1. Low Layer Multihoming
2. Layer 3 Network Layer multihoming
3. Layer 4 Transport layer multihoming
4. Higher Layer Multihoming techniques
5. Best Path Selection among multiple paths
   1. FIFO or Round Robin
   2. Weights ( predetermined or dynamically allocated)
   3. Prioritization algorithms
6. MultiPath protocol design
   1. Service discovery
   2. Unique Identifiers
      1. Path / Route and Network identifiers
      2. Header metadata
   3. Pre – Registration
   4. Handover / Failover from one path to another without disruption
   5. Security association for paths

A multihoming protocol maintains a simultaneous connection to multiple networks. Such a protocol enhances reliability, load balancing, and fault tolerance and makes an excellent candidate for signaling planes, which are lightweight packets managing the connection for the data plane. Dataplane acts as the actual data transfer protocol, which can be for multimedia such as audio-video content, games, streaming, and so on. Multihoming can be implemented at several levels of the network stack.

Low Layer Multihoming

While lower layers of the network stack do not generally have the logic to maintain statefulness and decision-making, they can still leverage multiple paths efficiently to provide redundancy and increased BDP.

Link Aggregation Control Protocol (LACP), an IEEE standard and part of the IEEE 802.3ad specification, bundles individual physical links of Ethernet connections into one logical link to increase throughput using multiple NICs.

Layer 3 Network Layer multihoming

BGP can be considered an example of multihoming since it process multiple paths via logical addressing and sets up the routes.

IP multihoming : Virtual Router Redundancy Protocol (VRRP) and Hot Standby Router Protocol (HSRP) enable routers to share IP and MAC address.

Layer 4 Transport layer multihoming

Some protocols simultaneously use these multiple paths for a single communication session such as Multi-path TCP. This can be used to improve bandwidth utilization ( sometimes unfairly) and build resilience. While simultaneous usage of multiple links/ paths provides great resilience, they can lead to asymmetrical routing issues.

SCTP (Stream Control Transmission Protocol) is another example of a transport layer protocol that provides connection-oriented reliable communication while having multiple IP addresses and interfaces. With multiple paths dynamically added, the SCTP session can switch from one path to another in the event of a failover or can load balance between paths.

Higher Layer Multihoming techniques

AnyCast is a networking technique that lets multiple endpoints share the same IP and dynamically select the best destination path.

SIP Forking is an application layer technique to have multiple VoIP endpoints receive an incoming call either parallelly or sequentially. The first SIP phone to answer the calls establishes the connection.

Equal-cost multipath (ECMP) is a routing technique that allows routers to distribute traffic across multiple equal-cost paths. This is often associated with OSPF (Open Shortest Path First) like implementations.

Best Path Selection among multiple paths

Selection of the best path and routing decision in a multihomed scenario can be made by :

FIFO or Round Robin

The first path to respond can be selected first to establish a connection in FIFO, while Round Robin can select paths in order of packet arrival to prevent starvation of a path.

Weights ( predetermined or dynamically allocated)

BGP can help provide weight and preferences to paths while exchanging routing and reachability. For example, AS( Autonomous systems ) that connect to multiple ISPs use BGP to advertise their IP prefixes through each link.

Prioritization algorithms

In an anycast scenario, where the same address is assigned to multiple destinations, the traffic can be assigned to the geographically nearest, shortest RTT, or best-performing path. For example, to manage ingress traffic to a service across multiple data centers. A more sophisticated algorithm can also add compound metrics such as a derivative of loss, jitter, etc.

Cost and load balancing are often top considerations for path selection. The more complex decisions can be based on instantaneous Qos collected or even forecasted. Resource utilization or carbon footprint is also a candidate for path selection. A fairness-based approach can also be built in to avoid starvation.

Mobility management

Mobility management has been long used to provide network continuity. From telecom devices’ handover between base stations ( home network and visitor network ) to mobile IPv6 ( via home agent and Foreign agent), mobility is crucial to meet the needs of mobile devices for seamless connectivity.

In cases of session-based protocols, when the source IP changes after an established connection, the mobility of the protocol should kick in to discover and migrate to another reachable address for the endpoint. This is especially crucial in the case of wireless networks and mobile devices. Examples of such cases may be

• when NATing changes and a host receives a new IP address.
•  A device with multiple interfaces or uplinks decides to tear down one of the interfaces to uplink and migrate to another one.
•  A user on a call travels through multiple networks, so the call is handed over from mobile data to wifi, etc.
•  Switching between ipv4 and ipv6 address

A mobility management protocol should be able to seamlessly update the existing sessions endpoint IP address without re-establishing all of the security protocol or handshakes or using a minimal subset of it.

MultiPath protocol design

To design a multipath protocol with multihoming and seamless session migration, any available network metrics and topologies should be identified. These help build mechanisms for discovery and selection. Multipath management is stateful, and the states are utilized to identify degradation, plan migration, and recovery. Multihoming protocol design can be divided into the following parts

Service discovery

The network endpoint needs to detect the presence of multiple paths to a common destination, such as a server. This can be achieved by gaining the link state information at the router, preferred policies and static routes, feedback packets sharing path alternatives, Hello’s, ICMP, or other monitoring techniques. Address Resolution Protocol(ARP), IPv6 Neighbor Discovery, and Neighbor Unreachability Detection are also instances of path discovery.

A successful discovery is followed by routability checks, which involve checking if a BGP session is established. If it is not, then check the advertised IP prefix by each link. At this stage, reviewing the routing table at the hub( or router) to ensure the entries corresponding to each IP prefix point to a valid next-hop address is also an option. Tools like ping and traceroute can be used to confirm that the packets can reach the multihomed network through each link.
This prevalidation leads to quick convergence from old to new path in case the network conditions change, and this avoids blackholing traffic and downtime.

Unique Identifiers

DHCP, a client-server protocol, assigns addresses dynamically to devices as they get joined to a network. Even with rapid network changes, DHCP automates the address assignment from a pool and even relays across subnets. Since the IPV4 address is not necessarily unique across networks (32 bits), a different UUID ( 128 bits) is needed to be generated per endpoint. In the case of IPv6-enabled endpoints, the IPv6 address itself can serve as a unique identifier.

Path / Route and Network identifiers

To identify migration from one network to another, it is important to address each network uniquely. While the details of most ISP infrastructure itself cannot be determined by analyzing a packet that traversed it, other means of analysis can help narrow the choices, such as IP address ranges and DNS resolutions. TTL ( Time to Live) and RTT (Round Trip Time) can also help make inferences on physical distance. Fragmentation and filtering can help narrow its behavior and policies. For all different kinds of networks identified between a common source and destination, there should be a unique identifier to address the paths. Network identifiers can be in the form of

• Link-layer identifiers for an interface include IEEE 802 addresses on Ethernet links.
• subnet prefixes, CIDR
• Gateway identification
• home address/care of address in mobile ipv6
• Index serialization 
• UUID generated from its characteristics so on

Append the router’s address to an array of path identifiers. Such a prefix helps to identify the path. Example

• SIP via header Via: SIP/2.0/UDP client.atlanta.com:5060;branch=z9hG45684bf9
• SIP Route header : <sip:proxy1.atlanta.com;lr>, <sip:proxy2.atlanta.com;lr>

Some techniques also rely on cookies to identify the route

Header metadata

In addition to the network identifiers themselves, there are more metadata that need to be part of the payload or the encapsulating header to determine that the packet is traversing the appropriate source and destination. This can be in the form of the following:

• unique pair of ip and port + optional magic number for NAT
•  tuple of transport type, ip, and port
•  unique ID derived from timestamp, ip , port, etc

Pre – Registration

As a mobile node detects a new network it registeres its current location with the network. This can be in form of authtication as a mobile node enters a network of sending its presence with pub/sub notification. Example :

• SIP registration
• acknowlege message for a new init
• hello handshake

Handover / Failover from one path to another without disruption

A shift from one network to another can be triggered by various factors such as signal strength, load balancing, or network congestion. The decision to move over to a new network may also be gauged by algorithmically analyzing signal quality, interference, performance metrics, or cost and usage.
L2 handover: Change in link layer connection such as disconnecting from a wireless access point and connecting to another. Another example of mobility management is the telecom network handover, which re-established link-layer connectivity instead of relying on upper layers to reconnect.
L3 handover: change of router to which the mobile node is connected to.
Assuming endpoint 1 of the session has M addresses and another endpoint, i.e., endpoint 2 has an N address, then the connection should be able to migrate between any one of the M*N address pairs.

Security association for paths

Although more related to IPsec tunnels, SAs( security associations) in this context refer to cryptographic certainty that the endpoints are authorized and authenticated. SA uses nonce to randomize the keygen tokens. Additionally, there is anti-replay and rekeys in place to detect if there is possible interference in the communication link. A successful security association should display that the binding is successful and the endpoints are now allowed to transfer data or establish a tunnel to transfer encapsulated encrypted dat

Token based validity : A token, often generated by nonce and session’s unique parameters, is often used to validate intermediate messages without having to cross-confirm every packet with the home network or profile database

ESP Header : In cases of encapsulated packets like Encapsulating Security Payload (ESP), a not null payload authentication header can provide information on the authenticity of the origin.

Pre – Registration / Validation : As a mobile node detects a new network, it registers its current location with the network. This can be in the form of authentication as a mobile node enters a network by sending its presence with pub/sub notification. Example: SIP registration, child SAs for various paths in IPSec

References :

• IKEv2 Mobility and Multihoming Protocol (MOBIKE) https://datatracker.ietf.org/doc/html/rfc4555
• IP mobility https://datatracker.ietf.org/doc/html/rfc3344
• Mobility Support in IPv6 https://datatracker.ietf.org/doc/html/rfc3775