each S under the same procedures discussed in . Once the (S,G) tree is established, the listeners can send (S, G) Prune messages to R so that multicast packets for group G sourced by S will only be delivered via the (S, G) tree and not from the (*, G) tree rooted at R. All mobility considerations discussed for SSM apply.

Bi-Directional PIM (BIDIR-PIM) provides an alternate approach to ASM that treats the Rendezvous Point (RP) as a Designated Forwarder (DF). Further considerations for BIDIR-PIM are out of scope.

An AERO Client can connect to multiple AERO links the same as for any data link service. In that case, the Client maintains a distinct AERO interface for each link, e.g., 'aero0' for the first link, 'aero1' for the second, 'aero2' for the third, etc. Each AERO link would include its own distinct set of Relays, Servers and Proxys, thereby providing redundancy in case of failures. The Relays, Servers and Proxys on each AERO link can assign AERO and SPAN addresses that use the same or different numberings from those on other links. Since the links are mutually independent there is no requirement for avoiding inter-link address duplication, e.g., the same AERO address such as fe80::1000 could be used to number distinct nodes that connect to different AERO links. Each AERO link could utilize the same or different ANET connections. The links can be distinguished at the link-layer via Virtual Local Area Network (VLAN) tagging (e.g., IEEE 802.1Q) and/or through assignment of distinct sets of MSPs on each link. This gives rise to the opportunity for supporting multiple redundant networked paths, where each VLAN is distinguished by a different label (e.g., colors such as Red, Green, Blue, etc.). In particular, the Client can tag its RS messages with the appropriate label to cause the network to select the desired VLAN. Clients that connect to multiple AERO interfaces can select the outgoing interface appropriate for a given Red/Blue/Green/etc. traffic profile while (in the reverse direction) correspondent nodes must have some way of steering their packets destined to a target via the correct AERO link. In a first alternative, if each AERO link services different MSPs, then the Client can receive a distinct MNP from each of the links. IP routing will therefore assure that the correct Red/Green/Blue/etc. network is used for both outbound and inbound traffic. This can be accomplished using existing technologies and approaches, and without requiring any special supporting code in correspondent nodes or Relays. In a second alternative, if each AERO link services the same MSP(s) then each link could assign a distinct "AERO Link Anycast" address that is configured by all Relays on the link. Correspondent nodes then include a "type 4" routing header with the Anycast address for the AERO link as the IPv6 destination and with the address of the target encoded as the "next segment" in the routing header . Standard IP routing will then direct the packet to the nearest Relay for the correct AERO link, which will replace the destination address with the target address then forward the packet to the target.

AERO Client MNs and INET correspondent nodes consult the Domain Name System (DNS) the same as for any Internetworking node. When correspondent nodes and Client MNs use different IP protocol versions (e.g., IPv4 correspondents and IPv6 MNs), the INET DNS must maintain A records for IPv4 address mappings to MNs which must then be populated in Gateway NAT64 mapping caches. In that way, an IPv4 correspondent node can send packets to the IPv4 address mapping of the target MN, and the Gateway will translate the IPv4 header and destination address into an IPv6 header and IPv6 destination address of the MN. When an AERO Client registers with an AERO Server, the Server can return the address(es) of DNS servers in RDNSS options . The DNS server provides the IP addresses of other MNs and correspondent nodes in AAAA records for IPv6 or A records for IPv4.

The SPAN ensures that dissimilar INET partitions can be joined into a single unified AERO link, even though the partitions themselves may have differing protocol versions and/or incompatible addressing plans. However, a commonality can be achieved by incrementally distributing globally routable (i.e., native) IP prefixes to eventually reach all nodes (both mobile and fixed) in all SPAN segments. This can be accomplished by incrementally deploying AERO Gateways on each INET partition, with each Gateway distributing its MNPs and/or discovering non-MNP prefixes on its INET links. This gives rise to the opportunity to eventually distribute native IP addresses to all nodes, and to present a unified AERO link view (bridged by the SPAN) even if the INET partitions remain in their current protocol and addressing plans. In that way, the AERO link can serve the dual purpose of providing a mobility service and a transition service. Or, if an INET partition is transitioned to a native IP protocol version and addressing scheme that is compatible with the AERO link MNP-based addressing scheme, the partition and AERO link can be joined by Gateways. Gateways that connect INETs/EUNs with dissimilar IP protocol versions must employ a network address and protocol translation function such as NAT64 .

In environments where rapid failure recovery is required, Servers and Relays SHOULD use Bidirectional Forwarding Detection (BFD) . Nodes that use BFD can quickly detect and react to failures so that cached information is re-established through alternate nodes. BFD control messaging is carried only over well-connected ground domain networks (i.e., and not low-end radio links) and can therefore be tuned for rapid response. Servers and Relays maintain BFD sessions in parallel with their BGP peerings. If a Server or Relay fails, BGP peers will quickly re-establish routes through alternate paths the same as for common BGP deployments. Similarly, Proxys maintain BFD sessions with their associated Relays even though they do not establish BGP peerings with them. Proxys SHOULD use proactive NUD for Servers for which there are currently active ANET Clients in a manner that parallels BFD, i.e., by sending unicast NS messages in rapid succession to receive solicited NA messages. When the Proxy is also sending RS messages on behalf of ANET Clients, the RS/RA messaging can be considered as equivalent hints of forward progress. This means that the Proxy need not also send a periodic NS if it has already sent an RS within the same period. If a Server fails, the Proxy will cease to receive advertisements and can quickly inform Clients of the outage by sending multicast RA messages on the ANET interface. The Proxy sends multicast RA messages with source address set to the Server's address, destination address set to All-Nodes multicast, and Router Lifetime set to 0. The Proxy SHOULD send MAX_FINAL_RTR_ADVERTISEMENTS RA messages separated by small delays . Any Clients on the ANET interface that have been using the (now defunct) Server will receive the RA messages and associate with a new Server.