| RAW | F. Theoleyre |
| Internet-Draft | CNRS |
| Intended status: Standards Track | G. Papadopoulos |
| Expires: January 11, 2021 | IMT Atlantique |
| G. Mirsky | |
| ZTE Corp. | |
| July 10, 2020 |
Operations, Administration and Maintenance (OAM) features for RAW
draft-theoleyre-raw-oam-support-03
Some critical applications may use a wireless infrastructure. However, wireless networks exhibit a bandwidth of several orders of magnitude lower than wired networks. Besides, wireless transmissions are lossy by nature; the probability that a packet cannot be decoded correctly by the receiver may be quite high. In these conditions, guaranteeing the network infrastructure works properly is particularly challenging, since we need to address some issues specific to wireless networks. This document lists the requirements of the Operation, Administration, and Maintenance (OAM) features recommended to construct a predictable communication infrastructure on top of a collection of wireless segments. This document describes the benefits, problems, and trade-offs for using OAM in wireless networks to achieve Service Level Objectives (SLO).
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 January 11, 2021.
Copyright (c) 2020 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 Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
Reliable and Available Wireless (RAW) is an effort that extends DetNet to approach end-to-end deterministic performances over a network that includes scheduled wireless segments. In wired networks, many approaches try to enable Quality of Service (QoS) by implementing traffic differentiation so that routers handle each type of packets differently. However, this differentiated treatment was expensive for most applications.
Deterministic Networking (DetNet) [RFC8655] has proposed to provide a bounded end-to-end latency on top of the network infrastructure, comprising both Layer 2 bridged and Layer 3 routed segments. Their work encompasses the data plane, OAM, time synchronization, management, control, and security aspects.
However, wireless networks create specific challenges. First of all, radio bandwidth is significantly lower than for wired networks. In these conditions, the volume of signaling messages has to be very limited. Even worse, wireless links are lossy: a layer 2 transmission may or may not be decoded correctly by the receiver, depending on a broad set of parameters. Thus, providing high reliability through wireless segments is particularly challenging.
Wired networks rely on the concept of links. All the devices attached to a link receive any transmission. The concept of a link in wireless networks is somewhat different from what many are used to in wireline networks. A receiver may or may not receive a transmission, depending on the presence of a colliding transmission, the radio channel's quality, and the external interference. Besides, a wireless transmission is broadcast by nature: any neighboring device may be able to decode it. The document includes detailed information on what the implications for the OAM features are.
Last but not least, radio links present volatile characteristics. If the wireless networks use an unlicensed band, packet losses are not anymore temporally and spatially independent. Typically, links may exhibit a very bursty characteristic, where several consecutive packets may be dropped. Thus, providing availability and reliability on top of the wireless infrastructure requires specific Layer 3 mechanisms to counteract these bursty losses.
Operations, Administration, and Maintenance (OAM) Tools are of primary importance for IP networks [RFC7276]. It defines a toolset for fault detection, isolation, and performance measurement.
The primary purpose of this document is to detail the specific requirements of the OAM features recommended to construct a predictable communication infrastructure on top of a collection of wireless segments. This document describes the benefits, problems, and trade-offs for using OAM in wireless networks to provide availability and predictability.
In this document, the term OAM will be used according to its definition specified in [RFC6291]. We expect to implement an OAM framework in RAW networks to maintain a real-time view of the network infrastructure, and its ability to respect the Service Level Objectives (SLO), such as delay and reliability, assigned to each data flow.
OAM Operations, Administration, and Maintenance
DetNet Deterministic Networking
SLO Service Level Objective
QoS Quality of Service
SNMP Simple Network Management Protocol
SDN Software-Defined Network
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.
RAW networks expect to make the communications reliable and predictable on top of a wireless network infrastructure. Most critical applications will define an SLO to be required for the data flows it generates. RAW considers network plane protocol elements such as OAM to improve the RAW operation at the service and the forwarding sub-layers.
To respect strict guarantees, RAW relies on an orchestrator able to monitor and maintain the network. Typically, a Software-Defined Network (SDN) controller is in charge of scheduling the transmissions in the deployed network, based on the radio link characteristics, SLO of the flows, the number of packets to forward. Thus, resources have to be provisioned a priori to handle any defect. OAM represents the core of the pre-provisioning process and maintains the network operational by updating the schedule dynamically.
Fault-tolerance also assumes that multiple paths have to be provisioned so that an end-to-end circuit keeps on existing whatever the conditions. The Packet Replication and Elimination Function ([PREF-draft]) on a node is typically controlled by a central controller/orchestrator. OAM mechanisms can be used to monitor that PREOF is working correctly on a node and within the domain.
To be energy-efficient, reserving some dedicated out-of-band resources for OAM seems idealistic, and only in-band solutions are considered here.
RAW supports both proactive and on-demand troubleshooting.
The specific characteristics of RAW are discussed below.
In wireless networks, a link does not exist. A common convention is to define a wireless link as a pair of devices that have a non-null probability of transmitting and decoding a packet. Similarly, we designate as neighbor any device which as a link with a specific transmitter.
Each wireless link is associated with a link quality, often measured as the Packet Delivery Ratio (PDR), i.e., the probability that the receiver can decode the packet correctly. It is worth noting that this link quality depends on many criteria, such as the level of external interference, the presence of concurrent transmissions, or the radio channel state. This link quality is even time-variant.
In modern switching networks, the unicast transmission is delivered uniquely to the destination. Wireless networks are much closer to the ancient shared access wireless networks. Unicast transmission is similar to a broadcast one and can be received by any neighbor.
However, contrary to wired networks, we cannot be sure that a packet is received by all the devices attached to the network. It depends on the radio channel state between the transmitter(s) and the receiver(s). In particular, concurrent transmissions may be possible or not, depending on the radio conditions.
Multiple neighbors may receive a transmission. Thus, anycast layer-2 forwarding helps to maximize the reliability by assigning multiple receivers to a single transmission. That way, the packet is lost only if none of the receivers decode it. Practically, it has been proven that different neighbors may exhibit very different radio conditions, and that reception independency may hold for some of them [anycast-property].
OAM features will enable RAW with robust operation both for forwarding and routing purposes.
Several solutions (e.g., Simple Network Management Protocol (SNMP), YANG-based data models) are already in charge of collecting the statistics. That way, we can encapsulate these statistics in specific monitoring packets, to send them to the controller.
We need to verify that two endpoints are connected. In other words, there exists "one" way to deliver the packets between two endpoints A and B. The solution may not here defer from those of detnet.
Additionally, to the Continuity Check, we have to verify the connectivity. This verification considers additional constraints, i.e., the absence of misconnection.
In particular, the resources have to be reserved by a given flow, and no packets from other flows steal the corresponding resources. Similarly, the destination does not receive packets from different flows through its interface.
Because of radio transmissions' broadcast nature, several receivers may be active at the same time to enable anycast Layer 2 forwarding. Thus, the connectivity verification must test any combination. We also consider priority-based mechanisms for anycast forwarding, i.e., all the receivers have different probabilities of forwarding a packet. To verify a delay SLO for a given flow, we must also consider all the possible combinations, leading to a probability distribution function for end-to-end transmissions. If this verification is implemented naively, the number of combinations to test may be exponential and too costly for wireless networks with low bandwidth.
It is worth noting that the control and data packets may not follow the same path. The connectivity verification has to be conducted in-band without impacting the data traffic. Test packets MUST share the fate with the monitored data traffic without introducing congestion in normal network conditions.
ICMP tools are comprehensive tools for diagnostic. They help to identify a subset of the list of routers in the route. To ensure predictable performance, resources are reserved per flow in RAW. Thus, we need to define route tracing tools able to track the route for a specific flow.
Wireless networks are meshed by nature: we have many redundant radio links. These meshed networks are both an asset and a drawback: while several paths exist between two endpoints, and we should choose the most efficient one(s), concerning specifically the reliability, and the delay.
Thus, multipath routing can be considered to make the network fault-tolerant. Even better, we can exploit the broadcast nature of wireless networks to exploit meshed multipath routing: we may have multiple Maintenance Intermediate Endpoints (MIE) for each hop in the path. In that way, each Maintenance Intermediate Endpoint has several possible next hops in the forwarding plane. Thus, all the possible paths between two maintenance endpoints should be retrieved, which may quickly become untractable if we apply a naive approach.
RAW expects to operate fault-tolerant networks. Thus, we need mechanisms able to detect faults, before they impact the network performance.
Wired networks tend to present stable performances. On the contrary, wireless networks are time-variant. We must consequently make a distinction between normal evolutions and malfunction.
The network has to detect when a fault occurred, i.e., the network has deviated from its expected behavior. While the network must report an alarm, the cause may not be identified precisely. For instance, the end-to-end reliability has decreased significantly, or a buffer overflow occurs.
The network has isolated and identified the cause of the fault. While detnet already expects to identify malfunctions, some problems are specific to wireless networks. We must consequently collect metrics and implement algorithms tailored for wireless networking. For instance, the quality of a specific link has decreased, requiring more retransmissions, or the level of external interference has locally increased.
The network has to expose a collection of metrics to support an operator making proper decisions, including:
These metrics should be collected:
We have to minimize the number of statistics / measurements to exchange:
Thus, localized and centralized mechanisms have to be combined together, and additional control packets have to be triggered only after a fault detection.
RAW aims to enable real-time communications on top of a heterogeneous architecture. Wireless networks are known to be lossy, and RAW has to implement strategies to improve reliability on top of unreliable links. Hybrid Automatic Repeat reQuest (ARQ) has typically to enable retransmissions based on the end-to-end reliability and latency requirements.
To make correct decisions, the controller needs to know the distribution of packet losses for each flow, and each hop of the paths. In other words, the average end-to-end statistics are not enough. They must allow the controller to predict the worst-case.
RAW targets also low-power wireless networks, where energy represents a key constraint. Thus, we have to take care of power and bandwidth consumption. The following techniques aim to reduce the cost of such maintenance:
RAW needs to implement a self-healing and self-optimization approach. The network must continuously retrieve the state of the network, to judge about the relevance of a reconfiguration, quantifying:
Thus, reconfiguration may only be triggered if the gain is significant.
When multiple paths are reserved between two maintenance endpoints, they may decide to replicate the packets to introduce redundancy, and thus to alleviate transmission errors and collisions. For instance, in Figure 1, the source node S is transmitting the packet to both parents, nodes A and B. Each maintenance endpoint will decide to trigger the replication/elimination process when a set of metrics passes through a threshold value.
===> (A) => (C) => (E) ===
// \\// \\// \\
source (S) //\\ //\\ (R) (root)
\\ // \\ // \\ //
===> (B) => (D) => (F) ===
Figure 1: Packet Replication: S transmits twice the same data packet, to its DP (A) and to its AP (B).
Wireless networks exhibit time-variant characteristics. Thus, the network has to provide additional resources along the path to fit the worst-case performance. This time-variant characteristics make the resource reservation very challenging: over-reaction waste radio and energy resources. Inversely, under-reaction jeopardize the network operations, and some SLO may be violated.
Wireless networks are known to be lossy. Thus, commands may be received or not by the node to reconfigure. Unfortunately, inconsistent states may create critical misconfigurations, where packets may be lost along a path because it has not been properly configured.
We have to propose mechanisms to guarantee that the network state is always consistent, even if some control packets are lost. Timeouts and retransmissions are not sufficient since the reconfiguration duration would be, in that case, unbounded.
This document has no actionable requirements for IANA. This section can be removed before the publication.
This section will be expanded in future versions of the draft.
TBD