| Internet-Draft | DTNMA | January 2024 |
| Birrane, et al. | Expires 1 August 2024 | [Page] |
The Delay-Tolerant Networking (DTN) architecture describes a type of challenged network in which communications may be significantly affected by long signal propagation delays, frequent link disruptions, or both. The unique characteristics of this environment require a unique approach to network management that supports asynchronous transport, autonomous local control, and a small footprint (in both resources and dependencies) so as to deploy on constrained devices.¶
This document describes a DTN management architecture (DTNMA) suitable for managing devices in any challenged environment but, in particular, those communicating using the DTN Bundle Protocol (BP). Operating over BP requires an architecture that neither presumes synchronized transport behavior nor relies on query-response mechanisms. Implementations compliant with this DTNMA should expect to successfully operate in extremely challenging conditions, such as over uni-directional links and other places where BP is the preferred transport.¶
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 1 August 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 Delay-Tolerant Networking (DTN) architecture, as described in [RFC4838], has been designed to cope with data exchange in challenged networks. Just as the DTN architecture requires new capabilities for transport and transport security, special consideration must be given for the management of DTN devices.¶
This document describes a logical DTN Management Architecture (DTNMA) providing configuration, monitoring, and local control of both application and network services on a managed device. The DTNMA is designed to provide for the management of devices operating either within or across a challenged network.¶
Fundamental properties of a challenged network are outlined in Section 2.2.1 of [RFC7228]. These properties include lacking end-to-end IP connectivity, having "serious interruptions" to end-to-end connectivity, and exhibiting delays longer than can be tolerated by end-to-end synchronization mechanisms (such as TCP). It is further noted that the DTN architecture was designed to cope with such networks.¶
Device management in these environments must occur without human interactivity, without system-in-the-loop synchronous function, and without requiring a synchronous underlying transport layer. This means that managed devices need to determine their own schedules for data reporting, their own operational configuration, and perform their own error discovery and mitigation.¶
Certain outcomes of device self-management should be determinable by a privileged external observer (such as a managing device). In a challenged network, these observers may need to communicate with a managed device after significant periods of disconnectedness. Non-deterministic behavior of a managed device may make establishing communication difficult or impossible.¶
The desire to define asynchronous and autonomous device management is not new. However, challenged networks (in general) and the DTN environment (in particular) represent unique deployment scenarios and impose unique design constraints. To the extent that these environments differ from more traditional, enterprise networks, their management may also differ from the management of enterprise networks. Therefore, existing techniques may need to be adapted to operate in the DTN environment or new techniques may need to be created.¶
This document describes the desirable properties of, and motivation for, a DTNMA. This document also provides a reference model, service descriptions, autonomy model, and use cases to better reason about ways to standardize and implement this architecture.¶
This document provides informative guidance to authors and users of such models, protocols, and implementations.¶
The selection of any particular transport or network layer is outside of the scope of this document. The DTNMA does not require the use of any specific protocol such as IP, BP, TCP, or UDP. In particular, the DTNMA design does not assume the use of either IPv4 or IPv6.¶
Network features such as naming, addressing, routing, and security are out of scope of the DTNMA. It is presumed that any operational network communicating DTNMA messages would implement these services for any payloads carried by that network.¶
The interactions between and amongst the DTNMA and other management approaches are outside of the scope of this document.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].¶
The remainder of this document is organized into the following nine sections, described as follows.¶
This section defines terminology that either is unique to the DTNMA or is necessary for understanding the concepts defined in this specification.¶
The DTNMA provides network management services able to operate in a challenged network environment, such as envisioned by the DTN architecture. This section describes what is meant by the term "challenged network", the important properties of such a network, and observations on impacts to conventional management approaches.¶
Constrained networks are defined as networks where "some of the characteristics pretty much taken for granted with link layers in common use in the Internet at the time of writing are not attainable." [RFC7228]. This broad definition captures a variety of potential issues relating to physical, technical, and regulatory constraints on message transmission. Constrained networks typically include nodes that regularly reboot or are otherwise turned off for long periods of time, transmit at low or asynchronous bitrates, and/or have very limited computational resources.¶
Separately, a challenged network is defined as one that "has serious trouble maintaining what an application would today expect of the end-to-end IP model" [RFC7228]. Links in such networks may be impacted by attenuation, propagation delays, mobility, occultation, and other limitations imposed by energy and mass considerations. Therefore, systems relying on such links cannot guarantee timely end-to-end data exchange.¶
By these definitions, a "challenged" network is a special type of "constrained" network, where constraints prevent timely end-to-end data exchange. As such, "all challenged networks are constrained networks ... but not all constrained networks are challenged networks ... Delay-Tolerant Networking (DTN) has been designed to cope with challenged networks" [RFC7228].¶
Solutions that work in constrained networks might not be solutions that work in challenged networks. In particular, challenged networks exhibit the following properties that impact the way in which the function of network management is considered.¶
The set of constraints that might be present in a challenged network impact both the topology of the network and the services active within that network.¶
Operational networks handle cases where nodes join and leave the network over time. These topology changes may or may not be planned, they may or may not represent errors, and they may or may not impact network services. Challenged networks differ from other networks not in the present of topological change, but in the likelihood that impacts to topology result in impacts to network services.¶
The difference between topology impacts and service impacts can be expressed in terms of connectivity. Topological connectivity usually refers to the existence of a path between an application message source and destination. Service connectivity, alternatively, refers to the existence of a path between a node and one or more services needed to process (often just-in-time) application messaging. Examples of service connectivity include access to infrastructure elements such as a Domain Name System (DNS) or a Certificate Authority (CA).¶
In networks that might be partitioned most of the time, it is less likely that a node would concurrently access both an application endpoint and one or more network service endpoints. For this reason, network services in a challenged network should be designed to allow for asynchronous operation. Accommodating this use case often involves the use of local caching, pre-placing information, and not hard-coding message information at a source that might change when a message reaches its destination.¶
Network management approaches must adapt to the topology and service impacts encountered in challenged networks. In particular, the ways in which "managers" and "agents" in a management architecture operate must consider how to operate with changes to topology and changes to service endpoints.¶
When connectivity to a manager cannot be guaranteed, agents must rely on locally available information and use local autonomy to react to changes at the node. Architectures that rely on external resources such as access to third-party oracles, operators-in-the-loop, or other service infrastructure may fail to operate in a challenged network.¶
In addition to disconnectivity, topological change can alter the associations amongst managed and managing devices. Different managing devices might be active in a network at different times or in different partitions. Managed devices might communicate with some, all, or none of these managing devices as a function of their own local configuration and policy.¶
Therefore, a network management architecture for challenged networks should:¶
The following special cases illustrate some of the operational situations that can be encountered in the management of devices in a challenged network.¶
These special cases highlight the need for managed devices to operate without presupposing a dedicated connection to a single managing device. To support this, managing devices must deliver instruction sets that govern the local, autonomous behavior of managed devices. These behaviors include (but are not limited to) collecting performance data, state, and error conditions, and applying pre-determined responses to pre-determined events. Managing devices in a challenged network might never expect a reply to a command, and communications from managed devices may be delivered much later than the events being reported.¶
This section describes those design properties that are desirable when defining a management architecture operating across challenged links in a network. These properties ensure that network management capabilities are retained even as delays and disruptions in the network scale. Ultimately, these properties are the driving design principles for the DTNMA.¶
The DTNMA should be agnostic of the underlying physical topology, transport protocols, security solutions, and supporting infrastructure of a given network. Due to the likelihood of operating in a frequently partitioned environment, the topology of a network may change over time. Attempts to stabilize an architecture around individual nodes can result in a brittle management framework and the creation of congestion points during periods of connectivity.¶
The DTNMA should not prescribe any association between a DM and a DA other than those defined in this document. There should be no logical limitation to the number of DMs that can control a DA, the number of DMs that a DA should report to, or any requirement that a DM and DA relationship implies a pair.¶
The DTNMA should use data models to define the syntactic and semantic contracts for data exchange between a DA and a DM. A given model should have the ability to "inherit" the contents of other models to form hierarchical data relationships.¶
Many network management solutions use data models to specify the semantic and syntactic representation of data exchanged between managed and managing devices. The DTNMA is not different in this regard - information exchanged between DAs and DMs should conform to one or more pre-defined, normative data models.¶
A common best practice when defining a data model is to make it cohesive. A cohesive model is one that includes information related to a single purpose such as managing a single application or protocol. When applying this practice, it is not uncommon to develop a large number of small data models that, together, describe the information needed to manage a device.¶
Another best practice for data model development is the use of inclusion mechanisms to allow one data model to include information from another data model. This ability to include a data model avoids repeating information in different data models. When one data model includes information from another data model, there is an implied model hierarchy.¶
Data models in the DTNMA should allow for the construction of both cohesive models and hierarchically related models. These data models should be used to define all sources of information that can be retrieved, configured, or executed in the DTNMA. This includes supporting DA autonomy functions such as parameterization, filtering, and event driven behaviors. These models will be used to both implement interoperable autonomy engines on DAs and define interoperable report parsing mechanisms on DMs.¶
DAs in the DTNMA architecture should determine when to push information to DMs as a function of their local state.¶
Pull management mechanisms require a managing device to send a query to a managed device and then wait for a response to that specific query. This practice implies some serialization mechanism (such as a control session) between entities. However, challenged networks cannot guarantee timely round-trip data exchange. For this reason, pull mechanisms must be avoided in the DTNMA.¶
Push mechanisms, in this context, refer to the ability of DAs to leverage local autonomy to determine when and what information should be sent to which DMs. The push is considered adaptive because a DA determines what information to push (and when) as an adaptation to changes to the DA's internal state. Once pushed, information might still be queued pending connectivity of the DA to the network.¶
Messages exchanged between a DA and a DM in the DTNMA should be defined in a way that allows for efficient on-the-wire encoding. DTNMA design decisions that result in smaller message sizes should be preferred over those that result in larger message sizes.¶
There is a relationship between message encoding and message processing time at a node. Messages with little or no encodings may simplify node processing whereas more compact encodings may require additional activities to generate/parse encoded messages. Generally, compressing a message takes processing time at the sender and decompressing a message takes processing time at a receiver. Therefore, there is a design tradeoff between minimizing message sizes and minimizing node processing.¶
There is no advantage to minimizing node processing time in a challenged network. The same sparse connectivity that benefits from store-and-forward transport provides time at a node for data processing prior to a future transmission opportunity.¶
However, there is a significant advantage to smaller message sizes in a challenged network. Smaller messages require smaller periods of viable transmission for communication, they incur less re-transmission cost, and they consume less resources when persistently stored en-route in the network.¶
Elements within the DTNMA should be uniquely identifiable so that they can be individually manipulated. Further, these identifiers should be universal - the identifier for a data element should be the same regardless of role, implementation, or network instance.¶
Identification schemes that are relative to a specific DA or specific system configuration might change over time. In particular, nodes in a challenged network may change their status or configuration during periods of partition from other parts of the network. Resynchronizing relative state or configuration should be avoided whenever possible.¶
The DTNMA should allow for the definition of new elements to a data model as part of the runtime operation of the management system. These definitions may represent custom data definitions that are applicable only for a particular device or network. Custom definitions should also be able to be removed from the system during runtime.¶
The custom definition of new data from existing data (such as through data fusion, averaging, sampling, or other mechanisms) provides the ability to communicate desired information in as compact a form as possible.¶
Custom data elements should be calculated and used both as parameters for DA autonomy and for more efficient reporting to DMs. Defining new data elements allows for DAs to perform local data fusion and defining new reporting templates allows for DMs to specify desired formats and generally save on link capacity, storage, and processing time.¶
The management of applications by a DA should be achievable using only knowledge local to the DA because DAs might need to operate during times when they are disconnected from a DM.¶
DA autonomy may be used for simple automation of predefined tasks or to support semi-autonomous behavior in determining when to run tasks and how to configure or parameterize tasks when they are run. In either case, a DA should provide the following features.¶
Features such as deterministic processing and engine-based behavior do not preclude the use of other Artificial Intelligence (AI) and Machine Learning (ML) approaches on a managed device.¶
Several network management solutions have been developed for both local-area and wide-area networks. Their capabilities range from simple configuration and report generation to complex modeling of device settings, state, and behavior. Each of these approaches are successful in the domains for which they have been built, but are not all equally functional when deployed in a challenged network.¶
Early network management tools designed for unchallenged networks provide synchronous mechanisms for communicating locally-collected data from devices to operators. Applications are managed using a "pull" mechanism, requiring a managing device to explicitly request the data to be produced and transmitted by a managed device.¶
More recent network management tools focus on message-based management, reduced state keeping by managed and managing devices, and increased levels of system autonomy.¶
This section describes some of the well-known, standardized protocols for network management and contrasts their purposes with the desirable properties of the DTNMA. The purpose of this comparison is to identify elements of existing approaches that can be adopted or adapted for use in challenged networks and where new elements must be created specifically for this environment.¶
The de facto example of a pull architecture is the Simple Network Management Protocol (SNMP) [RFC3410]. SNMP utilizes a request/response model to set and retrieve data values such as host identifiers, link utilization metrics, error rates, and counters between application software on managing and managed devices [RFC3411]. Data may be directly sampled or consolidated into representative statistics. Additionally, SNMP supports a model for unidirectional push notification messages, called event notifications, based on predefined triggering events.¶
SNMP managing devices can query agents for status information, send new configurations, and request to be informed when specific events have occurred. SNMP devices separate the representations for data modeling (Structure of Management Information (SMI) [RFC2578] and the Management Information Base (MIB) [RFC3418]) and messaging, sequencing and encoding (the SNMP protocol [RFC3411] [RFC3416]).¶
Separating data models from messaging and encoding is a best practice in subsequent management protocols and likely necessary for the DTNMA. In particular, SNMP MIBs provide well-organized, hierarchical Object Identifiers (OIDs) which support the compressibility necessary for challenged DTNs.¶
While there is a large installation base for SNMP, several aspects of the protocol make it inappropriate for use in a challenged network. SNMP relies on sessions with low round-trip latency to support its "pull" mechanism. Complex management can be achieved, but only through careful orchestration of real-time, end-to-end, managing-device-generated query-and-response logic.¶
There is existing work that uses the SNMP data model to support some low-fidelity Agent-side processing, to include the Distributed Management Expression MIB [RFC2982] and Definitions of Managed Objects for the Delegation of Management Scripts [RFC3165]. However, Agent autonomy is not an SNMP mechanism, so support for a local agent response to an initiating event is limited. In a challenged network where the delay between a managing device receiving an alert and sending a response can be significant, SNMP is insufficient for autonomous event handling.¶
Several network management protocols, including NETCONF [RFC6241], RESTCONF [RFC8040], and CORECONF [I-D.ietf-core-comi], share the same XML information set [xml-infoset] to describe the abstract data model necessary to manage the configuration of network devices. Each protocol, however, provides a different encoding of that XML information set.¶
YANG can be used to define the data model semantics and/or model syntax for the aforementioned network management protocols. YANG [RFC7950] is a data modeling language used to model the configuration and state data of managed devices and applications. A number of network management protocols have been developed around the definition, exchange, and reporting associated with YANG data models. Currently, YANG represents the standard for defining network management information.¶
The YANG model defines a schema for organizing and accessing a device's configuration or operational information. Once a model is developed, it is loaded to both the client and server, and serves as a contract between the two. A YANG model can be complex, describing many containers of managed elements, each providing methods for device configuration or reporting of operational state while differentiating implied and applied configuration [RFC8342].¶
The YANG module itself is a flexible data model that could be used for capturing the autonomy models and other behaviors needed by the DTNMA. The YANG schema provides flexibility in the organization of data to the model developer. The YANG schema supports a broad range of data types noted in [RFC6991]. YANG supports the definition of parameterized Remote Procedure Calls (RPCs) to be executed on managed nodes as well as the definition of push notifications within the model.¶
The YANG modeling language continues to evolve as new features are needed by adopting management protocols. Two evolving features that might be useful in the DTNMA are notifications and schema identifiers.¶
While the YANG model is currently the standard way to describe management data, there are concerns with its unmodified use in the DTNMA, as follows.¶
NETCONF [RFC6241], RESTCONF [RFC8040], and CORECONF [I-D.ietf-core-comi] each provide the mechanisms to install, manipulate, and delete the configuration of network devices. These network management protocols use the same XML information set, but provide different encodings of the abstract data model it describes.¶
NETCONF is a stateful, XML-based protocol that provides a RPC syntax to retrieve, edit, copy, or delete any data nodes or exposed functionality on a server. It requires that underlying transport protocols support long-lived, reliable, low-latency, sequenced data delivery sessions.¶
NETCONF connections are required to provide authentication, data integrity, confidentiality, and replay protection through secure transport protocols such as SSH or TLS. A bi-directional NETCONF session must be established before any data transfer can occur. All of these requirements make NETCONF a poor choice for operating in a challenged network.¶
RESTCONF is a stateless RESTful protocol based on HTTP. RESTCONF configures or retrieves individual data elements or containers within YANG data models by passing JSON over REST. This JSON encoding is used to GET, POST, PUT, PATCH, or DELETE data nodes within YANG modules.¶
RESTCONF is a stateless protocol because it presumes that it is running over a stateful secure transport (HTTP over TLS). Also, RESTCONF presumes that a single pull of information can be made in a single round-trip. In this way, RESTCONF is only stateless between queries - not internal to a single query.¶
CORECONF is an emerging stateless protocol built atop the Constrained Application Protocol (CoAP) [RFC7252] that defines a messaging construct developed to operate specifically on constrained devices and networks by limiting message size and fragmentation. CoAP also implements a request/response system and methods for GET, POST, PUT, and DELETE.¶
Currently, the CORECONF draft [I-D.ietf-core-comi] is archived and expired since 2021.¶
The future of network operations requires more autonomous behavior including self-configuration, self-management, self-healing, and self-optimization. One approach to support this is termed Autonomic Networking [RFC7575].¶
In particular, there is a large and growing set of work within the IETF focused on developing an Autonomic Networking Integrated Model and Approach (ANIMA). The ANIMA work has developed a comprehensive reference model for distributing autonomic functions across multiple nodes in an autonomic networking infrastructure [RFC8993].¶
This work, focused on learning the behavior of distributed systems to predict future events, is an exciting and emerging network management capability. This includes the development of signalling protocols such as GRASP [RFC8990] and autonomic control planes [RFC8368].¶
Both autonomic and challenged networks require similar degrees of autonomy. However, challenged networks cannot provide the complex coordination between nodes and distributed supporting infrastructure necessary for the frequent data exchanges for negotiation, learning, and bootstrapping associated with the above capabilities.¶
There is some emerging work in ANIMA as to how disconnected devices might join and leave the autonomic control plane over time. However, this work is solving an important, but different, problem than that encountered by challenged networks.¶
The future of network management will involve autonomous and autonomic functions operating on both managed and managing devices. However, the development of distributed autonomy for coordinated learning and event reaction is different from a managed device operating without connectivity to a managing node.¶
Management mechanisms that provide DTNMA desirable properties do not currently exist. This is not surprising since autonomous management in the context of a challenged networking environment is an emerging use case.¶
In particular, a management architecture is needed that provides the following new features.¶
Combining these new features with existing mechanisms for message data exchange (such as BP), data representations (such as CBOR) and data modeling languages (such as YANG) will form a pragmatic approach to defining challenged network management.¶
There are a multitude of ways in which both existing and emerging network management protocols, APIs, and applications can be integrated for use in challenged environments. However, expressing the needed behaviors of the DTNMA in the context of any of these pre-existing elements risks conflating systems requirements, operational assumptions, and implementation design constraints.¶
This section describes a network management concept for challenged networks (generally) and those conforming to the DTN architecture (in particular). The goal of this section is to describe how DTNMA services provide DTNMA desirable properties.¶
Similar to other network management architectures, the DTNMA draws a logical distinction between a managed device and a managing device. Managed devices use a DA to manage resident applications. Managing devices use a DM to both monitor and control DAs.¶
The DTNMA differs from some other management architectures in three significant ways, all related to the need for a device to self-manage when disconnected from a managing device.¶
A DTNMA reference model is provided in Figure 1 below. In this reference model, applications and services on a managing device communicate with a DM which uses pre-shared definitions to create a set of policy directives that can be sent to a managed device's DA via a command-based interface. The DA provides local monitoring and control (commanding) of the applications and services resident on the managed device. The DA also performs local data fusion as necessary to synthesize data products (such as reports) that can be sent back to the DM when appropriate.¶
DTNMA Reference Model¶
Managed Device Managing Device
+----------------------------+ +-----------------------------+
| +------------------------+ | | +-------------------------+ |
| |Applications & Services | | | | Applications & Services | |
| +----------^-------------+ | | +-----------^-------------+ |
| | | | | |
| +----------v-------------+ | | +-----------v-------------+ |
| | DTNMA +-------------+ | | | | +-----------+ DTNMA | |
| | AGENT | Monitor and | | |Commanding | | | Policy | MANAGER | |
| | | Control | | |<==========| | | Encoding | | |
| | +------+-------------+ | | | | +-----------+-------+ | |
| | |Admin | Data Fusion | | |==========>| | | Reporting | Admin | | |
| | +------+-------------+ | | Reporting | | +-----------+-------+ | |
| +------------------------+ | | +-------------------------+ |
+----------------------------+ +-----------------------------+
^ ^
| Pre-Shared Definitions |
| +---------------------------+ |
+--------| - Autonomy Model |--------+
| - Application Data Models |
| - Runtime Data Stores |
+---------------------------+
This model preserves the familiar concept of "managers" resident on managing devices and "agents" resident on managed devices. However, the DTNMA model is unique in how the DM and DA operate. The DM is used to pre-configure DAs in the network with management policies. it is expected that the DAs, themselves, perform monitoring and control functions on their own. In this way, a properly configured DA may operate without a reliable connection back to a DM.¶
The reference model illustrated in Figure 1 implies the existence of certain logical elements whose roles and responsibilities are discussed in this section.¶
By definition, managed applications and services reside on a managed device. These software entities can be controlled through some interface by the DA and their state can be sampled as part of periodic monitoring. It is presumed that the DA on the managed device has the proper data model, control interface, and permissions to alter the configuration and behavior of these software applications.¶
A DA resides on a managed device. As is the case with other network management approaches, this agent is responsible for the monitoring and control of the applications local to that device. Unlike other network management approaches, the agent accomplishes this task without a regular connection to a DTNMA Manager.¶
The DA performs three major functions on a managed device: the monitoring and control of local applications, production of data analytics, and the administrative control of the agent itself.¶
DAs monitor the status of applications running on their managed device and selectively control those applications as a function of that monitoring. The following components are used to perform monitoring and control on an agent.¶
DAs generate new data elements as a function of the current state of the managed device and its applications. These new data products may take the form of individual data values, or new collections of data used for reporting. The logical components responsible for these behaviors are as follows.¶
DAs must perform a variety of administrative services in support of their configuration. The significant such administrative services are as follows.¶
The DTNMA allows for a many-to-many relationship amongst DTNMA Agents and Managers. A single DM may configure multiple DAs, and a single DA may be configured by multiple DMs. Multiple managers may exist in a network for at least two reasons. First, different managers may exist to control different applications on a device. Second, multiple managers increase the likelihood of an agent encountering a manager when operating in a sparse or challenged environment.¶
While the need for multiple managers is required for operating in a dynamically partitioned network, this situation allows for the possibility of conflicting information from different managers. Implementations of the DTNMA should consider conflict resolution mechanisms. Such mechanisms might include analyzing managed content, time, agent location, or other relevant information to select one manager input over other manager inputs.¶
Managing applications and services reside on a managing device and serve as the both the source of DA policy statements and the target of DA reporting. They may operate with or without an operator in the loop.¶
Unlike management applications in unchallenged networks, these applications cannot exert closed-loop control over any managed device application. Instead, these applications must be built to exercise open-loop control by producing policies that can be configured and enforced on managed devices by DAs.¶
A DM resides on a managing device. This manager provides an interface between various managing applications and services and the DAs that enforce their policies. In providing this interface, DMs translate between whatever native interface exists to various managing applications and the autonomy models used to encode management policy.¶
The DM performs three major functions on a managing device: policy encoding, reporting, and administration.¶
DMs translate policy directives from managing applications and services into standardized policy expressions that can be recognized by DAs. The following logical components are used to perform this policy encoding.¶
DMs receive reports on the status of managed devices during periods of connectivity with the DAs on those devices. The following logical components are needed to implement reporting capabilities on a DM.¶
Managers in the DTNMA must perform a variety of administrative services in support of their proper configuration and operation. This includes the following logical components.¶
A consequence of operating in a challenged environment is the potential inability to negotiate information in real-time. For this reason, the DTNMA requires that managed and managing devices operate using pre-shared definitions rather than relying on data definition negotiation.¶
The three types of pre-shared definitions in the DTNMA are the DA autonomy model, managed application data models, and any runtime data shared by managers and agents.¶
A DTNMA autonomy model represents the data elements and associated autonomy structures that define the behavior of the agent autonomy engine. A standardized autonomy model allows for individual implementations of DAs, and DMs to interoperate. A standardized model also provides guidance to the design and implementation of both managed and managing applications.¶
This section provides a description of the services provided by DTNMA elements on both managing and managed devices. These service descriptions differ from other management descriptions because of the unique characteristics of the DTNMA operating environment.¶
DTNMA monitoring is associated with the agent autonomy engine. The term monitoring implies regular access to information such that state changes may be acted upon within some response time period. Within the DTNMA, connections between a managed and managing device are unable to provide such a connection and, thus, monitoring functions must be handled on the managed device.¶
Predicate autonomy on a managed device should collect state associated with the device at regular intervals and evaluate that collected state for any changes that require a preventative or corrective action. Similarly, this monitoring may cause the device to generate one or more reports destined to the managing device.¶
Similar to monitoring, DTNMA control results in actions by the agent to change the state or behavior of the managed device. All control in the DTNMA is local control. In cases where there exists a timely connection to a manager, received controls are still run through the autonomy engine. In this case, the stimulus is the direct receipt of the control and the response is to immediately run the control. In this way, there is never a dependency on a session or other stateful exchange with any remote entity.¶
DTNMA Fusion services produce new data products from existing state on the managed device. These fusion products can be anything from simple summations of sampled counters to complex calculations of behavior over time.¶
Fusion is an important service in the DTNMA because fusion products are part of the overall state of a managed device. Complete knowledge of this overall state is important for the management of the device, particularly in a stimulus-response system whose stimuli are evaluated against this state. In particular, the predicates of rules on a DA may refer to fused data.¶
In-situ data fusion is an important function as it allows for the construction of intermediate summary data, the reduction of stored and transmitted raw data, possibly fewer predicates in rule definitions, and otherwise insulates the data source from conclusions drawn from that data.¶
While some fusion is performed in any management system, the DTNMA requires fusion to occur on the managed device itself. If the network is partitioned such that no connection to a managing device is available, fusion must happen locally. Similarly, connections to a managing device might not remain active long enough for round-trip data exchange or may not have the bandwidth to send all sampled data.¶
DTNMA configuration services must update the local configuration of a managed device with the intent to impact the behavior and capabilities of that device. The change of device configurations is a common service provided by many network management systems. The DTNMA has a unique approach to configuration for the following reasons.¶
The DTNMA configuration service is unique in that the selection of managed device configurations must occur, itself, as a function of the state of the device. This implies that management proxies on the device store multiple configuration functions that can be applied as needed without consultation from a managing device.¶
When detecting stimuli, the agent autonomy engine must support a mechanism for evaluating whether application monitoring data or runtime data values are recent enough to indicate a change of state. In cases where data has not been updated recently, it may be considered stale and not used to reliably indicate that some stimulus has occurred.¶
DTNMA reporting services collect information known to the managed device and prepare it for eventual transmission to one or more managing devices. The contents of these reports, and the frequency at which they are generated, occurs as a function of the state of the managed device, independent of the managing device.¶
Once generated, it is expected that reports might be queued pending a connection back to a managing device. Therefore, reports must be differentiable as a function of the time they were generated.¶
When reports are sent to a managing device over a challenged network, they may arrive out of order due to taking different paths through the network or being delayed due to retransmissions. A managing device should not infer meaning from the order in which reports are received, nor should a given report be associated with a specific control or autonomy action on a given managed device.¶
Both local and remote services provided by the DTNMA affect the behavior of multiple applications on a managed device and may interface with multiple managing devices. It is expected that transport protocols used in any DTNMA implementation support security services such as integrity and confidentiality.¶
Authorization services enforce the potentially complex mapping of other DTNMA services amongst managed and managing devices in the network. For example, fine-grained access control can determine which managing devices receive which reports, and what controls can be used to alter which managed applications.¶
This is particularly beneficial in networks that either deal with multiple administrative entities or overlay networks that cross administrative boundaries. Allowlists, blocklists, key-based infrastructures, or other schemes may be used for this purpose.¶
An important characteristic of the DTNMA is the shift in the role of a managing device. In the DTNMA, managers configure the autonomy engines on agents, and it is the agents that provide local device management. One way to describe the behavior of the agent autonomy engine is to describe the characteristics of the autonomy model it implements.¶
This section describes a logical autonomy model in terms of the abstract data elements that would comprise the model. Defining abstract data elements allows for an unambiguous discussion of the behavior of an autonomy model without mandating a particular design, encoding, or transport associated with that model.¶
Managing autonomy on a potentially disconnected device must behave in both an expressive and deterministic way. Expressivity allows for the model to be configured for a wide range of future situations. Determinism allows for the forensic reconstruction of device behavior as part of debugging or recovery efforts.¶
The DTNMA autonomy model is a rule-based model in which individual rules associate a pre-identified stimulus with a pre-configured response to that stimulus.¶
Stimuli are identified using one or more predicate logic expressions that examine aspects of the state of the managed device. Responses are implemented by running one or more procedures on the managed device.¶
In its simplest form, a stimulus is a single predicate expression of a condition that examines some aspect of the state of the managed device. When the condition is met, a predetermined response is applied. This behavior can be captured using the construct:¶
IF <condition 1> THEN <response 1>;
¶
In more complex forms, a stimulus may include both a common condition shared by multiple rules and a specific condition for each individual rule. If the common condition is not met, the evaluation of the specific condition of each rule sharing the common condition can be skipped. In this way, the total number of predicate evaluations can be reduced. This behavior can be captured using the construct:¶
IF <common condition> THEN
IF <specific condition 1> THEN <response 1>
IF <specific condition 2> THEN <response 2>
IF <specific condition 3> THEN <response 3>
¶
DTNMA Autonomy Model¶
Managed Applications | DTNMA Agent | DTNMA Manager
+---------------------+--------------------------------+--------------+
| +---------+ |
| | Local | | Encoded
| | Rule DB |<-------------------- Policy
| +---------+ | Expressions
| ^ |
| | |
| v |
| +----------+ +---------+ |
Monitoring Data------>| Agent | | Runtime | |
| | Autonomy |<-->| Data |<---- Definitions
Application Control<------| Engine | | Store | |
| +----------+ +---------+ |
| | |
| +-------------------------> Reports
| |
The flow of data into and out of the agent autonomy engine is illustrated in Figure 2. In this model, the autonomy engine stores the combination of stimulus conditions and associated responses as a set of "rules" in a rules database. This database is updated through the execution of the autonomy engine and as configured from policy statements received by managers.¶
Stimuli are detected by examining the state of applications as reported through application monitoring interfaces and through any locally-derived data. Local data is calculated in accordance with definitions also provided by managers as part of the runtime data store.¶
Responses to stimuli are run as updated to the rules database, updated to the runtime data store, controls sent to applications, and the generation of reports.¶
There are several practical challenges to the implementation of a distributed rule-based system. Large numbers of rules may be difficult to understand, deconflict, and debug. Rules whose conditions are given by fused or other dynamic data may require data logging and reporting for deterministic offline analysis. Rule differences across managed devices may lead to oscillating effects. This section identifies those characteristics of an autonomy model that might help implementations mitigate some of these challenges.¶
There are a number of ways to represent data values, and many data modeling languages exist for this purpose. When considering how to model data in the context of the DTNMA autonomy model there are some modeling features that should be present to enable functionality. There are also some modeling features that should be prevented to avoid ambiguity.¶
Traditional network management approaches favor flexibility in their data models. The DTNMA stresses deterministic behavior that supports forensic analysis of agent activities "after the fact". As such, the following statements should be true of all data representations relating to DTNMA autonomy.¶
The expressive representation of simple data values is fundamental to the successful construction and evaluation of predicates in the DTNMA autonomy model. When defining such values, there are useful distinctions regarding how values are identified and whether values are generated internal or external to the autonomy model.¶
A DTNMA data value should combine a base type (e.g., integer, real, string) representation with relevant semantic information. Base types are used for proper storage and encoding. Semantic information allows for additional typing, constraint definitions, and mnemonic naming. This expanded definition of data value allows for better predicate construction and evaluation, early type checking, and other uses.¶
Data values may further be annotated based on whether their value is the result of a DA calculation or the result of some external process on the managed device. For example, operators may with to know which values can be updated by actions on the DA versus which values (such as sensor readings) cannot be reliably changed because they are calculated external to the DA.¶
The DTNMA autonomy model should, as required, report on the state of its managed device (to include the state of the model itself). This reporting should be done as a function of the changing state of the managed device, independent of the connection to any managing device. Queuing reports allows for later forensic analysis of device behavior, which is a desirable property of DTNMA management.¶
DTNMA data reporting consists of the production of some data report instance conforming to a data report schema. The use of schemas allows a report instance to identify the schema to which it conforms in lieu of carry that structure in the instance itself. This approach can significantly reduce the size of generated reports.¶
The agent autonomy engine requires that managed devices issue commands on themselves as if they were otherwise being controlled by a managing device. The DTNMA implements commanding through the use of controls and macros.¶
Controls represent parameterized, predefined procedures run by the DA either as directed by the DM or as part of a rule response from the DA autonomy engine. Controls are conceptually similar to RPCs in that they represent parameterized functions run on the managed device. However, they are conceptually dissimilar from RPCs in that they do not have a concept of a return code as they must operate over an asynchronous transport. The concept of return code in an RPC implies a synchronous relationship between the caller of the procedure and the procedure being called, which might not be possible within the DTNMA.¶
The success or failure of a control may be handled locally by the agent autonomy engine. Otherwise, the externally observable impact of a control can be understood through the generation and eventual examination of data reports produced by the managed device.¶
Macros represent ordered sequences of controls.¶
As discussed in Section 9.1, the DTNMA rule-based stimulus-response system associates stimulus detection with a predetermined response. Rules may be categorized based on whether their stimuli include generic statements of managed device state or whether they are optimized to only consider the passage of time on the device.¶
State-based rules are those whose stimulus is based on the evaluated state of the managed device. Time-based rules are a unique subset of state-based rules whose stimulus is given only by a time-based event. Implementations might create different structures and evaluation mechanisms for these two different types of rules to achieve more efficient processing on a platform.¶
Using the autonomy model mnemonics defined in Section 9, this section describes flows through sample configurations conforming to the DTNMA. These use cases illustrate remote configuration, local monitoring and control, multiple manager support, and data fusion.¶
The use cases presented in this section are documented with a shorthand notation to describe the types of data sent between managers and agents. This notation, outlined in Table 1, leverages the mnemonic definitions of autonomy model elements defined in Section 9.¶
| Term | Definition | Example |
|---|---|---|
| EDD# | Externally Defined Data - a data value defined external to the DA. | EDD1, EDD2 |
| V# | Variable - a data value defined internal to the DA. | V1 = EDD1 + 7 |
| EXPR | Predicate expression - used to define a rule stimulus. | V1 > 5 |
| ID | DTNMA Object Identifier. | V1, EDD2 |
| ACL# | Enumerated Access Control List. | ACL1 |
| DEF(ACL,ID,EXPR) | Define ID from expression. Allow managers in ACL to see this ID. | DEF(ACL1, V1, EDD1 + EDD2) |
| PROD(P,ID) | Produce ID according to predicate P. P may be a time period (1s) or an expression (EDD1 > 10). | PROD(1s, EDD1) |
| RPT(ID) | A report instance containing data named ID. | RPT(EDD1) |
These notations do not imply any implementation approach. They only provide a succinct syntax for expressing the data flows in the use case diagrams in the remainder of this section.¶
This nominal configuration shows a single DM interacting with multiple DAs. The control flows for this scenario are outlined in Figure 3.¶
Serialized Management Control Flow¶
+-----------+ +---------+ +---------+
| DTNMA | | DTNMA | | DTNMA |
| Manager A | | Agent A | | Agent B |
+----+------+ +----+----+ +----+----+
| | |
|-----PROD(1s, EDD1)--->| | (1)
|----------------------------PROD(1s, EDD1)-->|
| | |
| | |
|<-------RPT(EDD1)------| | (2)
|<----------------------------RPT(EDD1)-------|
| | |
| | |
|<-------RPT(EDD1)------| |
|<----------------------------RPT(EDD1)-------|
| | |
| | |
|<-------RPT(EDD1)------| |
|<----------------------------RPT(EDD1)-------|
| | |
In a serialized management scenario, a single DM interacts with multiple DAs.¶
In this figure, the DTNMA Manager A sends a policy to DTNMA Agents A and B to report the value of an EDD (EDD1) every second in (step 1). Each DA receives this policy and configures their respective autonomy engines for this production. Thereafter, (step 2) each DA produces a report containing data element EDD1 and sends those reports back to the DM.¶
This behavior continues without any additional communications from the DM and without requiring a connection between the DA and DM.¶
Building from the nominal configuration in Section 10.2, this scenario shows a challenged network in which connectivity between DTNMA Agent B and the DM is temporarily lost. Control flows for this case are outlined in Figure 4.¶
Challenged Management Control Flow¶
+-----------+ +---------+ +---------+
| DTNMA | | DTNMA | | DTNMA |
| Manager A | | Agent A | | Agent B |
+----+------+ +----+----+ +----+----+
| | |
|-----PROD(1s, EDD1)--->| | (1)
|----------------------------PROD(1s, EDD1)-->|
| | |
| | |
|<-------RPT(EDD1)------| | (2)
|<----------------------------RPT(EDD1)-------|
| | |
| | |
|<-------RPT(EDD1)------| |
|<----------------------------RPT(EDD1)-------|
| | |
| | |
|<-------RPT(EDD1)------| |
| | RPT(EDD1)| (3)
| | |
| | |
|<-------RPT(EDD1)------| |
| | RPT(EDD1)| (4)
| | |
| | |
|<-------RPT(EDD1)------| |
|<----------------RPT(EDD1), RPT(EDD1)--------| (5)
| | |
In a challenged network, DAs store reports pending a transmit opportunity.¶
In this figure, DTNMA Manager A sends a policy to DTNMA Agents A and B to produce an EDD (EDD1) every second in (step 1). Each DA receives this policy and configures their respective autonomy engines for this production. Produced reports are transmitted when there is connectivity between the DA and DM (step 2).¶
At some point, DTNMA Agent B loses the ability to transmit in the network (steps 3 and 4). During this time period, DA B continues to produce reports, but they are queued for transmission. This queuing might be done by the DA itself or by a supporting transport such as BP. Eventually, DTNMA Agent B is able to transmit in the network again (step 5) and all queued reports are sent at that time. DTNMA Agent A maintains connectivity with the DM during steps 3-5, and continues to send reports as they are generated.¶
This scenario illustrates the DTNMA open-loop control paradigm, where DAs manage themselves in accordance with policies provided by DMs, and provide reports to DMs based on these policies.¶
The control flow shown in Figure 5, includes an example of data fusion, where multiple policies configured by a DM result in a single report from a DA.¶
Consolidated Management Control Flow¶
+-----------+ +---------+ +---------+
| DTNMA | | DTNMA | | DTNMA |
| Manager A | | Agent A | | Agent B |
+----+------+ +----+----+ +----+----+
| | |
|-----PROD(1s, EDD1)--->| | (1)
|----------------------------PROD(1s, EDD1)-->|
| | |
| | |
|<-------RPT(EDD1)------| | (2)
|<----------------------------RPT(EDD1)-------|
| | |
| | |
|----------------------------PROD(1s, EDD2)-->| (3)
| | |
| | |
|<-------RPT(EDD1)------| |
|<--------------------------RPT(EDD1,EDD2)----| (4)
| | |
| | |
|<-------RPT(EDD1)------| |
|<--------------------------RPT(EDD1,EDD2)----|
| | |
A many-to-one mapping between management policy and device state reporting is supported by the DTNMA.¶
In this figure, DTNMA Manager A sends a policy statement in the form of a rule to DTNMA Agents A and B, which instructs the DAs to produce a report with EDD1 every second (step 1). Each DA receives this policy, which is stored in its respective Rule Database, and configures its Autonomy Engine. Reports are transmitted by each DA when produced (step 2).¶
At a later time, DTNMA Manager A sends an additional policy to DTNMA Agent B, requesting the production of a report for EDD2 every second (step 3). This policy is added to DTNMA Agent B's Rule Database.¶
Following this policy update, DTNMA Agent A will continue to produce EDD1 and DTNMA Agent B will produce both EDD1 and EDD2 (step 4). However, DTNMA Agent B may provide these values to the DM in a single report rather than as 2 independent reports. In this way, there is no direct mapping between the single consolidated report sent by DTNMA Agent B (step 4) and the two different policies sent to DTNMA Agent B that caused that report to be generated (steps 1 and 3).¶
The managed applications on a DA may be controlled by different administrative entities in a network. The DTNMA allows DAs to communicate with multiple DMs in the network, such as in cases where there is one DM per administrative domain.¶
Whenever a DM sends a policy expression to a DA, that policy expression may be associated with authorization information. One method of representing this is an ACL.¶
The ability of one DM to access the results of policy expressions configured by some other DM will be limited to the authorization annotations of those policy expressions.¶
An example of multi-manager authorization is illustrated in Figure 6.¶
Multiplexed Management Control Flow¶
+-----------+ +---------+ +-----------+
| DTNMA | | DTNMA | | DTNMA |
| Manager A | | Agent A | | Manager B |
+-----+-----+ +----+----+ +-----+-----+
| | |
|---DEF(ACL1,V1,EDD1*2)--->|<---DEF(ACL2, V2, EDD2*2)---| (1)
| | |
|---PROD(1s, V1)---------->|<---PROD(1s, V2)------------| (2)
| | |
|<--------RPT(V1)----------| | (3)
| |--------RPT(V2)------------>|
|<--------RPT(V1)----------| |
| |--------RPT(V2)------------>|
| | |
| |<---PROD(1s, V1)------------| (4)
| | |
| |----ERR(V1 no perm.)------->|
| | |
|--DEF(NULL,V3,EDD3*3)---->| | (5)
| | |
|---PROD(1s, V3)---------->| | (6)
| | |
| |<----PROD(1s, V3)-----------|
| | |
|<--------RPT(V3)----------|--------RPT(V3)------------>| (7)
|<--------RPT(V1)----------| |
| |--------RPT(V2)------------>|
|<-------RPT(V3)-----------|--------RPT(V3)------------>|
|<-------RPT(V1)-----------| |
| |--------RPT(V2)------------>|
Multiple DMs may interface with a single DA, particularly in complex networks.¶
In this figure, both DTNMA Managers A and B send policies to DTNMA Agent A (step 1). DM A defines a variable (V1) whose value is given by the mathematical expression (EDD1 * 2) and is associated with an ACL (ACL1) that restricts access to V1 to DM A only. Similarly, DM B defines a variable (V2) whose value is given by the mathematical expression (EDD2 * 2) and associated with an ACL (ACL2) that restricts access to V2 to DM B only.¶
Both DTNMA Managers A and B also send policies to DTNMA Agent A to report on the values of their variables at 1 second intervals (step 2). Since DM A can access V1 and DM B can access V2, there is no authorization issue with these policies and they are both accepted by the autonomy engine on Agent A. Agent A produces reports as expected, sending them to their respective managers (step 3).¶
Later (step 4) DM B attempts to configure DA A to also report to it the value of V1. Since DM B does not have authorization to view this variable, DA A does not include this in the configuration of its autonomy engine and, instead, some indication of permission error is included in any regular reporting back to DM B.¶
DM A also sends a policy to Agent A (step 5) that defines a variable (V3) whose value is given by the mathematical expression (EDD3 * 3) and is not associated with an ACL, indicating that any DM can access V3. In this instance, both DM A and DM B can then send policies to DA A to report the value of V3 (step 6). Since there is no authorization restriction on V3, these policies are accepted by the autonomy engine on Agent A and reports are sent to both DM A and B over time (step 7).¶
There are times where a single network device may serve as both a DM for other DAs in the network and, itself, as a device managed by someone else. This may be the case on nodes serving as gateways or proxies. The DTNMA accommodates this case by allowing a single device to run both a DA and DM.¶
An example of this configuration is illustrated in Figure 7.¶
Data Fusion Control Flow¶
---------------------------------------
| Node B |
| |
+-----------+ | +-----------+ +---------+ | +---------+
| DTNMA | | | DTNMA | | DTNMA | | | DTNMA |
| Manager A | | | Manager B | | Agent B | | | Agent C |
+---+-------+ | +-----+-----+ +----+----+ | +----+----+
| | | | | |
|---------------DEF(NULL,V0,EDD1+EDD2)->| | | (1)
|------------------PROD(EDD1&EDD2,V0)-->| | |
| | | | | |
| | | | | |
| | |--------------------PROD(1s, EDD2)->| (2)
| | | | | |
| | | | | |
| | |<--------------------RPT(EDD2)------| (3)
| | | | | |
|<------------------RPT(V0)-------------| | | (4)
| | | | | |
| | | | | |
| |
| |
---------------------------------------
A device can operate as both a DTNMA Manager and an Agent.¶
In this example, we presume that DA B is able to sample a given EDD (EDD1) and that DA C is able to sample a different EDD (EDD2). Node B houses DM B (which controls DA C) and DA B (which is controlled by DM A). DM A must periodically receive some new value that is calculated as a function of both EDD1 and EDD2.¶
First, DM A sends a policy to DA B to define a variable (V0) whose value is given by the mathematical expression (EDD1 + EDD2) without a restricting ACL. Further, DM A sends a policy to DA B to report on the value of V0 every second (step 1).¶
DA B can require the ability to monitor both EDD1 and EDD2. However, the only way to receive EDD2 values is to have them reported back to Node B by DA C and included in the Node B runtime data stores. Therefore, DM B sends a policy to DA C to report on the value of EDD2 (step 2).¶
DA C receives the policy in its autonomy engine and produces reports on the value of EDD2 every second (step 3).¶
DA B may locally sample EDD1 and EDD2 and uses that to compute values of V0 and report on those values at regular intervals to DM A (step 4).¶
While a trivial example, the mechanism of associating fusion with the Agent function rather than the Manager function scales with fusion complexity. Within the DTNMA, DAs and DMs are not required to be separate software implementations. There may be a single software application running on Node B implementing both DM B and DA B roles.¶
This informational document requires no registrations to be managed by IANA.¶
Security within a DTNMA MUST exist in at least two layers: security in the data model and security in the messaging and encoding of the data model.¶
Data model security refers to the confidentiality of elements of a data model and the authorization of DTNMA actors to access those elements. For example, elements of a data model might be available to certain DAs or DMs in a system, whereas the same elements may be hidden from other DAs or DMs.¶
The exchange of information between and amongst DAs and DMs in the DTNMA is expected to be accomplished through some messaging transport. As such, security at the transport layer is expected to address the questions of authentication, integrity, and confidentiality.¶
Brian Sipos of the Johns Hopkins University Applied Physics Laboratory (JHU/APL) provided excellent technical review of the DTNMA concepts presented in this document.¶