An NETCONF- and NETMOD-based Architecture for Network Management

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, .

Networks are increasing in complexity and capacity, as well as the density of the services deployed upon them. Uptime, reliability, and predictable latency requirements drive the need for automation. The problems with network management are not simple. They are complex and intricate. But these problems must be solved for networks to meet the stability needs of existing services while incorporating new services in a world where the growth of the networks is exhausting the supply of qualified networking engineers. We need to move from a CLI world into a world of automation, but that automation must be robust and trustworthy. This document presents an architecture based on NETCONF () and . NETCONF and YANG address the problems of network management with flexibility and expressiveness. NETCONF allows any manner of configuration and operational data to be carried with few rules governing the data. YANG allows data models to be defined that are flexible and extensible in ways that allow the data to be cohesive and structured, but not rigid. This approach allows the device to express its native capabilities in a way that is flexible and extensible. Evolution of devices and data models are permitted and managed.

The document mirrors terminology from NETCONF. NETCONF uses a simple RPC-based mechanism to facilitate communication between a client and a server. The client can be a script or application typically running as part of a network management system. The server is typically a network device. The terms "device" and "server" are used interchangeably in this document, as are "client" and "application".

NETCONF defines an XML-based RPC mechanism that leverages the simplicity and availability of high-quality XML parsers. XML gives a rich, flexible, hierarchical, standard representation of data that matches the needs of networking devices. NETCONF carries configuration data and operations encoded in XML using an RPC mechanism over a connection-oriented transport. XML's hierarchical data representation allows complex networking data to be rendered in a natural way. For example, the following configuration places interfaces in OSPF areas. The <ospf> element contains a list of <area> elements, each of which contain a list of <interface> elements. The <name> element identifies the specific area or interface. Additional configuration for each area or interface appears directly inside the appropriate element.

0.0.0.0 ge-0/0/0.0 30 100

120

ge-0/0/1.0 140 10.1.2.0 ge-0/0/2.0 100 ge-0/0/3.0 140

120

]]> NETCONF includes mechanisms for controlling configuration datastores, fetching state data, receiving notifications, and allows for additional RPC methods. Configuration operations include the ability to lock datastores to isolate one application from the actions of others, the ability to save and restore configuration data sets, and the ability to discover (via the <hello> message) the capabilities of the device.

YANG is the data model language for NETCONF that allows the description of hierarchies of data model nodes ("nodes") and the constraints that exist amongst them. YANG defines data models and how to manipulate those models via NETCONF protocol operations. Each YANG module defines a data model, uniquely identified by a namespace URI. These data models are extensible in a manner that allows tight integration of standard data models and proprietary data models. Models are built from organizational containers, lists of data instances and leaf data values.

A YANG module defines a data model in terms of the data, its hierarchical organization, and the constraints on that data. YANG defines how this data is represented in XML and how that data is used in NETCONF operations.

XML includes the concept of namespaces, allowing XML elements from different sources to be combined in the same hierarchy without risking collision. YANG modules define content for specific namespaces, but one module may augment the definition of another module, introducing elements from that module's namespace into the first module's hierarchy. Since one module can augment another module's definition, hierarchies of definitions are allowed to grow, as definitions from multiple sources are added to the base hierarchy. These augmentations are qualified using the namespace of the source module, helping to avoid issues with name conflicts as the modules change over time. For example, if the above OSPF configuration were the standard, a vendor module may augment this with vendor-specific extensions.

BL: need leading slash:

The <no‑neighbor‑down‑notification> element is then placed in the vendorx namespace:

0.0.0.0 ge-0/0/0.0 30

]]> Augmentations are seamlessly integrated with base modules, allowing them to be fetched, archived, loaded, and deleted within their natural hierarchy. If a client application asks for the configuration for a specific OSPF area, it will receive the sub-hierarchy for that area, complete with any augmentations.

In the NETMOD architecture, each device vendor implements a set of data models in their devices. These models are either standard data models, defined in YANG modules published by a standards body, or proprietary data models, defined in YANG modules published by vendors. Standard data models define content that is independent of the vendor, allowing client applications to request specific behavior without concern for the vendor, product line, or installed software revision. The translation between the standard model and the device specific behavior is performed by the device, freeing the application from such concerns. Proprietary data models allow the vendor to accurately describe the content and behavior of their devices in explicit detail. Applications may take advantage of these specifics to give their users complete control over the device. When a NETCONF session begins, the namespaces for all supported modules are announced as capabilities via the device's <hello> message. The device should also support the schema discovery mechanism [ref], enabling applications to discover the location from which the modules may be downloaded. The schema discovery for standard YANG modules should list a common, standard location for these modules, presumably one set by the organization that defined the standard. When an application connects with a device, it receives the list of capabilities supported by that device. The application may compare the set of capabilities announced by the device with the set of modules the application is aware of. Any new modules or new revisions of known modules may be downloaded as needed from the locations given via the schema discovery mechanism. Once the application has access to the YANG modules, it may manipulate the device as a "YANG data browser", capable of parsing the elements sent from the device with an understanding of the organization of the data. The module describes the syntax of the data and constraints on that data, allowing the application to create data that abides by those constraints. To have a real understanding of a module's content, the application may need to incorporate logic specific to that module. Semantic information contained in description statements is not machine readable, but module-specific custom work can be done to tailor the user interface to the particular semantic needs of a module. For example, a module could define the "location" of a device using longitude and latitude, and the application can use the "browser" style to display this data using input fields in a web form. Custom logic would be needed to take the value of these fields and place the device on a map of the world, and additional logic would be needed to update the data values when the user drags the device from Dallas to Dulles. If an application is meant to manage a specific problem, it may model the data internally in whatever form is most convenient to its organizational needs. When the application interacts with a device, it may choose one of two paths. If the device implements a standard module, the application may generate content for that standard by translating its internal form into the standard one. If the device doesn't implement such a standard or no such standard exists, the application may use a transformation that is particular to that device's vendor, product model, hardware, or software. Depending on the application, this transformation may be provided by the application vendor, the device vendor, a third-party, or the provider. For a popular application, the device vendor may wish to provide this transformation to increase market acceptance of their devices. For popular devices, the application may provide this transformation as a means of making the application useful in the maximum number of provider networks. For problem domains where the mapping from the application to the device is not straight-forward or requires tailoring to the specific provider or environment, the provider may wish to control this transformation. Additionally, other parties may make such transformations available via open source. Note that both cases may appear within a single application on an "as needed" basis. If the device announces the capability for the standard YANG module, the application may transmit to the device via NETCONF the content in the standard modules format. If the device does not announce the appropriate capability, the application may find a transformation that matches the device, perform the transformation on the standard data to produce device native configuration, and transmit via NETCONF that device configuration to the device. In both cases, the key is the ability to discover the capabilities of the specific device, download the YANG modules that support those capabilities, gain an understanding of those data models and their constraints, generate appropriate content, and transmit that content to the device.

The YANG data modeling language is the central piece of a group of related technologies. The YANG language itself, described in [ref], defines the syntax of the language and its statements, the meaning of those statements, and how to combine them to build the hierarchy of nodes that describe a data model. That document also defines the "on the wire" XML content for NETCONF operations on data models defined in YANG modules. This includes the basic mapping between YANG data tree nodes and XML elements, as well as mechanisms used in <edit‑config> content to manipulate that data, such as arranging the order of nodes within a list. YANG uses a syntax that is regular and easily described, primarily designed for human readability. YANG's syntax is friendly to email, diff, patch, and the constraints of RFC formatting.

In some environments, incorporating a YANG parser may not be an acceptable option. For those scenarios, an XML grammar for YANG is defined in YIN (YANG Independent Notation) [ref]. YIN allows the use of XML parsers which are readily available in both open source and commercial versions. Conversion between YANG and YIN is direct, loss-less and reversible. YANG statements are converted to XML elements, preserving the structure and content of YANG, but enabling the use of off-the-shelf XML parsers rather than requiring the integration of a YANG parser. YIN maintains complete semantic equivalence with YANG. BL: use of xslt is a key use

Since NETCONF content is encoded in XML, it is natural to use XML schema languages for their validation. To facilitate this, YANG offers a standardized mapping of YANG modules into Document Schema Description Languages (DSDL) [DSDL]. DSDL is considered to be the best choice for the given purpose because it addresses not only grammar and datatypes of XML documents but also semantic constraints and rules for modifying information set of the document. In addition, DSDL offers formal means for coordinating multiple independent schemas and specifying how to apply the schemas to the various parts of the document. This is useful since YANG content is typically composed of multiple vocabularies.

YANG supports a number of builtin types, and allows additional types to be derived from those types in an extensible manner. New types can add additional restrictions to allowable data values. A standard type library for use by YANG is available [ref]. These YANG modules define commonly used data types for IETF-related standards.

The data model in a YANG module yields value in five specific areas.

The YANG model tells the device developer what data is being modeled. The developer reads the YANG models, absorbs the zen of the model, and writes code that supports the model. The model describes the data hierarchy and associated constraints, and the description and reference material helps the developer understand how to transform the models view into the device native implementation.

The YANG model can be compiled into a YANG-based engine for either the client or server side. Incoming data can be validated, as can outgoing data. The complete configuration datastore may be validated in accordance with the constraints described in the data model. Serializers and deserializers for generating and receiving NETCONF content can be driven by the meta-data in the model. As data is received, the meta-data is consulted to ensure the validity of incoming XML elements.

The YANG module dictates the XML encoding sent "over the wire", though actual transmission should be encrypted so as not to appear as readable text on the physical media. The rules that define the encoding are fixed, so the YANG module can be used to ascertain whether a specific NETCONF payload is obeying the rules.

The YANG module tells the application developer what data can be modeled. Developers can inspect the modules and take one of three distinct views. In this section, we will consider them and the impact of YANG on their design. In the real world, most applications are a mixture of these approaches.

An application can be coded against the specific, well-known contents of YANG modules, implementing their organization, rules, and logic directly with explicit knowledge. For example, a script could be written to change the domain name of a set of devices using a standard YANG module that includes such a leaf node. This script takes the new domain name as an argument and insert it into a string containing the rest of the XML encoding as required by the YANG module. This content is then sent via NETCONF to the devices. This type of application is useful for small, fixed problems where the cost and complexity of flexibility is overwhelmed by the ease of hard coding direct knowledge into the application.

An application may take a generic, bottom up approach to configuration, concentrating on the device's data directly and treating that data without specific understanding. YANG modules may be used to drive the operation of the YANG equivalent of a "MIB Browser". Such an application manipulates the device's configuration data based on the data organization contained in the YANG module. For example, a GUI may present a straight-forward visualization where elements of the YANG hierarchy are depicted in a hierarchy of folders or GUI panels. Clicking on a line expands to the contents of the matching content. BL: GUI comments don't need to be here This type of GUI can easily be built by generating XSLT stylesheets from the YANG data models. An XSLT engine can then be used to turn configuration data into a set of web pages. The YANG modules allows the application to enforce a set of constraints without understanding the semantics of the YANG module.

In contrast to the bottom-up approach, the top-down approach allows the application to take a view of the configuration data which is distinct from the standard and/or proprietary YANG modules. The application is free to construct its own model for data organization and to present this model to the user. When the application needs to transmit data to a device, the application transforms its data from the problem-oriented view of the world into the data needed for that particular device. This transformation is under the control and maintenance of the application, allowing the transformation to be changed and updated without affecting the device. For example, an application could be written that models VPNs in a network-oriented view. The application would need to transform these high-level VPN definitions into the configuration data that would be handed to any particular device within a VPN.

In developing good data models, there are many conflicting interests the data modeler must keep in mind. Modelers need to be aware of four types of behavior in modeled device: [strict compliance] behavior that follow the model completely [modeled deviations] behavior that follows within deviations allowed by the model [allowable deviations] behavior that falls outside the model, but can still be handled [unacceptable deviations] behavior that is not at all consistent with the model Once the model is published, an implementer may decide to make a particular data model node configurable, where the standard model describes it a state data. The implementation reports the value normally and may have an "out of band" mechanism for reporting that this device behaves in a different manner than the standard. Applications capable of discovering such behavior can make allowances, but applications that do not discover such behavior can continue treating the implementation as if it were compliant. Rarely, implementations may make decisions that prevent compliance with the standard. Such occasions are regrettable, but they remain a part of reality, and modelers and application writers ignore them at their own risk. An implementation that emits an integer leaf as "cow" would be difficult to manage, but applications must expect to encounter such misbehaving devices in the field. Despite this, both client and server should view the YANG module as a contract, with both sides agreeing to abide by the terms. The modeler should be explicit about the terms of such a contract, and both client and server implementations should strive to faithfully and accurately implement the data model described in the YANG module.

This document defines a language with which to write and read descriptions of management information. The language itself has no security impact on the Internet.