| Internet-Draft | RAW Architecture/Framework | September 2022 |
| Thubert & Papadopoulos | Expires 20 March 2023 | [Page] |
Reliable and Available Wireless (RAW) provides for high reliability and availability for IP connectivity across any combination of wired and wireless network segments. The RAW Architecture extends the DetNet Architecture and other standard IETF concepts and mechanisms to adapt to the specific challenges of the wireless medium. This document defines an architecture element for the RAW data plane, in the form of an OODA loop, that optimizes the use of constrained spectrum and energy while maintaining the expected connectivity properties. It also introduces a new Control plane Function to prepare alternate paths to go around local failures. The loop involves OAM, PCE, and PREOF extensions, and a new component called the Path Selection Engine (PSE).¶
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 20 March 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.¶
Deterministic Networking is an attempt to emulate the properties of a serial link over a switched fabric, by providing a bounded latency and eliminating congestion loss, even when co-existing with best-effort traffic. It is getting traction in various industries including professional A/V, manufacturing, online gaming, and smartgrid automation, enabling cost and performance optimizations (e.g., vs. loads of P2P cables).¶
Bringing determinism in a packet network means eliminating the statistical effects of multiplexing that result in probabilistic jitter and loss. This can be approached with a tight control of the physical resources to maintain the amount of traffic within a budgeted volume of data per unit of time that fits the physical capabilities of the underlying network, and the use of time-shared resources (bandwidth and buffers) per circuit, and/or by shaping and/or scheduling the packets at every hop.¶
This innovation was initially introduced on wired networks, with IEEE 802.1 Time Sensitive networking (TSN) - for Ethernet LANs - and IETF DetNet. But the wired and the wireless media are fundamentally different at the physical level and in the possible abstractions that can be built for IPv6 [IPv6], more in [IPoWIRELESS]. Nevertheless, deterministic capabilities are required in a number of wireless use cases as well [RAW-USE-CASES]. With new scheduled radios such as TSCH and OFDMA [RAW-TECHNOS] being developed to provide determinism over wireless links at the lower layers, providing DetNet capabilities is now becoming possible.¶
Wireless networks operate on a shared medium where uncontrolled interference, including the self-induced multipath fading cause random transmission losses. Fixed and mobile obstacles and reflectors may block or alter the signal, causing transient and unpredictable variations of the throughput and packet delivery ratio (PDR) of a wireless link. This adds new dimensions to the statistical effects that affect the quality and reliability of the link. Multiple links and transmissions must be used, and the challenge is to provide enough diversity and redundancy to ensure the timely packet delivery while preserving energy and optimizing the use of the shared spectrum.¶
Reliable and Available Wireless (RAW) takes up the challenge of providing highly available and reliable end-to-end performances in a network with scheduled wireless segments. To defeat those additional causes of transmission delay and loss in wireless transmission, RAW requires and leverages deterministic Layer-2 capabilities. Operating at the Layer-3, RAW can further increase diversity in the spatial, time, code, and frequency domains by enabling multiple link-layer wired and wireless technologies in parallel or sequentially, for a higher resilience and a wider applicability. RAW can also provide homogeneous services to critical applications beyond the boundaries of a single subnetwork, e.g., controlling the use of diverse radio access technologies to optimize the end-to-end application experience.¶
While the generic "Deterministic Networking Problem Statement" [RFC8557] applies to both the wired and the wireless media, the methods to achieve RAW must extend those used to support time-sensitive networking over wires, as a RAW solution has to address less consistent transmissions, energy conservation and shared spectrum efficiency.¶
RAW provides DetNet elements that are specialized for transporting IP flows over deterministic radios technologies such as listed in [RAW-TECHNOS]. Conceptually, RAW is agnostic to the radio layer underneath though the capability to schedule transmissions is assumed. How the PHY is programmed to do so, and whether the radio is single-hop or meshed, are unknown at the IP layer and not part of the RAW abstraction. Nevertheless, cross-layer optimizations may take place to ensure proper link awareness (think, link quality) and packet handling (think, scheduling).¶
The "Deterministic Networking Architecture" [RFC8655] is composed of three planes: the Application (User) Plane, the Controller Plane, and the Network Plane. The DetNet Network Plane is composed of a DetNet service sub-layer that focuses on flow protection (e.g., using redundancy) and can be fully operated at Layer-3, and a DetNet forwarding sub-layer that associates the flows to the paths, ensures the availability of the necessary resources, and leverages Layer-2 functionalities for timely delivery to the next DetNet system.¶
The RAW Architecture extends the DetNet Network Plane, to accommodate one or multiple hops of homogeneous or heterogeneous wired and wireless technologies. RAW adds reactivity to the DetNet service sub-layer to compensate the dynamics for the radio links in terms of lossiness and bandwidth. This may apply for instance to mesh networks as illustrated in Figure 3, or diverse radio access networks as illustrated in Figure 8.¶
RAW and DetNet route application flows that require a special treatment along the paths that will provide that treatment. This may be seen as a form of Path Aware Networking and may be subject to impediments documented in [RFC9049].¶
The establishment of a path is not in-scope for RAW. It may be the product of a centralized Controller Plane Function like a Path computation Element (PCE) [RFC4655] or a distributed routing protocol. For the most part, the remainder of the document mentions centralized control and PCE, but conceptually, the same issues and needs would arise for a distributed protocol that would attempt to allocate constrained resources and optimize globally, and the distributed approach is considered in scope too.¶
As opposed to wired networks, the action of installing a path over a set of wireless links may be very slow relative to the speed at which the radio conditions vary, and it makes sense in the wireless case to provide redundant forwarding solutions along a complex path (see Section 2.1.3) and to leave it to the Network Plane to select which of those forwarding solutions are to be used for a given packet based on the current conditions.¶
RAW distinguishes the longer time scale at which routes are computed from the the shorter forwarding time scale where per-packet decisions are made. RAW operates within the Network Plane at the forwarding time scale on one DetNet flow over a complex path delineated by a Track (see Section 2.1.3.2). The Track is preestablished and installed by means outside of the scope of RAW; it may be strict or loose depending on whether each or just a subset of the hops are observed and controlled by RAW.¶
The RAW Architecture is based on an abstract OODA Loop (Observe, Orient, Decide, Act). The generic concept involves:¶
The overall OODA Loop optimizes the use of redundancy to achieve the required reliability and availability Service Level Agreement (SLA) while minimizing the use of constrained resources such as spectrum and battery.¶
This document presents the RAW problem and associated terminology in Section 2, and elaborates in Section 4 on the OODA loop based on the RAW conceptual model presented in Section 3.¶
RAW reuses terminology defined for DetNet in the "Deterministic Networking Architecture" [RFC8655], e.g., PREOF for Packet Replication, Elimination and Ordering Functions.¶
RAW also reuses terminology defined for 6TiSCH in [6TiSCH-ARCHI] such as the term Track. A Track associates a complex path with PAREO and shaping operations. The concept is agnostic to the underlaying technology and applies but is not limited to any fully or partially wireless mesh. RAW specifies strict and loose Tracks depending on whether the path is fully controlled by RAW or traverses an opaque network where RAW cannot observe and control the individual hops.¶
RAW uses the following terminology and acronyms:¶
Automatic Repeat Request, enabling an acknowledged transmission and retries. ARQ is a typical model at Layer-2 on a wireless medium. ARQ is typically implemented hop-by-hop and not end-to-end in wireless networks. Else, it introduces excessive indetermination in latency, but a limited number of retries within a bounded time may be used within end-to-end constraints.¶
OAM stands for Operations, Administration, and Maintenance, and covers the processes, activities, tools, and standards involved with operating, administering, managing and maintaining any system. This document uses the terms Operations, Administration, and Maintenance, in conformance with the 'Guidelines for the Use of the "OAM" Acronym in the IETF' [RFC6291] and the system observed by the RAW OAM is the Track.¶
Observe, Orient, Decide, Act. The OODA Loop is a conceptual cyclic model developed by USAF Colonel John Boyd, and that is applicable in multiple domains where agility can provide benefits against brute force.¶
Packet (hybrid) ARQ, Replication, Elimination and Ordering. PAREO is a superset Of DetNet's PREOF that includes radio-specific techniques such as short range broadcast, MUMIMO, PHY rate and other Modulation Coding Scheme (MCS) adaptation, constructive interference and overhearing, which can be leveraged separately or combined to increase the reliability. As can the case for other functions such as shaping, the PAREO functions may be actuated at the lower layers but controlled through abstractions from the RAW extensions in the DetNet Service sublayer.¶
In the context of RAW, a link flaps when the reliability of the wireless connectivity drops abruptly for a short period of time, typically of a subsecond to seconds duration.¶
Connection from end-devices to a data communication equipment. In the context of wireless, uplink refers to the connection between a station (STA) and a controller (AP) or a User Equipment (UE) to a Base Station (BS) such as a 3GPP 5G gNodeB (gNb).¶
The reverse direction from uplink.¶
Following the direction of the flow data path along a Track.¶
Against the direction of the flow data path along a Track.¶
Quoting section 1.1.3 of [INT-ARCHI]:¶
At a given moment, all the IP datagrams from a particular source host to a particular destination host will typically traverse the same sequence of gateways. We use the term "path" for this sequence. Note that a path is uni-directional; it is not unusual to have different paths in the two directions between a given host pair.¶
Section 2 of [I-D.irtf-panrg-path-properties] points to a longer, more modern definition of path, which begins as follows:¶
A sequence of adjacent path elements over which a packet can be transmitted, starting and ending with a node. A path is unidirectional. Paths are time-dependent, i.e., the sequence of path elements over which packets are sent from one node to another may change. A path is defined between two nodes.¶
It follows that the general acceptance of a path is a linear sequence of links and nodes, as opposed to a multi-dimensional graph, defined by the experience of the packet that went from a node A to a node B.¶
With DetNet and RAW, a packet may be duplicated, fragmented and network-coded, and the various byproducts may travel different paths that are not necessarily end-to-end between A and B; we refer to that experience as a complex path. The complex path does not fit the traditional description of a path, and is subject to change from a packet to the next. Therefore we introduce below the term of a Track as the overall topology where the possible complex paths are all contained.¶
In the context of this document, a path is observed by following one copy or one fragment of a packet that conserves its uniqueness and integrity. For instance, if C replicates to E and F and D eliminates on the way from A to B, a packet from A to B experiences 2 paths, A->C->E->D->B and A->C->F->D->B.¶
A networking graph that can be followed to transport packets with equivalent treatment; as opposed to the definition of a path above, a Track represents not an experience but a potential, is not necessarily a linear sequence, and is not necessarily fully traversed (flooded) by all packets of a flow. It may contain multiple paths that may overlap, fork and rejoin, for instance to enable the RAW PAREO operations.¶
+---------+
| IoT G/W |
+---------+
EGR <=== Elimination at Egress
| |
/------/ \-------\ Wired backbone
| |
+--|--+ +--|--+
| | | Backbone | | | Backbone
| | | Router | | | Router
+--|--+ +--|--+
| |
o \ o / Track branch
o o o---o---o o o o o
\ o / o o o
o o \ / o low power lossy network
\/ o o o
o IN <=== Replication at Track Ingress
|
o <- source device
In DetNet [RFC8655] terms, a Track has the following properties:¶
A Track within a Track. The RAW PSE selects a subTrack on a per-packet or a per-collection of packets basis to provide the desired reliability for the transported flows.¶
A serial path formed by a topological edge of a Track. East-West Segments are oriented from Ingress (East) to Egress (West). North/South Segments can be bidirectional; to avoid loops, measures must be taken to ensure that a given packet flows either Northwards or Southwards along a bidirectional Segment, but never bounces back.¶
This document reuses the terminology in section 2 of [RFC8557] and section 4.1.2 of [RFC8655] for deterministic networking and deterministic networks.¶
A collection of consecutive IP packets defined by the upper layers and signaled by the same 5 or 6-tuple, see section 5.1 of [RFC8939]. Packets of the same flow must be placed on the same Track to receive an equivalent treatment from Ingress to Egress within the Track. Multiple flows may be transported along the same Track. The subTrack that is selected for the flow may change over time under the control of the PSE.¶
A tuple identified by a stream_handle, and provided by a bridge, in accordance with IEEE 802.1CB. The tuple comprises at least source MAC, destination MAC, VLAN ID, and L2 priority. Continuous streams are characterized by bandwidth and max packet size; scheduled streams are characterized by a repeating pattern of timed transmissions.¶
See section 3.3 of [DetNet-DP]. The classical IP 5-tuple that identifies a flow comprises the source IP, destination IP, source port, destination port, and the upper layer protocol (ULP). DetNet uses a 6-tuple where the extra field is the DSCP field in the packet. The IPv6 flow label is not used for that purpose.¶
TSN stands for Time Sensitive Networking and denotes the efforts at IEEE 802 for deterministic networking, originally for use on Ethernet. Wireless TSN (WTSN) denotes extensions of the TSN work on wireless media such as the selected RAW technologies [RAW-TECHNOS].¶
In the context of the RAW work, Reliability and Availability are defined as follows:¶
In the context of RAW, an SLA (service level agreement) is a contract between a provider, the network, and a client, the application flow, about measurable metrics such as latency boundaries, consecutive losses, and packet delivery ratio (PDR).¶
A service level objective (SLO) is one term in the SLA, for which specific network setting and operations are implemented. For instance, a dynamic tuning of the packet redundancy will address an SLO of consecutive losses in a row by augmenting the chances of delivery of a packet that follows a loss.¶
A service level indicator (SLI) measures the compliance of an SLO to the terms of the contract. It can be for instance the statistics of individual losses and losses in a row as time series.).¶
Reliability is a measure of the probability that an item will perform its intended function for a specified interval under stated conditions (SLA). RAW expresses reliability in terms of Mean Time Between Failure (MTBF) and Maximum Consecutive Failures (MCF). More in [NASA].).¶
That is exempt of unscheduled outage or derivation from the terms of the SLA. A basic expectation for a RAW network is that the flow is maintained in the face of any single breakage or flapping.¶
Availability is a measure of the relative amount of time where a RAW Network operates in stated condition (SLA), expressed as (uptime)/(uptime+downtime). Because a serial wireless path may not be good enough to provide the required reliability, and even 2 parallel paths may not be over a longer period of time, the RAW availability implies a journey that is a lot more complex than following a serial path.¶
See [RFC7799]. In the context of RAW, Active OAM is used to observe a particular Track, subTrack, or Segment of a Track regardless of whether it is used for traffic at that time.¶
An active OAM packet is considered in-band for the monitored Track when it traverses the same set of links and interfaces and if the OAM packet receives the same QoS and PAREO treatment as the packets of the data flows that are injected in the Track.¶
Out-of-band OAM is an active OAM whose path is not topologically congruent to the Track, or its test packets receive a QoS and/or PAREO treatment that is different from that of the packets of the data flows that are injected in the Track, or both.¶
An active OAM packet is a Limited OAM packet when it observes the RAW operation over a node, a segment, or a subTrack of the Track, though not from Ingress to Egress. It is injected in the datapath and extracted from the datapath around the particular function or subnetwork (e.g., around a relay providing a Service sublayer replication point) that is being tested.¶
An upstream OAM packet is an Out-of-Band OAM packet that traverses the Track from egress to ingress on the reverse direction, to capture and report OAM measurements upstream. The collection may capture all information along the whole Track, or it may only learn select data across all, or only a particular subTrack, or Segment of a Track.¶
A residence time (RT) is defined as the time period between the reception of a packet starts and the transmission of the packet begins. In the context of RAW, RT is useful for a transit node, not ingress or egress.¶
[DetNet-OAM] provides additional terminology related to OAM in the context of DetNet and by extension of RAW, whereas [RFC7799] defines the Active, Passive, and Hybrid OAM methods.¶
The reliability criteria of a critical system pervades through its elements, and if the system comprises a data network then the data network is also subject to the inherited reliability and availability criteria. It is only natural to consider the art of high availability engineering and apply it to wireless communications in the context of RAW.¶
There are three principles [pillars] of high availability engineering:¶
These principles are common to all high availability systems, not just ones with Internet technology at the center. Examples of both non-Internet and Internet are included.¶
Physical and logical components in a system happen to fail, either as the effect of wear and tear, when used beyond acceptable limits, or due to a software bug. It is necessary to decouple component failure from system failure to avoid the latter. This allows failed components to be restored while the rest of the system continues to function.¶
IP Routers leverage routing protocols to compute alternate routes in case of a failure. There is a rather open-ended issue over alternate routes -- for example, when links are cabled through the same conduit, they form a shared risk link group (SRLG), and will share the same fate if the bundle is cut. The same effect can happen with virtual links that end up in a same physical transport through the games of encapsulation. In a same fashion, an interferer or an obstacle may affect multiple wireless transmissions at the same time, even between different sets of peers.¶
Intermediate network Nodes such as routers, switches and APs, wire bundles and the air medium itself can become single points of failure. For High Availability, it is thus required to use physically link- and Node-disjoint paths; in the wireless space, it is also required to use the highest possible degree of diversity (time, space, code, frequency, channel width) in the transmissions over the air to combat the additional causes of transmission loss.¶
From an economics standpoint, executing this principle properly generally increases capitalization expense because of the redundant equipment. In a constrained network where the waste of energy and bandwidth should be minimized, an excessive use of redundant links must be avoided; for RAW this means that the extra bandwidth must be used wisely and with parsimony.¶
Having a backup equipment has a limited value unless it can be reliably switched into use within the down-time parameters. IP Routers execute reliable crossover continuously because the routers will use any alternate routes that are available [RFC0791]. This is due to the stateless nature of IP datagrams and the dissociation of the datagrams from the forwarding routes they take. The "IP Fast Reroute Framework" [FRR] analyzes mechanisms for fast failure detection and path repair for IP Fast-Reroute, and discusses the case of multiple failures and SRLG. Examples of FRR techniques include Remote Loop-Free Alternate [RLFA-FRR] and backup label-switched path (LSP) tunnels for the local repair of LSP tunnels using RSVP-TE [RFC4090].¶
Deterministic flows, on the contrary, are attached to specific paths where dedicated resources are reserved for each flow. Therefore each DetNet path must inherently provide sufficient redundancy to provide the guaranteed SLA at all times. The DetNet PREOF typically leverages 1+1 redundancy whereby a packet is sent twice, over non-congruent paths. This avoids the gap during the fast reroute operation, but doubles the traffic in the network.¶
In the case of RAW, the expectation is that multiple transient faults may happen in overlapping time windows, in which case the 1+1 redundancy with delayed reestablishment of the second path will not provide the required guarantees. The Data Plane must be configured with a sufficient degree of redundancy to select an alternate redundant path immediately upon a fault, without the need for a slow intervention from the controller plane.¶
The execution of the two above principles is likely to render a system where the user will rarely see a failure. But someone needs to in order to direct maintenance.¶
There are many reasons for system monitoring (FCAPS for fault, configuration, accounting, performance, security is a handy mental checklist) but fault monitoring is sufficient reason.¶
"An Architecture for Describing Simple Network Management Protocol (SNMP) Management Frameworks" [STD 62] describes how to use SNMP to observe and correct long-term faults.¶
"Overview and Principles of Internet Traffic Engineering" [TE] discusses the importance of measurement for network protection, and provides abstract an method for network survivability with the analysis of a traffic matrix as observed by SNMP, probing techniques, FTP, IGP link state advertisements, and more.¶
Those measurements are needed in the context of RAW to inform the controller and make the long term reactive decision to rebuild a complex path based on statistical and aggregated information. RAW itself operates in the Network Plane at a faster time scale with live information on speed, state, etc... This live information can be obtained directly from the lower layer, e.g., using L2 triggers, read from a protocol such as the Dynamic Link Exchange Protocol (DLEP) [DLEP], or transported over multiple hops using OAM and reverse OAM, as illustrated in Figure 9.¶
The terms Reliability and Availability are defined for use in RAW in Section 2.1 and the reader is invited to read [NASA] for more details on the general definition of Reliability. Practically speaking a number of nines is often used to indicate the reliability of a data link, e.g., 5 nines indicate a Packet Delivery Ratio (PDR) of 99.999%.¶
This number is typical in a wired environment where the loss is due to a random event such as a solar particle that affects the transmission of a particular frame, but does not affect the previous or next frame, nor frames transmitted on other links. Note that the QoS requirements in RAW may include a bounded latency, and a packet that arrives too late is a fault and not considered as delivered.¶
For a periodic networking pattern such as an automation control loop, this number is proportional to the Mean Time Between Failures (MTBF). When a single fault can have dramatic consequences, the MTBF expresses the chances that the unwanted fault event occurs. In data networks, this is rarely the case. Packet loss cannot never be fully avoided and the systems are built to resist to one loss, e.g., using redundancy with Retries (HARQ) or Packet Replication and Elimination (PRE), or, in a typical control loop, by linear interpolation from the previous measurements.¶
But the linear interpolation method cannot resist multiple consecutive losses, and a high MTBF is desired as a guarantee that this will not happen, IOW that the number of losses-in-a-row can be bounded. In that case, what is really desired is a Maximum Consecutive Failures (MCF). If the number of losses in a row passes the MCF, the control loop has to abort and the system, e.g., the production line, may need to enter an emergency stop condition.¶
Engineers that build automated processes may use the network reliability expressed in nines or as an MTBF as a proxy to indicate an MCF, e.g., as described in section 7.4 of the "Deterministic Networking Use Cases" [RFC8578].¶
In contrast with wired networks, errors in transmission are the predominant source of packet loss in wireless networks.¶
The root cause for the loss may be of multiple origins, calling for the use of different forms of diversity:¶
A destructive interference by a reflection of the original signal.¶
A radio signal may be received directly (line-of-sight) and/or as a reflection on a physical structure (echo). The reflections take a longer path and are delayed by the extra distance divided by the speed of light in the medium. Depending on the frequency, the echo lands with a different phase which may add up to (constructive interference) or cancel the direct signal (destructive interference).¶
The affected frequencies depend on the relative position of the sender, the receiver, and all the reflecting objects in the environment. A given hop will suffer from multipath fading for multiple packets in a row till a physical movement changes the reflection patterns.¶
Energy in the spectrum used for the transmission confuses the receiver.¶
The wireless medium itself is a Shared Risk Link Group (SRLG) for nearby users of the same spectrum, as an interference may affect multiple co-channel transmissions between different peers within the interference domain of the interferer, possibly even when they use different technologies.¶
The optimal transmission happens when the Fresnel Zone between the sender and the receiver is free of obstacles.¶
As long as a physical object (e.g., a metallic trolley between peers) that affects the transmission is not removed, the quality of the link is affected.¶
In an environment that is rich of metallic structures and mobile objects, a single radio link will provide a fuzzy service, meaning that it cannot be trusted to transport the traffic reliably over a long period of time.¶
Transmission losses are typically not independent, and their nature and duration are unpredictable; as long as a physical object (e.g., a metallic trolley between peers) that affects the transmission is not removed, or as long as the interferer (e.g., a radar) keeps transmitting, a continuous stream of packets will be affected.¶
The key technique to combat those unpredictable losses is diversity. Different forms of diversity are necessary to combat different causes of loss and the use of diversity must be maximized to optimize the PDR.¶
A single packet may be sent at different times (time diversity) over diverse paths (spatial diversity) that rely on diverse radio channels (frequency diversity) and diverse PHY technologies, e.g., narrowband vs. spread spectrum, or diverse codes. Using time diversity will defeat short-term interferences; spatial diversity combats very local causes such as multipath fading; narrowband and spread spectrum are relatively innocuous to one another and can be used for diversity in the presence of the other.¶
With DetNet, the Controller Plane Function (CPF) handles the routing computation and maintenance. With RAW, the CPF also performs the PSE orientation, proposing SubTracks to use in response to network events. The CPF can be can be centralized in a PCE, and can reside outside the network. This is how the remainder of this document depicts it, though the CPF could be implemented otherwise without affecting the architecture. In a wireless mesh, the path to the PCE can be expensive and slow, possibly going across the whole mesh and back. Reaching to the PCE can also be slow in regards to the speed of events that affect the forwarding operation at the radio layer. In the same fashion, a distributed routing protocol may also take time and consume excessive wireless resources to reconverge to a new optimized state.¶
Due to that cost and latency, the Controller Plane is not expected to be sensitive/reactive to transient changes. The abstraction of a link at the routing level is expected to use statistical metrics that aggregate the behavior of a link over long periods of time, and represent its properties as shades of gray as opposed to numerical values such as a link quality indicator, or a boolean value for either up or down.¶
+----------------+
| Controller |
| [PCE] |
+----------------+
^
|
Slow
|
_-._-._-._-._-._-. | ._-._-._-._-._-._-._-._-._-._-._-._-
_-._-._-._-._-._-._-. | _-._-._-._-._-._-._-._-._-._-._-._-
|
Expensive
|
.... | .......
.... . | . .......
.... v ...
.. A-------B-------C---D ..
... / \ / \ ..
. I ----M-------N--***-- E ..
.. \ / / ...
.. P--***--Q-----M---R ....
.. ....
. <----- Fast -------> ....
....... ....
.................
*** = flapping at this time
In the case of wireless, the changes that affect the forwarding decision can happen frequently and often for short durations, e.g., a mobile object moves between a transmitter and a receiver, and will cancel the line of sight transmission for a few seconds, or a radar measures the depth of a pool and interferes on a particular channel for a split second.¶
There is thus a desire to separate the long term computation of the route and the short term forwarding decision. In that model, the routing operation computes a complex Track that enables multiple Non-Equal Cost Multi-Path (N-ECMP) forwarding solutions, and leaves it to the Data Plane to make the per-packet decision of which of these possibilities should be used.¶
In the wired world, and more specifically in the context of Traffic Engineering (TE), an alternate path can be used upon the detection of a failure in the main path, e.g., using OAM in MPLS-TP or BFD over a collection of SD-WAN tunnels. RAW formalizes a forwarding time scale that is an order(s) of magnitude shorter than the controller plane routing time scale, and separates the protocols and metrics that are used at both scales. Routing can operate on long term statistics such as delivery ratio over minutes to hours, but as a first approximation can ignore flapping. On the other hand, the RAW forwarding decision is made at the scale of the packet rate, and uses information that must be pertinent at the present time for the current transmission(s).¶
RAW inherits the conceptual model described in section 4 of the DetNet Architecture [RFC8655]. RAW extends the DetNet service layer to provide additional agility against transmission loss.¶
A RAW Network Plane may be strict (as illustrated in Figure 5 or loose (as illustrated in Figure 6, depending on whether RAW observes and takes actions on all hops or not. For instance, the packets between two wireless entities may be relayed over a wired infrastructure such as a Wi-Fi extended service set (ESS) or a 5G Core; in that case, RAW observes and controls the transmission over the wireless first and last hops, as well as end-to-end metrics such as latency, jitter, and delivery ratio. This operation is loose since the structure and properties of the wired infrastructure are ignored, and may be either controlled by other means such as DetNet/TSN, or neglected in the face of the wireless hops.¶
A Controller Plane Function (CPF) such as a PCE interacts with RAW Nodes over a Southbound API. The RAW Nodes are DetNet relays that are capable of additional diversity mechanisms and measurement functions related to the radio interface, in particular the PAREO diversity mechanisms. RAW leverages a CPF that operates inside the RAW Nodes (typically the Ingress Edge Nodes) to dynamically adapt the path of the packets and optimizes the resource usage.¶
The PCE defines a complex Track between an Ingress End System and an Egress End System, and indicates to the RAW Nodes where the PAREO operations may be actioned in the Network Plane. The Track may be strict, meaning that the DetNet forwarding sublayer operations are enforced end-to-end The Track may be expressed loosely to enable traversing a non-RAW subnetwork as in Figure 6. In that case, RAW can not leverage end-to-end DetNet and cannot provide latency guarantees. The non-RAW subnetwork is neglected in the RAW computation, that is, considered jitterless, and infinitely reliable and/or available in comparison with the links between RAW nodes, so loss and jitter that is measured end-to-end is attributed to the RAW hops (typically an access link).¶
CPF CPF CPF CPF
Southbound API
_-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-
_-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-._-
RAW --/ RAW --/ RAW --/ RAW
/-- Node /-- Node /-- Node /-- Node --/
Ingress --/ / / /-- Egress
End / / .. . End
Node ---/ / / .. .. . /-- Node
/-- RAW --/ RAW ( non-RAW ) -- RAW --/
Node /-- Node --- ( Nodes ) Node
... .
--/ wireless wired
/-- link --- link
The Link-Layer metrics are reported to the PCE in a time-aggregated, e.g., statistical fashion. Example Link-Layer metrics include typical Link bandwidth (the medium speed depends dynamically on the PHY mode), number of flows (bandwidth that can be reserved for a flow depends on the number and size of flows sharing the spectrum) and average and mean squared deviation of availability and reliability figures such as Packet Delivery Ratio (PDR) over long periods of time.¶
Based on those metrics, the PCE installs the Track with enough redundant forwarding solutions to ensure that the Network Plane can reliably deliver the packets within a System Level Agreement (SLA) associated to the flows that it transports. The SLA defines end-to-end reliability and availability requirements, where reliability may be expressed as a successful delivery in order and within a bounded delay of at least one copy of a packet.¶
Depending on the use case and the SLA, the Track may comprise non-RAW segments, either interleaved inside the Track, or all the way to the Egress End Node (e.g., a server in the Internet). RAW observes the Lower-Layer Links between RAW nodes (typically, radio links) and the end-to-end Network Layer operation to decide at all times which of the PAREO diversity schemes is actioned by which RAW Nodes.¶
Once a Track is established, per-segment and end-to-end reliability and availability statistics are periodically reported to the PCE to assure that the SLA can be met or have it recompute the Track if not.¶
RAW improves the reliability of transmissions and the availability of the communication resources, but does not provide scheduling and shaping, so RAW itself does not provide guarantees such as latency for the application payload. Rather, it should be seen as a dynamic optimization of the use of redundancy to maintain it within certain boundaries. For instance, ARQ, which is part of the PAREO capabilities (see Section 4.4) is operated by the lower layers and RAW will only abstract the concept and hint the lower layers on the desired outcome, as opposed to performing the retries at Layer-3.¶
Guarantees such as bounded latency depend on the upper layers (Transport or Application) to provide the payload in volumes and at times that match the contract with the DetNet sublayers and the layers below. Excess of incoming traffic at the DetNet Ingress will cause either dropping, queueing, or reclassification of the packets, and entail loss, latency, or jitter, and moot the guarantees that are provided inside the DetNet Network.¶
When the traffic from upper layers matches the expectation of the lower layers, RAW still depends on the lower layers to provide the timing and physical resources guarantees that are needed to match the traffic SLA. When the availability of the physical resource varies, RAW will act on the distribution of the traffic to leverage alternates within a finite set of potential resources.¶
RAW leverages the DetNet Forwarding sub-layer and requires the support of in-situ OAM in DetNet Transit Nodes (see fig 3 of [RFC8655] for the dynamic acquisition of link capacity and state to maintain a strict RAW service, end-to-end, over a DetNet Network. RAW enhances DetNet to improve the protection against link errors such as transient flapping that are far more common in wireless links. Nevertheless, the RAW methods are for the most part applicable to wired links as well, e.g., when energy savings are desirable and the available path diversity exceeds 1+1 linear redundancy.¶
RAW extends the DetNet Stack (see fig 4 of [RFC8655]) with additional functionality at the DetNet Service sub-layer for the PSE operation. Layer-3 in general and DetNet in particular operates on abstractions of the lower layers and through APIs to control those abstractions. For instance, DetNet already leverages lower layers for time-sensitive operations such as time synchronization and traffic shapers. Because the performances of the radio layers are subject to rapid changes, so RAW needs more dynamic gauges and knobs. To that effect, the DetNet PREOF is extended with the PAREO capabilities (see Section 4.4) and the RAW PAREO Actuator manages dynamically the PAREO operations, which may be performed either within the DetNet sublayers or at a lower layer, using a common radio abstraction and APIs in the latter case. In particular, PAREO needs the capability to push reliability and timing hints like suggest X retries (min, max) within a time window, or send unicast (one next hop) or multicast (for overhearing). The other way around RAW needs hints about the radio conditions like L2 triggers (RSSI, LQI, ETX...) over all the wireless hops. This information is useful in the controller plane for both the PCE and PSE.¶
The RAW Service sub-layer also adds the OAM Propagator that (re)generates the OAM information as it is formed and propagated In-Band or Out-of-Band. The RAW Service sub-layer may be present in DetNet Edge and Relay Nodes, though the PAREO Actuator has no operation in the Egress Edge Node.¶
RAW also adds a Control sub-layer that operates in the DetNet Controller Plane. The RAW Control sub-layer typically runs only in the DetNet Ingress Edge Node or End System, though it may also run in DetNet Relay Nodes when the RAW Control sub-layer is distributed along the Track. The RAW Control sub-layer functionality includes the PSE that decides the subTrack for the next packets of a flows and controls the PAREO Actuators along the subTrack through specific signaling, and the OAM Supervisor that triggers, and learns from, OAM observations, and feeds the PSE for its next decision.¶
+------------------------------+ +--------------------------------+
| | | |
.....................................................................
| | | |
| +----------+ +------------+ | | .-.-.-.-.-.--. .-.-.-.-.-.--. |
| | PSE | | OAM | | | | Distr. PSE | | Distr. OAM | |
| | | | Supervisor | | | | | | Supervisor | |
| +----------+ +------------+ | | .-.-.-.-.-.--. .-.-.-.-.-.--. |
| | | optional optional |
RAW Control sub-layer
.....................................................................
DetNet Service sub-layer
| | | |
| +----------+ +------------+ | | +------------+ +------------+ |
| | PAREO | | OAM | | | | PAREO | | OAM | |
| | Actuator | | Observer | | | | Actuator | | Observer | |
| +----------+ +------------+ | | +------------+ +------------+ |
| | | |
DetNet Service sub-layer
.....................................................................
DetNet Forwarding sub-layer
| | | |
| +------------+ | | +------------+ |
| |In-Situ OAM | | | |In-Situ OAM | |
| +------------+ | | +------------+ |
| | | |
+------------------------------+ +--------------------------------+
End System or Relay
Ingress Edge Node Node
There are 2 main proposed models to deploy RAW and DetNet. In the first model (strict) illustrated in Figure 5, RAW operates over a continuous DetNet Service end-to-end between the Ingress and the Egress Edge Nodes or End Systems.¶
A minimal Forwarding sub-layer service is provided at all DetNet Nodes to ensure that the OAM information flows. Relay Nodes may or may not support RAW services, and the Edge nodes do support RAW. DetNet guarantees such as latency are provided end-to-end, and RAW supports the DetNet Service to optimize the use of resources.¶
--------------------Flow Direction----------------------------------> +---------+ | RAW | | Control | +---------+ +---------+ +---------+ | RAW + | | RAW + | | RAW + | | DetNet | | DetNet | | DetNet | | Service | | Service | | Service | +---------+---------------------------+---------+--------+---------+ | DetNet | | Forwarding | +------------------------------------------------------------------+ Ingress Transit Relay Egress Edge ... Nodes ... Nodes ... Edge Node Node <--------------------Full Guarantees------------------------------->
In the second model (loose), illustrated in Figure 6, RAW operates over a partial DetNet Service where typically only the Ingress and the Egress End Systems support RAW. The DetNet Domain may extend beyond the Ingress node, or there may be a DetNet domain starting at an Ingress Edge Node at the first hop after the End System.¶
In the loose model, RAW cannot observe the hops in network, and the path beyond the first hop is opaque; RAW can still observe the end-to-end behavior and use Layer-3 measurements to decide whether to replicate a packet and select the first hop interface(s).¶
--------------------Flow Direction----------------------------------> +---------+ | RAW | | Control | +---------+ +---------+ +---------+ | RAW + | | DetNet | | RAW + | | DetNet | | Only | | DetNet | | Service | | Service | | Service | +---------+----------------------+---+ +---+---------+ | DetNet | | DetNet | | Forwarding | | Forwarding | +------------------------------------+ +-------------+ Ingress Transit Relay Internet Egress End ... Nodes ... Nodes ... ... End System System <----------------------No Guarantee-------------------------------->
The RAW Architecture is structured as an OODA Loop (Observe, Orient, Decide, Act). It involves:¶
+-------> Orient (PCE) --------+
| link stats, |
| pre-trained model |
| ... |
| v
Observe (OAM) Decide (PSE)
^ |
| |
| |
+-------- Act (PAREO) <--------+
At DetNet
Service sub-layer
The overall OODA Loop optimizes the use of redundancy to achieve the required reliability and availability Service Level Agreement (SLA) while minimizing the use of constrained resources such as spectrum and battery.¶
RAW In-situ OAM operation in the Network Plane may observe either a full Track or subTracks that are being used at this time. As packets may be load balanced, replicated, eliminated, and / or fragmented for Network Coding (NC) forward error correction (FEC), the RAW In-situ operation needs to be able to signal which operation occured to an individual packet.¶
Active RAW OAM may be needed to observe the unused segments and evaluate the desirability of a rerouting decision.¶
Finally, the RAW Service sublayer Assurance may observe the individual PAREO operation of a relay node to ensure that it is conforming; this might require injecting an OAM packet at an upstream point inside the Track and extracting that packet at another point downstream before it reaches the egress.¶
This observation feeds the RAW PSE that makes the decision on which PAREO function is actioned at which RAW Node, for one a small continuous series of packets.¶
... ..
RAN 1 ----- ... .. ...
/ . .. ....
+-------+ / . .. .... +------+
|Ingress|- . ..... |Egress|
| End |------ RAN 2 -- . Internet ....---| End |
|System |- .. ..... |System|
+-------+ \ . ...... +------+
\ ... ... .....
RAN n -------- ... .....
<------------------> <-------------------->
Observed by OAM Opaque to OAM
In the case of a End-to-End Protection in a Wireless Mesh, the Track is strict and congruent with the path so all links are observed.¶
Conversely, in the case of Radio Access Protection illustrated in Figure 8, the Track is Loose and only the first hop is observed; the rest of the path is abstracted and considered infinitely reliable. The loss if a packet is attributed to the first hop Radio Access Network (RAN), even if a particular loss effectively happens farther down the path. In that case, RAW enables technology diversity (e.g. Wi-Fi and 5G) which in turn improves the diversity in spectrum usage.¶
The Links that are not observed by OAM are opaque to it, meaning that the OAM information is carried across and possibly echoed as data, but there is no information capture in intermediate nodes. In the example above, the Internet is opaque and not controlled by RAW; still the RAW OAM measures the end-to-end latency and delivery ratio for packets sent via each if RAN 1, RAN 2 and RAN 3, and determines whether a packet should be sent over either or a collection of those access links.¶
RAW separates the long time scale at which a Track is elaborated and installed, from the short time scale at which the forwarding decision is taken for one or a few packets (see in Section 2.3) that will experience the same path until the network conditions evolve and another path is selected within the same Track.¶
The Track computation is out of scope, but RAW expects that the Controller plane protocol that installs the Track also provides related knowledge in the form of meta data about the links, segments and possible subTracks. That meta data can be a pre-digested statistical model, and may include prediction of future flaps and packet loss, as well as recommended actions when that happens.¶
The meta data may include:¶
The Track is installed with measurable objectives that are computed by the PCE to achieve the RAW SLA. The objectives can be expressed as any of maximum number of packet lost in a row, bounded latency, maximal jitter, maximum number of interleaved out of order packets, average number of copies received at the elimination point, and maximal delay between the first and the last received copy of the same packet.¶
The RAW OODA Loop operates at the path selection time scale to provide agility vs. the brute force approach of flooding the whole Track. The OODA Loop controls, within the redundant solutions that are proposed by the PCE, which will be used for each packet to provide a Reliable and Available service while minimizing the waste of constrained resources.¶
To that effect, RAW defines the Path Selection Engine (PSE) that is the counterpart of the PCE to perform rapid local adjustments of the forwarding tables within the diversity that the PCE has selected for the Track. The PSE enables to exploit the richer forwarding capabilities with PAREO and scheduled transmissions at a faster time scale over the smaller domain that is the Track, in either a loose or a strict fashion.¶
Compared to the PCE, the PSE operates on metrics that evolve faster, but that need to be advertised at a fast rate but only locally, within the Track. The forwarding decision may also change rapidly, but with a scope that is also contained within the Track, with no visibility to the other Tracks and flows in the network. This is as opposed to the PCE that must observe the whole network and optimize all the Tracks globally, which can only be done at a slow pace and using long-term statistical metrics, as presented in Table 1.¶
| PCE (Not in Scope) | PSE (In Scope) | |
|---|---|---|
| Operation | Typically Centralized | Source-Routed or Distributed |
| Communication | Slow, expensive | Fast, local |
| Time Scale | hours and above | seconds and below |
| Network Size | Large, many Tracks to optimize globally | Small, within one Track |
| Considered Metrics | Averaged, Statistical, Shade of grey | Instant values / boolean condition |
The PSE sits in the DetNet Service sub-Layer of Edge and Relay Nodes. On the one hand, it operates on the packet flow, learning the Track and path selection information from the packet, possibly making local decision and retagging the packet to indicate so. On the other hand, the PSE interacts with the lower layers and with its peers to obtain up-to-date information about its radio links and the quality of the overall Track, respectively, as illustrated in Figure 9.¶
|
packet | going
down the | stack
+==========v==========+=====================+=====================+
| (iOAM + iCTRL) | (L2 Triggers, DLEP) | (oOAM) |
+==========v==========+=====================+=====================+
| Learn from Learn from |
| packet tagging Maintain end-to-end |
+----------v----------+ Forwarding OAM packets |
| Forwarding decision < State +---------^-----------|
+----------v----------+ | Enrich or |
+ Retag Packet | Learn abstracted > Regenerate |
| and Forward | metrics about Links | OAM packets |
+..........v..........+..........^..........+.........^.v.........+
| Lower layers |
+..........v.....................^....................^.v.........+
frame | sent Frame | L2 Ack oOAM | | packet
over | wireless In | In | | and out
v | | v
RAW may control whether and how to use packet replication and elimination (PRE), fragmentation, and network coding, and how the lower layers performs Automatic Repeat reQuest (ARQ), Hybrid ARQ (HARQ) that includes Forward Error Correction (FEC), and other wireless-specific techniques such as overhearing and constructive interferences, in order to increase the reliabiility and availability of the end-to-end transmission. Because RAW may be leveraged on wired links, e.g., to save power, it is not expected that all lower layers support all RAW capabilities. Either way, RAW will manipulate the abstractions of the lower layer services and hint on the expected outcome, and the lower layer will act on those hints to provide the best approximation of the desired outcome, e.g., a level of reliability for one-hop transmission within a bounded budget.¶
Collectively, those function are called PAREO for Packet (hybrid) ARQ, Replication, Elimination and Ordering. By tuning dynamically the use of PAREO functions, RAW avoids the waste of critical resources such as spectrum and energy while providing that the guaranteed SLA, e.g., by adding redundancy only when a spike of loss is observed.¶
In a nutshell, PAREO establishes several paths in a network to provide redundancy and parallel transmissions to bound the end-to-end delay to traverse the network. Optionally, promiscuous listening between paths is possible, such that the Nodes on one path may overhear transmissions along the other path. Considering the scenario shown in Figure 10, many different paths are possible to traverse the network from ingress to egress. A simple way to benefit from this topology could be to use the two independent paths via Nodes A, C, E and via B, D, F. But more complex paths are possible by interleaving transmissions from the lower level of the path to the upper level.¶
(A) -- (C) -- (E)
/ \
Ingress = | | | = Egress
\ /
(B) -- (D) -- (F)
PAREO may also take advantage of the shared properties of the wireless medium to compensate for the potential loss that is incurred with radio transmissions.¶
For instance, when the source sends to Node A, Node B may listen promiscuously and get a second chance to receive the frame without an additional transmission. Note that B would not have to listen if it already received that particular frame at an earlier timeslot in a dedicated transmission towards B.¶
The PAREO model can be implemented in both centralized and distributed scheduling approaches. In the centralized approach, a Path Computation Element (PCE) scheduler calculates a Track and schedules the communication. In the distributed approach, the Track is computed within the network, and signaled in the packets, e.g., using BIER-TE, Segment Routing, or a Source Routing Header.¶
By employing a Packet Replication procedure, a Node forwards a copy of each data packet to more than one successor. To do so, each Node (i.e., Ingress and intermediate Node) sends the data packet multiple times as separate unicast transmissions. For instance, in Figure 11, the Ingress Node is transmitting the packet to both successors, nodes A and B, at two different times.¶
===> (A) => (C) => (E) ===
// \\// \\// \\
Ingress //\\ //\\ Egress
\\ // \\ // \\ //
===> (B) => (D) => (F) ===
An example schedule is shown in Table 2. This way, the transmission leverages with the time and spatial forms of diversity.¶
| Channel | 0 | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|---|
| 0 | S->A | S->B | B->C | B->D | C->F | E->R | F->R |
| 1 | A->C | A->D | C->E | D->E | D->F |
The replication operation increases the traffic load in the network, due to packet duplications. This may occur at several stages inside the Track, and to avoid an explosion of the number of copies, a Packet Elimination procedure must be applied as well. To this aim, once a Node receives the first copy of a data packet, it discards the subsequent copies.¶
The logical functions of Replication and Elimination may be collocated in an intermediate Node, the Node first eliminating the redundant copies and then sending the packet exactly once to each of the selected successors.¶
Considering that the wireless medium is broadcast by nature, any neighbor of a transmitter may overhear a transmission. By employing the Promiscuous Overhearing operation, the next hops have additional opportunities to capture the data packets. In Figure 12, when Node A is transmitting to its DP (Node C), the AP (Node D) and its sibling (Node B) may decode this data packet as well. As a result, by employing correlated paths, a Node may have multiple opportunities to receive a given data packet.¶
===> (A) ====> (C) ====> (E) ====
// ^ | \\ \\
Ingress | | \\ Egress
\\ | v \\ //
===> (B) ====> (D) ====> (F) ====
Variations on the same idea such as link-layer anycast and multicast may also be used to reach more than one next-hop with a single frame.¶
Constructive Interference can be seen as the reverse of Promiscuous Overhearing, and refers to the case where two senders transmit the exact same signal in a fashion that the emitted symbols add up at the receiver and permit a reception that would not be possible with a single sender at the same PHY mode and the same power level.¶
Constructive Interference was proposed on 5G, Wi-Fi7 and even tested on IEEE Std 802.14.5. The hard piece is to synchronize the senders to the point that the signals are emitted at slightly different time to offset the difference of propagation delay that corresponds to the difference of distance of the transmitters to the receiver at the speed of light to the point that the symbols are superposed long enough to be recognizable.¶
RAW uses all forms of diversity including radio technology and physical path to increase the reliability and availability in the face of unpredictable conditions. While this is not done specifically to defeat an attacker, the amount of diversity used in RAW makes an attack harder to achieve.¶
Radio networks typically encrypt at the MAC layer to protect the transmission. If the encryption is per pair of peers, then certain RAW operations like promiscuous overhearing become impossible.¶
RAW will typically select the cheapest collection of links that matches the requested SLA, for instance, leverage free WI-Fi vs. paid 3GPP access. By defeating the cheap connectivity (e.g., PHY-layer interference) the attacker can force an End System to use the paid access and increase the cost of the transmission for the user.¶
This document has no IANA actions.¶
The editor wishes to thank:¶
for their contributions to the text and ideas exposed in this document.¶
The authors wish to thank Dave Cavalcanti and Fabrice Theoleyre for their in-depth reviews during the development of this document.¶