Automatic Extended Route Optimization (AERO)

3.2.1. AERO/OMNI Reference Model

Figure 1 presents the basic OMNI link reference model:¶

                      +-----------------+
                      | AERO Gateway G1 |
                      | Nbr: S1, S2, P1 |
                      |(X1->S1; X2->S2) |
                      |      MSP M1     |
                      +--------+--------+
    +--------------+           |            +--------------+
    |  AERO P/S S1 |           |            |  AERO P/S S2 |
    |  Nbr: C1, G1 |           |            |  Nbr: C2, G1 |
    |  default->G1 |           |            |  default->G1 |
    |    X1->C1    |           |            |    X2->C2    |
    +-------+------+           |            +------+-------+
            |       OMNI link  |                   |
    X===+===+==================+===================+===+===X
        |                                              |
  +-----+--------+                            +--------+-----+
  |AERO Client C1|                            |AERO Client C2|
  |    Nbr: S1   |                            |   Nbr: S2    |
  | default->S1  |                            | default->S2  |
  |    MNP X1    |                            |    MNP X2    |
  +------+-------+                            +-----+--------+
         |                                          |
        .-.                                        .-.
     ,-(  _)-.     +-------+     +-------+      ,-(  _)-.
  .-(_  IP   )-.   |  AERO |     |  AERO |    .-(_  IP   )-.
(__    ENET     )--|Host H1|     |Host H2|--(__    ENET     )
   `-(______)-'    +-------+     +-------+     `-(______)-'

In this model:¶

• the OMNI link is an overlay network service configured over one or more underlay SRT segments which may be managed by different administrative authorities and have incompatible protocols and/or addressing plans.¶
• AERO Gateway G1 aggregates Mobility Service Prefix (MSP) M1, discovers Mobile Network Prefixes (MNPs) X* and advertises the MSP via BGP peerings over secured tunnels to Proxy/Servers (S1, S2). Gateways provide the backbone for an SRT spanning tree for the OMNI link.¶
• AERO Proxy/Servers S1 and S2 configure secured tunnels with Gateway G1 and also provide mobility, multilink, multicast and default router services for the MNPs of their associated Clients C1 and C2. (Proxy/Servers that act as Relays can also advertise non-MNP routes for non-mobile correspondent nodes the same as for MNP Clients.)¶
• AERO Clients C1 and C2 associate with Proxy/Servers S1 and S2, respectively. They receive MNP delegations X1 and X2, and also act as default routers for their associated physical or internal virtual ENETs.¶
• AERO Hosts H1 and H2 attach to the ENETs served by Clients C1 and C2, respectively.¶

An OMNI link configured over a single underlay network appears as a single unified link with a consistent addressing plan; all nodes on the link can exchange carrier packets via simple L2 encapsulation (i.e., following any necessary NAT traversal) since the underlay is connected. In common practice, however, OMNI links are often configured over an SRT spanning tree that bridges multiple distinct underlay network segments managed under different administrative authorities (e.g., as for worldwide aviation service providers such as ARINC, SITA, Inmarsat, etc.). Individual underlay networks may also be partitioned internally, in which case each internal partition appears as a separate segment.¶

The addressing plan of each SRT 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, proxys, packet filtering gateways, etc.), and disjoint segments often have no common physical link connections. Therefore, nodes can only be assured of exchanging carrier 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 Gateways.¶

The OMNI link spans multiple SRT segments using the OMNI Adaptation Layer (OAL) [I-D.templin-6man-omni] to provide the network layer with a virtual abstraction similar to a bridged campus LAN. The OAL is an OMNI interface sublayer that inserts a mid-layer IPv6 encapsulation header for inter-segment forwarding (i.e., bridging) without decrementing the network-layer TTL/Hop Limit of the original IP packet. An example OMNI link SRT is shown in Figure 2:¶

              . . . . . . . . . . . . . . . . . . . . . . .
            .                                               .
            .              .-(::::::::)                     .
            .           .-(::::::::::::)-.   +-+            .
            .          (:::: Segment A :::)--|G|---+        .
            .           `-(::::::::::::)-'   +-+   |        .
            .              `-(::::::)-'            |        .
            .                                      |        .
            .              .-(::::::::)            |        .
            .           .-(::::::::::::)-.   +-+   |        .
            .          (:::: Segment B :::)--|G|---+        .
            .           `-(::::::::::::)-'   +-+   |        .
            .              `-(::::::)-'            |        .
            .                                      |        .
            .              .-(::::::::)            |        .
            .           .-(::::::::::::)-.   +-+   |        .
            .          (:::: Segment C :::)--|G|---+        .
            .           `-(::::::::::::)-'   +-+   |        .
            .              `-(::::::)-'            |        .
            .                                      |        .
            .                ..(etc)..             x        .
            .                                               .
            .                                               .
            .    <-    Segment Routing Topology (SRT) ->    .
            .             (Spanned by OMNI Link)            .
              . . . . . . . . . . . . . .. . . . . . . . .

Gateway, Proxy/Server and Relay OMNI interfaces are configured over both secured tunnels and open INET underlay interfaces within their respective SRT segments. Within each segment, Gateways configure "hub-and-spokes" BGP peerings with Proxy/Servers and Relays as "spokes". Adjacent SRT segments are joined by Gateway-to-Gateway peerings to collectively form a spanning tree over the entire SRT. The "secured" spanning tree supports authentication and integrity for critical control plane messages. The "unsecured" spanning tree conveys ordinary carrier packets without security codes and that must be treated by destinations according to data origin authentication procedures. AERO nodes can employ route optimization to cause carrier packets to take more direct paths between OMNI link neighbors without having to follow strict spanning tree paths.¶

The AERO Multinet service concatenates SRT segments to form larger networks through Gateway-to-Gateway peerings as originally described in the "Catenet Model for Internetworking" [IEN48]; especially Figure 2 follows directly from the illustrations in [IEN48-2]. The original Catenet vision proposed a "network-of-networks" concatenation of independent and diverse Internetwork "segments" to form a much larger network supporting end-to-end services.¶

The original Catenet vision first articulated in the 1970's was distorted through the evolution of the Internet in later decades, since a critical element was missing from the architecture. As a result, the Internet evolved as a single, large public routing and addressing domain interconnecting private domains (i.e., instead of a true network-of-networks) which has impeded flexibility and inhibited end-to-end services. With the advent of the adaptation layer established by the AERO/OMNI services, however, the original Catenet "network-of-networks" vision is now made possible.¶

3.2.2. Addressing and Node Identification

AERO nodes on OMNI links use the Link-Local Address (LLA) prefix fe80::/64 [RFC4291] to assign LLAs to the OMNI interface in a latent state and do not employ LLAs for any operational purposes (instead, nodes assign LLAs solely to satisfy the requirements of [RFC4861]). AERO Clients configure LLAs constructed from MNPs (i.e., "LLA-MNPs") while AERO infrastructure nodes construct LLAs based on 56-bit random values ("LLA-RNDs") per [I-D.templin-6man-omni]. Non-MNP routes are also represented the same as for MNPs, but may include a prefix that is not properly covered by an MSP.¶

AERO nodes also use the Unique Local Address (ULA) prefix fd00::/8 followed by a pseudo-random 40-bit Global ID to form the prefix {ULA}::/48, then include a 16-bit Subnet ID '*' to form the prefix {ULA*}::/64 [RFC4291]. The AERO node then uses the prefix {ULA*}::/64 to form "ULA-MNPs" or "ULA-RNDs" as specified in [I-D.templin-6man-omni] to support OAL addressing. (The prefix {ULA*}::/64 appearing alone and with no suffix represents "default" for that prefix.)¶

AERO Clients also use Temporary Local Addresses (TLAs) and eXtended Local Addresses (XLAs) constructed per [I-D.templin-6man-omni], where TLAs are distinguished from ordinary ULAs based on the prefix fd00::/16 and XLAs are distinguished from ULAs/TLAs based on the prefix fd00::/64. Clients use TLA-RNDs only in initial control message exchanges until a stable MNP is assigned, but may sometimes also use them for sustained communications within a local routing region. AERO nodes use XLA-MNPs to provide forwarding information for the global routing table as well as IPv6 ND message addressing information.¶

AERO MSPs, MNPs and non-MNP routes are typically based on Global Unicast Addresses (GUAs), but in some cases may be based on IPv4 private addresses [RFC1918] or IPv6 ULA-C's [RFC4193]. A GUA block is also reserved for OMNI link anycast purposes. See [I-D.templin-6man-omni] for a full specification of LLAs, ULAs, TLAs, XLAs, GUAs and anycast addresses used by AERO nodes on OMNI links.¶

Finally, AERO Clients and Proxy/Servers configure node identification values as specified in [I-D.templin-6man-omni].¶

3.2.3. AERO Routing System

The AERO routing system comprises a private Border Gateway Protocol (BGP) [RFC4271] service coordinated between Gateways and Proxy/Servers (Relays also engage in the routing system as simplified Proxy/Servers). The service supports carrier packet forwarding at a layer below IP and does not interact with the public Internet BGP routing system, but supports redistribution of information for other links and networks connected by Relays.¶

In a reference deployment, each Proxy/Server is configured as an Autonomous System Border Router (ASBR) for a stub Autonomous System (AS) using a 32-bit AS Number (ASN) [RFC4271] that is unique within the BGP instance, and each Proxy/Server further uses eBGP to peer with one or more Gateways but does not peer with other Proxy/Servers. Each SRT segment in the OMNI link must include one or more Gateways in a "hub" AS, which peer with the Proxy/Servers within that segment as "spoke" ASes. All Gateways within the same segment are members of the same hub AS, and use iBGP to maintain a consistent view of all active routes currently in service. The Gateways of different segments peer with one another using eBGP.¶

Gateways maintain forwarding table entries only for ULA prefixes for infrastructure elements and XLA-MNPs corresponding to MNP and non-MNP routes that are currently active; Gateways also maintain black-hole routes for the OMNI link MSPs so that carrier packets destined to non-existent more-specific routes are dropped with a Destination Unreachable message returned. In this way, Proxy/Servers and Relays have only partial topology knowledge (i.e., they only maintain routing information for their directly associated Clients and non-AERO links) and they forward all other carrier packets to Gateways which have full topology knowledge.¶

Each OMNI link segment assigns a unique sub-prefix of {ULA}::/48 known as the "SRT prefix". For example, a first segment could assign {ULA}:1000::/56, a second could assign {ULA}:2000::/56, a third could assign {ULA}:3000::/56, etc. Within each segment, each Proxy/Server configures a ULA-RND within the segment's SRT prefix with a 56-bit random value in the interface identifier as specified in [I-D.templin-6man-omni].¶

The administrative authorities for each segment must therefore coordinate to assure mutually-exclusive ULA prefix assignments, but internal provisioning of ULAs is an independent local consideration for each administrative authority. For each ULA prefix, the Gateway(s) that connect that segment assign the all-zero's address of the prefix as a Subnet Router Anycast address. For example, the Subnet Router Anycast address for {ULA}:1023::/64 is simply {ULA}:1023::/64.¶

ULA prefixes are statically represented in Gateway forwarding tables. Gateways join multiple SRT segments into a unified OMNI link over multiple diverse network administrative domains. They support a virtual bridging service by first establishing forwarding table entries for their ULA prefixes either via standard BGP routing or static routes. For example, if three Gateways ('A', 'B' and 'C') from different segments serviced {ULA}:1000::/60, {ULA}:2000::/56 and {ULA}:3000::/56 respectively, then the forwarding tables in each Gateway appear as follows:¶

A:

{ULA}:1000::/56->local, {ULA}:2000::/56->B, {ULA}:3000::/56->C¶

B:

{ULA}:1000::/56->A, {ULA}:2000::/56->local, {ULA}:3000::/56->C¶

C:

{ULA}:1000::/56->A, {ULA}:2000::/56->B, {ULA}:3000::/56->local¶

These forwarding table entries rarely change, since they correspond to fixed infrastructure elements in their respective segments.¶

MNP (and non-MNP) routes are instead dynamically advertised in the AERO routing system by Proxy/Servers and Relays that provide service for their corresponding MNPs. The routes are advertised as XLA-MNP prefixes, i.e., as fd00::{MNP} (see: [I-D.templin-6man-omni]). For example, if three Proxy/Servers ('D', 'E' and 'F') service the MNPs 2001:db8:1000:2000::/56, 2001:db8:3000:4000::/56 and 2001:db8:5000:6000::/56 then the routing system would include:¶

D:

fd00::2001:db8:1000:2000/120¶

E:

fd00::2001:db8:3000:4000/120¶

F:

fd00::2001:db8:5000:6000/120¶

A full discussion of the BGP-based routing system used by AERO is found in [I-D.ietf-rtgwg-atn-bgp].¶

3.2.4. Segment Routing Topologies (SRTs)

The distinct {ULA}::/48 prefixes in an OMNI link domain identify distinct Segment Routing Topologies (SRTs). Each SRT is a mutually-exclusive OMNI link overlay instance using a distinct set of ULAs, and emulates a bridged campus LAN service for the OMNI link. In some cases (e.g., when redundant topologies are needed for fault tolerance and reliability) it may be beneficial to deploy multiple SRTs that act as independent overlay instances. A communication failure in one instance therefore will not affect communications in other instances.¶

Each SRT is identified by a distinct value in the 40-bit ULA Global ID field and assigns an OMNI IPv6 anycast address used for OMNI interface determination in Safety-Based Multilink (SBM) as discussed in [I-D.templin-6man-omni]. Each OMNI interface further applies Performance-Based Multilink (PBM) internally.¶

The Gateways and Proxy/Servers of each independent SRT engage in BGP peerings to form a spanning tree with the Gateways in non-leaf nodes and the Proxy/Servers in leaf nodes. The spanning tree is configured over both secured and unsecured underlay network paths. The secured spanning tree is used to convey secured control messages between Proxy/Servers and Gateways, while the unsecured spanning tree forwards data messages and/or unsecured control messages.¶

Each SRT segment is identified by a unique ULA prefix used by all Proxy/Servers and Gateways in the segment. Each AERO node must therefore discover an SRT prefix that correspondents can use to determine the correct segment, and must publish the SRT prefix in IPv6 ND messages.¶

Note: The distinct ULA prefixes in an OMNI link domain can be carried either in a common BGP routing protocol instance for all OMNI links or in distinct BGP routing protocol instances for different OMNI links. In some SBM environments, such separation may be necessary to ensure that distinct OMNI links do not include any common infrastructure elements as single points of failure. In other environments, carrying the ULAs of multiple OMNI links within a common routing system may be acceptable.¶

3.2.5. Segment Routing For OMNI Link Selection

Original IPv6 sources can direct IPv6 packets to an AERO node by including a standard IPv6 Segment Routing Header (SRH) [RFC8754] with the OMNI IPv6 anycast address for the selected OMNI link as either the IPv6 destination or as an intermediate hop within the SRH. This allows the original source to determine the specific OMNI link SRT an original IPv6 packet will traverse when there may be multiple alternatives.¶

When an AERO node processes the SRH and forwards the original IPv6 packet to the correct OMNI interface, the OMNI interface writes the next IPv6 address from the SRH into the IPv6 destination address and decrements Segments Left. If decrementing would cause Segments Left to become 0, the OMNI interface deletes the SRH before forwarding. This form of Segment Routing supports Safety-Based Multilink (SBM).¶

3.4.1. AERO Proxy/Server and Relay Behavior

When a Proxy/Server enables an OMNI interface, it assigns a ULA-RND appropriate for the given OMNI link SRT segment. The Proxy/Server also configures secured tunnels and engages in BGP routing protocol sessions with one or more neighboring Gateways.¶

The OMNI interface provides a single interface abstraction to the IP layer, but internally includes an NBMA nexus for sending carrier packets to OMNI interface neighbors over underlay INET interfaces and secured tunnels. The Proxy/Server further configures a service to facilitate IPv6 ND exchanges with AERO Clients and manages per-Client neighbor cache entries and IP forwarding table entries based on control message exchanges.¶

Relays are simply Proxy/Servers that run a dynamic routing protocol to redistribute routes between the OMNI interface and INET/ENET interfaces (see: Section 3.2.3). The Relay provisions MNPs to networks on the INET/ENET interfaces (i.e., the same as a Client would do) and advertises the MSP(s) for the OMNI link over the INET/ENET interfaces. The Relay further provides an attachment point of the OMNI link to a non-MNP-based global topology.¶

3.4.2. AERO Client Behavior

When a Client enables an OMNI interface, it assigns either an XLA-MNP or a TLA and sends OMNI-encapsulated RS messages over its ANET/INET underlay interfaces to an FHS Proxy/Server, which coordinates with a Hub Proxy/Server that returns an RA message with corresponding parameters. The RS/RA messages may pass through one or more NATs in the path between the Client and FHS Proxy/Server. (Note: if the Client used a TLA in its initial RS messages, it may discover ULA-MNPs in the corresponding RAs that it receives from FHS Proxy/Servers and begin using these new addresses. If the Client is operating outside the context of AERO infrastructure such as in a Mobile Ad-hoc Network (MANET), however, it may continue using TLAs for Client-to-Client communications until it encounters an infrastructure element that can delegate MNPs.)¶

A Client can further extend the OMNI link over its (downstream) ENET interfaces where it provides a first-hop router for Hosts and other AERO Clients connected to the ENET. A downstream Client that connects via the ENET serviced by an upstream Client can in turn service further downstream ENETs that connect other Hosts and Clients. This OMNI link extension can be applied recursively over a "chain" of ENET Clients.¶

3.4.3. AERO Host Behavior

When a Host enables an OMNI interface, it assigns an address taken from the ENET underlay interface which may itself be a GUA delegated by the upstream Client. The Host does not assign a link-local address to the OMNI interface, since no autoconfiguration is necessary on that interface. (As an implementation matter, the Host could instead configure the "OMNI interface" as a virtual sublayer of the ENET underlay interface itself.)¶

The Host sends OMNI-encapsulated RS messages over its ENET underlay interface to the upstream Client, which returns encapsulated RAs and provides routing services in the same fashion that Proxy/Servers provides services for Clients. Hosts represent the leaf end systems in recursively-nested chain of concatenated ENETs, i.e., they represent terminating endpoints for the OMNI link.¶

3.4.4. AERO Gateway Behavior

AERO Gateways configure an OMNI interface and assign a ULA-RND and corresponding Subnet Router Anycast address for each OMNI link SRT segment they connect to. Gateways configure secured tunnels with Proxy/Servers in the same SRT segment and other Gateways in the same (or an adjacent) SRT segment. Gateways then engage in a BGP routing protocol session with neighbors over the secured spanning tree (see: Section 3.2.3).¶

3.5.1. OMNI ND Messages

OMNI interfaces prepare IPv6 ND messages the same as for standard IPv6 ND, but also include a new option type termed the OMNI option [I-D.templin-6man-omni]. OMNI interfaces do not use LLAs as IPv6 ND message source and destination addresses, but instead use ULAs. This allows for multiple different OMNI links to be joined into a single link at some future time without requiring a global renumbering event.¶

For each IPv6 ND message, OMNI interfaces include one or more OMNI options (and any other ND message options) then completely populate all option information. If the OMNI interface includes an authentication signature, it sets the IPv6 ND message Checksum field to 0 and calculates the authentication signature over the entire length of the OAL packet or super-packet (beginning with a pseudo-header of the IPv6 header) but does not then proceed to calculate the IPv6 ND message checksum itself. Otherwise, the OMNI interface calculates the standard IPv6 ND message checksum over the OAL packet or super-packet and writes the value in the Checksum field. OMNI interfaces verify authentication and integrity of each IPv6 ND message received according to the specific check(s) included, and process the message further only following verification.¶

OMNI options include per-neighbor information that provides multilink forwarding, link-layer address and traffic selector information for the neighbor's underlay interfaces. This information is stored in the neighbor cache and provides the basis for the forwarding algorithm specified in Section 3.10. The information is cumulative and reflects the union of the OMNI information from the most recent IPv6 ND messages received from the neighbor; it is therefore not required that each IPv6 ND message contain all neighbor information.¶

The OMNI option is distinct from any Source/Target Link-Layer Address Options (S/TLLAOs) that may appear in an IPv6 ND message according to the appropriate IPv6 over specific link layer specification (e.g., [RFC2464]). If both an OMNI option and S/TLLAO appear, the former pertains to encapsulation addresses while the latter pertains to the native L2 address format of the underlay media.¶

OMNI interface IPv6 ND messages may also include other IPv6 ND options. In particular, solicitation messages may include a Nonce option if required for verification of advertisement replies. If an OMNI IPv6 ND solicitation message includes a Nonce option, the advertisement reply must echo the same Nonce. If an OMNI IPv6 ND advertisement message includes a Timestamp option, the recipient should check the Timestamp to determine if the message is current.¶

AERO Clients send RS messages to the link-scoped All-Routers multicast address or a ULA-RND while using unicast or anycast OAL/L2 addresses. AERO Proxy/Servers respond by returning unicast RA messages. During the RS/RA exchange, AERO Clients and Proxy/Servers include state synchronization parameters to establish Identification windows and other state.¶

AERO Hosts and Clients on ENET underlay networks send RS messages to the link-scoped All-Routers multicast address, a ULA-RND of a remote Hub Proxy/Server or the ULA-MNP of an upstream Client while using unicast or anycast OAL/L2 addresses. The upstream AERO Client responds by returning a unicast RA message.¶

AERO nodes use NS/NA messages for the following purposes:¶

• NS/NA(AR) messages are used for address resolution and optionally to establish sequence number windows. The ARS sends an NS(AR) to the solicited-node multicast address of the ART, and an ARR with addressing information for the ART returns a unicast NA(AR) that contains current, consistent and authentic target address resolution information. NS/NA(AR) messages must be secured.¶
• NS/NA(NUD) messages are used determine target reachability and optionally to also establish MFIB state. The source sends an NS(NUD) to the unicast address of the target while optionally including an OMNI Multilink Forwarding Parameters (MFP) sub-option naming a specific underlay interface pair, and the target returns a unicast NA(NUD) that echoes the MFP (if present). NS/NA(NUD) messages that use an in-window sequence number and do not update any other state need not include an authentication signature but instead must include an IPv6 ND message checksum. NS/NA(NUD) used to establish window synchronization and/or MFIB state must be secured.¶
• Unsolicited NA (uNA) messages are used to signal addressing and/or other neighbor state changes (e.g., address changes due to mobility, signal degradation, traffic selector updates, etc.). uNA messages that update state information must be secured.¶
• NS/NA(DAD) messages are not used in AERO, since Duplicate Address Detection is not required.¶

Additionally, nodes may set the OMNI option PNG flag in NA/RA messages to receive a uNA response from the neighbor. The uNA response MUST set the ACK flag (without also setting the SYN or PNG flags) with the Acknowledgement field set to the Identification used in the PNG message.¶

3.5.2. OMNI Neighbor Advertisement Message Flags

As discussed in Section 4.4 of [RFC4861] NA messages include three flag bits R, S and O. OMNI interface NA messages treat the flags as follows:¶

• R: The R ("Router") flag is set to 1 in the NA messages sent by all AERO/OMNI node types. Simple hosts that would set R to 0 do not occur on the OMNI link itself, but may occur on the downstream links of Clients and Relays.¶
• S: The S ("Solicited") flag is set exactly as specified in Section 4.4. of [RFC4861], i.e., it is set to 1 for Solicited NAs and set to 0 for uNAs (both unicast and multicast).¶
• O: The O ("Override") flag is set to 0 for solicited NAs returned by a Proxy/Server ARR and set to 1 for all other solicited and unsolicited NAs. For further study is whether solicited NAs for anycast targets apply for OMNI links. Since XLA-MNPs must be uniquely assigned to Clients to support correct IPv6 ND protocol operation, however, no role is currently seen for assigning the same XLA-MNP to multiple Clients.¶

3.5.3. OMNI Neighbor Window Synchronization

In secured environments (e.g., between secured spanning tree neighbors, between neighbors on the same secured ANET, etc.), OMNI interface neighbors can exchange OAL packets using randomly-initialized and monotonically-increasing Identification values (modulo 2**32) without window synchronization. In environments where spoofing is considered a threat, OMNI interface neighbors instead invoke window synchronization by including OMNI Neighbor Coordination sub-options in NS/NA message exchanges to maintain send/receive window state in their respective neighbor cache entries as specified in [I-D.templin-6man-omni].¶

3.10.1. Host Forwarding Algorithm

When an original IP packet enters a Host's OMNI interface from the network layer the Host searches for a NCE that matches the destination. If there is a matching NCE, the Host performs L2 encapsulation, fragments the encapsulated packet if necessary and forwards the packets into the ENET addressed to the L2 address of the neighbor.¶

After sending a packet, the Host may receive a Redirect message from its upstream Client to inform it of another AERO node on the same ENET that would provide a better first hop. The Host authenticates the Redirect message, then updates its neighbor cache accordingly.¶

3.10.2. Client Forwarding Algorithm

When an original IP packet enters a Client's OMNI interface from the network layer the Client searches for a NCE that matches the destination. If there is a matching NCE on an ANET/INET interface (i.e., an upstream interface), the Client selects one or more "reachable" neighbor interfaces in the entry for forwarding purposes. Otherwise, the Client invokes address resolution and multilink forwarding procedures per Section 3.13. If there is a matching NCE on an ENET interface (i.e., a downstream interface), the Client instead performs OAL and/or L2 encapsulation and forwards the packet to the downstream Host or Client.¶

When a carrier packet enters a Client's OMNI interface from the link-layer, if the OAL destination matches one of the Client's ULAs the Client (acting as an OAL destination) verifies that the Identification is in-window for this OAL source, then reassembles and decapsulates as necessary and delivers the original IP packet to the network layer. If the OAL destination matches a NCE for a Host or Client on an ENET interface, the Client instead forwards the carrier packet to the Host/Client. If the OAL destination does not match, the Client drops the original IP packet and MAY return a network-layer ICMP Destination Unreachable message subject to rate limiting (see: Section 3.11).¶

When a Client forwards a carrier packet from an ENET Host to a neighbor connected to the same ENET, it also returns a Redirect message to inform the source that it can reach the neighbor directly as an ENET peer.¶

Note: Clients and their FHS Proxy/Server (and other Client) peers can exchange original IP packets over ANET underlay interfaces without invoking the OAL, since the ANET is secured at the link and physical layers. By forwarding original IP packets without invoking the OAL, however, the ANET peers can engage only in classical path MTU discovery since the packets are subject to loss and/or corruption due to the various per-link MTU limitations that may occur within the ANET. Moreover, the original IP packets do not include either the OAL integrity check or per-packet Identification values that can be used for data origin authentication and link-layer retransmissions. The tradeoff therefore involves an assessment of the per-packet encapsulation overhead saved by bypassing the OAL vs. inheritance of classical network "brittleness". (Note however that ANET peers can send small original IP packets without invoking the OAL, while invoking the OAL for larger packets. This presents the beneficial aspects of both small packet efficiency and large packet robustness, with delay variance and reordering as possible side effects.)¶

Note: The forwarding table entries established in peer Clients of a multihop forwarding region are based on ULA-MNPs and/or TLAs used to seed the multihop routing protocols. When ULA-MNPs are used, the ULA /64 prefix provides topological relevance for the multihop forwarding region, while the 64-bit Interface Identifier encodes the Client MNP. Therefore, Clients can forward atomic fragments with compressed OAL headers that do not include ULA or MFVI information by examining the MNP-based addresses in the actual IP packet header. In other words, each forwarding table entry contains two pieces of forwarding information - the ULA information in the prefix and the MNP information in the interface identifier.¶

3.10.3. Proxy/Server and Relay Forwarding Algorithm

When a Proxy/Server receives an original IP packet from the network layer, it drops the packet if routing indicates that it should be forwarded back to the network layer to avoid looping. Otherwise, the Proxy/Server regards the original IP packet the same as if it had arrived as carrier packets with OAL destination set to its own ULA. When the Proxy/Server receives carrier packets on underlay interfaces with OAL destination set to its own ULA, it performs OAL reassembly if necessary to obtain the original IP packet. The Proxy/Server then supports multilink forwarding procedures as specified in Section 3.13.2 and/or acts as an ARS to initiate address resolution as specified in Section 3.13.¶

When the Proxy/Server receives a carrier packet with OAL destination set to a ULA-MNP that does not match the MSP, it accepts the carrier packet only if data origin authentication succeeds and if there is a network layer routing table entry for a GUA route that matches the ULA-MNP. If there is no route, the Proxy/Server drops the carrier packet; otherwise, it reassembles and decapsulates to obtain the original IP packet then acts as a Relay to present it to the network layer where it will be delivered according to standard IP forwarding.¶

When a Proxy/Server receives a carrier packet from one of its Client neighbors with OAL destination set to another node, it forwards the packets via a matching NCE or via the spanning tree if there is no matching entry. When the Proxy/Server receives a carrier packet with OAL destination set to a ULA-MNP of one of its Client neighbors established through RS/RA exchanges, it accepts the carrier packet only if data origin authentication succeeds. If the NCE state is DEPARTED, the Proxy/Server changes the OAL destination address to the ULA of the new Proxy/Server, then re-encapsulates the carrier packet and forwards it to a Gateway which will eventually deliver it to the new Proxy/Server. If the neighbor cache state for the Client is REACHABLE, the Proxy/Server forwards the carrier packets to the Client which then must reassemble. (Note that the Proxy/Server does not reassemble carrier packets not explicitly addressed to its own ULA, since some of the carrier packets of the same original IP packet could be forwarded through a different Proxy/Server.) In that case, the Client may receive fragments from multiple Proxy/Servers, but that can still be reassembled.¶

Proxy/Servers process carrier packets with OAL destinations that do not match their ULA in the same manner as for traditional IP forwarding within the OAL, i.e., nodes use IP forwarding to forward packets not explicitly addressed to themselves. (Proxy/Servers include a special case that accepts and reassembles carrier packets destined to a ULA-MNP of one of their Clients received over the secured spanning tree.) Proxy/Servers process carrier packets with their ULA as the destination by first examining the packet for a CRH-32 header or an OCH header. In that case, the Proxy/Server examines the next MFVI in the carrier packet to locate the MFV entry in the MFIB for next hop forwarding (i.e., without examining IP addresses). When the Proxy/Server forwards the carrier packet, it changes the destination address according to the MFVI value for the next hop found either in the CRH-32 header or in the node's own MFIB. Proxy/Servers must verify that the L2 addresses of carrier packets not received from the secured spanning tree are "trusted" before forwarding according to an MFV (otherwise, the carrier packet must be dropped).¶

Note: Proxy/Servers may receive carrier packets addressed to their own ULA with CRH-32s that include additional forwarding information. Proxy/Servers use the forwarding information to determine the correct NCE and underlay interface for forwarding to the target Client, then remove the CRH-32 and forward the carrier packet. If necessary, the Proxy/Server reassembles first before re-encapsulating (and possibly also re-fragmenting) then forwards to the target Client.¶

Note: Clients and their FHS Proxy/Server peers can exchange original IP packets over ANET underlay interfaces without invoking the OAL, since the ANET is secured at the link and physical layers. By forwarding original IP packets without invoking the OAL, however, the Client and Proxy/Server can engage only in classical path MTU discovery since the packets are subject to loss and/or corruption due to the various per-link MTU limitations that may occur within the ANET. Moreover, the original IP packets do not include either the OAL integrity check or per-packet Identification values that can be used for data origin authentication and link-layer retransmissions. The tradeoff therefore involves an assessment of the per-packet encapsulation overhead saved by bypassing the OAL vs. inheritance of classical network "brittleness". (Note however that ANET peers can send small original IP packets without invoking the OAL, while invoking the OAL for larger packets. This presents the beneficial aspects of both small packet efficiency and large packet robustness.)¶

Note: When a Proxy/Server receives a (non-OAL) original IP packet from an ANET Client, or a carrier packet with OAL destination set to its own ULA from any Client, the Proxy/Server reassembles if necessary then performs ARS functions on behalf of the Client. The Client may at some later time begin sending carrier packets to the OAL address of the actual target instead of the Proxy/Server, at which point it may begin functioning as an ARS on its own behalf and thereby "override" the Proxy/Server's ARS role.¶

Note; Proxy/Servers drop any original IP packets (received either directly from an ANET Client or following reassembly of carrier packets received from an ANET/INET Client) with a destination that corresponds to the Client's delegated MNP. Similarly, Proxy/Servers drop any carrier packet received with both a source and destination that correspond to the Client's delegated MNP regardless of their OMNI link point of origin. These checks are necessary to prevent Clients from either accidentally or intentionally establishing endless loops that could congest Proxy/Servers and/or ANET/INET links.¶

Note: Proxy/Servers forward secure control plane carrier packets via the SRT secured spanning tree and forward other carrier packets via the unsecured spanning tree. When a Proxy/Server receives a carrier packet from the secured spanning tree, it considers the message as authentic without having to verify upper layer authentication signatures. When a Proxy/Server receives a carrier packet from the unsecured spanning tree, it applies data origin authentication itself and/or forwards the unsecured message toward the destination which must apply data origin authentication on its own behalf.¶

Note: If the Proxy/Server has multiple original IP packets to send to the same neighbor, it can concatenate them in a single OAL super-packet [I-D.templin-6man-omni].¶

3.10.4. Gateway Forwarding Algorithm

Gateways forward spanning tree carrier packets while decrementing the OAL header Hop Count but not the original IP header Hop Count/TTL. Gateways convey carrier packets that encapsulate critical IPv6 ND control messages or routing protocol control messages via the SRT secured spanning tree, and may convey other carrier packets via the secured/unsecured spanning tree or via more direct paths according to MFIB information. When the Gateway receives a carrier packet, it removes the L2 headers and searches for an MFIB entry that matches an MFVI or an IP forwarding table entry that matches the OAL destination address.¶

Gateways process carrier packets with OAL destinations that do not match their ULA or the SRT Subnet Router Anycast address in the same manner as for traditional IP forwarding within the OAL, i.e., nodes use IP forwarding to forward packets not explicitly addressed to themselves. Gateways process carrier packets with their ULA or the SRT Subnet Router Anycast address as the OAL destination by first examining the packet for a full OAL header with a CRH-32 extension or an OCH header. In that case, the Gateway examines the next MFVI in the carrier packet to locate the MFV entry in the MFIB for next hop forwarding (i.e., without examining IP addresses). When the Gateway forwards the carrier packet, it changes the OAL destination address according to the MFVI value for the next hop found either in the CRH-32 header or in the node's own MFIB. If the Gateway has a NCE for the target Client with an entry for the target underlay interface and current L2 addresses, the Gateway instead forwards directly to the target Client while using the final hop MFVI instead of the next hop (see: Section 3.13.4).¶

Gateways forward carrier packets received from a first segment via the secured spanning tree to the next segment also via the secured spanning tree. Gateways forward carrier packets received from a first segment via the unsecured spanning tree to the next segment also via the unsecured spanning tree. Gateways use a single IPv6 routing table that always determines the same next hop for a given OAL destination, where the secured/unsecured spanning tree is determined through the selection of the underlay interface to be used for transmission (i.e., a secured tunnel or an open INET interface).¶

As for Proxy/Servers, Gateways must verify that the L2 addresses of carrier packets not received from the secured spanning tree are "trusted" before forwarding according to an MFV (otherwise, the carrier packet must be dropped).¶

3.12.1. AERO Service Model

Each AERO Proxy/Server on the OMNI link is configured to facilitate Client prefix delegation/registration requests. Each Proxy/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 OMNI link and securely distributed to all Proxy/Servers, e.g., via the Lightweight Directory Access Protocol (LDAP) [RFC4511], via static configuration, etc. Clients receive the same service regardless of the Proxy/Servers they select.¶

Clients associate each of their ANET/INET underlay interfaces with a FHS Proxy/Server. Each FHS Proxy/Server locally services one or more of the Client's underlay interfaces, and the Client typically selects one among them to serve as the Hub Proxy/Server (the Client may instead select a "third-party" Hub Proxy/Server that does not directly service any of its underlay interfaces). All of the Client's other FHS Proxy/Servers forward proxyed copies of RS/RA messages between the Hub Proxy/Server and Client without assuming the Hub role functions themselves.¶

Each Client associates with a single Hub Proxy/Server at a time, while all other Proxy/Servers are candidates for providing the Hub role for other Clients. An FHS Proxy/Server assumes the Hub role when it receives an RS message with its own ULA or link-scoped All-Routers multicast as the destination. An FHS Proxy/Server assumes the proxy role when it receives an RS message with the ULA of another Proxy/Server as the destination. (An FHS Proxy/Server can also assume the proxy role when it receives an RS message addressed to link-scoped All-Routers multicast if it can determine the ULA of another Proxy/Server to serve as a Hub.)¶

Hosts and Clients on ENET interfaces associate with an upstream Client on the ENET the same as a Client would associate with an ANET Proxy/Server. In particular, the Host/Client sends an RS message via the ENET which directs the message to the upstream Client. The upstream Client returns an RA message. In this way, the downstream nodes see the ENET as an ANET and see the upstream Client as a Proxy/Server for that ANET.¶

AERO Hosts, Clients and Proxy/Servers use IPv6 ND messages to maintain neighbor cache entries. AERO Proxy/Servers configure their OMNI interfaces as advertising NBMA 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 Proxy/Server state alive.¶

AERO Clients and Hub Proxy/Servers include prefix delegation and/or registration parameters in RS/RA messages. The IPv6 ND messages are exchanged between the Client and Hub Proxy/Server (via any FHS Proxy/Servers acting as proxys) according to the prefix management schedule required by the service. If the Client knows its MNP in advance, it can employ prefix registration by including its XLA-MNP as the source address of an RS message and with an OMNI option with valid prefix registration information for the MNP. If the Hub Proxy/Server accepts the Client's MNP assertion, it injects the MNP into the routing system and establishes the necessary neighbor cache state. If the Client does not have a pre-assigned MNP, it can instead employ prefix delegation by including a TLA as the source address of an RS message and with an OMNI option with prefix delegation parameters to request an MNP.¶

The following sections outlines Host, Client and Proxy/Server behaviors based on the Router Discovery and Prefix Registration specifications found in Section 15 of [I-D.templin-6man-omni]. These sections observe all of the OMNI specifications, and include additional specifications of the interactions of Client-Proxy/Server RS/RA exchanges with the AERO mobility service.¶

3.12.2. AERO Host and Client Behavior

AERO Hosts and Clients discover the addresses of candidate Proxy/Servers by resolving the Potential Router List (PRL) in a similar manner as described in [RFC5214]. Discovery methods include static configuration (e.g., a flat-file map of Proxy/Server addresses and locations), or through an automated means such as Domain Name System (DNS) name resolution [RFC1035]. Alternatively, the Host/Client can discover Proxy/Server addresses through a layer 2 data link login exchange, or through an RA response to a multicast/anycast RS as described below. In the absence of other information, the Host/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 OMNI link (e.g., "example.com"). The name resolution returns a set of resource records with Proxy/Server address information.¶

The Host/Client then performs RS/RA exchanges over each of its underlay interfaces to associate with (possibly multiple) FHS Proxy/Serves and a single Hub Proxy/Server as specified in Section 15 of [I-D.templin-6man-omni]. The Host/Client then sends each RS (either directly via Direct interfaces, via a VPN for VPNed interfaces, via an access router for ANET interfaces or via INET encapsulation for INET interfaces) and waits up to RetransTimer milliseconds for an RA message reply (see Section 3.12.3) while retrying up to MAX_RTR_SOLICITATIONS if necessary. If the Host/Client receives no RAs, or if it receives an RA with Router Lifetime set to 0, the Client SHOULD abandon attempts through the first candidate Proxy/Server and try another Proxy/Server.¶

After the Host/Client registers its underlay interfaces, it may wish to change one or more registrations, e.g., if an interface changes address or becomes unavailable, if traffic selectors change, etc. To do so, the Host/Client prepares an RS message to send over any available underlay interface as above. The RS includes an OMNI option with prefix registration/delegation information and with an Interface Attributes sub-option specific to the selected underlay interface. When the Host/Client receives the Hub Proxy/Server's RA response, it has assurance that both the Hub and FHS Proxy/Servers have been updated with the new information.¶

If the Host/Client wishes to discontinue use of a Hub Proxy/Server it issues an RS message over any underlay interface with an OMNI option with a prefix release indication (i.e., by setting the OMNI Neighbor Coordination sub-option Preflen to 0). When the Hub Proxy/Server processes the message, it releases the MNP, sets the NCE state for the Host/Client to DEPARTED and returns an RA reply with Router Lifetime set to 0. After a short delay (e.g., 2 seconds), the Hub Proxy/Server withdraws the MNP from the routing system. (Alternatively, when the Host/Client associates with a new FHS/Hub Proxy/Server it can include an OMNI "Proxy/Server Departure" sub-option in RS messages with the ULAs of the Old FHS/Hub Proxy/Servers.)¶

3.12.3. AERO Proxy/Server Behavior

AERO Proxy/Servers act as both IP routers and IPv6 ND proxys, and support a prefix delegation/registration service for Clients. Proxy/Servers arrange to add their ULAs to the PRL maintained in a static map of Proxy/Server addresses for the link, the DNS resource records for the FQDN "linkupnetworks.[domainname]", etc. before entering service. The PRL should be arranged such that Clients can discover the addresses of Proxy/Servers that are geographically and/or topologically "close" to their underlay network connections.¶

When a FHS/Hub Proxy/Server receives a prospective Client's RS message, 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, it processes the RS as specified in Section 15 of [I-D.templin-6man-omni]. When the Hub Proxy/Server receives the RS, it determines the correct MNPs to provide to the Client by processing the XLA-MNP prefix parameters and/or the DHCPv6 OMNI sub-option. When the Hub Proxy/Server returns the MNPs, it also creates an XLA-MNP forwarding table entry for the MNP resulting in a BGP update (see: Section 3.2.3). The Hub Proxy/Server then returns an RA to the Client with destination set to the source of the RS (if an FHS Proxy/Server on the return path proxys the RA, it changes the destination to the Client's ULA-MNP).¶

After the initial RS/RA exchange, the Hub Proxy/Server maintains a ReachableTime timer for each of the Client's underlay interfaces individually (and for the Client's NCE collectively) set to expire after ReachableTime seconds. If the Client (or an FHS Proxy/Server) issues additional RS messages, the Hub Proxy/Server sends an RA response and resets ReachableTime. If the Hub Proxy/Server receives an IPv6 ND message with a prefix release indication it sets the Client's NCE to the DEPARTED state and withdraws the MNP route from the routing system after a short delay (e.g., 2 seconds). If ReachableTime expires before a new RS is received on an individual underlay interface, the Hub Proxy/Server marks the interface as DOWN. If ReachableTime expires before any new RS is received on any individual underlay interface, the Hub Proxy/Server sets the NCE state to STALE and sets a 10 second timer. If the Hub Proxy/Server has not received a new RS or uNA message with a prefix release indication before the 10 second timer expires, it deletes the NCE and withdraws the XLA-MNP from the routing system.¶

The Hub Proxy/Server processes any IPv6 ND messages pertaining to the Client while forwarding to the Client or responding on the Client's behalf as necessary. The Hub Proxy/Server may also issue unsolicited RA messages, e.g., with reconfigure parameters to cause the Client to renegotiate its prefix delegation/registrations, with Router Lifetime set to 0 if it can no longer service this Client, etc. The Hub Proxy/Server may also receive carrier packets via the secured spanning tree that contain initial data packets sent while route optimization is in progress. The Hub Proxy/Server reassembles, then re-encapsulates/re-fragments and forwards the packets to the target Client via an FHS Proxy/Server if necessary. Finally, If the NCE is in the DEPARTED state, the old Hub Proxy/Server forwards any carrier packets it receives from the secured spanning tree and destined to the Client to the new Hub Proxy/Server, then deletes the entry after DepartTime expires.¶

Note: Clients SHOULD arrange to notify former Hub Proxy/Servers of their departures, but Hub Proxy/Servers are responsible for expiring neighbor cache entries and withdrawing XLA-MNP routes even if no departure notification is received (e.g., if the Client leaves the network unexpectedly). Hub Proxy/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 RS/RA messaging between Clients and FHS Proxy/Servers will keep any NAT state alive (see above).¶

Note: All Proxy/Servers on an OMNI 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 Proxy/Servers on the same link advertised different values.¶

3.13.1. Multilink Address Resolution

When one or more original IP packets from a source node destined to a target node arrives, the ARS checks for a NCE with an XLA-MNP that matches the target destination. If there is a NCE in the REACHABLE state, the ARS invokes the OAL and forwards the resulting carrier packets according to the cached state then returns from processing. Otherwise, if there is no NCE the ARS creates one in the INCOMPLETE state.¶

The ARS next prepares an NS message for Address Resolution (NS(AR)) to send toward an ART while including the original IP packet(s) as trailing data following the NS(AR) in an OAL super-packet [I-D.templin-6man-omni]. The resulting NS(AR) message must be sent securely, and includes:¶

• the ULA of the ARS as the source address.¶
• the XLA corresponding to the original IP packet's destination as the Target Address, e.g., for 2001:db8:1:2::10:2000 the Target Address is fd00::2001:db8:1:2.¶
• the Solicited-Node multicast address [RFC4291] formed from the lower 24 bits of the original IP packet's destination as the destination address, e.g., for 2001:db8:1:2::10:2000 the NS(AR) destination address is ff02:0:0:0:0:1:ff10:2000.¶

The NS(AR) message also includes an OMNI option with an authentication sub-option if necessary and with an OMNI Neighbor Coordination sub-option with Preflen set to the prefix length associated with the NS(AR) source. The ARS also includes Interface Attributes and Traffic Selectors for all of the source Client's underlay interfaces, calculates the authentication signature or checksum, then selects an Identification value and submits the NS(AR) message for OAL encapsulation with OAL source set to its own ULA and OAL destination set to the XLA-MNP corresponding to the ART and with window synchronization parameters. The ARS then inserts a fragment header, calculates the OAL checksum performs fragmentation and L2 encapsulation, then sends the resulting carrier packets into the SRT secured spanning tree without decrementing the network-layer TTL/Hop Limit field.¶

When the ARS is a Client, it must instead use the ULA of one of its FHS Proxy/Servers as the OAL destination. The ARS Client then fragments, performs L2 encapsulation and forwards the carrier packets to the FHS Proxy/Server. The FHS Proxy/Server then reassembles, verifies the NS(AR) authentication signature or checksum, changes the OAL source to its own ULA, changes the OAL destination to the XLA-MNP corresponding to the ART, selects an appropriate Identification, calculates the OAL checksum, then re-fragments and forwards the resulting carrier packets into the secured spanning tree on behalf of the Client.¶

Note: both the ART Client/Relay and its Hub Proxy/Server include current and accurate information for the ART's multilink Interface Attributes profile. The Hub Proxy/Server can be trusted to provide an authoritative ARR response on behalf of the ART should the need arise. While the ART has no such trust basis, any attempt by the ART to mount an attack by providing false Interface Attributes information would only result in black-holing of return traffic, i.e., the "attack" could only result in denial of service to the ART itself. Therefore, the ART's asserted Interface Attributes need not be validated by the Hub Proxy/Server.¶

3.13.2. Multilink Forwarding

Following address resolution, the ARS and ART can assert multilink forwarding paths through underlay interface pairs serviced by the same source/destination ULAs by sending unicast NS/NA messages with OMNI Multilink Forwarding Parameter (MFP) and/or Neighbor Coordination sub-options. The unicast NS/NA messages establish multilink forwarding state in OAL intermediate nodes in the path between the ARS and ART. Note that either the ARS or ART can independently initiate multilink forwarding by sending unicast NS messages on behalf of specific underlay interface pairs.¶

Nodes that configure OMNI interfaces include an additional forwarding table termed the Multilink Forwarding Information Base (MFIB) that supports carrier packet forwarding based on OMNI neighbor underlay interface pairs. The MFIB contains Multilink Forwarding Vectors (MFVs) identified by locally-unique 4-octet values known as MFV Indexes (MFVIs). The MVFs cache uncompressed OAL header information as well as the previous/next-hop addressing and MFVI information.¶

OAL source, intermediate and target nodes create MFVs/MFVIs when they process an NS message with an MFP sub-option with Job code '00' (Initialize; Build B) or a solicited NA message with Job code '01' (Follow B; Build A) (see: [I-D.templin-6man-omni]). The OAL source of the NS (which is also the OAL destination of the solicited NA) is considered to reside in the "First Hop Segment (FHS)", while the OAL destination of the NS (which is also the OAL source of the solicited NA) is considered to reside in the "Last Hop Segment (LHS)".¶

When an OAL node initiates or forwards an NS with Job code '00', it creates an MFV, records the NS source and destination ULAs then generates and assigns a locally-unique "B" MFVI (while also caching the "B" values for all previous OAL hops on the path from the FHS OAL source). When the OAL node receives future carrier packets that include "B", it can unambiguously locate the correct MFV and determine directionality without examining addresses. When the MFV is indexed by its "B" MFVI, it returns the ULAs in (dst,src) order the opposite of how they appeared in the OAL header of the original NS to support full header reconstruction for reverse-path forwarding. (If the NS message included a nested OAL encapsulation, the ULAs of both OAL headers are returned.)¶

When an OAL node initiates or forwards a solicited NA with Job code '01', it uses the "B" MFVI to locate the MFV created by the NS then generates and assigns a locally-unique "A" MFVI (while also caching the "A" values for all previous OAL hops on the path from the LHS OAL source). When the OAL node receives future carrier packets that include "A", it can unambiguously locate the correct MFV and determine directionality without examining addresses. When the MFV is indexed by its "A" MFVI, it returns the ULAs in (src,dst) order the same as they appeared in the OAL header of the original NS to support full header reconstruction for forward-path forwarding. (If the NS message included a nested OAL encapsulation, the ULAs of both OAL headers are returned.)¶

OAL nodes generate random non-zero 32-bit values as candidate MFVIs which must first be tested for local uniqueness. If a candidate MFVI s already in use, the OAL node repeats the random generation process until it obtains a unique non-zero value. Since the number of MFVs in service at each OAL node is likely to be much smaller than 2**32, the process will generate a unique value after a small number of tries. Since the uniqueness property is node-local only, an MFVI locally generated by a first OAL node must not be tested for uniqueness by other OAL nodes.¶

OAL nodes cache MFVs for up to ReachableTime seconds following their initial creation. If the node processes another NS or NA message specific to an MFV, it resets ReachableTime to REACHABLE_TIME seconds, i.e., the same as for NCEs. If ReachableTime expires, the node deletes the MFV and frees its associated MFVIs so they can be reused for future MFVs.¶

The following sections provide the detailed specifications of these NS/NA exchanges for all nodes along the forward and reverse paths.¶

3.13.3. Rapid Commit Route Optimization

When the LHS peer receives an NS(AR) with a set of Interface Attributes for the source Client, it can perform "rapid commit" by immediately invoking multilink route optimization as above instead of returning an NA(AR). In order to perform rapid commit, the LHS peer prepares a unicast NS message with an OMNI option with window synchronization information responsive to the NS(AR), with an MFP sub-option selected for a specific underlay interface pair and with Interface Attributes for all of the LHS peer's other underlay interfaces. The LHS peer can also include ordinary IP packets as OAL super-packet extensions to the NS message if it has immediate data to send to the FHS peer. The LHS peer then returns the NS to the FHS peer the same as for the NA(AR) case.¶

When the NS message traverses the return path to the FHS peer, all intermediate nodes in the path establish state exactly the same as for an ordinary NS/NA multilink route optimization exchange. When the NS message arrives at the FHS peer, the window synchronization parameters confirm that the NS is taking the place of the NA(AR), thereby eliminating an extraneous message transmission and associated delay. The FHS peer then completes the route optimization by returning a responsive NA.¶

Note: The LHS peer must accept unicast NS messages with an ACK matching the SYN included in the NS(AR) as an equivalent message replacement for the NA(AR). Address resolution and multilink forwarding coordination can therefore be coordinated in a single three-way handshake connection with minimal messaging and delay (i.e., as opposed to a four-message exchange).¶

3.13.4. Client/Gateway Route Optimization

Following multilink route optimization for specific underlay interface pairs, FHS/LHS Clients located on open INETs can invoke Client/Gateway route optimization to improve performance and reduce load and congestion on their respective Proxy/Servers. To initiate Client/Gateway route optimization, the Client prepares an NS message with its own XLA address as the source and the ULA of its Gateway as the destination while creating a NCE for the Gateway if necessary. The NS message must be no larger than the minimum MPS and encapsulated as an atomic fragment.¶

The Client then includes an Interface Attributes sub-option for its underlay interface as well as an authentication signature but does not include window synchronization parameters. The Client then performs OAL encapsulation with its own ULA-MNP as the source and the ULA of the Gateway as the destination while including a randomly-chosen Identification value, then performs L2 encapsulation on the atomic fragment and sends the resulting carrier packet directly to the Gateway.¶

When the Gateway receives the carrier packet, it verifies the authentication signature then creates a NCE for the Client. The Gateway then caches the L2 encapsulation addresses (which may have been altered by one or more NATs on the path) as well as the Interface Attributes for this Client omIndex, and marks this Client underlay interface as "trusted". The Gateway then prepares an NA reply with its own ULA as the source and the XLA of the Client as the destination where the NA again must be no larger than the minimum MPS.¶

The Gateway then echoes the Client's Interface Attributes, includes an Origin Indication with the Client's observed L2 addresses and includes an authentication signature. The Gateway then performs OAL encapsulation with its own ULA as the source and the ULA-MNP of the Client as the destination while using the same Identification value that appeared in the NS, then performs L2 encapsulation on the atomic fragment and sends the resulting carrier packet directly to the Client.¶

When the Client receives the NA reply, it caches the carrier packet L2 source address information as the Gateway target address via this underlay interface while marking the interface as "trusted". The Client also caches the Origin Indication L2 address information as its own (external) source address for this underlay interface.¶

After the Client and Gateway have established NCEs as well as "trusted" status for a particular underlay interface pair, each node can begin forwarding ordinary carrier packets intended for this multilink route optimization directly to one another while omitting the Proxy/Server from the forwarding path while the status is "trusted". The NS/NA messaging will have established the correct state in any NATs in the path so that NAT traversal is naturally supported. The Client and Gateway must maintain a timer that watches for activity on the path; if no carrier packets and/or NS/NA messages are sent or received over the path before NAT state is likely to have expired, the underlay interface pair status becomes "untrusted".¶

Thereafter, when the Client forwards a carrier packet with an MFVI toward the Gateway as the next hop, the Client uses the MFVI for the Gateway (discovered during multilink route optimization) instead of the MFVI for its Proxy/Server; the Gateway will accept the packet from the Client if and only if the MFVI matches the correct MFV and the underlay interface status is trusted. (The same is true in the reverse direction when the Gateway sends carrier packets directly to the Client.)¶

Note that the Client and Gateway each maintain a single NCE, but that the NCE may aggregate multiple underlay interface pairs. Each underlay interface pair may use differing source and target L2 addresses according to NAT mappings, and the "trusted/untrusted" status of each pair must be tested independently. When no "trusted" pairs remain, the NCE is deleted.¶

Note that the above method requires Gateways to participate in NS/NA message authentication signature application and verification. In an alternate approach, the Client could instead exchange NS/NA messages with authentication signatures via its Proxy/Server but addressed to the ULA of the Gateway, and the Proxy/Server and Gateway could relay the messages over the secured spanning tree. However, this would still require the Client to send additional messages toward the L2 address of the Gateway to populate NAT state; hence the savings in complexity for Gateways would result in increased message overhead for Clients.¶

3.13.5. Client/Client Route Optimization

When the FHS/LHS Clients are both located on the same SRT segment, Client-to-Client route optimization is possible following the establishment of any necessary state in NATs in the path. Both Clients will have already established state via their respective shared segment Proxy/Servers (and possibly also the shared segment Gateway) and can begin forwarding packets directly via NAT traversal while avoiding any Proxy/Server and/or Gateway hops.¶

When the FHS/LHS Clients on the same SRT segment perform the initial NS/NA exchange to establish MFIB state, they also include an Origin Indication (i.e., in addition to an MFP sub-option) with the mapped addresses discovered during the RS/RA exchanges with their respective Proxy/Servers. After the MFV paths have been established, both Clients can begin sending packets via strict MFV paths while establishing a direct path for Client-to-Client route optimization.¶

To establish the direct path, either Client (acting as the source) transmits a bubble to the mapped L2 address for the target Client which primes its local chain of NATs for reception of future packets from that L2 address (see: [RFC4380] and [I-D.templin-6man-omni]). The source Client then prepares an NS message with its own XLA as the source, with the XLA of the target as the destination and with an OMNI option with an Interface Attributes sub-option. The source Client then encapsulates the NS in an OAL header with its own ULA-MNP as the source, with the ULA-MNP of the target Client as the destination and with an in-window Identification for the target. The source Client then fragments and encapsulates in L2 headers addressed to its Proxy/Server then forwards the resulting carrier packets to the Proxy/Server.¶

When the Proxy/Server receives the carrier packets, it re-encapsulates and forwards them as unsecured carrier packets according to MFIB state where they will eventually arrive at the target Client which can verify that the identifications are within the acceptable window and reassemble if necessary. Following reassembly, the target Client prepares an NA message with its own XLA as the source, with the XLA of the source Client as the destination and with an OMNI option with an Interface Attributes sub-option. The target Client then encapsulates the NA in an OAL header with its own ULA-MNP as the source, with the ULA-MNP of the source Client as the destination and with an in-window Identification for the source Client. The target Client then fragments and encapsulates in L2 headers addressed to the source Client's Origin addresses then forwards the resulting carrier packets directly to the source Client.¶

Following the initial NS/NA exchange, both Clients mark their respective (source, target) underlay interface pairs as "trusted" for no more than ReachableTime seconds. While the Clients continue to exchange carrier packets via the direct path avoiding all Proxy/Servers and Gateways, they should perform additional NS/NA exchanges via their local Proxy/Servers to refresh NCE state as well as send additional bubbles to the peer's Origin address information if necessary to refresh NAT state.¶

Note that these procedures are suitable for a widely-deployed but basic class of NATs. Procedures for advanced NAT classes are outlined in [RFC6081], which provides mechanisms that can be employed equally for AERO using the corresponding sub-options specified by OMNI.¶

Note also that each communicating pair of Clients may need to maintain NAT state for peer to peer communications via multiple underlay interface pairs. It is therefore important that Origin Indications are maintained with the correct peer interface and that the NCE may cache information for multiple peer interfaces.¶

Note that the source and target Client exchange Origin information during the secured NS/NA multilink route optimization exchange. This allows for subsequent NS/NA exchanges to proceed using only the Identification value as a data origin confirmation. However, Client-to-Client peerings that require stronger security may also include authentication signatures for mutual authentication.¶

3.13.6. Client-to-Client OMNI Link Extension

Clients may be recursively nested within the ENETs of other Clients. When a Client is the downstream-attached ENET neighbor of an upstream Client, it still supports the route optimization functions discussed above by maintaining an MFIB and assigning MFVI values. When the Client processes an IPv6 ND NS/NA message that includes an MFP sub-option, it writes its MFVI information as the first/last MFVI list entry the same as for the single Client case discussed above.¶

The Client then forwards the NS/NA message to the next Client in the extended OMNI link toward the FHS/LHS Proxy/Server, which records the MVFI value then overwrites the MFVI list entry with its own MFVI value. This process iteratively continues until the Client that will forward the NS/NA message to the FHS/LHS Proxy/Server is reached, at which point the NS/NA MFVI list entries are populated by the intermediate nodes on the path to the LHS/FHS the same as discussed above.¶

In this way, each Client in the extended OMNI link discovers the A/B MVFIs of the next/previous Client without intruding into the MFP sub-option MFVI list. Therefore the list can remain fixed at 5 entries even though the Client-to-Client OMNI link extension can be arbitrarily long. Therefore, route optimization is not possible between consecutive Client members of the extended OMNI link but becomes possible at the Internetworking border that separates the FHS and LHS elements.¶

3.13.7. Intra-ANET/ENET Route Optimization for AERO Peers

When a Client forwards a packet from a Host or another Client connected to one of its downstream ENETs to a peer within the same downstream ENET, the Client returns an IPv6 ND Redirect message to inform the source that that target can be reached directly. The contents of the Redirect message are the same as specified in [RFC4861].¶

In the same fashion, when a Proxy/Server forwards a packet from a Host or Client connected to one of its downstream ANETs to a peer within the same downstream ANET, the Proxy/Server returns an IPv6 ND Redirect message.¶

All other route optimization functions are conducted per the NS/NA messaging discussed in the previous sections.¶

3.15.1. Mobility Update Messaging

Hub Proxy/Servers (and/or the mobile Clients themselves) accommodate mobility and/or multilink change events by sending secured uNA messages to each of the Client's active neighbors. When a node sends a uNA message to each specific neighbor on behalf of a mobile Client, it sets the IPv6 source address to its own ULA or XLA, sets the destination address to the neighbor's ULA or XLA and sets the Target Address to the mobile Client's XLA. The uNA also includes an OMNI option with OMNI Neighbor Coordination sub-option Preflen set to the prefix length associated with the mobile Client's XLA, includes Interface Attributes and Traffic Selectors for the mobile Client's underlay interfaces and includes an authentication signature if necessary. The node then sets the uNA R flag to 1, S flag to 0 and O flag to 1, then encapsulates the message in an OAL header with source set to its own ULA and destination set to either the specific neighbor's ULA or the FHS Proxy/Server's ULA. The uNA message will then follow the secured spanning tree and arrive at the specific neighbor.¶

As discussed in Section 7.2.6 of [RFC4861], the transmission and reception of uNA messages is unreliable but provides a useful optimization. In well-connected Internetworks with robust data links uNA messages will be delivered with high probability, but in any case the node can optionally send up to MAX_NEIGHBOR_ADVERTISEMENT uNAs to each neighbor to increase the likelihood that at least one will be received. Alternatively, the node can set the PNG flag in the uNA OMNI option header to request a uNA acknowledgement as specified in [I-D.templin-6man-omni].¶

When the FHS/LHS Proxy/Server receives a uNA message prepared as above, if the uNA destination was its own ULA the Proxy/Server uses the included OMNI option information to update its NCE for the target but does not reset ReachableTime since the receipt of a uNA message does not provide confirmation that any forward paths to the target Client are working. If the destination was the XLA of the FHS/LHS Client, the Proxy/Server instead changes the OAL source to its own ULA, includes an authentication signature if necessary, and includes an in-window Identification for this Client. Finally, if the uNA message PNG flag was set, the node that processes the uNA returns a uNA acknowledgement as specified in [I-D.templin-6man-omni].¶

3.15.2. Announcing Link-Layer Information Changes

When a Client needs to change its underlay Interface Attributes and/or Traffic Selectors (e.g., due to a mobility event), the Client sends an RS message to its Hub Proxy/Server (via a first-hop FHS Proxy/Server, if necessary). The RS includes an OMNI option with an Interface Attributes sub-option with the omIndex and with new link quality and any other information.¶

Note that the first FHS Proxy/Server may change due to the underlay interface change. If the Client supplies the address of the former FHS Proxy/Server, the new FHS Proxy/Server can send a departure indication (see below); otherwise, any stale state in the former FHS Proxy/Server will simply expire after ReachableTime expires with no effect on the Hub Proxy/Server.¶

Up to MAX_RTR_SOLICITATIONS RS messages MAY be sent in parallel with sending carrier packets containing user data in case one or more RAs are lost. If all RAs are lost, the Client SHOULD re-associate with a new Proxy/Server.¶

After performing the RS/RA exchange, the Client sends uNA messages to all neighbors the same as described in the previous section.¶

3.15.3. Bringing New Links Into Service

When a Client needs to bring new underlay interfaces into service (e.g., when it activates a new data link), it sends an RS message to the Hub Proxy/Server via a FHS Proxy/Server for the underlay interface (if necessary) with an OMNI option that includes an Interface Attributes sub-option with appropriate link quality values and with link-layer address information for the new link. The Client then again sends uNA messages to all neighbors the same as described above.¶

3.15.4. Deactivating Existing Links

When a Client needs to deactivate an existing underlay interface, it sends a uNA message toward the Hub Proxy/Server via an FHS Proxy/Server with an OMNI option with appropriate Interface Attributes values for the deactivated link - in particular, the link quality value 0 assures that neighbors will cease to use the link.¶

If the Client needs to send uNA messages over an underlay interface other than the one being deactivated, it MUST include Interface Attributes with appropriate link quality values for any underlay interfaces being deactivated. The Client then again sends uNA messages to all neighbors the same as described above.¶

Note that when a Client deactivates an underlay interface, neighbors that receive the ensuing uNA messages need not purge all references for the underlay interface from their neighbor cache entries. The Client may reactivate or reuse the underlay interface and/or its omIndex at a later point in time, when it will send new RS messages to an FHS Proxy/Server with fresh interface parameters to update any neighbors.¶

3.15.5. Moving Between Proxy/Servers

The Client performs the procedures specified in Section 3.12.2 when it first associates with a new Hub Proxy/Server or renews its association with an existing Hub Proxy/Server.¶

When a Client associates with a new Hub Proxy/Server, it sends RS messages to register its underlay interfaces with the new Hub while including the old Hub's ULA in the "Old Hub Proxy/Server ULA" field of a Proxy/Server Departure OMNI sub-option. When the new Hub Proxy/Server returns the RA message via the FHS Proxy/Server (acting as a Proxy), the FHS Proxy/Server sends a uNA to the old Hub Proxy/Server (i.e., if the ULA is non-zero and different from its own). The uNA has the XLA of the Client as the source and the ULA of the old hub as the destination and with OMNI Neighbor Coordination sub-option Preflen set to 0. The FHS Proxy/Server encapsulates the uNA in an OAL header with the ULA of the new Hub as the source and the ULA of the old Hub as the destination, the fragments and sends the carrier packets via the secured spanning tree.¶

When the old Hub Proxy/Server receives the uNA, it changes the Client's NCE state to DEPARTED, resets DepartTime and caches the new Hub Proxy/Server ULA. After a short delay (e.g., 2 seconds) the old Hub Proxy/Server withdraws the Client's MNP from the routing system. While in the DEPARTED state, the old Hub Proxy/Server forwards any carrier packets received via the secured spanning tree destined to the Client's ULA-MNP to the new Hub Proxy/Server's ULA. After DepartTime expires, the old Hub Proxy/Server deletes the Client's NCE.¶

Mobility events may also cause a Client to change to a new FHS Proxy/Server over a specific underlay interface at any time such that a Client RS/RA exchange over the underlay interface will engage the new FHS Proxy/Server instead of the old. The Client can arrange to inform the old FHS Proxy/Server of the departure by including a Proxy/Server Departure sub-option with a ULA for the "Old FHS Proxy/Server ULA", and the new FHS Proxy/Server will issue a uNA using the same procedures as outlined for the Hub above while using its own ULA as the source address. This can often result in successful delivery of packets that would otherwise be lost due to the mobility event.¶

Clients SHOULD NOT move rapidly between Hub Proxy/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 Hub Proxy/Server include a Hub Proxy/Server that has gone unreachable, topological movements of significant distance, movement to a new geographic region, movement to a new OMNI link segment, etc.¶

3.16.1. Source-Specific Multicast (SSM)

When an ARS "X" (i.e., either a Client or Proxy/Server) 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-OMNI interfaces, X then forwards the (S, G) Join/Prune to any PIM routers on those interfaces per [RFC7761].¶

For each S belonging to a prefix reachable via X's OMNI interface, X sends an NS(AR) message (see: Section 3.13) using its own ULA or XLA as the source address, the solicited node multicast address corresponding to S as the destination and the XLA of S as the target address. X then encapsulates the NS(AR) in an OAL header with source address set to its own ULA and destination address set to the ULA for S, then forwards the message into the secured spanning tree which delivers it to ARR "Y" that services S. The resulting NA(AR) will return an OMNI option with Interface Attributes for any underlay interfaces that are currently servicing S.¶

When X processes the NA(AR) it selects one or more underlay interfaces for S and performs an NS/NA multilink route optimization exchange over the secured spanning tree while including a PIM Join/Prune message for each multicast group of interest in the OMNI option. If S is located behind any Proxys "Z"*, each Z* then updates its MRIB accordingly and maintains the XLA of X as the next hop in the reverse path. Since Gateways forward messages not addressed to themselves without examining them, this means that the (reverse) multicast tree path is simply from each Z* (and/or S) to X with no other multicast-aware routers in the path.¶

Following the initial combined Join/Prune and NS/NA messaging, X maintains a NCE 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 NCE alive for up to t_periodic seconds [RFC7761]. If no new Joins are received within t_periodic seconds, X allows the NCE to expire. Finally, if X receives any additional Join/Prune messages for (S,G) it forwards the messages over the secured spanning tree.¶

Client C that holds an MNP for source S may later depart from a first Proxy/Server Z1 and/or connect via a new Proxy/Server Z2. In that case, Y sends a uNA message to X the same as specified for unicast mobility in Section 3.15. When X receives the uNA message, it updates its NCE for the XLA for source S and sends new Join messages in NS/NA exchanges addressed to the new target Client underlay interface connection for S. There is no requirement to send any Prune messages to old Proxy/Server Z1 since source S will no longer source any multicast data traffic via Z1. Instead, the multicast state for (S,G) in Proxy/Server Z1 will soon expire since no new Joins will arrive.¶

3.16.2. Any-Source Multicast (ASM)

When an ARS X acting as a PIM router receives Join/Prune messages from a node on its downstream interfaces containing one or more (*,G) pairs, it updates its Multicast Routing Information Base (MRIB) accordingly. X first performs an NS/NA(AR) exchange to receive route optimization information for Rendezvous Point (RP) R for each G. X then includes a copy of each Join/Prune message in the OMNI option of an NS message with its own ULA or XLA as the source address and the ULA or XLA for R as the destination address, then encapsulates the NS message in an OAL header with its own ULA as the source and the ULA of R's Proxy/Server as the destination then sends the message into the secured spanning tree.¶

For each source S that sends multicast traffic to group G via R, Client S* that aggregates S (or its Proxy/Server) encapsulates the original IP packets in PIM Register messages, includes the PIM Register messages in the OMNI options of uNA messages, performs OAL encapsulation and fragmentation then forwards the resulting carrier packets with Identification values within the receive window for Client R* that aggregates R. Client R* may then elect to send a PIM Join to S* in the OMNI option of a uNA over the secured spanning tree. This will result in an (S,G) tree rooted at S* with R as the next hop so that R will begin to receive two copies of the original IP packet; one native copy from the (S, G) tree and a second copy from the pre-existing (*, G) tree that still uses uNA PIM Register encapsulation. R can then issue a uNA PIM Register-stop message over the secured spanning tree to suppress the Register-encapsulated stream. At some later time, if Client S* moves to a new Proxy/Server, it resumes sending original IP packets via uNA PIM Register encapsulation via the new Proxy/Server.¶

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 Section 3.16.1. Once the (S,G) tree is established, the listeners can send (S, G) Prune messages to R so that multicast original IP 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.¶

3.16.3. Bi-Directional PIM (BIDIR-PIM)

Bi-Directional PIM (BIDIR-PIM) [RFC5015] 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.¶