Internet-Draft DTNMA AMM July 2024
Birrane, et al. Expires 22 January 2025 [Page]
Workgroup:
Delay-Tolerant Networking
Internet-Draft:
draft-ietf-dtn-amm-01
Published:
Intended Status:
Standards Track
Expires:
Authors:
E.J. Birrane
JHU/APL
B. Sipos
JHU/APL
J. Ethier
JHU/APL

DTNMA Application Management Model (AMM) and Data Models

Abstract

This document defines a data model that captures the information necessary to asynchronously manage applications within the Delay-Tolerant Networking Management Architecture (DTNMA). This model provides a set of common type definitions, data structures, and a template for publishing standardized representations of model elements.

Status of This Memo

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 22 January 2025.

Table of Contents

1. Introduction

The Delay-Tolerant Networking Management Architecture (DTNMA) [I-D.ietf-dtn-dtnma] defines a concept for the open-loop control of applications (and protocols) in situations where timely, highly-available connections cannot exist among managing and managed nodes in a network. While the DTNMA provides a conceptual information model, it does not include details necessary to produce interoperable data models.

1.1. Scope

This document defines a two-level data model suitable for managing applications in accordance with the DTNMA. The two levels of model are:

  1. A meta-model for the DTNMA, called the Application Management Model (AMM), which defines the object structures and literal-value types used in the DTNMA in a concrete way.
  2. An object model, based on the AMM meta-model, which is used in static Application Data Models (ADMs) and dynamic Operational Data Models (ODMs) as instances of the AMM within an Agent.

This document does not define any specific encodings of AMM values or of ADM or ODM contents. In order to communicate data models and values between DTNMA Agents and Managers in a network, they must be encoded for transmission. Any such encoding scheme is outside of the scope of this document. Generally, the encoding of the model is a separate concern from the specification of data within the model.

Because different networks may use different encodings for data, mandating an encoding format would require incompatible networks to encapsulate data in ways that could introduce inefficiency and obfuscation. It is envisioned that different networks would be able to encode values in their native encodings such that the translation of ADM data from one encoding to another can be completed using mechanical action taken at network borders.

Since the specification does not mandate an encoding format, the AMM and ADM must provide enough information to make encoding (and translating from one encoding to another) an unambiguous process. Therefore, where necessary, this document provides identification, enumeration and other schemes that ensure ADMs contain enough information to prevent ambiguities caused by different encoding schemes.

1.2. Terminology

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 terms "Actor", "Agent", "Externally Defined Data", "Variable", "Constant", Control", "Literal", "Macro", "Manager", "Operator", "Report", "Report Template", "Rule", "State-Based Rule", "Table", "Table Template", and "Time-Based Rule" are used without modification from the definitions provided in [I-D.ietf-dtn-dtnma].

Additional terms defined in this document are as follows.

Application:
A software implementation running on an Agent and being managed by a Manager. This includes software that implements protocol processing on an Agent.
Application Management Model (AMM):
The object and literal-value model defined by this document in Section 3 and implemented as instances in ADMs and ODMs.
Application Resource Identifier (ARI):
A unique identifier for any AMM object, namespace, or literal value. The text form of an ARI is conformant to the Uniform Resource Identifier (URI) syntax documented in [RFC3986] and using the scheme name "ari".
Application Data Model (ADM):
The set of statically-defined objects necessary to manage an application asynchronously. This is also the name for the specific syntax used to express the contents of that ADM, as defined in another document.
Operational Data Model (ODM):
The set of dynamically-defined objects created and controlled by Managers in the network. There is currently no specific syntax used to express the entire contents of an ODM outside of data from introspection reports generated by an Agent.
Namespace:
Each ADM and ODM has a universally unique identifier and acts as a namespace for a set of AMM objects.

2. Data Modeling Concept of Operations

In order to asynchronously manage an application in accordance with the [I-D.ietf-dtn-dtnma], an application-specific data model must be created containing static structure for that application. This model is termed the Application Data Model (ADM) and forms the core set of information for that application in whichever network it is deployed. ADM structure and base ADMs are discussed in detail in Section 4.

The objects codified in the ADM represents static configurations and definitions that apply to any deployment of the application, regardless of the network in which it is operating. Within any given network, Managers supplement the information provided by ADMs with dynamic objects. Each namespace of dynamic objects is termed an Operational Data Model (ODM) and is discussed in detail in Section 5.

Both the ADMs and ODMs rely on a common meta-model, the Application Management Model of Section 3, which defines the basic structure of what kinds of types and objects are available to use within the DTNMA. The relationships among the AMM, ADM, and ODM are illustrated in Figure 1. Together, the set of objects in the union of all supported ADMs with dynamic ODM objects forms the data model used to manage an Agent.

      +---------------+
      |      AMM      |
      | (object types)|
      +-------^-------+
              |
      +-------+-------+
      |               |
+-----+-----+   +-----+-----+
|    ADM    |   |    ODM    |
|(instances)|   |(instances)|
+-----------+   +-----------+
Figure 1: Data Model Relationships

The AMM defines a strict separation between long-lived object instances and ephemeral value instances. While an Agent hosts the object instances, each manager must contain the corresponding ADM and ODM definitions in order to identify and interact with those objects. Those interactions are performed using Application Resource Identifiers (ARIs) as depicted in Figure 2.

       +-------------+            +-------------+
       |  AMM Value  |            |  AMM Object |
       |  (in ARI)   |            |  (in model) |
       +------^------+            +-------------+
              |                         ^
              | subtypes                |
      +-------+-------+                 |
      |               |                 | referenced
+-----+-----+   +-----+-----+           | object
|  Literal  |   |   Object  |-----------+
|           |   | Reference |
+-----------+   +-----------+
Figure 2: AMM Value and Object Relationships

While an agent hosts the actual object instances, each manager must contain the corresponding ADM and ODM definitions in order to identify and interact with those objects. Those interactions are performed using Application Resource Identifiers (ARIs) as depicted in Figure 3.

+------------------------------------------+
|                                          |
|  +-------+  +-------+         +-------+  |
|  | ADM 1 |  | ADM 2 |         | ODM 1 |  |
|  | def'n |  | def'n |         | def'n |  |'
|  +-------+  +-------+         +-------+  |
|           ...                    ...     |
|        +-------+              +-------+  |
|        | ADM N |              | ODM M |  |
|        | def'n |              | def'n |  |
|        +-------+              +-------+  |
|            Manager instance              |
+------------------------------------------+
                     ^^
                     ||
                 ARI values
                     ||
                     vv
+------------------------------------------+
|                                          |
|  +-------+  +-------+         +-------+  |
|  | ADM 1 |  | ADM 2 |         | ODM 1 |  |
|  | objs  |  | objs  |         | objs  |  |
|  +-------+  +-------+         +-------+  |
|           ...                    ...     |
|        +-------+              +-------+  |
|        | ADM N |              | ODM M |  |
|        | objs  |              | objs  |  |
|        +-------+              +-------+  |
|              Agent instance              |
+------------------------------------------+
Figure 3: Agent and Manager Interaction

2.1. Values and Value-Producing Objects

The ARI of [I-D.ietf-dtn-ari] is used as the basis for the values used internally for Agent Processing activities and for the basis of Agent-Manager Messaging contents.

Of the value-producing object types discussed in Section 3.4 and Section 6.5, the functions of these objects are summarized and compared with literals in Table 1 and the following list. In that table, "internal" means values are managed by the Agent itself and "external" means the source of values is outside the Agent.

Literal Values:

ARI literals are, by definition, immutable and fully self-contained values.

For example, the number 4 is a literal value. The name "4" and the value 4 represent the same thing and are inseparable. Literal values cannot change ("4" could not be used to mean 5) and they are defined external to the autonomy model (the autonomy model is not expected to redefine what 4 means).

Constant (CONST):

These objects are named values which are defined in specific revisions of an ADM and produced directly by the Agent implementing the ADM. Both the name and the value of the constant are fixed and cannot be changed (within a revision).

An example of a constant would be defining the numerical value pi to some predetermined precision.

Variable (VAR):

These objects are named value storage entities which are defined in ADMs or ODMs and managed by the Agent implementing the ADM or ODM. While the name is constant the value can change over time due to controls acting upon the Agent. One standard interface is an ADM-defined initial state expression with a control available to reset to that initial state (which can itself reference other value producing objects and operators). Another standard interface is a control to set a variable to a specific value.

An example of a variable using just its initial expression would be an accumulator summing together a list of counter values produced by other objects. An example of a manager-controlled variable would be a threshold value used to compare against a sensor value in a rule predicate.

Externally Defined Data (EDD):

These objects are named entities which are defined in an ADM but produce values based on data provided to an Agent from its environment. These values are the foundation of state-based autonomy as they capture the state of the managed device. The autonomy model treats these values as read-only state. It is an implementation matter to determine how external data is transformed into values of the specific type specified for an EDD.

Examples of externally defined values include temperature sensor readings and the instantaneous data rate from a modem or radio.

Table 1: Value-Producing Object Types
Immutable Mutable
Internal CONST VAR
External Literal EDD

2.2. Agent Processing

Based on the reasoning described in [I-D.ietf-dtn-dtnma], much of the closed-loop processing of the state of the DTNMA Agent is performed on the Agent itself using rule objects. The different types of processing performed on the Agent are separated into Execution, Evaluation, and Reporting with corresponding AMM object types related to each of these as indicated in Table 2 (e.g., execution relates to CTRL objects but not OPER objects). Some of the objects defined in the Agent ADM (Section 4.3) combine the use of these processing activities, but they are still independent of each other. There is no mixing of activities such as executing a control within an expression; although the execution of a control can result in an expression being evaluated they are independent activities.

Within the runtime of an Agent any input, output, and intermediate values can use the concept of a semantic type (Section 3.3) to do things like restrict the valid numeric range of a value or allow a union of disjoint types to be present (e.g., a certain value can be a boolean or an unsigned integer). This is combined with the built-in types available (see Section 3.2) to allow complex type information to be present in an ADM or ODM without requiring additional over-the-wire encoding size. A Type Conversion activity is defined for when implicit or explicit type conversion is needed.

Table 2: Processing Activities and Object Types
Activity Objects Values
Execution CTRL MAC
Evaluation OPER, TYPEDEF EXPR
Reporting N/A RPTT
Value Production CONST, EDD, VAR N/A
Type Casting TYPEDEF N/A
Rule Autonomy SBR, TBR N/A

2.3. Agent-Manager Messaging

This document does not define a messaging protocol between agents and managers but full functioning of this data model behavior does rely on the following types of messages being available. Because each message is based on ARI value types, they can be implemented in Agent and Manager by encoding the associated ARI according to a network-specific transport profile. The choice of encoding form, framing, or transport are implementation matters outside of this specific document.

Execution:
This message causes an Execution of a referenced parameterized CTRL object (Section 3.4.6) or MAC value (Section 4.2.3.2). The form of this message is an Execution-Set (EXECSET) value. This type of message is only sent from Manager to Agent. Each message can contain multiple execution targets but all must be associated with the same nonce value. It is an implementation detail whether a Manager sends fewer messages with more targets or more timely messages with fewer targets.
Reporting:
This message carries the reports generated by Reporting activities and as the result of Execution when the Manager provides a correlator nonce. The form of this message is an Reporting-Set (RPTSET) value. This type of message is only sent from Agent to Manager. Each message can contain multiple report containers but all must be associated with the same nonce value. It is an implementation detail whether an Agent sends fewer messages with more reports or more timely messages with fewer reports.

The contents of these messages, individual fields, are representable by ARI values so require only small additional message-type identifying and framing overhead to bind to whatever transport is being used (e.g. the Bundle Protocol).

3. Application Management Model (AMM)

This section describes the Application Management Model, which is the meta-model used to implement the DTNMA. This section also provides additional information necessary to work with this model, such as: literal value types, object structure, naming conventions, and processing semantics.

The overall AMM is decomposed into two categories:

Objects:
These are the structural and behavioral elements of the AMM, present in ADMs and ODMs. Objects implement the actual purpose of the applications being managed; they extract values from the Agent's environment, operate on expressions, store variables, and execute controls to affect the Agent's environment. Because objects are behavioral they have no complete static representation, objects are only ever described and identified. AMM object types are defined in Section 3.4 and objects are instantiated as part of an ADM or ODM.
Values:
These are the runtime state and intermediates of the AMM, present in the Agent's state but not directly in ADMs or ODMs. Objects produce, operate on, and store or consume values and these values are what are present in messaging between Managers and Agents. AMM values are explained in more detail in Section 3.1. One subset of AMM values are object references used to identify (and parameterize) individual AMM objects.

3.1. AMM Values

Values within the AMM have two top-level classes: literal values, and object reference values. Each of these is discussed more detail in the following subsections. Both classes of AMM values are related to what can be represented externally as an ARI, as described in Section 3.1.3.

3.1.1. Literal Values

Literal values are those whose value and identifier are equivalent. These are the most simple values in the AMM. For example, the literal "4" serves as both an identifier and a value.

Because the Literal value serves as its own identifier, there is no concept of a parent namespace or parameters. A literal can be completely identified by its data type and data value.

Literals have two layers of typing:

Built-In Type:
This is the lower layer which defines the syntax of the literal value and bounds the domain of the value (e.g. a BOOL has two possible values, while an INT has a large domain of integers). There are a small number of built-in types as described in Section 3.2 and they are managed with an IANA registry defined in Section 9.3 of [I-D.ietf-dtn-ari].
Semantic Type:
This is the higher layer which defines additional semantics to a value, such as a restricted domain of values or a human-friendly text unit or limits on the items of an ARI collection. Semantic types are defined within ADMs (see Section 3.3 and Section 3.4.2), so there can be an arbitrary number of them and they are managed outside of a central authority.

All literal values have a concrete and stand-alone representation independent of any ADM or ODM behavior in the form of an ARI (Section 3.1.3) and a value itself has no direct association with a semantic type outside of a specific context in which that type is used (e.g., Section 6.10 and Section 6.11).

3.1.2. Object Reference Values

Every object in the AMM is uniquely identifiable, regardless of whether the item is defined statically in an ADM or dynamically in an ODM. Object reference values are composed of four parts: a namespace, an object type, an object name, and object-specific optional parameters.

AMM objects are identified within unique namespaces to prevent conflicting names within network deployments, particularly in cases where network operators are allowed to define their own object names. In this capacity, namespaces exists to eliminate the chance of a conflicting object name. They MUST NOT be used as a security mechanism. An Agent or Manager MUST NOT infer security information or access control based solely on namespace information.

Two categories of namespaces available within the object reference value and the ARI syntax:

ADM Namespace:
These are defined, with text-name and enumeration, in an IANA registry of ADMs in Section 9.3 of [I-D.ietf-dtn-ari]. There is also a reservation of private-use code points for domain- and mission-specific ADMs. In ARI form, ADM namespaces are present as either their name or enumeration directly.
ODM Namespace:
These are defined, with text-name and enumeration, in an IANA registry of ODMs in Section 9.3 of [I-D.ietf-dtn-ari]. It is expected that most ODM use will be domain- and mission-specific. In ARI form, ODM namespaces are present as a "!" prefixed name or as a negative-value enumeration.

Object types, each with a text-name and enumeration, are defined in an IANA registry by Section 9.3 of [I-D.ietf-dtn-ari].

Object names are text strings and enumerations whose value is determined by the creator of the object. For those objects defined in an ADM, the structure of the object name is given in Section 3.4.1.

3.1.2.1. Parameters

Parameterization is used in the AMM to enable expressive autonomous function and reduce the amount of traffic communicated between Managers and Agents. In the AMM, most objects can be parameterized and the meaning of parameterization for each object type is defined in Section 3.4 with behaviors related to parameters defined in Section 6.

There are three notions of parameters defined in the AMM, which take their name from computer programming vernacular used for discussing function declarations and function calls, those are: formal parameters, given parameters, and actual parameters. Formal parameters are discussed in Section 3.4.1 while given and actual parameters are discussed here in relation to the object reference.

Given parameters represent the data values passed to a parameterized AMM Object at runtime. They "fulfill" the parameter requirements defined by the formal parameters for that object. Each object type can have a slightly different notion of how its parameters affect its processing activities.

A given parameter MUST include a value and MAY include a type. If a type is provided it MUST be consistent with the type provided by the corresponding formal parameter.

There are two ways in which the value of an given parameter can be used:

Parameter by Value:
This method involves directly supplying the value as part of the given parameter. It is the most direct method for supplying values.
Parameter by Label:
This method involves supplying the LABEL of some other processing-context-specific value and substituting, at runtime, that named value as the value of this parameter. This method is useful when a parameterized AMM Object produces a value that references a parameter of the producing object. The produced value's given parameter can be given as the LABEL of the producing object's formal parameter. In this way, a value-producing object's parameters can "flow down" to all of the values that it produces.

In cases where a formal parameter contains a default value, the associated given parameter may be omitted. Default values in formal parameters (and, thus, optional given parameters) are encouraged as they reduce the size of data items communicated between Managers and Agents in a network.

Finally, actual parameters are the result of applying the Parameter Handling procedure to normalize a set of given parameters based on a set of formal parameters from a processing context.

3.1.3. The Application Resource Identifier (ARI)

The Application Resource Identifier (ARI) is used to represent AMM values outside of an Agent or Manager (i.e. in messaging between them) and is defined in [I-D.ietf-dtn-ari]. Another function of the ARI is for diagnostic or configuration purposes within either Managers or Agents. It is important to make the distinction that within an AMM entity (Agent or Manager) the semantic type of a value is kept, but when exchanged via ARI the semantic type is lost. The AMM defines type compression and reconstruction rules in Section 6.12 to handle this.

3.2. Built-In Types

This section describes the built-in types used for AMM values, which are those usable directly with ARI syntax. By definition, literal values are self-contained and literal types restrict the form and function of those values.

All built-in types within the AMM exit within a flat namespace, but some types have complex relationships with other types beyond the "is a" concept of type inheritance. Built-in types are defined within the "Literal Types" and "Managed Object Types" sub-registries of [IANA-DTNMA] and explained in this section. The following subsections divide the types into groups to simplify their explanation, not because of an intrinsic relationship within each group.

These lists of built-in type names are not fixed in any single specification, and require standards action to add to and update URI processors to handle them, so it is expected that this list will be relatively static (compared to the expected rate of addition or changes to ADMs).

3.2.1. Simple Types

Simple types are those which cannot be subdivided and represent an "atomic" value within the AMM type system. They correspond roughly with the CBOR primitive types Section 3.3 of [RFC8610]. The simple types are summarized in Table 3.

Table 3: Simple Literal Types
Type Description
NULL The singleton null value.
BOOL A native boolean true or false value.
BYTE An 8-bit unsigned integer.
INT A 32-bit signed integer.
UINT A 32-bit unsigned integer.
VAST A 64-bit signed integer.
UVAST A 64-bit unsigned integer.
REAL32 A 32-bit [IEEE.754-2019] floating point number.
REAL64 A 64-bit [IEEE.754-2019] floating point number.
TEXTSTR A text string composed of (unicode) characters.
BYTESTR A byte string composed of 8-bit values.
TP An absolute time point (TP).
TD A relative time difference (TD) with a sign.
LABEL A text label of a parent object parameter. This is only valid in a nested parameterized ARI.
CBOR A byte string containing an encoded CBOR item. The structure is opaque to the Agent but guaranteed well-formed for the ADM using it.
ARITYPE An integer value representing one of the code points in this Literal Types table.

The following subsections discuss nuances in sub-groups of these simple types.

3.2.1.1. Discrete Value Types

The NULL and BOOL types are used to limit to specific discrete values. Because there are CBOR primitive types corresponding exactly with these AMM types, generators of ARIs with these types can always be compressed by eliding the literal type as defined in Section 6.12.

The NULL type has only a single value, null, which is not useful for expressions or type casting but is useful for defining union types which have "optional value" semantics where the null value is used to indicate the absence of a normal value.

The BOOL type is useful for type casting (Section 6.11.1) where an arbitrary value is treated as "truthy" or "falsey" in a context such as a State-Based Rule (SBR) condition.

3.2.1.2. Numeric Types

All of the numeric types (BYTE, UINT, INT, UVAST, VAST, REAL32, and REAL64) exist within a domain where values can be converted between types (Section 6.11.2). Some cases of implicit casting is done for type promotion as necessary for arithmetic operations.

3.2.1.3. Absolute (TP) and Relative (TD) Time Types

The TP type represents an instant in time in the UTC datum. When in text form it is formatted in accordance with the date-time symbol of Appendix A of [RFC3339] and always in the "Z" time-offset.

The TD type represents an offset in time from a relative epoch instant, either later than (a positive offset) or earlier than (a negative offset). When in text form it is formatted in accordance with the duration symbol of Appendix A of [RFC3339] with a positive or negative sign prefix. The epoch instant of a relative time MUST be unambiguously defined in the context using the time value.

3.2.2. Containers

AMM objects, or parameters associated with those objects, often need to represent groups of related data or more complex nesting of data. These are the literal types for AMM value containers, which can only be present in a typed-literal ARI form.

The AMM defines three collection literal types (AC, AM, and TBL) and allows ADMs to combine these literal types with a complex pattern syntax to create semantic types constraining their contents (e.g., for macros and expressions in Section 4.2.3.1).

3.2.2.1. ARI Collection (AC)

An ARI Collection (AC) is an ordered list of ARI elements. The contents of an AC can be restricted in size and type by the use of a semantic type (Section 3.3).

An AC is used when there is a need to refer to multiple AMM values as a single unit. For example, when defining a Report Template, the definition has an AC that defines the ordered ARIs whose values constitute that report.

3.2.2.2. ARI Map (AM)

An ARI Map (AM) is a mapping from a set of "key" ARIs to arbitrary-typed "value" ARIs. As defined in [I-D.ietf-dtn-ari] the AM keys are limited to untyped literals, while the AM values can be any type. The contents of an AM can be restricted in size and type by the use of a semantic type (Section 3.3).

An AM is used when there is a need to define data structures with complex, optionally present attributes. For example, as control parameters used to define new objects in an ODM.

3.2.2.3. ARI Table (TBL)

An ARI Table (TBL) is a collection of values which are logically structured as a two dimensional table of rows and columns, with each cell of the table containing an AMM value.

Although the contents of a TBL can be handled independently of any data model, the meaning of a TBL can only be interpreted within the context of a Table Template (TBLT) defined within an ADM. The TBLT takes the form of a structured type definition on a value-producing object which defines the columns of the table, including each of their column names and types and optional constraints on the number of and uniqueness of rows in the TBL.

A TBL is used when an EDD represents a set or list of complex items as rows in a table. For example, the Agent ADM reports its own set of supported ADMs and features as a TBL (see the "capability" object).

3.2.2.4. Execution-Set (EXECSET)

An Execution-Set (EXECSET) is a collection of values used as targets for the Execution activity. Each message can reference multiple execution sources (CTRL and MAC) and, unlike the MAC execution itself, can be handled by executing multiple items in parallel.

The contents of an EXECSET value are as follows:

Correlator nonce:
This field is an optional opaque correlator "nonce" which can be used to indicate that the result of the corresponding CTRL execution is desired to be reported back to the Manager. The value is limited to match the NONCE (Section 4.2.4) type.
Targets:
This is an unordered list of targets to be executed by an Agent. Each execution target is limited to match the exec-tgt (Section 4.2.3.4) type.
3.2.2.5. Reporting-Set (RPTSET)

A Reporting-Set (RPTSET) is a collection of report containers, where each report container consists of a timestamp and an ordered list of data values populated in conformance to a source object being reported on. Reporting-Set values and reports themselves do not have individual identifiers, rather they are identified by their source and the timestamp at which their data values were collected.

The contents of an RPTSET value are as follows:

Correlator nonce:
This field is an optional opaque correlator "nonce" which is used to associate report containers with specific EXECSET messages which caused the reports to be generated. The value is limited to match the NONCE type (Section 4.2.4).
Reference time:
This field is used as an absolute reference time for all reports contained in the RPTSET. The value is limited to match the TP built-in type. It is used as an storage optimization when a large number of reports are generated around the same time.
Report list:
The main content of the RPTSET are the reports themselves, which are defined below. The order of reports within the RPTSET are not significant, and the presence of a report in any particular RPTSET is not significant. The RPTSET itself is only a container.

The contents of each report within a RPTSET are as follows:

Source:
The source of the report in the form of an ARI with an object-reference for one of the following types: VALUE-OBJ (Section 4.2.4), or CTRL. If the source was parameterized, this ARI SHALL contain the actual parameters used at the time of reporting.
Generation Time:
The timestamp at which the report items were sampled, relative to the Reference Time of the containing RPTSET. The value is limited to match the TD built-in type.
Correlator nonce:
This value is identical to the nonce from the EXECSET which caused the associated reports to be generated, or the null value if the report is not associated with an execution activity. The value is limited to match the NONCE type (Section 4.2.4).
Items:

A list of values corresponding to the source object, with cardinality according to the following:

  • For a VALUE-OBJ source the item list SHALL be the result of reporting (Section 6.8.2) on that object.
  • For a CTRL-REF source there SHALL be a single value representing the Result of the execution. A result of undefined indicates a failure executing the CTRL.

3.2.3. Object Reference Types

For each of the AMM Object Types there is a corresponding object reference type. The object type names and enumerations from the "Managed Object Types" sub-registry of [IANA-DTNMA] are used as names for the built-in type of the corresponding object reference.

For example, a reference value for a CTRL object is typed as CTRL. It is important to understand the distinction, illustrated in Figure 2, between the built-in type and the object type both of which have the same name but used in two completely independent contexts.

3.2.4. Value-Class Types

As a special case of built-in type which act as a union or class of types are those listed in Table 4. These are implemented as a built-in type rather than a semantic type because they behave differently than a Type Union (Section 3.3.6) because they will match any value in the associated class and conversions within these types will not affect the value.

Table 4: Value-Class Types
Type Description
LITERAL Any possible literal value.
OBJECT Any possible object reference value.

3.2.5. Custom Types

When an application requires a more complex or specialized literal type than one already available the preferred design procedure is as follows:

  1. If an existing ADM already defines a semantic typedef (see Section 3.4.2) it is RECOMMENDED to import that ADM and use its typedef.
  2. Otherwise, when it is possible to use an ADM-defined semantic typedef to achieve the desired goals it is RECOMMENDED to do so.
  3. Otherwise, when the desired behavior cannot be accomplished by a semantic typedef, it is RECOMMENDED to use the opaque CBOR type with interface documentation to explain the syntax of the encoded CBOR item.
  4. Otherwise, the application MAY make use of the private-use block of literal type code points.

Implementing a custom literal type requires implementation effort on both an Agent and its associated Manager(s) as well as being more opaque to diagnostic tools and middleboxes.

3.3. Semantic Types

While built-in types control the basic syntax and domain of AMM values, the concept of semantic type is to provide a means to augment literal types by expanding (via union), narrowing (via constraints), and adding human-friendly annotation (such as references to defining documents, or explanations of purpose).

Semantic types can be defined in two ways: a named Semantic Type Definition (TYPEDEF) or an anonymous semantic type defined at the point of use (e.g., within an AMM object definition). The specific syntax used to define semantic types within an ADM are defined and explained in a separate document.

When a "type" is needed for an AMM value in an object definition it SHALL be either one of the built-in types, a namespace-qualified semantic type, or an anonymous semantic type just for that value.

The actual mechanics of semantic typing are based on the classes defined in the following subsections.

3.3.1. Named Type Use

The simplest case is where an existing named type is referenced to be used in a specific context. The form of a named type use SHALL be an AMM value containing either an ARITYPE literal, for a built-in type (Section 3.2), or a TYPEDEF object reference, for a data-model-provided semantic type (Section 3.4.2).

Even an unconstrained type reference can be used within a TYPEDEF to provide a human-friendly name or associated documentation for the use of a simple type. While the tooling might not care about direct type use, it can greatly improve human interpretation of a data model.

An implementation SHALL be able to handle situations where type references create a loop. This would allow values to follow a recursive structure, but it does not mean the values themselves would be of an indefinite size.

Within a named type use, annotations can be added which enhance human understanding of the type. Annotations possible within a named type use SHALL consist of:

  • A free-form text reference to a specific document defining the type in more detail.
  • A free-form text description of the type, which can be in addition to a reference.
  • A units name for NUMERIC values to make the interpretation of values more explicit.
  • A display hint to enable type-specific handling of values, such as IP addresses within BYTESTR values.

Within a named type use, constraints can be added to some of the Simple Types in order to limit what values are considered valid within the type domain. Constraints possible within a named type use SHALL consist of:

  • Limits on ranges of valid NUMERIC types
  • Limits on length of TEXTSTR, BYTESTR, or CBOR
  • Labels of enumerated values or bit positions for INTEGER types
  • Regular expression patterns for TEXTSTR
  • Structural patterns for CBOR items using Concise Data Definition Language (CDDL)

3.3.2. Uniform List

This is the case of a list of AMM values within an ARI Collection (AC) where the type of each value is uniform for the whole list. Only the AC type MAY be refined by a uniform list. Each item of a uniform list SHALL be constrained to a single semantic or built-in type. The number of items in the list MAY be constrained within a range of valid sizes.

3.3.3. Diverse List

This is the case of a list of AMM values within an ARI Collection (AC) where the type of each value is different throughout the list. Only the AC type MAY be refined by a diverse list. Each part of a uniform list SHALL be constrained as either: an item with a single semantic or built-in type or a Sequence (Section 3.3.7).

3.3.4. Uniform Map

This is the case of a map of AMM values within an ARI Map (AM) where the type of each key and each value is uniform for the whole map. Only the AM type MAY be refined by a uniform map use.

3.3.5. Table Template

This is the case of a table of AMM values within a ARI Table (TBL) where each column is annotated with a text name and the type of each value in a column is uniform across all rows. Only the TBL type MAY be refined by a table template. The number of rows in the table MAY be constrained within a range of valid sizes. A single "key" column SHOULD be identified as the unique identifier for each row. One or more column tuples MAY be identified as unique among all rows.

3.3.6. Type Union

This creates a choice between a combination of multiple semantic types. Each of the types in a union SHOULD be exclusive to avoid ambiguity in interpretation by a value processor. The order of choices within a union SHALL be used as the order to check for Type Matching and Type Conversion procedures.

3.3.7. Sequence

This creates a subset of a Diverse List (Section 3.3.3) which matches multiple sequential elements of the list. A sequence is similar to a Uniform List except that it doesn't specify an AC container, it is used to specify items within a container. Each item of a sequence SHALL be constrained to a single semantic or built-in type. The number of items in the sequence MAY be constrained within a range of valid sizes.

A sequence can also be used with formal parameters to create a form of variadic parameter, where multiple given parameters are matched and combined into a single actual parameter (see Section 6.4).

3.4. AMM Object Types

This section identifies the types of objects that make up the AMM and which are instantiated within each ADM and ODM. Each object type is defined by its logical structure and its behavior in Value Production, Execution, or Evaluation contexts within Agents. Each type can allow or disallow parameters within objects and, due to processing behaviors, can either allow or disallow use within an ADM or ODM.

The names for the types of objects defined in this section can be used in two different and separate contexts: as a name for the type of the object itself (written as plain text within this document) when in the context of the AMM object model, or as the name of an object reference (Section 3.1.2) type (written in typewriter text within this document) when used in the context of the AMM value model.

Unless explicitly specified in the object type subsection, an object SHALL NOT be parameterized.

3.4.1. Common Object Fields

Every object type in the AMM includes a set of fields providing annotative or otherwise user-friendly descriptive information for the object. This information may be used as documentation (for example, only present in ADMs and on operator consoles) and/or encoded and transmitted over the wire as part of a management protocol.

The metadata supported by the AMM for all objects is as follows:

Name:
An object name is a text string associated with the object, but does not constitute the sole identifier for the object. Names provide human-readable and/or user-friendly ways to refer to objects with the text form of an ARI. Each object definition SHALL contain a name field. An object's name SHALL NOT change between ADM revisions. Each name SHALL conform to the id-text ABNF symbol of Section 4 of [I-D.ietf-dtn-ari]. Within each namespace and object type, the name of an object SHALL be unique.
Enumeration:
An object enumeration is an integer associated with the object, which identifies the object just like its name. Object enumerations provide a stable and concise identifier for the binary encoded form of an ARI. Each object definition SHOULD contain an enumeration field. An object's enumeration SHALL NOT change between ADM revisions. When present, each enumeration SHALL be an unsigned integer value. Within each namespace and object type, the enumeration of an object SHALL be unique.
Status:
Each object definition MAY contain a status field. The valid status values of an object are the same as the valid status values for an ADM in Section 4.1.1. In the absence of a status field, the status of the object SHALL be considered the same as the status of the ADM which contains it.
Reference:
Each object definition MAY contain a reference field. A reference is a text string referring to a specification or other document which details the source or purpose of the object.
Description:
Each object definition MAY contain a description field. A description is a text string explaining the purpose or usage of the object in a human-readable format. There is no minimum or maximum size of description text for an object. The description serves as documentation for the object and SHOULD be the same regardless of how the object might be parameterized. For example, the description of a CTRL object should document the purpose of the CTRL in a way that is independent of the value of any particular parameter value passed to that CTRL.

Formal parameters define a method to customize an AMM object. When used by an object definition, it's formal Parameters SHALL be an ordered list of individual formal parameter definitions. Each formal parameter SHALL include type and name. Each formal parameter MAY include an optional default value. The application of default parameters and relationship of actual parameters (Section 3.1.2.1) to formal parameters is defined in Section 6.4.

3.4.2. Semantic Type Definition (TYPEDEF)

An ADM can define a semantic type definition (TYPEDEF) to give a name to a semantic type (Section 3.3). This TYPEDEF name can then be used as a type anywhere else in the same ADM or another one which imports it.

The definition of a TYPEDEF consists of the following:

Name, Enumeration, Status, Reference, Description:
As defined in Common Object Fields.
Type:
A TYPEDEF definition SHALL include the type being named, as described in Section 3.3. The type of a TYPEDEF is fixed and SHALL NOT change between ADM revisions. The type SHALL be either a union of other types or a restriction of or annotation upon another type.

As defined in this document, TYPEDEFs and semantic types can only be defined within an ADM. Future capability could allow the use of TYPEDEFs within ODMs.

3.4.3. Identity Object (IDENT)

An ADM can use an identity object (IDENT) to define a unique, abstract, and untyped identity. The only purpose of an IDENT is to denote its name, parameters, and existance semantics. This allows an extensible but controlled enumeration of possible values when using an IDENT object reference.

Each IDENT object MAY be derived from one or more other base IDENT objects to form a directed graph. An IDENT which is not derived from any other is referred to as a "root" object. Any chain of derived IDENT objects SHALL NOT form a loop.

The definition of a IDENT consists of the following:

Name, Enumeration, Status, Reference, Description:
As defined in Common Object Fields.
Parameters:
A non-root IDENT definition MAY include formal parameters to be used when the IDENT is referenced. Parameterized objects are discussed in Section 7. The formal parameters of an IDENT are fixed and SHALL NOT change between ADM revisions.

As defined in this document, IDENTs can only be defined within an ADM.

3.4.4. Externally Defined Data (EDD)

Externally defined data (EDD) objects represent data values that are produced based on a source external to the Agent itself. The Value Production occurs at the moment the value is needed, by either an Evaluation or a Reporting activity. The actual value could come from outside of the Agent proper, or be derived from data outside of the Agent.

The value production of an EDD SHOULD be nilpotent and have no side-effects in the processor. This property is not enforced by the Agent but requires consideration of the ADM designers, see Section 7.

The value produced by an EDD is allowed to, but not required to, change over time. Because EDDs can be referenced by condition expressions of Time-Based Rules (Section 3.4.9) or elsewhere, an Agent implementation could be optimized by allowing an EDD to indicate when its produced value would change. It is an implementation matter for if and how an application can provide that indication.

For values managed entirely within the Agent use a Variable (VAR) or for constant-values use a Constant (CONST). For complex tabular data, use an EDD with a type which produces an ARI Table (TBL).

The definition of an EDD consists of the following:

Name, Enumeration, Status, Reference, Description:
As defined in Common Object Fields.
Parameters:
An EDD definition MAY include formal parameters to be used when the EDD is used to produce a value. Parameterized objects are discussed in Section 7. The formal parameters of an EDD are fixed and SHALL NOT change between ADM revisions.
Type:
An EDD definition SHALL include the type of the value produced by the object, as described in Section 3.3. The type of an EDD is fixed and SHALL NOT change between ADM revisions.

As defined in this document, EDDs can only be defined within an ADM. Future capability could allow the use of EDDs within ODMs.

3.4.5. Constant (CONST)

A Constant (CONST) represents a named literal value, but unlike an Externally Defined Data (EDD) or Variable (VAR) a CONST always produces the same value. Examples include common mathematical values such as PI or well-known time epochs such as the UNIX Epoch. A CONST typed to produce a simple value can be used within an expression (see Section 6.7), where the object is used to produce a value at the moment of evaluation. A CONST can also be typed to produce an EXPR value to evaluate, or MAC value to execute.

The definition of a CONST consists of the following:

Name, Enumeration, Status, Reference, Description:
As defined in Common Object Fields.
Type:
A CONST definition SHALL include the type of the value produced by the object, as described in Section 3.3. The type of a CONST is fixed and SHALL NOT change between ADM revisions.
Value:
A CONST definition SHALL include the literal value produced during evaluation.

As defined in this document, CONSTs can only be defined within an ADM. Allowing network operators to define constants dynamically means that a Constant could be defined, removed, and then re-defined at a later time with a different value, which defeats the purpose of having Constants. When adding new "fixed" values to an ODM, a Variable (VAR) MUST be used instead of a Constant.

3.4.6. Control (CTRL)

A Control (CTRL) represents a predefined function that can be executed on an Agent. Controls are not able to be defined as part of dynamic network configuration since their execution is typically part of the firmware or other implementation outside of the Agent proper.

The execution of a CTRL SHOULD be idempotent and have no effect if executed multiple times in sequence. This property is not enforced by the Agent but requires consideration of the ADM designers, see Section 7.

Controls can be executed in a "one shot" manner as part of messaging from a Manager to an Agent. Network operators that wish to autonomously execute functions on an Agent may use a State-Based Rule (SBR) or Time-Based Rule (TBR). When an execution involves the ordered sequence of controls, a Macro (MAC) SHOULD be used instead of a more fragile use of CTRL directly.

The definition of a CTRL consists of the following:

Name, Enumeration, Status, Reference, Description:
As defined in Common Object Fields.
Parameters:

A CTRL definition MAY include formal parameters to be used when the CTRL is executed. Parameterized objects are discussed in Section 7. The formal parameters of a CTRL are fixed and SHALL NOT change between ADM revisions.

Result:
A CTRL definition MAY include the definition of a result. The result SHALL have a name and a type. The result MAY have a default value. The result of a CTRL is separate from the execution status as being successful or failed.

As defined in this document, CTRLs can only be defined within an ADM. Future capability could allow the use of CTRLs within ODMs if there was some mechanism to bind a CTRL definition to some platform-specific execution specification (e.g., a command line sequence).

3.4.7. Operator (OPER)

An Operator (OPER) represents a user-defined, typically mathematical, function that operates within the evaluation of an Expression (EXPR). It is expected that operators are implemented in the firmware of an Agent.

The AMM separates the concepts of Operators and Controls to prevent side-effects in Expression evaluation (e.g. to avoid constructs such as A = B + GenerateReport()). For this reason, Operators are given their own object type and Controls do not interact with operators.

The definition of an OPER consists of the following:

Name, Enumeration, Status, Reference, Description:
As defined in Common Object Fields.
Parameters:
An OPER definition MAY include formal parameters to be used when the OPER is evaluated. Parameterized objects are discussed in Section 7. The formal parameters of an OPER are distinct from the operands from the expression stack.
Operands:
An OPER definition MAY include definitions of operand values to be popped from the expression stack when the OPER is evaluated. Each operand SHALL consist of a name, a type, and a cardinality. Any non-trivial OPER will have one or more operands. An OPER can have a non-fixed operand count which is based on a parameter value (e.g., an operator can average the top N values from the stack, where N is a parameter).
Result:
An OPER definition SHALL include definition of a result value to be pushed onto the expression stack after the OPER is evaluated. The result SHALL have a name and a type. The result SHALL NOT have a default value.

As defined in this document, OPERs can only be defined within an ADM. Future capability could allow the use of OPERs within ODMs if there was some mechanism to bind an OPER definition to some platform-specific evaluation specification.

3.4.8. State-Based Rule (SBR)

A State-Based Rule (SBR) is a form of autonomy in which the Agent performs an action upon the change of state to meet a specific condition. The execution model of the SBR is to evaluate the Condition (as often as necessary to handle changes in its expression evaluation) and when it evaluates to a truthy (Section 6.11.1) value and it has been no shorter than the Minimum Interval since the last execution, the Action is executed. When the Maximum Count of executions is reached the TBR is disabled. The execution occurs concurrently with any time processing and may take longer than the Minimum Interval, so it is possible that multiple executions are requested to overlap in time.

Each SBR has an enabled state to allow rules to be retained in an ADM or ODM but not enabled during Manager-controlled time periods or under certain Manager-desired conditions. See Section 4.3 for details about what SBR-related controls are in the Agent ADM.

The definition of an SBR consists of the following:

Name, Enumeration, Status, Reference, Description:
As defined in Common Object Fields.
Action:
An SBR definition SHALL include an action in the form of a Macro (MAC). When triggered, the action execution SHALL be executed in accordance with Section 6.6 in an execution context with no parameters.
Condition:
An SBR definition SHALL include a condition in the form of an Expression (EXPR). The condition SHALL be evaluated in accordance with Section 6.7 in an evaluation context with no parameters. The result of the condition SHALL be converted to a BOOL value after evaluation and used to determine when to execute the action of the SBR.
Minimum Interval:
An SBR definition SHALL include a minimum execution interval in the form of a non-negative TD value. The interval MAY be zero to indicate that there is no minimum. This is not a limit on the interval of evaluations of the condition. This value can be used to limit potentially high processing loads on an Agent.
Maximum Count:
An SBR definition SHALL include a maximum execution count in the form of a non-negative UVAST value. The count sentinel value zero SHALL be interpreted as having no maximum. This is not a limit on the number of evaluations of the condition.
Initial Enabled:
An SBR definition MAY include an initial value for its enabled state. If not provided, the initial enabled state SHALL be true.

3.4.9. Time-Based Rule (TBR)

A Time-Based Rule (TBR) is a form of autonomy in which the Agent performs an action at even intervals of time. The execution model of the TBR is to start a timer at the Start Time of the TBR ticking at an even Period; each time the timer expires the Action is executed. When the Maximum Count of executions is reached the TBR is disabled. The execution occurs concurrently with any time processing and may take longer than the TBR Period, so it is possible that multiple executions are requested to overlap in time.

Each TBR has an enabled state to allow rules to be retained in an ADM or ODM but not enabled during Manager-controlled time periods or under certain Manager-desired conditions. See Section 4.3 for details about what TBR-related controls are in the Agent ADM.

The definition of a TBR consists of the following:

Name, Enumeration, Status, Reference, Description:
As defined in Common Object Fields.
Action:
A TBR definition SHALL include an action in the form of a Macro (MAC). When triggered, the action execution SHALL be executed in accordance with Section 6.6 in an execution context with no parameters.
Start Time:
A TBR definition SHALL include a start time in the form of a TIME (Section 4.2.4) value. A relative start time SHALL be interpreted relative to the absolute time at which the Agent is initialized (for ADM rules) or the rule is created (for ODM rules). The start time MAY be the relative time zero to indicate that the TBR is always active. This is not a limit on the interval of evaluations of the condition.
Period:
A TBR definition SHALL include a period in the form of a positive TD value. The period SHALL NOT be zero but any non-zero small period is valid.
Maximum Count:
A TBR definition SHALL include a maximum execution count in the form of a non-negative UVAST value. The count sentinel value zero SHALL be interpreted as having no maximum. This is not a limit on the number of evaluations of the condition.
Initial Enabled:
A TBR definition MAY include an initial value for its enabled state. If not provided, the initial enabled state SHALL be true.

3.4.10. Variable (VAR)

A Variable (VAR) is a stateful store of a value in an Agent. The use of a VAR is similar to an EDD (Section 3.4.4) except that all the behavior of a VAR is entirely within an Agent, while the ultimate source of an EDD value is outside of the Agent.

The value production of a VAR into a value SHALL be nilpotent and have no side-effects in the processor.

The value produced by an VAR is allowed to, but not required to, change over time. Because VARs can be referenced by condition expressions of Time-Based Rules (Section 3.4.9) or elsewhere, an Agent implementation could be optimized by allowing a VAR to indicate when its produced value would change. It is an implementation matter for if and how an application can provide that indication.

A VAR has an initializer, which is used at Agent initialization and to reset the VAR (see Section 4.3), but the VAR is otherwise stateful and will retain its last value between any actions which modify it.

The definition of a VAR consists of the following:

Name, Enumeration, Status, Reference, Description:
As defined in Common Object Fields.
Parameters:
A VAR definition MAY include ARI parameters to be used when the VAR is evaluated. Parameterized objects are discussed in Section 7. Parameters for a VAR are only meaningful when the VAR contains a value with actual parameters themselves containing a LABEL value.
Type:
An VAR definition SHALL include the data type of the value produced during evaluation, as described in Section 3.3. The type of a VAR is fixed and SHALL NOT change between ADM revisions.
Initializer:
An VAR definition MAY include an initializer in the form of an Expression (EXPR). The only times the initializer are evaluated are at Agent Initialization and when a CTRL is used to reset the state of the VAR.

4. Application Data Models (ADMs)

An ADM is a logical entity for defining static AMM object instances, which are discussed in detail in Section 3.4. Each ADM exists as a separate namespace for its contained objects, but allows importing object names from other ADMs to reuse them. Each Agent can support any number of ADMs at one time (subject to implementation limitations) and each Manager can operate with ADMs of different revisions to support diverse Agents.

The following subsections define what is present in an ADM generally and what objects necessary to operate a DTNMA Agent are present in two base ADMs.

4.1. ADM Definitions

An ADM is "static" in the sense that it is revision-controlled and a released revision of an ADM does not change. Besides AMM object definitions there are metadata and handling rules for the ADM itself, which are discussed in this section.

4.1.1. ADM Metadata

This section explains the purposes of the metadata fields of an ADM, while the specific syntax for how these fields fit into an ADM module is left to another document.

Name:
Each ADM definition SHALL contain a name field. An ADM's name SHALL NOT change between ADM revisions. The name SHALL conform to the id-text ABNF symbol of Section 4 of [I-D.ietf-dtn-ari].
Enumeration:
An ADM enumeration is an integer associated with the object, which identifies the object just like its name. ADM enumerations provide a stable and concise identifier for the binary encoded form of an ARI. Each ADM definition SHALL contain an enumeration field. An ADM's enumeration SHALL NOT change between ADM revisions. The enumeration SHALL be an unsigned integer value.
Revision History:
Each ADM SHALL contain a history of dated revisions. At least one revision SHALL be present and mark the date at which the ADM was released for use. During development and testing an ADM need not have updated revisions, only when a release occurs should a revision be added.
Status:
Each ADM definition SHOULD contain a status field. The valid status value of an ADM SHALL be identical to the Status field of YANG Section 7.21.2 of [RFC7950]. In the absence of a status field, the status of the ADM SHALL be considered the same as the status of the ADM which contains it.
Reference:
Each ADM definition SHOULD contain a reference field. A reference is a text string referring to a specification or other document which details the source or purpose of the ADM.
Description:
Each ADM definition SHOULD contain a description field. A description is a text string explaining the purpose or usage of the ADM in a human-readable format.
Features:
Each ADM definition MAY contain a set of feature definitions, see Section 4.1.2 for details. Each feature SHALL have a name that is unique within the namespace of the ADM. Each name SHALL conform to the id-text ABNF symbol of Section 4 of [I-D.ietf-dtn-ari].

4.1.2. Features and Conformance

Following in the pattern of YANG features from Section 5.6.2 of [RFC7950] and SMIv2 conformance groups from [RFC2580], the AMM has the concept of ADM features and Agent conformance to those features. Each feature is a simple qualified name and each object in an ADM can be conditional on the conformance to a set of features.

In the same way that an Agent instance can choose to implement or omit any particular ADM (assuming its dependencies are satisified), an Agent instance can choose to implement or omit particular features within an ADM. This allows more fine-grained control of what an Agent supports at runtime and also provides a standard mechanism for naming and indicating that support.

4.2. Contents of an AMM ADM

This base ADM is a necessary part of the AMM typing, execution, and evaluation models. Rather than having some Agent logic defined purely by specification, this document uses an "AMM" ADM to define semantic types and controls needed for normal Agent operations. The needed types are still set by specification and are unchanging within an ADM revision, but this avoids having a separate, intermediate typing system between the AMM-defined semantic types and the ARI-defined literal types. This is also in-line with how YANG [RFC6991] and SMIv2 [RFC2578] both rely on base modules for some core behavior.

4.2.1. Display Hint Root

Rather than using fixed enumerations for the display hint of a Named Type Use, the AMM uses an IDENT (Section 3.4.3) hierarchy, where each leaf object represents a specific form of display for one of the built-in types. The root IDENT object for this hierarcy is defined in this ADM, but the leaf objects will be managed outside the ADM.

4.2.2. Simple Semantic Types

The most basic use of a semantic type is to provide additional meaning to simple types. None of these types associates a unit with the value, which it is expected that a derived type or an anonymous type (at the point of use) would add for additional clarity.

These are summarized below:

counter32 and counter64:
An unsigned integer value with an arbitrary initial value which increments over time and wraps around the maximum value. These correspond with the same names defined in YANG [RFC6991] and SMIv2 [RFC2578].
gauge32 and gauge64:
An integer value sampling some measurement which can increase or decrease arbitrarily over time. These correspond with the same names defined in YANG [RFC6991] and SMIv2 [RFC2578].
timestamp:
An absolute time at which an event happened. This corresponds with the same name defined in YANG [RFC6991] and SMIv2 [RFC2578].

4.2.3. Container Semantic Types

This section contains more complex semantic types which constrain the contents of a container (Section 3.2.2) so that the value as a whole has a specific semantic.

4.2.3.1. Expression (EXPR)

An Expression (EXPR) is an ordered collection of references to Operators or operands. An EXPR takes the form of a semantic typedef refining an AC to be a list of ARIs referencing OPERs, ARIs referencing evaluate-able objects (see Section 6.7), or literal value ARIs (with Simple Types). These operands and operators form a mathematical expression that is used to compute a resulting value.

The evaluation procedure of an EXPR is defined in Section 6.7. Expressions are used within an ADM for defining the initializer of a Variable (VAR) and for defining the condition of a State-Based Rule (SBR).

Since the Expression is an AC, there are no annotative constructs such as parenthesis to enforce certain orders of operation. To preserve an unambiguous calculation of values, the ARIs that form an Expression MUST be represented in postfix order. Postfix notation requires no additional symbols to enforce precedence, always results in a more efficient encoding, and post-fix engines can be implemented efficiently in embedded systems.

For example, the infix expression A * (B * C) is represented as the postfix A B C * *.

4.2.3.2. Macro (MAC)

A Macro (MAC) is an ordered collection of references to Controls or other Macros. A Macro takes the form of a semantic typedef refining an AC to be a list of ARIs referencing Controls or objects which produce other Macros.

The execution procedure of an MAC is defined in Section 6.6. Macros are used within an ADM for defining the action of a State-Based Rule (SBR) or Time-Based Rule (TBR).

In cases where a Macro references another Macro, Agent implementations MUST implement some mechanism for preventing infinite recursions, such as defining maximum nesting levels, performing Macro inspection, and/or enforcing maximum execution times.

4.2.3.3. Report Template (RPTT)

A Report Template (RPTT) is an ordered list of object references or expression values used as a source for generating items for report (Section 3.2.2.5) containers. A RPTT takes the form of a semantic typedef refining an AC to be a list of references to value-producing objects (VALUE-OBJ (Section 4.2.4)) or expressions (EXPR (Section 4.2.3.1)). An object which produces an RPTT can itself be parameterized so that the object flows down parameters as described in Section 3.1.2.1.

A RPTT can be viewed as a schema that defines how to generate and interpret a Report; they contain no direct values. RPTT values either defined in an ADM or configured between Managers and Agents in an ODM. Reports themselves are ephemeral and represented within ARI built-in type RPTSET, not as part of the AMM object model. The procedure for reporting on a RPTT is defined in Section 6.8.1.

RPTT values SHOULD be used within a CONST where possible. RPTT values MAY be used within a VAR where necessary. This makes correlating a RPT value with its associated RPTT easier over time. Rather than having a VAR object's RPTT value changing over time, it is RECOMMENDED to deprecate earlier RPTT-producing CONST objects and create new objects.

4.2.3.4. Execution Target Type

A convenience typedef exec-tgt is defined to codify the type of values allowed to be used as input for an Execution (or within an Execution-Set (EXECSET) value) or produced by objects referenced as execution targets. The execution target type is defined to be either a direct CTRL reference, a direct MAC value, or a reference to a value-producing object which itself is typed as exec-tgt.

4.2.3.5. Evaluation Target Type

A convenience typedef eval-tgt is defined to codify the type of values allowed to be used as input for an Evaluation activity or produced by objects referenced as evaluation targets. The execution target type is defined to be either a direct SIMPLE value, a direct EXPR value, or a reference to a value-producing object which itself is typed as eval-tgt.

4.2.4. Type Unions

All of the literal types defined in [I-D.ietf-dtn-ari] have a flat structure, with some types sharing the same CBOR primitive encoding but using distinct built-in type code points to distinguish them. In order to allow types to fit into a more logical taxonomy, the AMM ADM defines some specific semantic typedefs to group literal types. These groups are not a strict logical hierarchy and are intended only to simplify the effort of an ADM designer when choosing type signatures.

These are summarized below:

TYPE-REF:
The union of ARITYPE and TYPEDEF types.
INTEGER:
The union of BYTE, UINT, INT, UVAST, and VAST types.
FLOAT:
The union of REAL32 and REAL64 types.
NUMERIC:
The union of INTEGER and FLOAT types.
PRIMITIVE:
The union of NULL, BOOL, NUMERIC, TEXTSTR, and BYTESTR types. This matches any untyped literal value.
TIME:
The union of TP and TD types.
SIMPLE:
The union of PRIMITIVE, and TIME types. This matches any non-container literal value (typed or untyped).
ANY:
The union of LITERAL and OBJECT value-class types (Section 3.2.4). This matches all values that can be in an ARI.
VALUE-OBJ:
The union of CONST, EDD, and VAR reference types. This matches any reference to an object that can produce a value (Section 6.5).
NONCE:
The union of BYTESTR, UINT64, and NULL types. This is used by EXECSET and RPTSET values to correlate Agent-Manager messages (see Section 2.3).

4.3. Contents of an Agent ADM

While the AMM ADM described in Section 4.2 contains definitions of static aspects of the AMM, the DTNMA Agent ADM is needed to include necessary dynamic aspects of the operation of an Agent. This separation is also helpful in order to allow the dynamic behaviors of an Agent to be modified over time while the AMM definitions stay stable and unchanging.

4.3.1. Agent State Introspection

The Agent ADM contains the following EDD objects used to introspect the Agent's state, all of which can change over time within an Agent.

  • The ADMs supported by the Agent, including the unique name and revision of each. By indicating specific revision and supported feature set, the contained objects in each ADM can be derived. Because of this, the ADM-contained objects do not require additional introspection.
  • The set of SBRs and TBRs in the Agent's ODMs, along with controls to ensure a specific object is either present or absent. These are all conditioned on whether the Agent actually supports the built-in rule feature.
  • The set of VARs in the Agent's ODMs, along with controls to ensure a specific object is either present or absent.
  • Visibility into the execution state(s) of an Agent, including counters for the total number of successful and failed executions.
  • Counters for the total number of messages sent or received by the agent, including reception failures.

4.3.2. Macro Helper Controls

The Agent ADM contains a set of controls which implement behaviors to macro execution logic.

Branching Control:
This control has a condition parameter to evaluate and two optional parameters to define sub-macros, one of which is executed depending upon the condition result truthy-ness.
Failure Catching Control:
This control has one parameter of a macro to execute normally and a second parameter of a macro to execute on the condition that the normal execution fails for some reason.
Waiting Controls:
This family of controls is to be embedded at the start (or anywhere within) a macro and pauses execution to wait on either: a specific absolute time, a relative time from start-of-control, or a condition to evaluate to truthy.

4.3.3. Basic Operators

The Agent ADM contains a set of operators which provide logical and mathematical functions to macro execution.

Numeric Operators:
These perform operations related to negation, addition, subtraction, multiplication, division, and remainder (modulo) of their operands. These perform numeric promotion of their operands in accordance with Section 6.11.2.1.
Boolean Operators:
These perform operations related to boolean NOT, AND, OR, and XOR of their operands. These perform boolean casting of their operands in accordance with Section 6.11.1.
Bitwise Operators:
These perform bitwise NOT, AND, OR, and XOR operations on only unsigned integer operands.
Comparison Operators:
These perform pairwise comparison between their operands. Equality and inequality can be performed on any operand types, but ordered comparison (e.g., greater than) can only be performed on numeric operands.
Table Filtering:
This operator is used to process tables produced within an expression to filter by row contents and specific columns. This is an example of a parameterized operator because the parameters control the filtering while the operand is the table-to-be-filtered.

5. Operational Data Models (ODMs)

An ODM is a logical entity for containing AMM objects, similar to an ADM (Section 4) but in an ODM the objects are not static. An ODM's objects can be added, removed, and (with some restrictions) modified during the runtime of an Agent. Like an ADM, each ODM exists as a separate namespace for its contained objects and an Agent can contain any number of ODMs.

Some object types, those which require implementation outside of the Agent proper, are not available to be created in an ODM. These include the CTRL, EDD, and OPER.

The actions for inspecting and manipulating the contents of an ODM are available through EDDs and CTRLs of the Agent ADM (Section 4.3.1).

6. Processing Activities

This section discusses logic and requirements for processing of AMM objects and values. Each subsection is a separate class of processing that is performed by an Agent.

A Manager (or any other entity) MAY perform some of the same processing, e.g. evaluating an expression, in order to validate values or configurations before sending them to an Agent. That kind of behavior is effectively creating a "digital twin" of the managed Agent to ensure that the processing will behave as expected before it is sent. For this reason, the subject noun used in all of these activities is the "processor".

6.1. Agent Initialization

The initialization of the Agent state can be associated with a power-on event or, due to the use of volatile memory, can be an explicit activity initiated from outside the Agent runtime. If volatile memory is used the contents of the ODMs on an Agent will be present for the initialization procedure; otherwise the ODMs will be considered empty or absent.

The procedure to initialize an Agent is as follows:

  1. All ADM-defined VAR objects SHALL have their value set to one of the following:

    • If an Initializer is defined for the VAR, the value is the result of evaluating the associated Initializer expression and then converting (Section 6.11) to the VAR type.
    • Otherwise, the value is undefined.

    Any ODM-defined VAR objects MAY retain their state.

  2. All ADM-defined TBR and SBR objects SHALL have their Enabled state set to the Initial Enabled value. Any ODM-defined TBR and SBR objects MAY retain their Enabled state. Any rules which are enabled are ready for processing.

6.2. ARI Resolving

Within an ADM, ARIs present in the various fields of object definitions are URI References, which can take the form of Relative URIs (see Section 4.2 of [RFC3986]). Any ARIs within an ADM definition SHALL be handled as URI References and resolved in accordance with the procedure of Section 5 of [RFC3986] with the following used as a Base URI:

  • For ARIs within a single AMM object definition, the non-parameterized ARI of that object SHALL be the Base URI. This includes ARIs used in nested structures under the object definition; the object is the anchor point.
  • For all other ARIs, the default Base URI ari:/ SHALL be the Base URI. This means that all ARIs within an ADM do not require a URI scheme part.

6.3. Dereferencing

An Object Reference Values contains an object path and a parameter part. Dereferencing an OBJECT value uses the object path to look up a specific defined object available to the agent.

The process of dereferencing a value is as follows:

  1. The value has to contain an object reference. If the value is not an object reference, this procedure stops and is considered failed.
  2. The OBJECT value namespace (whether text or enumeration) is used to search for a defined ADM or ODM namespace. A text form namespace SHALL be compared within the UTF-8 character set in accordance with [RFC3629]. An integer namespace SHALL be compared numerically. If no corresponding namespace is available, this procedure stops and is considered failed.
  3. Within the namespace the object type and object name (whether text or enumeration) is used to search for a specific defined object. A text form object name SHALL be compared within the UTF-8 character set in accordance with [RFC3629]. An integer object name namespace SHALL be compared numerically. If no corresponding object is available, this procedure stops and is considered failed.

6.4. Parameter Handling

An Object Reference Values contains an object path and a parameter part. The parameter part of an OBJECT value represents the given parameters (Section 3.1.2.1) being used. The given parameters are present either as a (possibly empty) ARI list or an ARI map. Due to nuances of the AMM value system, the given parameters are not themselves either AC or AM values but similar to untyped ARI values.

The process to validate and normalize given parameters against an object's formal parameters to produce actual parameters is as follows.

  1. For each formal parameter, the processor performs the following:

    If the given parameters are a list, the formal parameter is correlated to the list by its position in the formal parameters list. If the given parameters list does not contain a corresponding position the given parameter is treated as the undefined value. If the last formal parameter is a Sequence (Section 3.3.7), it can correlate with multiple given parameters.

    If the given parameters are a map, the formal parameter is correlated to a map key by either its position (as an integer) or its name (as a text string) but not both. If both integer and name are present in the given parameters map the procedure stops and is considered failed. If the given parameters map does not contain a corresponding key the given parameter is treated as the undefined value.

  2. If any of the given parameters is not correlated with a formal parameter the procedure stops and is considered failed.
  3. For each correlated pair of formal parameter and given parameter(s), the processor performs the following:

    1. If the given parameter is undefined (whether explicitly or implicitly) and the formal parameter defines a default value, that default is used as the actual parameter value. If there is no default value, the actual parameter is left as the undefined value.

    2. If the given parameter is a TYPEDEF and the object reference itself has a parameter, the given parameter is treated as the result of a type conversion (Section 6.11.3) to the semantic type of the TYPEDEF. If the conversion fails this procedure stops and is considered failed.

    3. The given parameter is converted to an actual parameter using the type of the formal parameter in accordance with Section 6.11. If the conversion fails, this procedure stops and is considered failed.

The actual parameters resulting from this procedure are intended to be able to be looked up by an implementation either by ordinal position in the formal parameters list or by unique name of the formal parameter. It is an implementation matter whether or not to provide both accessing methods and the specifics of how, for example, and EDD or CTRL runtime accesses actual parameter values.

An implementation MAY perform deferred "lazy" processing of any of the above steps, causing a failure when the actual parameter value is needed. One caveat about deferred processing is that it will not fail if the parameter is unused, which is not necessarily a problem but could mask other issues in whatever provided the given parameters.

6.5. Value Production

Value production can be thought of as a common behavior used for Execution, Evaluation, and Reporting activities. Within the AMM the following entities have a value production procedure: CONST, EDD, and VAR object references.

This activity relies on an object reference value to have been dereferenced in accordance with Section 6.3 and its parameters handled in accordance with Section 6.4. After that, each of the object types is treated differently as defined in the following subsections.

6.5.1. CONST and VAR Objects

Both CONST and VAR objects act as a store of a single literal value within the Agent. Formal parameters on either CONST or VAR objects are applicable only when the objects store a value which itself contains parameters with at least one LABEL type.

The value production for these objects takes the stored value from the object and augments it by label substitution based on the following:

  1. The processor identifies all LABEL type within the stored value, descending into container (Section 3.2.2) contents and object reference parameter contents as necessary.
  2. If any LABEL text is not present in the formal parameters of the value-producing object then this procedure stops and is considered failed.
  3. For each LABEL value the corresponding actual parameter is not the undefined value, the LABEL value is replaced by the actual parameter.

This augmentation has no effect on the stored value, it occurs only in the produced value. It is valid both for an actual parameter to have no substitution occur with its value and for an undefined actual value not be sued in substitution.

6.5.2. EDD Objects

For EDD objects, the actual parameters are used by the underlying implementation to produce the value in an arbitrary way. The produced value is typically either a SIMPLE (Section 4.2.4) value or an ARI Table (Section 3.2.2.3).

The value production for these objects occurs outside of the Agent proper within an implementation of the EDD being produced from.

The context given to the implementation is the following:

Object Path:
This gives visibility into the EDD object reference which was dereferenced during the production.
Actual Parameters:
The set of actual parameters used for the production.
Result Type and Storage:
The result of the production is placed here before completion.

The initial state of the Result Storage is the undefined value. It is an implementation matter and author consideration (Section 7) to enforce that the produced value is consistent with the type of the object.

6.6. Execution

Within the AMM only two entities can be the target of an execution procedure: controls and macros. Controls are executed by reference, while macros are executed both by value and by reference. This means the execution target value SHALL match the exec-tgt (Section 4.2.3.4) semantic type.

The procedure for executing is divided into phases to ensure that it does not fail due to invalid references or produced values after some controls have already been executed. The phases are processed as follows:

  1. In the expansion phase the target value is processed to dereference all references, handle all parameters, and expand any produced values.

    If the target is a literal value, the following is performed:

    1. The value needs to match the MAC (Section 4.2.3.2) semantic type. If it does not match, this procedure stops and is considered failed.
    2. The processor then iterates through all elements of the MAC value and performs the expansion step on each in turn. If any sub-expansion fails, this procedure stops and is considered failed.

    If the target is an object reference, the following is performed:

    1. The value is dereferenced in accordance with Section 6.3 and its parameters are handled in accordance with Section 6.4. If either fails, this procedure stops and is considered failed.
    2. If the target object is a value-producing object, a value is produced in accordance with Section 6.5. This includes substitution of any LABEL parameters within the value.
    3. The processor then performs the expansion step on the produced value. If sub-expansion fails, this procedure stops and is considered failed.

    After expansion the target is either a dereferenced CTRL object, or a (possibly nested) macro expanded to contain only dereferenced CTRL objects.

  2. The processor then executes the top-level expanded value as either a macro in accordance with Section 6.6.1 or as a control in accordance with Section 6.6.2

6.6.1. Expanded MAC Values

The execution of an Macro (MAC) value after expansion is as follows:

  1. The processor iterates through all items of the expanded MAC in order and performs the following:

    If the item is an CTRL-REF it is executed in accordance with Section 6.6.2. If the execution fails, this procedure stops and is considered failed.

    Otherwise the item is an expanded sub-macro and it is executed in accordance with this procedure.

An effect of this procedure is that if any referenced CTRL fails during execution the processing fails immediately and subsequent CTRLs or MACs are not executed.

6.6.2. CTRL Objects

This activity relies on an object reference value to have been dereferenced in accordance with Section 6.3 and its parameters handled in accordance with Section 6.4.

The execution of a Control (CTRL) object occurs outside of the Agent proper within an implementation of the CTRL behavior.

The context given to the implementation is the following:

Manager:
The manager which directly caused this execution, if available, is provided as context.
Object Path:
This gives visibility into the CTRL object reference which was dereferenced during the execution.
Actual Parameters:
The set of actual parameters augmented for the execution.
Result Type and Storage:
The result of the execution is placed here before completion.

The initial state of the Result Storage is the null value.

If the execution fails, the result value SHALL be treated as the undefined value for the purposes of any subsequent reporting.

6.7. Evaluation

Within the AMM the following entities can be the target of an evaluation procedure: references to value-producing objects, OPERs, and TYPEDEFs and EXPR or SIMPLE literal values.

The procedure for evaluation is divided into phases to ensure that it does not fail due to invalid references or produced values after some expressions have already been evaluated. The phases are processed as follows:

  1. In the expansion phase the target value is processed to dereference all references, handle all parameters, and expand any produced values.

    If the target is a literal value, the following is performed:

    1. The value needs to match the SIMPLE (Section 4.2.4) or EXPR (Section 4.2.3.2) semantic type. If it does not match, this procedure stops and is considered failed.
    2. If the value is an EXPR, the processor then iterates through all elements of the EXPR value and performs the expansion step on each in turn. If any sub-expansion fails, this procedure stops and is considered failed.

    If the target is an object reference, the following is performed:

    1. The value is dereferenced in accordance with Section 6.3 and its parameters are handled in accordance with Section 6.4. If either fails, this procedure stops and is considered failed.
    2. If the target object is a value-producing object, a value is produced in accordance with Section 6.5. This includes substitution of any LABEL parameters within the value.
    3. The processor then performs the expansion step on the produced value. If sub-expansion fails, this procedure stops and is considered failed.

    After expansion the target is either a SIMPLE value, or a (possibly nested) expression expanded to contain only SIMPLE or ARITYPE values, or dereferenced OPER or TYPEDEF objects.

  2. If the expanded evaluation target is already a SIMPLE value, then that is the result of the evaluation. Otherwise, the expanded expression is evaluated in accordance with Section 6.7.1.

6.7.1. Expanded EXPR Values

The reduction of an Expression (EXPR) value after expansion is as follows:

  1. Any sub-expressions are first reduced to their result values which are substituted back into the corresponding expression item. If any sub-evaluation fails this procedure stops and is considered failed. At this point the expression consists of only SIMPLE or ARITYPEvalues, the result value of sub-expression reduction, or dereferenced OPER or TYPEDEF objects.
  2. An empty value stack is initialized for this reduction.
  3. The expression is treated as a Reverse Polish Notation (RPN) sequence, where the following is performed on each item in the AC in sequence:

    If the item is an ARITYPE value or dereferenced OPER or TYPEDEF object it is evaluated in accordance with Section 6.7.4, Section 6.7.2 or Section 6.7.3 respectively. If the evaluation fails, this procedure stops and is considered failed.

    Otherwise, the item is pushed onto the stack.

  4. After RPN processing if the value stack is empty or has more than one item, this procedure stops and is considered failed. Otherwise, the result of the evaluation is the single literal value in the stack.

One effect of this procedure is that if any referenced values cannot be produced the procedure fails before any OPER is evaluated. Another effect of this procedure is that if any referenced OPER fails during evaluation or any value production fails the EXPR processing fails immediately and subsequent OPER values, EXPR values, or VALUE-OBJ references are not evaluated.

6.7.2. OPER Objects

This procedure applies only during the evaluation of a containing expression (Section 6.7.1); an OPER cannot be evaluated in isolation.

The evaluation of an OBJECT value referencing a Operator (OPER) is as follows:

  1. The value is dereferenced in accordance with Section 6.3 and its parameters are handled in accordance with Section 6.4. If either fails, this procedure stops and is considered failed.
  2. The processor passes the evaluation on to the underlying implementation of the OPER being evaluated.

    The context available to the implementation is the following:

    Object Path:
    This gives visibility into the OPER object reference which was dereferenced during the evaluation.
    Parameters:
    The set of actual parameters augmented for the evaluation.
    Expression Stack:
    The operands are popped from this stack and the result is pushed here before completion.

If the evaluation procedure fails, the failure SHALL propagate up to any expression evaluation.

6.7.3. TYPEDEF Objects

This procedure applies only during the evaluation of a containing expanded expression; a TYPEDEF object cannot be evaluated in isolation. The evaluation of a TYPEDEF is handled similarly to a unary OPER but it occurs entirely within the Agent and does not rely on an object-specific implementation.

The evaluation of a TYPEDEF object is as follows:

  1. If the TYPEDEF value itself has no parameters, the input value is popped from the stack. If the TYPEDEF value itself has one parameter, the input value is that parameter. If the TYPEDEF value itself has more than parameter, this procedure stops and is considered failed.
  2. The result value is a type conversion (Section 6.11.3) on the input value. If the conversion fails this procedure stops and is considered failed.
  3. The result value is pushed onto the stack.

6.7.4. ARITYPE Values

This procedure applies only during the evaluation of a containing expanded expression; an ARITYPE value cannot be evaluated in isolation. The evaluation of an ARITYPE is handled similarly to a TYPEDEF but with no possibility of a parameterized conversion.

The evaluation of an ARITYPE value is as follows:

  1. The input value is popped from the stack.
  2. The result value is a type conversion (Section 6.11) on the input value. If the conversion fails this procedure stops and is considered failed.
  3. The result value is pushed onto the stack.

6.8. Reporting

Within the AMM the following entities have a reporting context: RPTT and EXPR values and CONST, EDD, and VAR objects. The value-producing objects are reported-on by reference, while RPTT are reported-on both by value and by reference.

6.8.1. RPTT Values

The reporting on a Report Template (RPTT) value, which is structured as an AC, is as follows:

  1. An empty item list is initialized for this template.
  2. The processor iterates through all items of the AC, performing the following:

    If the item is an EXPR value it is replaced by the result of evaluation in accordance with Section 6.7. If the evaluation fails the undefined value is used as a substitute.

    Otherwise, if the item is a VALUE-OBJ it is replaced by the value produced in accordance with Section 6.5. If the production fails the undefined value is used as a substitute.

    Otherwise, the result of the production is the value appended to the item list.

Because this procedure acts on an RPTT value and not an object reference, the report itself cannot be assembled within this context. One effect of this procedure is that if any item of the RPTT cannot be reported on, the undefined value is used as a sentinel and the other report items are still generated.

6.8.2. Value-Producing Objects

This activity relies on an object reference value to have been dereferenced in accordance with Section 6.3 and its parameters handled in accordance with Section 6.4.

The reporting on an object producing a value of any type is as follows:

  1. The value is produced from the object in accordance with Section 6.5. This step includes substitution of any LABEL parameters within the value.
  2. If the value is an RPTT (Section 4.2.3.3) type, this value is used to generate an AC which contains report items in accordance with Section 6.8.1.

    Otherwise, the produced value is used as the single RPT item.

  3. The report (Section 3.2.2.5) is produced by combining: the source ARI used for this procedure, the current timestamp, and the items generated in the previous step.

6.9. Agent-Manager Message Handling

6.9.1. Execution-Set Aggregation

Managers SHOULD aggregate multiple Execution-Set (EXECSET) values associated with the same Agent and Correlator Nonce into a single Execution-Set. The aggregation MAY be based on a size limit (e.g., number of targets), time limit, or an event (e.g., network availability). This avoids the overhead of transport and processing multiple executions on the same Agent, and due to the requirements in Section 6.9.2 makes no difference to (lack of) guarantees in execution order.

6.9.2. Execution-Set Processing

An Agent SHALL process an Execution-Set through the independent Execution of each item in the target list. Execution order is not guaranteed and failures on one target do not affect other target, so targets MAY be executed in any order or concurrently. This is not the same behavior as the execution of a macro, where execution of items is ordered and a failure of any execution causes subsequent items to not be executed.

6.9.3. Reporting-Set Aggregation

Agents SHOULD aggregate multiple Reporting-Set (RPTSET) values associated with the same Manager and Correlator Nonce into a single Reporting-Set. The aggregation MAY be based on a size limit (e.g., number of reports or number of total report items), time limit, or an event (e.g., network availability or power-saving wake-up). This avoids the overhead of transport and processing multiple messages on a Manager and improves timestamp compression in the reports, but it does require that all of the items are associated with the same manager and nonce.

6.9.4. Reporting-Set Processing

A Manager SHALL process each report within a Reporting-Set independently. Failures in processing any one report do not affect other reports, so reports MAY be processed in any order or concurrently. After using a Report Template to correlate report items with source objects, a Manager SHALL treat each (timestamp, object, item value) tuple independently from its containing Reporting-Set or Report.

6.10. Type Matching

Type matching is done through pattern matching and does not affect the AMM value. AMM values are not strictly typed, and as long as an AMM value matches the pattern for a type, that value can be used where that type is needed. If there is any overlap in the patterns for different semantic types, then there will be ambiguity in the sense that the same value can be used as different types.

6.10.1. Built-In Types

The matching of a built-in literal type to any object reference value SHALL be considered to fail. The built-in LITERAL type SHALL match any typed or untyped literal value.

The matching of a built-in literal type to a typed literal value as follows:

  1. If the value type differs from the built-in type, the match fails.
  2. Otherwise, the literal type is matched to the value primitive as defined below.

The matching of a built-in literal type to an untyped literal value as follows:

NULL:
This type only matches the null primitive value.
BOOL:
This type only matches the true and false primitive values.
BYTE:
This type only matches uint primitive values in the domain 0 to 2^8-1 inclusive.
INT:
This type only matches int primitive values in the domain -2^31 to 2^31-1 inclusive.
UINT:
This type only matches uint primitive values in the domain 0 to 2^32-1 inclusive.
VAST:
This type only matches int primitive values in the domain -2^63 to 2^63-1 inclusive.
UVAST:
This type only matches uint primitive values in the domain 0 to 2^64-1 inclusive.
REAL32:
This type only matches float primitive values in the domain of a 32-bit [IEEE.754-2019] floating point number.
REAL64:
This type only matches float primitive values in the domain of a 64-bit [IEEE.754-2019] floating point number.
TEXTSTR:
This type matches tstr primitive values.
BYTESTR:
This type matches bstr primitive values.
TP:
This type matches TBD values.
TD:
This type matches TBD values.
LABEL:
This type matches int and tstr values.
CBOR:
This type matches bstr primitive values.
ARITYPE:
This type matches int and tstr values.

The matching of a built-in object reference type to any literal value SHALL be considered to fail. The built-in OBJECT type SHALL match any object reference values.

The matching of a built-in object reference type to an object reference value SHALL be considered to succeed if the object reference value Type ID is identical to the type.

6.10.2. Semantic Types

The matching of an input value to each class of semantic type (Section 3.3) is as follows:

Named Type Use:
Matching for a named type use SHALL be identical to the matching for the type being named, whether that is TYPEDEF or built-in.
Uniform List:
Matching for this class SHALL require the value to be an AC, with an item count optionally constrained by minimum and maximum size from the type, and with each item of the AC itself matching the specific sub-type for the list.
Diverse List:
Matching for this class SHALL require the value to be an AC, with an item count optionally constrained by minimum and maximum size from the type, and with each item of the AC itself matching the specific sub-type or sequence for the list.
Uniform Map:
Matching for this class SHALL require the value to be an AM, with a size optionally constrained by minimum and maximum size from the type, and with each key and value of the AM itself matching the specific respective sub-type for the map.
Table Template:
Matching for this class SHALL require the value to be an TBL, with a column count matching exactly the number of columns present in the table template and each row containing items matching the column-specific sub-type for the template. If the table template contains a limit on minimum or maximum size, the row count SHALL conform with those limits to match. If the table template contains a key column or unique column-set then all rows SHALL satisfy the uniqueness of those constraints to match.
Type Union:

The matching for a type union SHALL be performed as follows:

  1. The underlying literal types usable with a semantic type are obtained by recursively flattening the type union(s) down to built-in types and removing duplicate built-in types after the first instance of them. This defines a list of acceptable built-in types for the result.
  2. The input value is matched to each built-in type in the list, and the first successful match results in a success of matching the whole semantic type. If none of the built-in types can successfully match the input value, the match is considered failed.
  3. If successful, the specific base type which matched is also part of the result of this procedure.

6.11. Type Conversion

The type system of the AMM allows conversions of values between different literal and semantic types in a way which is supposed to preserve the "meaning" of the value.

In some cases, type conversion is performed implicitly by the Agent while other cases the conversion is explicitly part of an expression. One example of implicit casting is during Parameter Handling to ensure each processed parameter meets the formal parameter type signature. Another example of implicit conversion is for numeric operators in the Agent ADM (Section 4.3.3).

6.11.1. BOOL Type

The AMM has the concepts of "truthy" and "falsey" as being the result of casting to BOOL type. Similar to the ToBoolean() function from [ECMA-262], the AMM casting treats the following as falsey and every other value as truthy:

  • The undefined value
  • The null value (of NULL)
  • The false value (of BOOL)
  • Zero value of BYTE, UINT, INT, UVAST, and VAST
  • Positive and negative zero, and NaN values of REAL32 and REAL64
  • Empty value of TEXTSTR and BYTESTR
  • Zero value of TD

When casting a value to BOOL type, the processor SHALL use the result value false if the original value is falsey and true otherwise.

6.11.2. NUMERIC Types

The casting of a value to a NUMERIC type is intended to easily allow mixed-type expressions while keeping the number of operators and parameter unions small.

When casting a value to an INTEGER type from any other NUMERIC type, the processor SHALL perform the following:

  1. If the input is one of the FLOAT types and is not finite, the conversion is considered failed.
  2. If the input is one of the FLOAT types, the value is truncated to an integer by rounding toward zero.
  3. If the input value is outside the domain of the output type, the conversion is considered failed.

When casting a value to an FLOAT type from any other NUMERIC type, the processor SHALL perform the following:

  1. If the input value is outside the domain of the output type, the conversion is considered failed.
6.11.2.1. Numeric Promotion

While the earlier discussion of numeric type casting is about converting from an input type to an output type, the concept of a type promotion is about finding a "least compatible type" which can accommodate most, if not all, of the input type range. Converting to a promoted type is called an "up" conversion, and from a promoted type a "down" conversion.

The promotion order for NUMERIC types is as follows:

  • A promoted type has a larger span of values (the difference between largest and smallest representable value).
  • A promoted type can gain signed-ness but not lose it.
  • A promoted type can lose precision for some values.

This promotion logic does not guarantee that an up-conversion will always succeed (e.g. some large UVAST values will not fit within a VAST or REAL32) but does provide a strict ordering for finding a compatible type between two NUMERIC values. The "least compatible type" between two types SHALL be defined as the smallest up-conversion that will accommodate the input types, as indicated in Table 5. This is almost a strict ordering except for the conversion of INT and UVAST to VAST to accommodate both the signed-ness and the size of the inputs.

Table 5: NUMERIC Type Promotion
BYTE UINT INT UVAST VAST REAL32 REAL64
BYTE BYTE UINT INT UVAST VAST REAL32 REAL64
UINT UINT INT UVAST VAST REAL32 REAL64
INT INT VAST VAST REAL32 REAL64
UVAST UVAST VAST REAL32 REAL64
VAST VAST REAL32 REAL64
REAL32 REAL32 REAL64
REAL64 REAL64

6.11.3. Semantic Types

The converting of an input value to each class of semantic type (Section 3.3) is as follows. Similar to built-in type conversion, each semantic type class has the potential to change a value as needed to conform to type limitations.

Named Type Use:
Uniform List:
Converting to this class SHALL require the value to be an AC; if the input is not an AC this procedure stops and is considered failed. If the uniform list contains limits on the number of items and the value does not satisfy those limits this procedure stops and is considered failed. For each item of the input value the uniform sub-type SHALL be used to convert to a corresponding output value; if any of those conversions fail this procedure stops and is considered failed.
Diverse List:
Converting to this class SHALL require the value to be an AC; if the input is not an AC this procedure stops and is considered failed. If the diverse list contains limits on the number of items and the value does not satisfy those limits this procedure stops and is considered failed. For each item of the input value the corresponding diverse list sub-type SHALL be used to convert to a corresponding output value; if any of those conversions fail this procedure stops and is considered failed.
Uniform Map:
Converting to this class SHALL require the value to be an AM, if the input is not an AC this procedure stops and is considered failed. If the uniform map contains limits on the number of pairs and the value does not satisfy those limits this procedure stops and is considered failed. For each key--value pair of the input value the uniform sub-types SHALL be used to convert to a corresponding output pair; if any of those conversions fail this procedure stops and is considered failed. If the conversion results in any duplicate keys this procedure stops and is considered failed.
Table Template:
Converting to this class SHALL require the value to be a TBL, if the input is not an TBL this procedure stops and is considered failed. If the table template contains limits on the number of rows and the value does not satisfy those limits this procedure stops and is considered failed. If the value contains a different number of columns from the table template this procedure stops and is considered failed. For each item across each row of the input the corresponding column sub-type SHALL be used to convert to a corresponding output table item; if any of those conversions fail this procedure stops and is considered failed.
Type Union:

Converting a value for a type union SHALL be performed as follows:

  1. The underlying built-in types usable with a semantic type are obtained by recursively flattening type union(s) down to built-in types and removing duplicate built-in types after the first instance of them. This defines a priority list of acceptable built-in types for the result. Because a TYPEDEF union is unchanging within an ADM (see Section 3.4.2) a processor MAY cache this flattened type list.
  2. The input value is cast to each built-in type in the list, and the first successful cast is taken as the built-in type of the result value. If none of the built-in types can successfully cast the input value, this procedure stops and is considered failed.
  3. The result value is associated with the the type being cast to, and that is the result of this procedure.

6.12. Translating ARIs and Semantic Types

NOTE: These procedures need to be validated by a trial implementation.

The procedures in this section allow AMM values, which can have a semantic type (Section 3.3), to be translated into and out of the ARI syntax (Section 3.1.3), which has no semantic type information. They also describe a way to use type-less literal values to give further compression of values in certain circumstances.

The compression of removing type information is possible only when the context in which the value is being used has a specific semantic or built-in type associated with it. For example, when a formal object parameter or a report item is typed to either a built-in type or a semantic type that doesn't represent a type union then a value being used for the parameter or item can only have that specific type; any other value type will be mismatched and invalid. Another way of looking at this compression is when the value has the same type as its context requires, then the value's type is redundant and can be elided without loss of information.

In addition to compression by eliding semantic type within a context, there are also some built-in types which have values which only exist in that type. For example, the BOOL value true exists only within that type while the UINT value 5 is also within the domain of INT and several others.

For the procedures below, the contexts which provide type information SHALL be: all formal parameters, VALUE-OBJ values, report template items, tabular columns.

When translating from an AMM value into an ARI, the processor performs the following:

  1. If the use context is associated with a TYPEDEF and the value is associated with an ambiguous or incompatible TYPEDEF, the ARI form SHALL be wrapped in a TYPEDEF-as-cast ARI. An ambiguous type is one where the presence of the inner type cast produces a different result than if the cast were absent. An incompatible type is one where the presence of the inner cast results in a failure of the outer cast.
  2. If the use context is associated with a single built-in type (no unions) and the built-in type is present in Table 2 "Literal Implied Types" of [I-D.ietf-dtn-ari], the ARI form MAY have its literal type removed.

When translating from an ARI into an AMM value, the processor performs the following:

  1. If the use context is associated with a type and the value is associated with a different TYPEDEF or no TYPEDEF, the value SHALL be cast to the context type in accordance with Section 6.11.

7. ADM Author Considerations

The AMM model provides multiple ways to represent certain types of data. This section provides informative guidance on how to express application management constructs efficiently when authoring an ADM document.

7.1. CTRL Definitions Need to Consider Idempotency

All CTRLs SHOULD be given names and behaviors that reflect the idempotency requirements of Section 3.4.6. For example the term "ensure" is preferable to "add or modify". Likewise "discard" is encouraged instead of "remove if necessary".

Agent behavior SHOULD be reasonable even if duplicate and concurrent CTRL executions are performed. Consider an "add_" CTRL that fails if a value already exists as opposed to an "ensure_" CTRL that checks a precondition and stops, thus guaranteeing idempotency.

7.2. EDD Definitions Need to Consider Nilpotency

All EDDs SHOULD be given names and behaviors that reflect the nilpotency requirements of Section 3.4.4. Agent behavior SHOULD be reasonable even if duplicate and concurrent EDD value production is performed.

7.3. Use Parameters for Dynamic Information

Parameters provide a powerful mechanism for expressing associative look-ups of EDD data. EDDs SHOULD be parameterized when the definition of the EDD is dependent upon run-time information.

For example, if requesting the number of bytes through a specific endpoint, the construct num_bytes("endpoint_name") is simpler to understand and more robust to new endpoint additions than attempting to enumerate the number and name of potential endpoints when defining the ADM.

7.4. Do Not Use Parameters for Static Information

Parameters incur transport and processing costs (see Section 6.4) and should only be used where necessary. If an EDD object can be parameterized, but the set of parameters is known and unchanging it may be more efficient to define multiple non-parameterized EDD objects instead.

For example, consider a single parameterized EDD object reporting the number of bytes of data received for a specific, known set of priorities and a request to report on those bytes for the "low", "med", and "high" priorities. Below are two ways to represent these data: using parameters and not using parameters.

Table 6: Example Parameterized EDDs
Parameterized Uses Non-Parameterized Uses
./EDD/num_bytes_by_pri(low) ./EDD/num_bytes_by_low_pri
./EDD/num_bytes_by_pri(med) ./EDD/num_bytes_by_med_pri
./EDD/num_bytes_by_pri(high) ./EDD/num_bytes_by_high_pri

The use of parameters in this case only incurs the overhead of type checking, parameter encoding/decoding, and associative lookup. This situation should be avoided when deciding when to parameterize AMM objects.

In cases where multiple EDD or VAR values are likely to be produced and evaluated together, then that information SHOULD be placed in an Table Template (Section 3.3.5) rather than defining multiple EDD and/or VAR objects. By making a Table Template, the relationships among various data values are preserved. Otherwise, Managers would need to remember to query multiple EDD and/or VAR objects together which is burdensome, but also results in increased transport and processor utilization and the potential for non-synchronized access across multiple value productions.

8. IANA Considerations

This section provides guidance to the Internet Assigned Numbers Authority (IANA) regarding registration of schema and namespaces related to core ADMs, in accordance with BCP 26 [RFC1155].

8.1. DTN Management Architecture Parameters

This document relies on existing ARI-defined sub-registries defined in [IANA-DTNMA] by Section 9.3 of [I-D.ietf-dtn-ari].

This document registers the following entries within the "Application Data Models" sub-registry of the "DTN Management Architecture Parameters" registry [IANA-DTNMA].

Table 7: Application Data Models
Enumeration Name Reference Notes
1 ietf-dtnma-agent [This document]

9. Security Considerations

This document does not describe any on-the-wire encoding or other messaging syntax. It is assumed that the exchange of AMM objects between Agents and Managers occurs within the context of an appropriate network environment.

The Access Control Lists (ACLs) functionality presented in this document would be implemented separately from network security mechanisms.

ACL groups are expected to be associated with Managers. However, the form of Manager identification must be provided by separate transport-specific ADMs. The AMM provides no general purpose identifier, such as peer name and address, that would be required to uniquely describe each Manager.

10. References

10.1. Normative References

[IANA-DTNMA]
IANA, "Delay-Tolerant Networking Management Architecture (DTNMA) Parameters", <https://www.iana.org/assignments/TBA/>.
[IEEE.754-2019]
IEEE, "IEEE Standard for Floating-Point Arithmetic", IEEE IEEE 754-2019, DOI 10.1109/IEEESTD.2019.8766229, , <https://ieeexplore.ieee.org/document/8766229>.
[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, , <https://www.rfc-editor.org/info/rfc2119>.
[RFC3629]
Yergeau, F., "UTF-8, a transformation format of ISO 10646", STD 63, RFC 3629, DOI 10.17487/RFC3629, , <https://www.rfc-editor.org/info/rfc3629>.
[RFC3986]
Berners-Lee, T., Fielding, R., and L. Masinter, "Uniform Resource Identifier (URI): Generic Syntax", STD 66, RFC 3986, DOI 10.17487/RFC3986, , <https://www.rfc-editor.org/info/rfc3986>.
[RFC3339]
Klyne, G. and C. Newman, "Date and Time on the Internet: Timestamps", RFC 3339, DOI 10.17487/RFC3339, , <https://www.rfc-editor.org/info/rfc3339>.
[RFC7950]
Bjorklund, M., Ed., "The YANG 1.1 Data Modeling Language", RFC 7950, DOI 10.17487/RFC7950, , <https://www.rfc-editor.org/info/rfc7950>.
[RFC8341]
Bierman, A. and M. Bjorklund, "Network Configuration Access Control Model", STD 91, RFC 8341, DOI 10.17487/RFC8341, , <https://www.rfc-editor.org/info/rfc8341>.
[I-D.ietf-dtn-ari]
Birrane, E. J., Annis, E., and B. Sipos, "DTNMA Application Resource Identifier (ARI)", Work in Progress, Internet-Draft, draft-ietf-dtn-ari-01, , <https://datatracker.ietf.org/doc/html/draft-ietf-dtn-ari-01>.

10.2. Informative References

[ECMA-262]
Ecma International, "ECMA-262 12th Edition, June 2021. ECMAScript 2021 language specification", , <https://262.ecma-international.org/12.0/>.
[RFC1155]
Rose, M. and K. McCloghrie, "Structure and identification of management information for TCP/IP-based internets", STD 16, RFC 1155, DOI 10.17487/RFC1155, , <https://www.rfc-editor.org/info/rfc1155>.
[RFC2578]
McCloghrie, K., Ed., Perkins, D., Ed., and J. Schoenwaelder, Ed., "Structure of Management Information Version 2 (SMIv2)", STD 58, RFC 2578, DOI 10.17487/RFC2578, , <https://www.rfc-editor.org/info/rfc2578>.
[RFC2580]
McCloghrie, K., Ed., Perkins, D., Ed., and J. Schoenwaelder, Ed., "Conformance Statements for SMIv2", STD 58, RFC 2580, DOI 10.17487/RFC2580, , <https://www.rfc-editor.org/info/rfc2580>.
[RFC6991]
Schoenwaelder, J., Ed., "Common YANG Data Types", RFC 6991, DOI 10.17487/RFC6991, , <https://www.rfc-editor.org/info/rfc6991>.
[RFC8610]
Birkholz, H., Vigano, C., and C. Bormann, "Concise Data Definition Language (CDDL): A Notational Convention to Express Concise Binary Object Representation (CBOR) and JSON Data Structures", RFC 8610, DOI 10.17487/RFC8610, , <https://www.rfc-editor.org/info/rfc8610>.
[I-D.ietf-dtn-dtnma]
Birrane, E. J., Heiner, S., and E. Annis, "DTN Management Architecture", Work in Progress, Internet-Draft, draft-ietf-dtn-dtnma-14, , <https://datatracker.ietf.org/doc/html/draft-ietf-dtn-dtnma-14>.

Appendix A. Access Control Lists

This section presents an overview of fine-grained application security using Access Control Lists (ACLs).

Access Control shall be a function of the Agent. A table of entries associating permission tags with groups of objects shall be queried at runtime to ensure privileged access, while simultaneously allowing efficient implementation on an embedded device.

The concepts presented are in agreement with the Network Configuration Access Control Model (NACM) documented in [RFC8341].

A.1. Tags

Permissions are defined using an Access Control Tag (AT). Each AT is a bit mask corresponding to a series of flags. For example three bits would be required for read, write, and execute permissions. Tags are similar in principle to file permissions on Unix, which tracks flags for Read/Write/Execute.

An AT could be stored in a CBOR unsigned integer. For example 0x800C would be a valid tag for an AT with four access controls of four bits each.

A.2. Groups

Groups provide a general-purpose configuration to map an AT to a set of objects.

An annotative name may be associated with a group, and a numeric group ID is used for for cross-referencing.

A.2.1. Associations

The following entities within an Agent may retain group associations:

  • The Agent shall be associated with a group.
  • Each node in the network may have a group association. A transport-specific ADM shall define how to map from authenticated Manager identifier to an access control group.
  • ADM objects shall be associated with the group of the Agent.
  • ODM objects shall belong to the group associated with their execution context of creation. This prevents a Manager from exploiting permissions by, for example creating a one-second TBR to execute a task requiring elevated permissions.

A CTRL is provided to allow an object's group to be re-assigned. If a group is deleted all permissions associated with the group shall also be deleted, and objects previously belonging to the group shall inherit the default permissions.

A.2.2. Permissions

Access control permissions shall be assigned using the combination of a Group, ARI Pattern (defined in [I-D.ietf-dtn-ari]), and Access Control Tag. At runtime the Agent shall retain a table of these tuples to store all necessary permissions. The table shall be queried by the Agent to find the corresponding AT for each managed object. If an object does not have an AT in the table then the default AT shall be used.

The Agent shall use the Group corresponding to the appropriate execution context when querying for permissions. For more information see Appendix A.2.3.

An object's ARI shall be used by the Agent when querying the table. An important consideration is that an ARI contains a namespace-id and object-id, both of which may be expressed in either a text or numeric form. When looking up permissions the Agent must use the ARI of the object itself - which can match all four possible forms - and NOT limit lookup to a particular ARI used by the Manager to reference an object.

If the Agent discovers multiple tuples that correspond to an object, the AT with least permission should be applied to the object. For example, if three tuples would allow an operation and the fourth would not, the Agent should deny permission.

Permissions shall be loaded during agent initialization and may be changed by an operator with sufficient permission. The default access level shall deny all operations.

An agent implementation should provide a way to audit assignment of permissions.

A.2.3. Execution Context

Each execution context shall be associated with a group. For direct execution the group is expected to correspond to the Manager that caused the execution to occur. For delayed execution such as a TBR the execution context shall refer to the group of the applicable object. When creating reports the ability to produce the report and send to a Manager is driven by permissions of the group of the Manager receiving the report.

TBD There may be cases where an object is initialized as protected but assigned to a variable that is not protected, allowing another Manager a means of working around access controls. In this early version of the document we do not protect against actions of a malicious Agent, or a privileged manager abusing its privileges. An Agent shall prevent an unprivileged Manager from abusing permissions to perform an unprivileged action.

A.3. Enforcement

Access control shall be enforced in the following way for processing activities described in Section 6.

A.3.1. Dereferencing

When a variable is dereferenced the Agent shall look up the AT associated with the object. This is similar to other name-based access control systems such as AppArmor in Linux.

A.3.2. Parameter Handling

Read permission shall be required for an object to be passed as a parameter.

A.3.3. Value Production, Execution, and Evaluation

Execute permission shall be required for the object producing a value, executing, or being evaluated. There is one exception: if an OBJ-REF produces itself then Read permission is required.

Write permission is required for any object that could be modified by an operation. Note that result storage is ephemeral and parameters are passed by value, so any modifications to a VAR would be made by a limited number of special CTRLs in the agent ADM such as store_var.

A.3.4. Reporting

Read permission is required if the OBJ-REF is a single value that will be reported directly.

Execute permission is required if the OBJ-REF is a RPTT that will be used to generate a RPT.

A.4. Roles

Manager roles are implementation-specific and do not need to be specified in the ADM. However the likely manger roles are Trusted (all permissions) and Read-Only.

Acknowledgments

The following participants contributed technical material, use cases, and useful thoughts on the overall approach captured in this document: David Linko, Sarah Heiner, and Jenny Cao of the Johns Hopkins University Applied Physics Laboratory.

Authors' Addresses

Edward J. Birrane, III
The Johns Hopkins University Applied Physics Laboratory
11100 Johns Hopkins Rd.
Laurel, MD 20723
United States of America
Brian Sipos
The Johns Hopkins University Applied Physics Laboratory
Justin Ethier
The Johns Hopkins University Applied Physics Laboratory