| Internet-Draft | RFC 3535, 20 Years Later | March 2024 |
| Boucadair, et al. | Expires 21 September 2024 | [Page] |
The IAB has organized an important workshop to establish a dialog between network operators and protocol developers, and to guide the IETF focus on work regarding network management. The outcome of that workshop was documented in the "IAB Network Management Workshop" (RFC 3535) which was instrumental for developing NETCONF and YANG, in particular.¶
20 years later, it is time to evaluate what has been achieved since then and identify the operational barriers for making these technologies widely implemented. Also, this document intends to capture new requirements for network management operations.¶
This note is to be removed before publishing as an RFC.¶
Source for this draft and an issue tracker can be found at https://github.com/boucadair/rfc3535-20years-later.¶
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 21 September 2024.¶
Copyright (c) 2024 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 IAB has organized a workshop (June 4-June 6, 2002) to establish a dialog between network operators and protocol developers, and to guide the IETF focus on work regarding network management. The outcome of that workshop was documented in the "IAB Network Management Workshop" [RFC3535] which was instrumental for developing NETCONF [RFC6241], YANG [RFC6020][RFC7950], and RESTCONF [RFC8040].¶
20 years later, new requirements on network management operations are emerging from the operators. This document intends to capture these requirements that reflect the progress in this area. The document also provide an assessment of the RFC3535 recommendations and to what extend that roadmap was driving network management efforts within the IETF.¶
Early version of the document includes many placeholders on purpose as the intent is to collect inputs from interested parties. Items listed in Section 4 are provided to exemplify candidate items to discuss in that section.¶
To be further elaborated:¶
Containerization [I-D.ietf-bmwg-containerized-infra]¶
Network APIs¶
CoAP Management Interface (CORECONF) [I-D.ietf-core-comi]¶
YANG Schema Item iDentifier (YANG SID) [I-D.ietf-core-sid]¶
See also "An Overview of the IETF Network Management Standards" [RFC6632].¶
Section 6 of [RFC3535] includes the following recommendations:¶
The workshop recommends that the IETF stop forcing working groups to provide writable MIB modules. It should be the decision of the working group whether they want to provide writable objects or not.¶
In 2014, the IESG published a statement Writable MIB Module, which states that:¶
SNMP MIB modules creating and modifying configuration state should only be produced by working groups in cases of clear utility and consensus to use SNMP write operations for configuration, and in consultation with the OPS ADs/MIB doctors.¶
The workshop recommends that a group be formed to investigate why current MIB modules do not contain all the objects needed by operators to monitor their networks.¶
Status Update: xxx¶
The workshop recommends that a group be formed to investigate why the current SNMP protocol does not satisfy all the monitoring requirements of operators.¶
Status Update: xxx¶
The workshop recommends, with strong consensus from both protocol developers and operators, that the IETF focus resources on the standardization of configuration management mechanisms.¶
NETCONF [RFC6241], RESTCONF [RFC8040], CORECONF [I-D.ietf-core-comi], YANG.¶
YANG is a transport-independent data modeling language. It can be used independently of NETCONF/RESTCONF. For example, YANG can be used to define abstract data structures [RFC8791] that can be manipulated by other protocols (e.g., [RFC9132]).¶
The workshop recommends, with strong consensus from the operators and rough consensus from the protocol developers, that the IETF/IRTF should spend resources on the development and standardization of XML-based device configuration and management technologies (such as common XML configuration schemas, exchange protocols and so on).¶
The workshop recommends, with strong consensus from the operators and rough consensus from the protocol developers, that the IETF/IRTF should not spend resources on developing HTML-based or HTTP-based methods for configuration management.¶
The IETF deviated from this recommendation, e.g., RESTCONF [RFC8040] or CoAP Management Interface (CORECONF) [I-D.ietf-core-comi].¶
The workshop recommends, with rough consensus from the operators and strong consensus from the protocol developers, that the IETF should continue to spend resources on the evolution of the SMI/SPPI data definition languages as being done in the SMIng working group.¶
SMIng WG was concluded in 2003-04-04.¶
The workshop recommends, with split consensus from the operators and rough consensus from the protocol developers, that the IETF should spend resources on fixing the MIB development and standardization processs.¶
Section 6 of [RFC3535] also includes the following but without tagging them as recommendations:¶
The workshop had split consensus from the operators and rough consensus from the protocol developers, that the IETF should not focus resources on CIM extensions.¶
The IETF didn't dedicate any resources on CIM extensions.¶
The workshop had rough consensus from the protocol developers that the IETF should not spend resources on COPS-PR development. So far, the operators have only very limited experience with COPS-PR. In general, however, they felt that further development of COPS-PR might be a waste of resources as they assume that COPS-PR does not really address their requirements.¶
The workshop had rough consensus from the protocol developers that the IETF should not spend resources on SPPI PIB definitions. The operators had rough consensus that they do not care about SPPI PIBs.¶
The current YANG device models ecosystem is fragmented: some standards models are defined in the IETF while similar ones are defined in other fora such as Openconfig or ONF. Unlike service and network models, IETF-defined device models are not widely implemented. There is a need to rationalize this space and avoid redundant efforts.¶
Many NETCONF-related features are (being) specified by the IETF, but these features are not widely supported (e.g., Push). Editing a profile document with a set of recommendations about core/key features with the appropriate justification will help the emergence of more implementations that meet the operators’ needs.¶
Examples of such profile documents are the various RFCs that were published by the behave WG [BCP127]. Another approach is to consider an approach similar to the "Roadmap for Transmission Control Protocol (TCP) Specification Documents" [RFC7414]. Such a document would serve as a guide and reference for implementers and any other parties who desire information contained in the 'NETCONF/RESTCONF/YANG'-related RFCs.¶
Likewise, reassess the value of some IETF proposals vs. competing/emerging solutions would be useful (e.g., gRPC vs. YANG-Push).¶
RFCs might not be suited for documenting YANG modules (it takes much too long, especiallly for updates). In the meantime, there is a need for "reference models" and "sufficiently stable models". An hybrid approach might be investigated for documenting IETF- endorsed YANG modules, such as considering an RFC to describe the initial module sketch and objectives and an official IETF repository for maintaining intermediate YANG versions.¶
Section 3 of [RFC3535] describes a set of network operator requirements. One of the requirements is the ease of use which, according to Section 3.2 of [RFC6244], is addressed by NETCONF and YANG. For configuration this holds true, for network observability it is unfortunately not yet. This has been confirmed with a set of network operators asking how long it takes from subscribing YANG data to make it accessible to the operator. Minutes, Hours, Days, or Weeks. None of them answered Minutes or Hours. All of them responded Days or Weeks. Hinting manual post processing of YANG data.¶
Collecting YANG metrics from networks is already a struggle due to late arrival of [RFC8639], [RFC8640], [RFC8641], [I-D.ietf-netconf-https-notif], and [I-D.ietf-netconf-udp-notif] for configured subscription transport protocols which defined YANG-Push in the industry. This caused network vendors to implement alternative solutions to collect real-time streaming data in the meanwhile, such as gNMI which was proposed in 2018 in [I-D.openconfig-rtgwg-gnmi-spec] to the IETF but not followed up on. Unfortunately, these implementations differ between network Operating Systems due to the lack of standardization, specifically for the metadata which would ensure machine readability.¶
When a set of network operators where asked to where operational YANG data needs to be integrated to, the answer homogeneously was Apache Kafka Message Broker and Time Series Databases. There is a need to specify how YANG-Push can be integrated into Apache Kafka and references needed YANG-Push extensions and YANG schema registry development. The YANG-Push extensions addressing needs to make YANG-Push messages machine readable and against semantic validate able to ensure a consistent data processing.¶
Another challenge is that the subscribed YANG data referenced with datastore-subtree-filter or datastore-xpath-filter breaks semantic integrity which needs to be addressed by either updating Section 4 of [RFC8641] or proposing a new YANG module being used at the YANG-Push receiver.¶
The use of a flat tree hierarchy in YANG models may induce some performance issues compared to other graph models. See, for example, [ODL].¶
Network Telemetry, as described in [RFC9232], involve a set of protocols. Due to the different requirements, one Network Telemetry protocol doesn't address all needs. This is mainly due to the nature of the subscribed data. BGP Monitoring Protocol (BMP) [RFC7854] adds monitoring and tracing capabilities natively to the BGP process to minimize the processing overhead. While IPFIX [RFC7011][RFC7012] can be applied according to [RFC5472] to gain visibility into the data and forwarding planes, due to the amount of data, sampling as defined in [RFC5476] and applied to IPFIX in [RFC5477] and aggregation as defined in [RFC7015] for IPFIX is needed to reduce the amount of exposed data. While YANG-Push focuses on exposing already YANG modelled data, which eases the correlation among network configuration and operational data.¶
[RFC9232] is an informational document and does not specify what these Network Telemetry protocols should have in common to ensure consistent data structures for data export. While data types are fairly good aligned, a lack of metadata standardization among the Network Telemetry protocols is observed. In particular describing from where the metrics has been exported from and timestamping. In Section 4.2 of [RFC7854] timestamps are optional and sysName [RFC1213] is only carried in the BMP initiation message (Section 4.3 of [RFC7854]), while the message header of IPFIX defined in Section 4.3 of [RFC7011] lacks the sysName definition.¶
The lack of information from where the data is being pushed from is only known to the Network Telemetry data collection due to the transport session being established from the network node exporting the information. When Network Telemetry messages are being transformed and forwarded, this information is being lost. Therefore, it is common among network operators to augment sysName and other metadata at the data collection.¶
The same common principle applies to when observation timestamping is missing in the Network Telemetry message. Since the data collection is the closest element to the network, a time stamp is added to give the network operator at least the information when the Network Telemetry message was collected. However, since Network Telemetry addresses real-time streaming needs, this is often not accurate enough for data correlation.¶
Pluggins/Proxy YANG/CLI is still the rule in many operations.¶
Complexity in dev the pluggins (as you need to cover many OS/vendors).¶
Network models for the realization provides some "level" of abstraction and then automations.¶
TBC.¶
From the time RFC 3535 was released up to now, new kind of services and applications have been developed and deployed over the time, with very diverse, and some times contradicting, requirements. Those services have been engineered on top of multi-service networks for the sake of efficiency and simplicity, accommodating such a variety of needs. As a result, services requiring mobility, data replication, large capacity, adaptability, multi-path support, determinism, etc., coexist on the same shared network, needing from it mechanisms for graceful operation.¶
Likewise, such diversity of services also require different management capabilities. For example, session continuity, distribution trees, traffic engineering, congestion status notification, reordering, or on-time delivery impose very different management needs to be satisfied.¶
This reality is different from the one existing at the time of [RFC3535], and as such, the new identified needs can require from novel approaches to guarantee the aforementioned co-existence of services.¶
Also, some networks have specific network management requirements such as the need for asynchronous operations or constraints on data compactness. An example of such networks is Delay-Tolerant Networking (DTN) [RFC838].¶
Networks are being updated to abandon the silo approach from the past towards an increasing convergence. Specifically, there are trends towards a tighter interaction and integration of different technologies previously considered as totally separated from an operational perspective. Examples of that trends are the IP and Optical integration (e.g., the introduction of colored interfaces on routers), or the extension of deterministic-behavior features to Layer 3 networks. This kind of convergence in most cases creates dependencies on the conventional network management features, which require to incorporate or integrate functionality from other technological domains.¶
Furthermore, such convergence is also reflected on the need of interacting and interworking with distinct network parts participating in the end-to-end service delivery. Mobile access, fixed access, data center, enterprise, radio functional split (i.e., fronthaul and midhaul), neutral exchanges, intensive data networks (e.g., scientific academic networks), content distribution, etc., represent network parts constituent of end-to-end services that can impose dependencies of the management of an intermediate network.¶
That convergence shown the last years also implies the need of an up-to-date refresh of management capabilities and tooling of the conventional networks. Also, it highlights the need to easily map the data models that are used to manage each specific segment.¶
For example, [RFC8667] (IS-IS extensions for SR) was published in December 2019, while [I-D.ietf-isis-sr-yang] will be published ~5 years after.¶
Consider having YANG as part of the protocol specification/change where possible, or have the YANG document progress in parallel. That may slow down the protocol specification, though.¶
While there are open-source implementations for NETCONF (e.g., NETOPEER), the gRPC/gNMI suite seems to have more support for tools on the client side. For example, "ygot" generates structures from YANG models and these can easily be used by a client to configure a device with gNMI. NETCONF is not supported though (we need the XML tags).¶
TBC.¶
This document has no IANA actions.¶
TODO acknowledge.¶