| Internet-Draft | Semantic Networking Survey | November 2022 |
| King & Farrel | Expires 28 May 2023 | [Page] |
The Internet Protocol (IP) has become the global standard in any computer network, independent of the connectivity to the Internet. Generally, an IP address is used to identify an interface on a network device. Routing protocols are also required and developed based on the assumption that a destination address has this semantic with routing decisions made on addresses and additional fields in the packet headers.¶
Over time, routing decisions were enhanced to route packets according to additional information carried within the packets and dependent on policy coded in, configured at, or signaled to the routers. Many proposals have been made to add semantics to IP addresses. The intent is usually to facilitate routing decisions based solely on the address and without finding and processing information carried in other fields within the packets.¶
This document is presented as a survey to support the study and further research into clarifying and understanding the issues. It does not pass comment on the advisability or practicality of any of the proposals and does not define any technical solutions¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 28 May 2023.¶
Copyright (c) 2022 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
The Internet continues to expand rapidly, and the Internet Protocol (IP) has become the global standard in many types of computer network independent of whether or what connectivity to the Internet it has. At the same time, there are increasingly varied expectations of the services and service level objectives that can be required from networks. For example, packet-delivery quality expectations beyond best effort is a growth area: throughput, latency, error recovery, and (absence of) packet or connectivity loss, reordering, or jitter. Requirements include relative or absolute guarantees or predictable elastic changes under contention on these performance factors. This places significant pressure on Service Providers to be aware of the type of services being delivered, and to have access to sufficient information about how individual packets should be treated to meet the user, application and application instance requirements.¶
IP addresses facilitate the identification of how a device is attached to the Internet and how it is distinguished from every other device. Addresses are used to direct packets to a destination (destination address) and indicate to where the receiver and network replies and error messages should be sent (source address). An IP address may be assigned to each network interface of a device connected to a network that uses IP. Applications use IP addresses to both identify a host and to indicate the physical or virtual location of the host.¶
This document presents a brief survey of proposals to extend the semantics of IP addresses by assigning additional meanings to some parts of the address, or by partitioning the address into a set of subfields that give scoped addressing instructions. Some of these proposals are intended to be deployed in limited domains [RFC8799] that are IP-based, while other proposals are intended for use across the Internet. Limited domains may present their own challenges in terms of ensuring the perimeter of the domains, and connecting domains across the Internet.¶
The impact that some proposals may have on routing systems could require clean-slate solutions, hybrid solutions, extensions to existing routing protocols, or potentially no changes at all. A separate document ([I-D.king-rtgwg-challenges-in-routing]) describes the challenges to the routing system presented by changes to IP address semantics, and sets out research questions that should be investigated by those proposing new semantic address schemes.¶
Two approaches are typically used for network path selection. Firstly, a priori assessment by having the feasible paths and constraints computed in advance. Secondly, real-time computation in response to changing network conditions.¶
The first approach may be conducted offline and allows for concurrent or global optimization based on constraints and policy. However. as network size and complexity increase, the required computing power may increase exponentially for this type of computation.¶
The second approach must consider the speed of calculation where complex constraints are applied to the path selection. This processing may delay service setup and the responsiveness to changes (such as failures) in the network. Network topology filters may be applied to reduce the complexity of the network data and the computation algorithm, however, the path computation accuracy and optimality may be negatively affected.¶
In both approaches, the amount of information that needs to be imported and processed can become very large (e.g., in large networks, with many possible paths and route metrics), which might impede the scalability of either method both in terms of the storage and the distribution of the information.¶
In the last decade, significant research has been conducted into the architecture of the future Internet (for example, [RESEARCHFIAref] and [ITUNET2030ref]). During this research, several techniques emerged, highlighting the benefits of path awareness and path selection for end-hosts, and multiple path-aware network architectures have been proposed, including SCION [SCIONref] and RINA [RINAref], and the work of the Path Aware Networking Research Group (PANRG) as discussed later in this document.¶
When choosing the best paths or topology structures, the following may need to be considered:¶
The current architecture for IP networking is built using a best-effort philosophy. There are techniques discussed in this document that attempt better-than-best-effort delivery. The start-point and end-point of a path are identified using IP addresses, and traffic is steered along the path that does not necessarily follow the "shortest path first" route through the network. Furthermore, the path might not run all the way from a packet's source to its destination. The assumption is that a packet reaching the end of a path is forwarded to its destination using best-effort techniques.¶
Evaluating and building paths that respect requirements beyond the simple best-effort model is particularly challenging and computationally heavy since numerous quality-related parameters need to be considered.¶
Networks are often divided into addressing regions for various administrative or technological reasons. Different routing paradigms may be applied in each region, and specific "private" semantics may be applied to the IP addresses within a single region¶
These address semantics are established using customer types, customer connections, topological constraints, performance groups, and security, etc. Service Providers or network operators will apply local policies to user and application packets as they enter the network possibly mapping addresses, or encapsulating them with an additional IP header. In some case, the packet has its source and destination within a single network and the network operator can apply address semantics policies across the whole network. In other cases (such as general IP-based traffic), a packet will require a path across multiple networks, and each may apply its own set of traffic forwarding policies. In these cases, there is often no consistency or guaranteed performance unless a Service Level Agreement (SLA) is applied to traffic traversing multiple networks.¶
Semantic networking proposals may apply to addresses in a specific domain, or domain set. In this context, a "limited domain" means that the interpretation of the address, in a semantic networking domain, is only applicable to a well-defined set of network nodes, or specific points in the network. If a packet bearing an address with a modified semantic were to escape from the domain, the special meaning of the address would be lost. Additionally (or alternatively), the meaning of "specific points in the network" may be applied to the source and destination nodes of a packet, while all transit nodes are unaware of the special semantic. However, it could be the case that some key transit nodes are able to access the meaning of the address and so apply special routing or other functions to the packet.¶
Such proposals include the following:¶
In many cases, it may be argued that existing mechanisms applied on top of the common address semantic defined in [RFC4291] can deliver the correct functionality for these scenarios. That is, packets may be tunneled over IP using several existing encapsulation techniques. Nevertheless, there is pressure to reduce the amount of encapsulation (partly to resist reduction in the maximum transmission unit (MTU) over the network, and partly to achieve a flatter and more transparent network architecture). This leads to investigations into whether the current IP addresses can be "overloaded" (without any negative connotations being attached to that word) by adding semantics to the addresses.¶
Semantic Networking is the process of routing and forwarding packets that contain IP addresses with additional semantics, possibly using that information to perform policy-based routing or other enhanced routing functions. Thus, faciliating enhanced routing decisions based on these additional semantics and provide differentiated paths for different packet flows, distinct from simple shortest path first routing. The process known as Semantic Networking is described further in [I-D.farrel-rtgwg-intro-to-semantic-networking].¶
Key use cases exist for semantic networking, typically for specific applications and deplyments, including low earth orbit (LEO) satellite constellations [I-D.lhan-satellite-semantic-addressing].¶
Based on a variety of use cases, key technical challenges exist for semantic networking: these are discussed further in [I-D.king-rtgwg-challenges-in-routing].¶
Semantics may be applied in multiple ways to integrate with existing routing architectures. The most obvious is to build an overlay such that IP is used only to route packets between network nodes that utilize the semantics at a higher layer. There are several uses of this approach, including Service Function Chaining (SFC) (see Section 4.2.3) and Information Centric Networking (ICN) (see Section 5.4.1). An overlay may be achieved in a higher layer, or may be performed using tunneling techniques (such as IP-in-IP) to traverse the areas of the IP network that cannot parse additional semantics and so join together those nodes that use the semantics.¶
The application of semantics may also be constrained to within a limited domain. In some cases, such a domain will use IP, but be disconnected from Internet. In those cases, the challenges are limited to enabling the desired function within the domain. In other cases, traffic from within the domain is exchanged with other domains that are connected across an IP-based network using tunnels or via application gateways. And in another case traffic from the domain is routed across the Internet to other nodes and this requires backward-compatible routing approaches, tunnels, or gateway functions.¶
Limited domains [RFC8799] are a fact of networking life. They are used to safely deploy or test features and functions in a controlled environment so that they cannot contaminate other networks or the Internet in general. Examples of a limited domain in use today include:¶
Limited domains are also used as a driver for innovation. They provide a safe space to run experiments and deploy new functions such as advances in traffic steering, improvements in security, and new routing protocols. A limited domain is a way to achieve incremental deployability on an isolated island, and this enables innovation that may (or may not) percolate to the whole Internet at a later stage. For example, experiments to increase the programmability of network forwarding functions need to be carried out in networks of similarly capable nodes (to avoid the risks of broken interoperability or forwarding loops), yet these experiments need to use real user data that is flowing between hosts and servers.¶
Because limited domains don't always operate in isolation they may need to be connected to other domains over the Internet, or other nodes within the wider Internet.¶
Several IETF-based approaches are available to allow service providers to perform policy-based routing, including identifying and marking IP traffic either by changing the semantic of IP addresses or by adding such a semantic in other fields/namespaces, enabling differentiated handling by transit routers (queuing, dropping, forwarding, etc.). The sections below distinguish between those schemes that perform routing based on information other than IP addresses, those that establish an overlay network in which to apply semantics, and those that add semantics to the addresses. A further separate group of approaches is presented here to cover the concept of group semantics where a single address identifies more than one endpoint.¶
Many routing schemes examine the destination address field and other fields in the packet header to make routing decisions. These approaches (sometimes referred to as "policy-based routing") allow packets to follow different paths through the network depending on semantics assigned to these other fields or based on hashing algorithms operating on the values of those fields.¶
Deep Packet Inspection (DPI) may be used by a router to learn the characteristics of packets in order to forward them differently. This involves looking into the packet beyond the top-level network-layer header to identify the payload. Once identified, the traffic type can be used as an input for marking the packets for network handling, or for performing specific policies on the packets.¶
However, DPI may be expensive both in processing costs and latency. The processing costs means that dedicated infrastructure is necessary to carry out the function, and this may have an associated financial cost. The latency incurred may be too much for use with any delay/jitter sensitive applications. As a result, DPI is difficult for large-scale deployment and its usage is often limited to specific functions at the edge of the network.¶
Despite this, "shallow DPI" is commonplace in routers today as they examine the five-tuple of source address, destination address, payload protocol, source port, and destination port to perform a hash function for ECMP purposes (a form of policy-based routing).¶
Quality of Service (QoS) based on Differentiated Services [RFC2474] is a widely deployed framework specifying a simple and scalable coarse-grained mechanism for classifying and managing network traffic. However, in a service providers network, DiffServ codepoint (DSCP) values cannot be trusted when they are set by the customer, and may have different meanings as packets are passed between networks.¶
In real-world scenarios, Service Providers deploy "remarking" points at the edges of their network, re-classifying received packets by rewriting the DSCP field according to local policy using information such as the source/destination address, IP protocol number, transport layer source/destination ports, and possibly applying DPI as described in Section 4.1.1.¶
The traffic classification process and node-by-node processing leads to increased packet processing overhead and complexity at the edge of the Service Providers network.¶
[RFC8200] defines the IPv6 header and also a number of extension headers. These extension headers can be used to carry additional information that may be used by transit routers (the hop-by-hop options header) or by the destination identified by the destination address field (the destination options header). In addition, these extension headers could encode additional semantics that might enable routing decisions and determine what functions and operations should be performed on a packet.¶
[RFC7872] and [I-D.ietf-v6ops-ipv6-ehs-packet-drops] provide some discussions about the operational problems of using IPv6 extension headers, especially in multi-domain environments, while [I-D.bonica-6man-ext-hdr-update] proposes to update RFC 8200 with guidance regarding the processing, insertion and deletion of IPv6 extension headers.¶
An overlay network is built on top of an underlay or transport network. Packets are encapsulated with the header for the overlay network to carry the additional information needed to provide the desired function, and then the packets are encapsulated for transport through the underlay network. In this case, no changes are made to the meaning of the IP addresses in the underlay, but the destination address identifies the next hop in the overlay network rather than the ultimate destination of the packet. In this way, packets can be steered through different overlay nodes where routing decisions can be made.¶
Application-Layer Traffic Optimization (ALTO) [RFC7285] is an architecture and protocol. ALTO defines abstractions and services to provide simplified network views and network services to guide the application usage of network resources, including cost.¶
An ALTO server gathers information about the network and answers queries from an ALTO client that wants to find a suitable path for traffic. ALTO responses are typically used to route whole flows (not individual packets) either to suitable destinations (such as network functions) or onto paths that have specific qualities.¶
Multipath TCP (MPTCP) [RFC8684] enables the use of TCP in a multipath network using multiple host addresses. A Multipath TCP connection provides a bidirectional bytestream between two hosts communicating like normal TCP and thus does not require any change to the applications. However, Multipath TCP enables the hosts to use different paths with different IP addresses to exchange packets belonging to the MPTCP connection.¶
MPTCP increases the available bandwidth, and so provides shorter delays; it increases fault tolerance, by allowing the use of other routes when one or more routes become unavailable; and it enables traffic engineering and load balancing.¶
Service Function Chaining (SFC) [RFC7665] is the process of sending traffic through an ordered set (a sequence) of abstract service functions. This may be achieved using an overlay encapsulation such as the Network Service Header (NSH) [RFC8300] or MPLS [RFC8596] that rely on tunneling through an underlay without any additional semantics applied to the IP addresses.¶
Alternatively, SFC can be performed by adding semantics to the addresses, for example, as in Section 4.3.3.¶
The Path Computation Element (PCE) [RFC4655] is an architecture and protocol [RFC5440] that can be used to assist with network path selection. A PCE is an entity capable of computing paths for a single or set of services. A PCE might be a network node, network management station, or dedicated computational platform that is resource-aware and has the ability to consider multiple constraints for sophisticated path computation. PCE applications compute label switched paths for MPLS and GMPLS traffic engineering, but the PCE has been extended for a variety of additional traffic engineering problems.¶
In semantic networking, additional information or meaning is placed into the IP address, and this is used to route packets within the network.¶
The Locator/ID Separation Protocol (LISP) [RFC6830] was published by the IETF as an Experimental RFC in 2013 and is now being moved to the Standards Track [I-D.ietf-lisp-rfc6830bis] and [I-D.ietf-lisp-rfc6833bis]. LISP separates IP addresses into two numbering spaces: Endpoint Identifiers (EIDs) and Routing Locators (RLOCs). The former, the EIDs, are used to identify communication end-points (as the name states) as well as local routing and forwarding in the edge network. The latter, RLOCs, are used to locate the EIDs in the Internet topology end are usually the address of ASBRs (Autonomous System Border Routers). IP packets addressed with EIDs are encapsulated with RLOCs for routing and forwarding over the Internet.¶
As end-to-end packet forwarding includes both EIDs and RLOCs an additional control-plane is needed. This control plane provides a mapping system and basic traffic engineering capabilities. Multihoming becomes easier because one EID can be associated to more than one RLOC or even to a local network address prefix.¶
The Identifier-Locator Network Protocol (ILNP) [RFC6740] is an experimental network protocol designed to separate the two functions of network addresses: identification of network endpoints, topology or location information. Differently from LISP, ILNP encodes both locator and identifier in the IPv6 address format (128 bits). More specifically, the most significant 64 bits of the 128 bits IPv6 address is the locator, while the less significant 64 bits form the identifier. Upon reaching the destination network, a cache is used to find the corresponding node. Furthermore, DNS can be dynamically updated, which is essential for mobility and also for provider-independent addresses. Similar to LISP, multihoming can be set by assigning multiple locators to the same identifier. In addition, identifiers can also be encrypted for privacy reasons. It was intended that ILNP should be backwards-compatible with existing IP, and that it should be incrementally deployable.¶
Segment Routing (SR) [RFC8402] leverages the source routing paradigm. A node steers a packet through an ordered list of instructions, called "segments". A segment can represent any instruction, topological or service based. A segment can have a semantic local to an SR node or global within an SR domain. SR provides a mechanism that allows a flow to be restricted to a specific topological path, while maintaining per-flow state only at the ingress node(s) to the SR domain.¶
In SR for IPv6 networks (SRv6) segment routing functions are used to achieve a networking objective that goes beyond packet routing, in order to provide "network programming" [RFC8986]. The network program is expressed as a list of instructions, which are represented as 128-bit segments, called Segment Identifiers (SID) - encoded and presented in the form of an IPv6 address. The first instruction of the network program is placed in the Destination Address field of the packet. If the network program requires more than one instruction, the remaining list of instructions is placed in the Segment Routing Extension Header (SRH)[RFC8754].¶
An SRv6 instruction can represent any topological or service-based instruction. The SRv6 domain is the service provider domain where SRv6 services are built to transport any kind of customer traffic including IPv4, IPv6, or frames. SRv6 is the instantiation of Segment Routing deployed on the IPv6 data plane. Therefore, in order to support SRv6, the network must first be enabled for IPv6.¶
The SRH in the IPv6 header is only processed for nodes forwarding traffic if the destination address identifies the local node. In this case, the node must take several actions, including reading the SRH, performing any node-specific actions identified by the destination address or the next SIDs in the SRH, and re-writing the IPv6 destination address field using information from the SRH before forwarding the packet.¶
Preferred Path Routing (PPR) [I-D.chunduri-lsr-isis-preferred-path-routing] is a proposed routing protocol mechanism where alternate forwarding state is installed for a set of different preferred paths. Each preferred path is described as an ordered linear list of nodes, links, and network functions, and the path is identified by a network-global preferred path identifier. If a packet is marked with preferred path identifier, it is forwarded according to the preferred path that has been installed on the router. If a packet is not marked or if the preferred path is not installed on the router, the packet is forwarded using the normal shortest path first algorithm.¶
In PPR, the preferred path identifier is encoded in an IP address, but the address is only used in an encapsulation of the end-to-end packet. This approach is a hybrid in that it is applying a different meaning to the IP addresses, using that meaning in an encapsulation, but routing the packets through an existing IP network.¶
The Connectionless Network Protocol (CLNP) [CLNPref] is a network layer encoding that supports variable length, hierarchical addressing. It is widely deployed in many communications networks and is the ITU-T's standardized encoding for packets in the management plane for Synchronous Digital Hierarchy (SDH) networks. For a while, CLNP was considered in competition with IP as the network layer encoding for the Internet, but IP (in conjunction with TCP) won out.¶
Many of the considerations for semantic addressing can be handled using CLNP, and it is particularly well suited to applications that demand variable-length addresses or that structure addresses hierarchically for routing or geo-political reasons.¶
Routing for CLNP can be achieved using the IS-IS routing protocol in its full form as documented in [ISISref] rather than its IP-only form [RFC1195]. While this may make it possible to use CLNP alongside IP in some routed networks, it does not integrate the use of IP addresses with additional semantics with the historic use of IP addresses except in "ships that pass in the night" fashion. Alternatively, [RFC1069] explains how to carry regular IP addresses in CLNP.¶
A mayor enhanced addressing semantic in IP is called "group semantics". Here, an IP address identifies more than one individual interface or node. This facilitates the delivery of a packet to any one of a group of destinations, or to all group members.¶
Multicast address semantics support delivery to all members of a group of destinations. This is a controlled variant of broadcasting where packets are delivered to all possible receivers in a particular (static) scope such as a multi-access link. Membership of a multicast link is dynamically signalled by the group members, and a group is identified by a specific address.¶
IP multicast [RFC1112], based on the protocol and service definition aspects of Steve Deering's PhD, is widely deployed for IPv4. It is equally adopted and used in IPv6 using the addressing architecture specified in [RFC4291]. In IP multicast (Any Source Multicast - ASM) any node can send to the multicast group and have its packets delivered to all members of the group.¶
Research deployments in the 1990s (the so called 'MBone' [MBONEref]) indicated that IP multicast gave rise to a number of issues related to address assignment, implementation, scale, and security. The problem of allocation and management of IP multicast (group) addresses led to several proposals, including Multicast Address Dynamic Client Allocation Protocol (MADCAP) [RFC2730], the Multicast Address Allocation Architecture (MALLOC) [RFC6308], the Multicast Address-Set Claim Protocol (MASC) [RFC2909], and the Multicast-Scope Zone Announcement Protocol (MZAP) [RFC2776], but none was widely adopted. Attempts to create a complete routing protocol suite for IP multicast service model within the IETF resulted in the Multicast Source Discovery Protocol (MSDP) being published as an experimental RFC [RFC3618].¶
The popularity of multicast as a concept and the widespread deployment of commercial IPv4 multicast led to the development of "Source Specific Multicast" service (SSM) [RFC4607]. In SSM, the combination of the Source and Group addresses (S,G) of an IP multicast packet form a so-called SSM channel address, which identifies group of receivers and implies a single permitted sender. Receivers subscribe to every SSM channel.¶
From a service user's perspective, SSM solves the security issue (only valid sources can send traffic) and the address assignment issue (all group addresses are relative to the source address). For the operator, SSM also eliminates the complex operational requirements of ASM.¶
Automatic Multicast Tunneling (AMT) [RFC7450] is a protocol for delivering multicast traffic from sources in a multicast-enabled network to receivers that lack multicast connectivity to the source network. The protocol uses UDP encapsulation and unicast replication to provide this functionality as a hybrid solution using both multicast routing and an overlay approach.¶
The IETF standardized or otherwise deployed protocol solutions in support of ASM and SSM in about 2015 relied all on per-hop, per ASM-group/per-SSM-channel stateful hop-by-hop forwarding/replication. Service Provider at that time were starting to removing or reduce heavy-weight control and per-hop forwarding processing in unicast caused by MPLS LDP/RSVP-TE driven designs, replacing it with more lightweight MPLS-SR and later SRv6 forwarding and associated control planes. But to reduce the cost for multicast service, the only transit-hop stateless solution available was ingress-replication, tunnel multicast across unicast, hence trading hop-by-hop state (and its control and management plane cost) in the network against traffic overhead and (under congestion) higher latency.¶
Bit Index Explicit Replication (BIER) [RFC8279] addresses these problems. BIER does not contain the notion of ASM or SSM groups. Instead, a sender enumerates the set of receivers to which the packet is to be delivered. The network routers forward packets and replicate them onto the shortest paths to the destinations. As the packets are replicated, so the enumeration of the receivers is pruned on each copy of the packet.¶
BIER is able to use existing routing protocols without modification, but requires enhancements in the forwarding plane to encode, parse, and act on the set of receivers. The BIER information is carried in new encapsulations [RFC8296] that is carried hop-by-hop in IP. Thus, the additional semantic is in an overlay.¶
This section presents a limited survey of techniques and projects that provide mechanisms to facilitate path and forwarding decisions based on contextual information.¶
More recently, the proceedings of the June 2021 Semantic Addressing and Routing for Future Networks (SARNET-21) Workshop was compiled and published as [I-D.galis-irtf-sarnet21-report]. It captures the views and positions of the participants as expressed during the workshop.¶
Some research work is engaged in examining the emerging set of new requirements that exceed the network and transport services of the current Internet, which only delivers "best effort" service. This work aims to determine what features can be built on top of existing solutions by adding additional new components or features. A starting point for this discussion can be found in [I-D.bryant-arch-fwd-layer-ps].¶
Several additional techniques for improving IP-based routing have been proposed, some of these are highlighted below.¶
The IRTF's Path Aware Networking Research Group [PANRGref] aims to support research in bringing path aware techniques into use in the Internet. This research overlaps with many past and existing IETF and IRTF efforts, including multipath transport protocols, congestion control in multiply-connected environments, traffic engineering, and alternate routing architectures.¶
[I-D.irtf-panrg-path-properties] offers a vocabulary of path properties. By doing so it gives some clarity of the distinction between path aware routing and semantic networking as considered in this document.¶
[I-D.irtf-panrg-what-not-to-do] provides a catalog and analysis of past efforts to develop and deploy Path Aware techniques. Most, but not all, of these mechanisms were considered at higher levels, although some apply at the IP routing and forwarding layer.¶
The SCION (Scalability, Control, and Isolation on Next-Generation Networks) [SCIONref] inter-domain network architecture has been designed to address security and scalability issues and provides an alternative to current Border Gateway Protocol (BGP) solutions. The SCION proposal combines a globally distributed public key infrastructure, a way to efficiently derive symmetric keys between any network entities, and the forwarding approach of packet-carried forwarding state.¶
SCION End-hosts fetch viable path segments from the path server infrastructure, and construct the exact forwarding route themselves by combining those path segments. The architecture ensures that a variety of combinations among the path segments are feasible, while cryptographic protections prevent unauthorized combinations or path-segment alteration. The architecture further enables path validation, providing per-packet verifiable guarantees on the path traversed.¶
Recursive Inter Network Architecture (RINA) [RINAref] builds upon the principle that applications communicate through Inter-process Communication (IPC) facilities. For an application to communicate through the distributed IPC facility, it only needs to know the name of the destination application and to use the IPC interface to request communication.¶
By leveraging IPC concepts, RINA allows two processes to communicate, IPC requires certain functions such as locating processes, determining permission, passing information, scheduling, and managing memory. Similarly, two applications on different end-hosts should communicate by utilizing the services of a distributed IPC facility (DIF). A DIF is an organizing structure, generally referred to as a "layer".¶
The scope and functions provided by the different IPC facilities may vary given the different type of network and performance goals. Moreover, an IPC layer may recursively request services from other IPC layers. The idea of recursively using multiple inter-process communication services creates a multilayer structure repeated until an IPC facility can fit well for physical technologies, e.g., wired or wireless networks.¶
Information-Centric Networking (ICN) [ICNref] is an approach to evolve the Internet infrastructure away from a host-centric paradigm, based on perpetual connectivity and the end-to-end principle, to a network architecture in which the focal point is information (or content or data) that is assigned specific identifiers.¶
Several scenarios exist for semantic-based networking, providing reachability based on Content Routing [CONTENTref] and Name Data Networking [NDNref]. The technology area of ICN is now reaching maturity, after many years of research and commercial investigation. A technical discussion into the deployment and operation of ICNs continues in the IETF: [RFC8763] provides several important deployment considerations for facilitating ICN and practical deployments.¶
Although ICN is primarily an overlay technology, a more recently concept, Hybrid-Information-Centric Networking (hICN), has been introduced [HICNref]. In an hICN environment the ICN aspect is integrated into the IPv6 architecture, reusing existing IPv6 packet formats with the intention of maintaining compatibility with existing and deployed IP network technology without creating overlays that might require a new packet format or additional encapsulations. The work is described in [I-D.muscariello-intarea-hicn].¶
Dyncast (Dynamic anycast) addresses the problem of directing traffic from a client to one service instance among several available, while considering decision metrics beyond shortest path when doing so. Those service instances are therefore possible destinations for a specific service demand. [I-D.liu-dyncast-ps-usecases] outlines several use cases where such traffic steering requirement is desirable and may occur, such as in edge computing scenarios but also in distribution of video content in scenarios like autonomous driving. The draft also outlines problems with existing solutions, most notably latency in changing relations from one service instance to another due to a change in metric, which defines that decision (e.g., load in servers, latency, or a combination of several such metrics).¶
Key to the proposed dyncast [I-D.li-dyncast-architecture] architecture is to build on the notion of (IP) anycast, while changing the addressing semantic from a locator-based addressing to a service-oriented one. Here, the initial "service demand" packet is being identified through a service identifier as destination address. This identifier is then mapped onto a binding IP (locator-based) address at the ingress of the network, allowing for locator-based routing to be used throughout the network. The ingress-based architecture is designed in such a way that ingress nodes upon arrival of a new service demand can determine which instance (i.e., which binding IP address) to use considering both network- and service-related metrics. Furthermore, these metrics can be distributed among ingress nodes in various ways, including over a routing protocol solution.¶
The overall forwarding decision is based on the adherence to what is termed "instance affinity", i.e., the need to adhere to a previous routing decision for more than one packet, unlike IP forwarding on locator addresses. This affinity is created, by means of a binding table on the ingress nodes, since often more than one packet is needed for the overall service-level transaction with a specific service instance. For instance, HTTP requests may span more than one routed packet. Also, a service instance may also create ephemeral state, which requires the client to continue communicating with this instance for the duration of this state. While the affinity is entirely defined by the application layer protocol, the network layer takes the affinity marking as input into the decision to renew its routing decision.¶
A modification to anycast that can be instantiated by additional engineering in the routing system is called "prioritycast". Instead of relying on the shortest path forwarding semantic, prioritycast directs all traffic to the anycast address instance that is reachable and has the highest priority. This approach only requires small modifications to routing protocols so that priorities are advertised along side the addresses.¶
Prioritycast was originally introduced as a recommended operational practice for deployments of Bidirectional PIM (Bidir-PIM) [RFC8736] which requires a single active instance of its Rendezvous Point (RP) service. The RP is the root of a bidirectional tree and prioritycast addresses for RP allow fast failover without additional redundancy protocols beside the routing protocol, which would otherwise be necessary for such a redundancy service.¶
The Internet uses DNS for single-level name resolution, converting user-level domain names into IP addresses. However, techniques are being proposed for multiple levels of name resolution; these would include: application-level and user-level descriptors, service identifiers, function identifiers and endpoint identifiers, which may then be mapped to IP addresses. These additional levels of naming and resolution would allow services and components to construct the service to be easily identifiable and directly and persistently named.¶
TBD¶
Information-centric networking (ICN) is an approach to evolve the Internet infrastructure to directly support this use by introducing uniquely named data as a core Internet principle. Data becomes independent from location, application, storage, and means of transportation, enabling in-network caching and replication.¶
The OpenSRN [OPENSRNref] project proposed a Content-Based Routing Scheme (CBR) that uses packet content and header information to forward traffic conetextually. This proposal uses a novel software defined networking architecture to provide a semantic routing for big data network applications.¶
The Expedited Internet Bypass Protocol (EIBP) [EIBPref] is a clean slate approach to routing and forwarding in the Internet using the Internet infrastructure, but bypassing the Internet Protocol (IP). The EIBP method may be deployed in current routers and when invoked for a specific end to end IP hosts or networks, EIBP bypasses the heavy traffic and security challenges faced at Layer-3. EIBP does not require routing protocols, instead it abstracts network structural (physical or logical) information into intelligent forwarding addresses that are acquired by EIBP routers automatically.¶
The Forwarding tables used by EIBP are proportional to the connectivity (degree) at a routing device making the protocol scalable. The EIBP routing system does not require network-wide dissemination. Topology change impacts are local and thus instabilities on topology changes are minimal. EIBP is a low configuration protocol, which can be deployed in an AS and extended to multiple ASes independently. EIBP evaluations were conducted using GENI testbeds and compared to IP using Open Shortest Path First and Border Gateway Protocol. Significant performance improvements in terms of convergence and churn rates highlight the capabilities of EIBP.¶
It is entirely possible that some forms of modified address semantic will work perfectly well with existing routing protocols and mechanisms either across the whole Internet or within limited and carefully controlled domains. Claims for this sort of functionality need to be the subject of careful research and analysis as the existing protocols were developed with a different view of the meaning of IP addresses, and because routing systems are notoriously fragile.¶
Several documents are availible that discuss the requirements for applications and services that may benefit from Semantic Networking techniques, including:¶
Improving IP-based semantic network routing capabilities and capacity so that they scale and address a set of growing requirements presents significant research challenges, and will require contributions from the networking research community.¶
As research into the scenarios and possible uses of semantic networking progresses, a number of questions need to be addressed in the scope of routing. These questions go beyond "Why do we need this function?" and "What could we achieve by carrying this additional semantic in an IP address?" The questions are also distinct from issues of how the additional semantics can be encoded within an IP address. All of those issues are, of course, important considerations in the debate about semantic networking, but they form part of the essential groundwork of research into semantic routing itself.¶
[I-D.king-rtgwg-challenges-in-routing] sets out the challenges for the routing system, and how it might be impacted by the use of semantic networking.¶
This document is a survey of existing work and so introduces no security considerations of itself. However, many of the proposals referenced either are intended to improve security or have their own security implications. For example:¶
This document makes no requests for IANA action.¶
Thanks to Stewart Bryant for useful conversations. Luigi Iannone, Robert Raszuk, Ron Bonica, Marie-Jose; Montpetit, Yizhou Li, Toerless Eckert, Tony Li, Joel Halpern, and Carsten Bormann made helpful suggestions. The text on Dyncast is based on suggestions from Dirk Trossen, Luigi Iannone, and Yizhou Li. Toerless Eckert suggested text for the multicast sections.¶
This work is partially supported by the European Commission under Horizon 2020 grant agreement number 101015857 Secured autonomic traffic management for a Tera of SDN flows (Teraflow).¶
Joanna Dang
Email: dangjuanna@huawei.com
¶
Dirk Trossen
Email: dirk.trossen@huawei.com
¶