RFC 6740
Identifier-Locator Network Protocol (ILNP) Architectural Description
3. Architectural Changes Introduced by ILNP
In this section, we describe the key changes that are made to the current Internet Architecture. These key changes impact end-systems, rather than routers.3.1. Identifiers
Identifiers, also called Node Identifiers (NIDs), are non-topological values that identify an ILNP node. A node might be a physical node or a virtual node. For example, a single physical device might contain multiple independent virtual nodes. Alternately, a single virtual device might be composed from multiple physical devices. In the case of a Multi-Level Secure (MLS) system [DIA] [DoD85] [DoD87] [RFC5570], each valid Sensitivity Label of that system might be a separate virtual node. A node MAY have multiple Identifier values associated with it, which MAY be used concurrently. In normal operation, when a node is responding to a received ILNP packet that creates a new network-layer session, the correct NID value to use for that network-layer session with that correspondent node will be learned from the received ILNP packet. In normal operation, when a node is initiating communication with a correspondent node, the correct I value to use for that session with that correspondent node will be learned either through the application-layer naming, through DNS name resolution, or through
some alternative name resolution system. Another option is an application may be able to select different I values directly -- as Identifiers are visible above the network layer via the transport protocol.3.1.1. Node Identifiers Are Immutable during a Session
Once a Node Identifier (NID) value has been used to establish a transport-layer session, that Node Identifier value forms part of the end-to-end (invariant) transport-layer session state and so MUST remain fixed for the duration of that session. This means, for example, that throughout the duration of a given TCP session, the Source Node Identifier and Destination Node Identifier values will not change. In normal operation, a node will not change its set of valid Identifier values frequently. However, a node MAY change its set of valid Identifier values over time, for example, in an effort to provide identity obfuscation, while remaining subject to the architectural rule of the preceding paragraph. When a node has more than one Node Identifier value concurrently, the node might have multiple concurrent ILNP sessions with some correspondent node, in which case Node Identifier values MAY differ between the different concurrent ILNP sessions.3.1.2. Syntax
ILNP Identifiers have the same syntax as IPv6 interface identifiers [RFC4291], based on the EUI-64 format [IEEE-EUI], which helps with backwards compatibility. There is no semantic equivalent to an ILNP Identifier in IPv4 or IPv6 today. The Modified EUI-64 syntax used by both ILNP Identifiers and IPv6 interface identifiers contains a bit indicating whether the value has global scope or local scope [IEEE-EUI] [RFC4291]. ILNP Identifiers have either global scope or local scope. If they have global scope, they SHOULD be globally unique. Regardless of whether an Identifier is global scope or local scope, an Identifier MUST be unique within the scope of a given Locator value to which it is bound for a given ILNP session or packet flow. As an example, with ILNPv6, the ordinary IPv6 Neighbour Discovery (ND) processes ensure that this is true, just as ND ensures that no two IPv6 nodes on the same IPv6 subnetwork have the same IPv6 address at the same time.
Both the IEEE EUI-64 specification and the Modified EUI-64 syntax also has a 'Group' bit [IEEE-EUI] [RFC4291]. For both ILNP node Identifiers and also IPv6 interface identifiers, this Group bit is set to 0.3.1.3. Semantics
Unicast ILNP Identifier values name the node, rather than naming a specific interface on that node. So ILNP Identifiers have different semantics than IPv6 interface identifiers.3.2. Locators
Locators are topologically significant names, analogous to (sub)network routing prefixes. The Locator names the IP subnetwork that a node is connected to. ILNP neither prohibits nor mandates in- transit modification of Locator values. A host MAY have several Locators at the same time, for example, if it has a single network interface connected to multiple subnetworks (e.g., VLAN deployments on wired Ethernet) or has multiple interfaces each on a different subnetwork. Locator values normally have Locator Preference Indicator (LPI) values associated with them. These LPIs indicate that a specific Locator value has higher or lower preference for use at a given time. Local LPI values may be changed through local policy or via management interfaces. Remote LPI values are normally learned from the DNS, but the local copy of a remote LPI value might be modified by local policy relating to preferred paths or prefixes. Locator values are used only at the network layer. Locators are not used in end-to-end transport state. For example, Locators are not used in transport-layer session state or application-layer session state. However, this does not preclude an end-system setting up local dynamic bindings for a single transport flow to multiple Locator values concurrently. The routing system only uses Locators, not Identifiers. For unicast traffic, ILNP uses longest-prefix match routing, just as the IP Internet does. Section 4 below describes in more detail how Locators are used in forwarding and routing packets from a sending node on a source subnetwork to one or more receiving nodes on one or more destination subnetworks.
A difference from earlier proposals [GSE] [8+8] is that, in normal operation, the originating host supplies both Source Locator and Destination Locator values in the packets it sends out. Section 4.3 describes packet forwarding in more detail, while Section 4.4 describes packet routing in more detail.3.2.1. Locator Values Are Dynamic
The ILNP architecture recognises that Locator values are topologically significant, so the set of Locator values associated with a node normally will need to change when the node's connectivity to the Internet topology changes. For example, a mobile or multihomed node is likely to have connectivity changes from time to time, along with the corresponding changes to the set of Locator values. When a node using a specific set of Locator values changes one or more of those Locator values, then the node (1) needs to update its local knowledge of its own Locator values, (2) needs to inform all active Correspondent Nodes (CNs) of those changes to its set of Locator values so that ILNP session continuity is maintained, and (3) if it expects incoming connections the node also needs to update its Locator-related entries in the Domain Name System. [RFC6741] describes the engineering and implementation details of this process.3.2.2. Locator Updates
As Locator values can be dynamic, and they could change for a node during an ILNP session, correspondents need to be notified when a Locator value for a node changes for any existing ILNP session. To enable this, a node that sees its Locator values have changed MUST send a Locator Update (LU) message to its correspondent nodes. The details of this procedure are discussed in other ILNP documents -- [RFC6741], [RFC6743], and [RFC6745]. (The change in Locator values may also need to be notified to DNS but that is discussed elsewhere.)3.2.3. Syntax
ILNP Locators have the same syntax as an IP unicast routing prefix.3.2.4. Semantics
ILNP unicast Locators have the same semantics as an IP unicast routing prefix, since they name a specific subnetwork. ILNP neither prohibits nor requires in-transit modification of Locator values.
3.3. IP Address and Identifier-Locator Vector (I-LV)
Historically, an IP Address has been considered to be an atomic datum, even though it is recognised that an IP Address has an internal structure: the network prefix plus either the host ID (IPv4) or the interface identifier (IPv6). However, this internal structure has not been used in end-system protocols; instead, all the bits of the IP Address are used. (Additionally, in IPv4 the IPv4 subnet mask uses bits from the host ID, a further confusion of the structure, even thought it is an extremely useful engineering mechanism.) In ILNP, the IP Address is replaced by an "Identifier-Locator Vector" (I-LV). This consists of a pairing of an Identifier value and a Locator value for that packet, written as [I, L]. All ILNP packets have Source Identifier, Source Locator, Destination Identifier, and Destination Locator values. The I value of the I-LV is used by upper-layer protocols (e.g., TCP, UDP, SCTP), so needs to be immutable. Locators are not used by upper-layer protocols (e.g., TCP, UDP, SCTP). Instead, Locators are similar to IP routing prefixes, and are only used to name a specific subnetwork. While it is possible to say that an I-LV is an approximation to an IP Address of today, it should be understood that an I-LV: a) is not an atomic datum, being a pairing of two data types, an Identifier and a Locator. b) has different semantics and properties to an IP Address, as is described in this document. In our discussion, it will be convenient sometimes to refer to an I-LV, but sometimes to refer only to an Identifier value, or only to a Locator value. ILNP packets always contain a source I-LV and a destination I-LV.3.4. Notation
In describing how capabilities are implemented in ILNP, we will consider the differences in end-systems' state between IP and ILNP in order to highlight the architectural changes.
We define a formal notation to represent the data contained in the
transport-layer session state. We define:
A = IP Address
I = Identifier
L = Locator
P = Transport-layer port number
To differentiate the local and remote values for the above items, we
also use suffixes, for example:
_L = local
_R = remote
With IPv4 and IPv6 today, the invariant state at the transport-layer
for TCP can be represented by the tagged tuple:
<TCP: A_L, A_R, P_L, P_R> --- (1)
Tag values that will be used are:
IP Internet Protocol
ILNP Identifier-Locator Network Protocol
TCP Transmission Control Protocol
UDP User Datagram Protocol
So, for example, with IP, a UDP packet would have the tagged tuple:
<UDP: A_L, A_R, P_L, P_R> --- (2)
A TCP segment carried in an IP packet may be represented by the
tagged tuple binding:
<TCP: A_L, A_R, P_L, P_R><IP: A_L, A_R> --- (3)
and a UDP packet would have the tagged tuple binding:
<UDP: A_L, A_R, P_L, P_R><IP: A_L, A_R> --- (4)
In ILNP, the transport-layer state for TCP is:
<TCP: I_L, I_R, P_L, P_R> --- (5)
The binding for a TCP segment within an ILNP packet:
<TCP: I_L, I_R, P_L, P_R><ILNP: L_L, L_R> --- (6)
When comparing tuple expressions (3) and (6), we see that for IP, any change to network addresses impacts the end-to-end state, but for ILNP, changes to Locator values do not impact end-to-end state. This provides end-system session state invariance, a key feature of ILNP compared to IP as it is used in some situations today. ILNP adopts the end-to-end approach for its architecture [SRC84]. As noted previously, nodes MAY have more than one Locator concurrently, and nodes MAY change their set of active Locator values as required. While these documents do not include SCTP examples, the same notation can be used, simply substituting the string "SCTP" for the string "TCP" or the string "UDP" in the above examples.3.5. Transport-Layer State and Transport Pseudo-Headers
In ILNP, protocols above the network layer do not use the Locator values. Thus, the transport layer uses only the I values for the transport-layer session state (e.g., TCP pseudo-header checksum, UDP pseudo-header checksum), as is shown, for example, in expression (6) above. Additionally, from a practical perspective, while the I values are only used in protocols above the network layer, it is convenient for them to be carried in network packets, so that the namespace for the I values can be used by any transport-layer protocols operating above the common network layer.3.6. Rationale for This Document
This document provides an architectural description of the core ILNP capabilities and functions. It is based around the use of example scenarios so that practical issues can be highlighted. In some cases, illustrative suggestions and light discussion are presented with respect to engineering issues, but detailed discussion of engineering issues are deferred to other ILNP documents. The order of the examples presented below is intended to allow an incremental technical understanding of ILNP to be developed. There is no other reason for the ordering of the examples listed below. Many of the descriptions are based on the use of an example site network as shown in Figure 3.1.
site . . . . +----+ network . .-----+ CN | . . . . +------+ link1 . . +----+ . . | +------. . . D . | | . . . .----+ SBR | . Internet . . H . | | . . . . | +------. . . . . . +------+ link2 . . . . . . . . CN = Correspondent Node D = Device H = Host SBR = Site Border Router Figure 3.1: A Simple Site Network for ILNP Examples In some cases, hosts (H) or devices (D) act as end-systems within the site network, and communicate with (one or more) Correspondent Node (CN) instances that are beyond the site. Note that the figure is illustrative and presents a logical view. For example, the CN may itself be on a site network, just like H or D. Also, for formulating examples, we assume ILNPv6 is in use, which has the same packet header format (as viewed by routers) as IPv6, and can be seen as a superset of IPv6 capabilities. For simplicity, we assume that name resolution is via the deployed DNS, which has been updated to store DNS records for ILNP [RFC6742]. Note that, from an engineering viewpoint, this does NOT mean that the DNS also has to be ILNP capable: existing IPv4 or IPv6 infrastructure can be used for DNS transport.3.7. ILNP Multicasting
Multicast forwarding and routing are unchanged, in order to avoid requiring changes in deployed IP routers and routing protocols. ILNPv4 multicasting is the same as IETF Standards Track IPv4 multicasting [RFC1112] [RFC3376]. ILNPv6 multicasting is the same as IETF Standards Track IPv6 multicasting [RFC4291] [RFC2710] [RFC3810].
4. ILNP Basic Connectivity
In this section, we describe basic packet forwarding and routing in ILNP. We highlight areas where it is similar to current IP, and also where it is different from current IP. We use examples in order to illustrate the intent and show the feasibility of the approach. For this section, in Figure 4.1, H is a fixed host in a simple site network, and CN is a remote Correspondent Node outside the site; H and CN are ILNP-capable, while the Site Border Router (SBR) does not need to be ILNP-capable. site . . . . +----+ network . .-----+ CN | . . . . +------+ . . +----+ . . | +------. . . . | | . . . .----+ SBR | . Internet . . H . | | . . . . | | . . . . . . +------+ . . . . . . . . CN = Correspondent Node H = Host SBR = Site Border Router Figure 4.1: A Simple Site Network for ILNP Examples4.1. Basic Local Configuration
This section uses the term "address management", in recognition of the analogy with capabilities present in IP today. In this document, address management is about enabling hosts to attach to a subnetwork and enabling network-layer communication between and among hosts, also including: a) enabling identification of a node within a site. b) allowing basic routing/forwarding from a node acting as an end- system. If we consider Figure 4.1, imagine that host H has been connected to the site network. Administratively, it needs at least one I value and one L value in order to be able to communicate.
Today, local administrative procedures allocate IP Addresses, often using various protocol mechanisms (e.g., NETCONF-based router configuration, DHCP for IPv4, DHCP for IPv6, IPv6 Router Advertisements). Similarly, local administrative procedures can allocate I and L values as required, e.g., I_H and L_H. This may be through manual configuration. Additionally, if it is expected or desired that H might have incoming communication requests, e.g., it is a server, then the values I_H and L_H can be added to the relevant name services (e.g., DNS, NIS/YP), so that FQDN lookups for H resolve to the appropriate DNS resource records (e.g., NID, L32, L64, and LP [RFC6742]) for node H. From a network operations perspective, this whole process also can be automated. As an example, consider that in Figure 3.1 the Site Border Router (SBR) is an IPv6-capable router and is connected via link1 to an ISP that supports IPv6. The SBR will have been allocated one (or more) IPv6 prefixes that it will multicast using IPv6 Routing Advertisements (RAs) into the site network, e.g., prefix L_1. L_1 is actually a local IPv6 prefix (/64), which is formed from an address assignment by the upstream ISP, according to [RFC3177] or [RFC6177]. Host H will see these RAs, for example, on its local interface with name eth0, will be able to use that prefix as a Locator value, and will cache that Locator value locally. Also, node H can use the mechanism documented in either Section 2.5.1 of [RFC4291], in [RFC3972], [RFC4581], [RFC4982], or in [RFC4941] in order to create a default I value (say, I_H), just as an IPv6 host can. For DNS, the I_H and L_1 values may be pre-configured in DNS by an administrator who already has knowledge of these, or added to DNS by H using Secure DNS Dynamic Update [RFC3007] to add or update the correct NID and L64 records to DNS for the FQDN for H.4.2. I-L Communication Cache
For the purposes of explaining the concept of operations, we talk of a local I-L Communication Cache (ILCC). This is an engineering convenience and does not form part of the ILNP architecture, but is used in our examples. More details on the ILCC can be found in [RFC6741]. The ILCC contains information that is required for the operation of ILNP. This will include, amongst other things, the current set of valid Identifier and Locator values in use by a node, the bindings between them, and the bindings between Locator values and interfaces.
4.3. Packet Forwarding
When the SBR needs to send a packet to H, it uses local address resolution mechanisms to discover the bindings between interface addresses and currently active I-LVs for H. For our example of Figure 3.1, IPv6 Neighbour Discovery (ND) can be used without modification, as the I-LV for ILNPv6 occupies the same bits as the IPv6 address in the IPv6 header. For packets from H to SBR, the same basic mechanism applies, as long as SBR supports IPv6 and even if it is not ILNPv6-capable, as IPv6 ND is used unmodified for ILNPv6. For Figure 3.1, assuming: - SBR advertises prefix L_1 locally, uses I value I_S, and has an Ethernet MAC address M_S on interface with local name sbr0 - H uses I value I_H, and has an Ethernet MAC address of M_H on the interface with local name eth0 then H will have in its ILCC: [I_H, L_1] --- (7a) L_1, eth0 --- (7b) After the IPv6 RA and ND mechanism has executed, the ILCC at H would contain, as well as expressions (7a) and (7b), the following entry for SBR: [I_S, L_1], M_S --- (8) For ILNPv6, it does not matter that the SBR is not ILNPv6-capable, as the I-LV [I_S, L_1] is physically equivalent to the IPv6 address for the internal interface sbr0. At SBR, which is not ILNP-capable, there would be the following entries in its local cache and configuration: L_1:I_S --- (9a) L_1, sbr0 --- (9b) Expression (9a) represents a valid IPv6 ND entry: in this case, the I_S value (which is 64 bits in ILNPv6) and the L_1 values are, effectively, concatenated and treated as if they were a single IPv6 address. Expression (9b) binds transmissions for L_1 to interface sbr0. (Again, sbr0 is a local, implementation-specific name, and such a binding is possible with standard tools today, for example, ifconfig(8).)
4.4. Packet Routing
If we assume that host H is configured as in the previous section, it is now ready to send and receive ILNP packets. Let us assume that, for Figure 4.1, it wishes to contact the node CN, which has FQDN cn.example.com and is ILNP-capable. A DNS query by H for cn.example.com will result in NID and L64 records for CN, with values I_CN and L_CN, respectively, being returned to H and stored in its ILCC: [I_CN, L_CN] --- (10) This will be considered active as long as the TTL values for the DNS records are valid. If the TTL for an I or L value is zero, then the value is still usable but becomes stale as soon as it has been used once. However, it is more likely that the TTL value will be greater than zero [BA11] [SBK01]. Once the CN's I value is known, the upper-layer protocol, e.g., the transport protocol, can set up suitable transport-layer session state: <UDP: I_H, I_CN, P_H, P_CN> --- (11) For routing of ILNP packets, the destination L value in an ILNPv6 packet header is semantically equivalent to a routing prefix. So, once a packet has been forwarded from a host to its first-hop router, only the destination L value needs to be used for getting the packet to the destination network. Once the packet has arrived at the router for the site network, local mechanisms and the packet- forwarding mechanism, as described above in Section 4.3, allow the packet to be delivered to the host. For our example of Figure 4.1, H will send a UDP packet over ILNP as: <UDP: I_H, I_CN, P_H, P_CN><ILNP: L_1, L_CN> --- (12a) and CN will send UDP packets to H as: <UDP: I_CN, I_H, P_CN, P_H><ILNP: L_CN, L_1> --- (12b) The I value for H used in the transport-layer state (I_H in expression (12a)) selects the correct L value (L_1 in this case) from the bindings in the ILCC (expression (7a)), and that, in turn, selects the correct interface from the ILCC (expression (7b)), as
described in Section 4.2. This gets the packet to the first hop router; beyond that, the ILNPv6 packet is treated as if it were an IPv6 packet.5. Multihoming and Multi-Path Transport
For multihoming, there are three cases to consider: a) Host Multihoming (H-MH): a single host is, individually, connected to multiple upstream links, via separate routing paths, and those multiple paths are used by that host as it wishes. That is, use of multiple upstream links is managed by the single host itself. For example, the host might have multiple valid Locator values on a single interface, with each Locator value being associated with a different upstream link (provider). b) Multi-Path Transport (MTP): This is similar to using ILNP's support for host multihoming (i.e., H-MH), so we describe multi-path transport here. (Indeed, for ILNP, this can be considered a special case of H-MH.) c) Site Multihoming (S-MH): a site network is connected to multiple upstream links via separate routing paths, and hosts on the site are not necessarily aware of the multiple upstream paths. That is, the multiple upstream paths are managed, typically, through a site border router, or via the providers. Essentially, for ILNP, multihoming is implemented by enabling: a) multiple Locator values to be used simultaneously by a node b) dynamic, simultaneous binding between one (or more) Identifier value(s) and multiple Locator values With respect to the requirements for hosts [RFC1122], the multihoming function provided by ILNP is very flexible. It is not useful to discuss ILNP multihoming strictly within the confines of the exposition presented in Section 3.3.4 of [RFC1122], as that text is couched in terms of relationships between IP Addresses and interfaces, which can be dynamic in ILNP. The closest relationship between ILNP multihoming and [RFC1122] would be that certainly ILNP could support the notion of "Multiple Logical Networks", "Multiple Logical Hosts", and "Simple Multihoming".
5.1. Host Multihoming (H-MH)
At present, host multihoming is not common in the deployed Internet. When TCP or UDP are in use with an IP-based network-layer session, host multihoming cannot provide session resilience, because the transport protocol's pseudo-header checksum binds the transport-layer session to a single IP Address of the multihomed node, and hence to a single interface of that node. SCTP has a protocol-specific mechanism to support node multihoming; SCTP can support session resilience both at present and also without change in the proposed approach [RFC5061]. Host multihoming in ILNP is supported directly in each host by ILNP. The simplest explanation of H-MH for ILNP is that an ILNP-capable host can simultaneously use multiple Locator values, for example, by having a binding between an I value and two different L values, e.g., the ILCC may contain the I-LVs: [I_1, L_1] --- (14a) [I_1, L_2] --- (14b) Additionally, a host may use several I values concurrently, e.g., the ILCC may contain the I-LVs: [I_1, L_1] --- (15a) [I_1, L_2] --- (15b) [I_2, L_2] --- (15c) [I_3, L_1] --- (15d) Architecturally, ILNP considers these all to be cases of multihoming: the host is connected to more than one subnetwork, each subnetwork being named by a different Locator value. In the cases above, the selection of which I-LV to use would be through local policy or through management mechanisms. Additionally, suitably modified transport-layer protocols, such as multi-path transport-layer protocol implementations, may make use of multiple I-LVs. Note that in such a case, the way in which multiple I-LVs are used would be under the control of the higher-layer protocol. Recall, however, that L values also have preference -- LPI values -- and these LPI values can be used at the network layer, or by a transport-layer protocol implementation, in order make use of L values in a specific manner. Note that, from a practical perspective, ILNP dynamically binds L values to interfaces on a node to indicate the SNPA for that L value, so the multihoming is very flexible: a node could have a single
interface and have multiple L values bound to that interface. For
example, for expressions (14a) and (14b), if the end-system has a
single interface with local name eth0, then the entries in the ILCC
will be:
L_1, eth0 --- (16a)
L_2, eth0 --- (16b)
And, if we assume that for expressions (15a-c) the end-system has two
interfaces, eth0 and eth1, then these ILCC entries are possible:
L_1, eth0 --- (17a)
L_2, eth1 --- (17b)
Let us consider the network in Figure 5.1.
site . . . .
network . .
. . . . +------+ L_1 . .
. . | +------. .
. . | | . .
. .----+ SBR | . Internet .
. . | | . .
. H . | +------. .
. . . . +------+ L_2 . .
. .
. . . .
L_1 = global Locator value 1
L_2 = global Locator value 2
SBR = Site Border Router
Figure 5.1: A Simple Multihoming Scenario for ILNP
We assume that H has a single interface, eth0. SBR will advertise
L_1 and L_2 internally to the site. Host H will configure these as
both reachable via its single interface, eth0, by using ILCC entries
as in expressions (16a) and (16b). When packets from H that are to
egress the site network reach SBR, it can make appropriate decisions
on which link to use based on the source Locator value (which has
been inserted by H) or based on other local policy.
If, however, H has two interfaces, eth0 and eth1, then it can use
ILCC entries as in expressions (17a) and (17b).
Note that the values L_1 and L_2 do not need to be PI-based Locator
values, and can be taken from ISP-specific PA routing prefix
allocations from the upstream ISPs providing the two links.
Of course, this example is illustrative: many other configurations
are also possible, but the fundamental mechanism remains the same, as
described above.
If any Locator values change, then H will discover this when it sees
new Locator values in RAs from SBR, and sees that L values that were
previously used are no longer advertised. When this happens, H will:
a) maintain existing active network-layer sessions: based on its
current ILCC entries and active sessions, send Locator Update
(LU) messages to CNs to notify them of the change of L values.
(LU messages are synonymous to Mobile IPv6 Binding Updates.)
b) if required, update its relevant DNS entries with the new L
value in the appropriate DNS records, to enable correct
resolution for new incoming session requests.
From an engineering viewpoint, H also updates its ILCC data, removing
the old L value(s) and replacing with new L value(s) as required.
Depending on the nature of the physical change in connectivity that
the L value change represents, this may disrupt upper-level
protocols, e.g., a fibre cut. Dealing with such physical-level
disruption is beyond the scope of ILNP. However, ILNP supports
graceful changes in L values, and this is explained below in Section
6 in the discussion on mobility support.
5.2. Support for Multi-Path Transport Protocols
ILNP supports deployment and use of multi-path transport protocols,
such as the Multi-Path extensions to TCP (MP-TCP) being defined by
the IETF TCPM Working Group. Specifically, ILNP will support the use
of multiple paths as it allows a single I value to be bound to
multiple L values -- see Section 5.1, specifically expressions (15a)
and (15b).
Of course, there will be specific mechanisms for:
- congestion control
- signalling for connection/session management
- path discovery and path management
- engineering and implementation issues
These transport-layer mechanisms fall outside the scope of ILNP and
would be defined in the multi-path transport protocol specifications.
As far as the ILNP architecture is concerned, the transport protocol
connection is simply using multiple I-LVs, but with the same I value
in each, and different L values, i.e., a multihomed host.
5.3. Site Multihoming (S-MH)
At present, site multihoming is common in the deployed Internet. This is primarily achieved by advertising the site's routing prefix(es) to more than one upstream Internet service provider at a given time. In turn, this requires de-aggregation of routing prefixes within the inter-domain routing system. This increases the entropy of the inter-domain routing system (e.g., RIB/FIB size increases beyond the minimal RIB/FIB size that would be required to reach all sites). Site multihoming, in its simplest form in ILNP, is an extension of the H-MH scenario described in Section 5.1. If we consider Figure 5.1, and assume that there are many hosts in the site network, then each host can choose (a) whether or not to manage its own ILNP connectivity, and (b) whether or not to use multiple Locator values. This allows maximal control of connectivity for each host. Of course, with ILNPv6, just as any IPv6 router is required to generate IPv6 Router Advertisement messages with the correct routing prefix information for the link the RA is advertised upon, the SBR is also required to generate RAs containing the correct Locator value(s) for the link that the RA is advertised upon. The correct values for these RA messages are typically configured by system administration, or might be passed down from the upstream provider. To avoid a DNS Update burst when a site or (sub)network changes location, a DNS record optimisation is possible by using the new LP record for ILNP. This would change the number of DNS Updates required from Order(Number of nodes within the site/subnetwork that moved) to Order(1) [RFC6742].5.3.1. A Common Multihoming Scenario - Multiple SBRs
The scenario of Figure 5.1 is an example to illustrate the architectural operation of multihoming for ILNP. For site multihoming, a scenario such as the one depicted in Figure 5.2 is also common. Here, there are two SBRs, each with its own global connectivity.
site . . . . network . . . . . . +-------+ L_1 . . . . | +------. . . . | | . . . .---+ SBR_A | . . . . | | . . . . | | . . . . +-------+ . . . . ^ . . . . | CP . Internet . . . v . . . . +-------+ L_2 . . . . | +------. . . . | | . . . .---+ SBR_B | . . . . | | . . . . | | . . . . . . +-------+ . . . . . . . . CP = coordination protocol L_1 = global Locator value 1 L_2 = global Locator value 2 SBR_A = Site Border Router A SBR_B = Site Border Router B Figure 5.2: A Dual-Router Multihoming Scenario for ILNP The use of two physical routers provides an extra level of resilience compared to the scenario of Figure 5.1. The coordination protocol (CP) between the two routers keeps their actions in synchronisation according to whatever management policy is in place for the site network. Such capabilities are available today in products. Note that, logically, there is little difference between Figures 5.1 and 5.2, but with two distinct routers in Figure 5.2, the interaction using CP is required. Of course, it is also possible to have multiple interfaces in each router and more than two routers.5.4. Multihoming Requirements for Site Border Routers
For multihoming, the SBR does NOT need to be ILNP-capable for host multihoming or site multihoming. This is true provided the multihoming is left to individual hosts as described above. In this deployment approach, the SBR need only issue Routing Advertisements
(RAs) that are correct with respect to its upstream connectivity; that is, the SBR properly advertises routing prefixes (Locator values) to the ILNP hosts. In such a scenario, when hosts in the site network see new Locator values, and see that a previous Locator value is no longer being advertised, those hosts can update their ILCCs, send Locator Updates to CNs, and change connectivity as required.6. Mobility
ILNP supports mobility directly, rather than relying upon special- purpose mobility extensions as is the case with both IPv4 [RFC2002] (which was obsoleted by [RFC5944]) and IPv6 [RFC6275]. There are two different mobility cases to consider: a) Host Mobility: individual hosts may be mobile, moving across administrative boundaries or topological boundaries within an IP-based network, or across the Internet. Such hosts would need to independently manage their own mobility. b) Network (Site) Mobility: a whole site, i.e., one or more IP subnetworks may be mobile, moving across administrative boundaries or topological boundaries within an IP-based network, or across the Internet. The site as a whole needs to maintain consistency in connectivity. Essentially, for ILNP, mobility is implemented by enabling: a) Locator values to be changed dynamically by a node, including for active network-layer sessions. b) use of Locator Updates to allow active network-layer sessions to be maintained. c) for those hosts that expect incoming network-layer or transport-layer session requests (e.g., servers), updates to the relevant DNS entries for those hosts. It is possible that a device is both a mobile host and part of a mobile network, e.g., a smartphone in a mobile site network. This is supported in ILNP as the mechanism for mobile hosts and mobile networks are very similar and work in harmony.
For mobility, there are two general features that must be supported:
a) Handover (or Hand-off): when a host changes its connectivity
(e.g., it has a new SNPA as it moves to a new ILNP subnetwork),
any active network-layer sessions for that host must be
maintained with minimal disruption (i.e., transparently) to the
upper-layer protocols.
b) Rendezvous: when a host that expects incoming network-layer or
transport-layer session requests has new connectivity (e.g., it
has a new SNPA as it moves to a new ILNP subnetwork), it needs
to update its relevant DNS entries so that name resolution will
provide the correct I and L values to remote nodes.
6.1. Mobility / Multihoming Duality in ILNP
Mobility and multihoming present the same set of issues for ILNP.
Indeed, mobility and multihoming form a duality: the set of Locators
associated with a node or site changes. The reason for the change
might be different for the case of mobility and multihoming, but the
effects on the network-layer session state and on correspondents is
identical.
With ILNP, mobility and multihoming are supported using a common set
of mechanisms. In both cases, different Locator values are used to
identify different IP subnetworks. Also, ILNP nodes that expect
incoming network-layer or transport-layer session requests are
assumed to have a Fully Qualified Domain Name (FQDN) stored in the
Domain Name System (DNS), as is already done within the deployed
Internet. ILNP mobility normally relies upon the Secure Dynamic DNS
Update standard for mobile nodes to update their location information
in the DNS. This approach of using DNS for rendezvous with mobile
systems was proposed earlier by others [PHG02].
Host Mobility considers individual hosts that are individually mobile
-- for example, a mobile telephone carried by a person walking in a
city. Network (Site) Mobility considers a group of hosts within a
local topology that move jointly and periodically change their
uplinks to the rest of the Internet -- for example, a ship that has
wired connections internally but one or more wireless uplinks to the
rest of the Internet.
For ILNP, Host Mobility is analogous to host multihoming (H-MH) and
Network Mobility is analogous to site multihoming (S-MH). So,
mobility and multihoming capabilities can be used together, without
conflict.
6.2. Host Mobility
With Host Mobility, each individual end-system manages its own connectivity through the use of Locator values. (This is very similar to the situation described for H-MH in Section 5.1.) Let us consider the network in Figure 6.1. site . . . . network A . . . . . . +-------+ L_A . . . . | +------. . . . | | . . . .---+ SBR_A | . . . . | | . . . H(1) . | | . . . . +-------+ . . . . . . . . . . . H(2) . . Internet . . . . . . . . . . . +-------+ L_B . . . H(3) . | +------. . . . | | . . . .---+ SBR_B | . . . . | | . . . . | | . . . . . . +-------+ . . site . . network B . . . . H(X) = host H at position X L_A = global Locator value A L_B = global Locator value B SBR = Site Border Router Figure 6.1: A Simple Mobile Host Scenario for ILNP A host H is at position (1), hence H(1) in a site network A. This site network might be, for example, a single radio cell under administrative domain A. We assume that the host will move into site network B, which might be a single radio cell under administrative domain B. We also assume that the site networks have a region of overlap so that connectivity can be maintained; else, of course, the host will lose connectivity. Also, let us assume that the host already has ILNP connectivity in site network A.
If site network A has connectivity via Locator value L_A, and H uses
Identifier value I_H with a single interface ra0, then the host's
ILCC will contain:
[I_H, L_A] --- (18a)
L_A, ra0 --- (18b)
Note the equivalence of expressions (18a) and (18b), respectively,
with the expressions (15a) and (16a) for host multihoming.
The host now moves into the overlap region of site networks A and B,
and has position (2), hence H(2) as indicated in Figure 6.1. As this
region is now in site network B, as well as site network A, H should
see RAs from SBR_B for L_B, as well as the RAs for L_A from SBR_A.
The host can now start to use L_B for its connectivity. The host H
must now:
a) maintain existing active upper-layer sessions: based on its
current ILCC entries and active sessions, send Locator Update
(LU) messages to CNs to notify them of the change of L values.
(LU messages are synonymous to Mobile IPv6 Binding Updates.)
b) if required, update its relevant DNS entries with the new L
value in the appropriate DNS records, to enable correct
resolution for new incoming network-layer or transport-layer
session requests.
However, it can opt to do this one of two ways:
1) immediate handover: the host sends Locator Update (LU) messages
to CNs, immediately stops using L_A, and switches to using L_B
only. In this case, its ILCC entries change to:
[I_H, L_B] --- (19a)
L_B, ra0 --- (19b)
There might be packets in flight to H that use L_A, and H MAY
choose to ignore these on reception.
2) soft handover: the host sends Locator Update (LU) messages to
CNS, but it uses both L_A and L_B until (i) it no longer
receives incoming packets with destination Locator values set
to L_A within a given time period and (ii) it no longer sees
RAs for L_A (i.e., it has left the overlap region and so has
left site network A). In this case, its ILCC entries change
to:
[I_H, L_A] --- (20a)
L_A, ra0 --- (20b)
[I_H, L_B] --- (20c)
L_B, ra0 --- (20d)
ILNP does not mandate the use of one handover option over another.
Indeed, a host may implement both and decide, through local policy or
other mechanisms (e.g., under the control of a particular transport
protocol implementation), to use one or other for a specific
transport-layer session, as required.
Note that if using soft handover, when in the overlap region, the
host is multihomed. Also, soft handover is likely to provide a less
disruptive handover (e.g., lower packet loss) compared to immediate
handover, all other things being equal.
There is a case where both the host and its correspondent node are
mobile. In the unlikely event of simultaneous motion that changes
both nodes' Locators within a very small time period, there is the
possibility that communication may be lost. If the communication
between the nodes was direct (i.e., one node initiated communication
with another, through a DNS lookup), a node can use the DNS to
discover the new Locator value(s) for the other node. If the
communication was through some sort of middlebox providing a relay
service, then communication is more likely to disrupted only if the
middlebox is also mobile.
It is also possible that high packet loss results in Locator Updates
being lost, which could disrupt handover. However, this is an
engineering issue and does not impact the basic concept of operation;
additional discussion on this issue is provided in [RFC6741].
Of course, for any handover, the new end-to-end path through SBR_B
might have very different end-to-end path characteristics (e.g.,
different end-to-end delay, packet loss, throughput). Also, the
physical connectivity on interface ra0 as well as through SBR_B's
uplink may be different. Such impacts on end-to-end packet transfer
are outside the scope of ILNP.
6.3. Network Mobility
For network mobility, a whole site may be mobile, e.g., the SBRs of
Figure 6.1 have a radio uplink on a moving vehicle. Within the site,
individual hosts may or may not be mobile.
In the simplest case, ILNP deals with mobile networks in the same way
as for site multihoming: the management of mobility is delegated to
each host in the site, so it needs to be ILNP-capable. Each host,
effectively, behaves as if it were a mobile host, even though it may not actually be mobile. Indeed, in this way, the mechanism is very similar to that for site multihoming. Let us consider the mobile network in Figure 6.2. site ISP_1 network SBR . . . . . . . +------+ L_1 . . . . | ra1+------. . . .----+ | . . . H . | ra2+-- . . . . . . +------+ . . . . . Figure 6.2a: ILNP Mobile Network before Handover site ISP_1 network SBR . . . . . . . +------+ L_1 . . . . | ra1+------. . . . . . .----+ | . . . H . | ra2+------. . . . . . +------+ L_2 . . . . . . . . . . ISP_2 Figure 6.2b: ILNP Mobile Network during Handover site ISP_2 network SBR . . . . . . . +------+ . . . . | ra1+-- . . . .----+ | . . . H . | ra2+------. . . . . . +------+ . . . . . Figure 6.2c: ILNP Mobile Network after Handover H = host L_1 = global Locator value 1 L_2 = global Locator value 2 SBR = Site Border Router Figure 6.2: A Simple Mobile Network Scenario for ILNP
In Figure 6.2, we assume that the site network is mobile, and the SBR has two radio interfaces ra1 and ra2. However, this particular figure is chosen for simplicity and clarity for our scenario, and other configurations are possible, e.g., a single radio interface which uses separate radio channels (separate carriers, coding channels, etc.). In the figure, ISP_1 and ISP_2 are separate, radio- based service providers, accessible via ra1 and ra2. In Figure 6.2a, the SBR has connectivity via ISP_1 using Locator value L_1. The host H, with interface ra0 and Identifier I_H, has an established connectivity via the SBR and so has ILCC entries as shown in (21): [I_H, L_1] --- (21a) L_1, ra0 --- (21b) Note the equivalence to expressions (18a) and (18b). As the whole network moves, the SBR detects a new radio provider, ISP_2, and connects to it using ra2, as shown in Figure 6.2b, with the service areas of ISP_1 and ISP_2 overlapping. ISP_2 provides Locator L_2, which the SBR advertises into the site network along with L_1. As with the mobile host scenario above, individual hosts may decide to perform immediate handover or soft handover. So, the ILCC state for H will be as for expressions (19a) and (19b) and (20a)-(20d), but with L_1 in place of L_A, and L_2 in place of L_B. Finally, as in Figure 6.2c, the site network moves and is no longer served by ISP_1, and handover is complete. Note that during the handover the site is multihomed, as in Figure 6.2b.6.4. Mobility Requirements for Site Border Routers
As for multihoming, the SBR does NOT need to be ILNP-capable: it simply needs to advertise the available routing prefixes into the site network. The mobility capability is handled completely by the hosts.6.5. Mobility with Multiple SBRs
Just as Section 5.3.1 describes the use of multiple routers for multihoming, so it is possible to have multiple routers for mobility for ILNP, for both mobile hosts and mobile networks.
(page 36 continued on part 3)
