1 Introduction

oBIX is designed to provide access to the embedded software systems which sense and control the world around us. Historically integrating to these systems required custom low level protocols, often custom physical network interfaces. But now the rapid increase in ubiquitous networking and the availability of powerful microprocessors for low cost embedded devices is weaving these systems into the very fabric of the Internet. Generically the term M2M for Machine-to-Machine describes the transformation occurring in this space because it opens a new chapter in the development of the Web - machines autonomously communicating with each other. The oBIX specification lays the groundwork for building this M2M Web using standard, enterprise friendly technologies like XML, HTTP, and URIs.Design Concerns

The following design points illustrate the problem space oBIX attempts to solve:

XML: representing M2M information in a standard XML syntax;
Networking: transferring M2M information in XML over the network;
Normalization: standard representations for common M2M features: points, histories, and alarms;
Foundation: providing a common kernel for new standards;

1.1.1 XML

The principal requirement of oBIX is to develop a common XML syntax for representing information from diverse M2M systems. The design philosophy of oBIX is based on a small but extensible data model which maps to a simple fixed XML syntax. This core object model and its XML syntax is simple enough to capture entirely in one illustration provided in Section 4. The object model’s extensibility allows for the definition of new abstractions through a concept called contracts. The majority of the oBIX specification is actually defined in oBIX itself through contracts.

1.1.2 Networking

Once we have a way to represent M2M information in XML, the next step is to provide standard mechanisms to transfer it over networks for publication and consumption. oBIX breaks networking into two pieces: an abstract request/response model and a series of protocol bindings which implement that model. Version 1.1 of oBIX defines two protocol bindings designed to leverage existing Web Service infrastructure: an HTTP REST binding and a SOAP binding.

1.1.3 Normalization

There are a few concepts which have broad applicability in systems which sense and control the physical world. Version 1.1 of oBIX provides a normalized representation for three of these:

Points: representing a single scalar value and its status – typically these map to sensors, actuators, or configuration variables like a setpoint;
Histories: modeling and querying of time sampled point data. Typically edge devices collect a time stamped history of point values which can be fed into higher level applications for analysis;
Alarming: modeling, routing, and acknowledgment of alarms. Alarms indicate a condition which requires notification of either a user or another application.

1.1.4 Foundation

The requirements and vertical problem domains for M2M systems are immensely broad – too broad to cover in one single specification. oBIX is deliberately designed as a fairly low level specification, but with a powerful extension mechanism based on contracts. The goal of oBIX is to lay the groundwork for a common object model and XML syntax which serves as the foundation for new specifications. It is hoped that a stack of specifications for vertical domains can be built upon oBIX as a common foundation.

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.

1.3 Normative References

RFC2119 Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels”, BCP 14, RFC 2119, March 1997. http://www.ietf.org/rfc/rfc2119.txt.

RFC2246 Dierks, T., Allen, C., “Transport Layer Security (TLS) Protocol Version 1.0”, IETF RFC 2246, January 1999. http://www.ietf.org/rfc/rfc2246.txt.

RFC3986 Berners-Lee, T., Fielding, R., Masinter, L., “Uniform Resource Identifier (URI): Generic Syntax”, IETF RFC 3986, January 2005. http://www.ietf.org/rfc/rfc3986.txt.

SOA-RM Reference Model for Service Oriented Architecture 1.0, October 2006. OASIS Standard. http://docs.oasis-open.org/soa-rm/v1.0/soa-rm.pdf

oBIX REST Bindings for oBIX: REST Bindings Version 1.0.
See link in "Related work" section on cover page.

oBIX SOAP Bindings for oBIX: SOAP Bindings Version 1.0.
See link in "Related work" section on cover page.

oBIX Encodings Encodings for oBIX: Common Encodings Version 1.0.
See link in "Related work" section on cover page.

WSDL Christensen, E., Curbera, F., Meredith, G., Weerawarana, S., “Web Services Description Language (WSDL), Version 1.1”, W3C Note, 15 March 2001. http://www.w3.org/TR/wsdl.

XLINK DeRose, S., Maler, E., Orchard, D., Walsh, N. “XML Linking Language (XLink) Version 1.1”, May 2010. http://www.w3.org/TR/xlink11/ .

XPOINTER DeRose, S., Maler, E., Daniel Jr., R., “XPointer xpointer() Scheme”, December 2002. http://www.w3.org/TR/xptr-xpointer/.

XML Schema Biron, P.V., Malhotra, A., “XML Schema Part 2: Datatypes Second Edition”, October 2004. http://www.w3.org/TR/xmlschema-2/.

1.4 Non-Normative References

REST Fielding, R.T., “Architectural Styles and the Design of Network-based Software Architectures”, Dissertation, University of California at Irvine, 2000. http://www.ics.uci.edu/~fielding/pubs/dissertation/top.htm

SOAP Gudgin, M., Hadley, M., Mendelsohn, N., Moreau, J., Nielsen, H., Karmarkar, A., Lafon, Y., “SOAP Version 1.2 (Second Edition)”, W3C Recommendation 27 April 2007. http://www.w3.org/TR/soap12/.

UML Unified Modeling Language (UML), Version 2.2, Object Management Group, February, 2009. http://www.omg.org/technology/documents/formal/uml.htm .

1.5 Changes from Version 1.0

Changes to this specification since the initial version 1.0 are listed below, along with a brief description.

· Add date, time primitive types and tz facet to the core object model.

· Add binary encoding – Note this is now part of the Encodings for oBIX document.

· Add support for History Append operation.

· Add HTTP content negotiation – Note this is now part of the oBIX REST document.

· Add the of attribute to the ref element type and specify usage of the is attribute for ref.

· Add metadata inclusion for alternate hierarchies (tagging).

· Add compact history record encoding.

· Add support for alternate history formats.

· Add support for concise encoding of long contract lists.

· Add Delete request semantics.

· Clean up references and usage in text.

· Add conformance clauses.

2 Quick Start

This chapter is for those eager beavers who want to immediately jump right into oBIX and all its angle bracket glory. The best way to begin is to take a simple example that anybody is familiar with – the staid thermostat. Let’s assume we have a very simple thermostat. It has a temperature sensor which reports the current space temperature and it has a setpoint that stores the desired temperature. Let’s assume our thermostat only supports a heating mode, so it has a variable that reports if the furnace should currently be on. Let’s take a look at what our thermostat might look like in oBIX XML:

</obj>

The first thing to notice is that there are three element types. In oBIX there is a one-to-one mapping between objects and elements. Objects are the fundamental abstraction used by the oBIX data model. Elements are how those objects are expressed in XML syntax. This document uses the term object and sub-objects, although you can substitute the term element and sub-element when talking about the XML representation.

The root obj element models the entire thermostat. Its href attribute identifies the URI for this oBIX document. There are three child objects for each of the thermostat’s variables. The real objects store our two floating point values: space temperature and setpoint. The bool object stores a boolean variable for furnace state. Each sub-element contains a name attribute which defines the role within the parent. Each sub-element also contains a val attribute for the current value. Lastly we see that we have annotated the temperatures with an attribute called unit so we know they are in Fahrenheit, not Celsius (which would be one hot room). The oBIX specification defines a bunch of these annotations which are called facets.

In real life, sensor and actuator variables (called points) imply more semantics than a simple scalar value. In other cases such as alarms, it is desirable to standardize a complex data structure. oBIX captures these concepts into contracts. Contracts allow us to tag objects with normalized semantics and structure.

Let’s suppose our thermostat’s sensor is reading a value of -412°F? Clearly our thermostat is busted, so we should report a fault condition. Let’s rewrite the XML to include the status facet and to provide additional semantics using contracts:

<real name="spaceTemp" is="obix:Point"

val="-412.0" status="fault"

unit="obix:units/fahrenheit"/>

<real name="setpoint" is="obix:Point"

val="72.0"

unit="obix:units/fahrenheit"/>

</obj>

Notice that each of our three scalar values are tagged as obix:Points via the is attribute. This is a standard contract defined by oBIX for representing normalized point information. By implementing these contracts, clients immediately know to semantically treat these objects as points.

Contracts play a pivotal role in oBIX for building new abstractions upon the core object model. Contracts are slick because they are just normal objects defined using standard oBIX. In fact, the following sections defining the core oBIX object model are expressed using Contracts. One can see how easily this approach allows for definition of the key parts of this model, or any model that builds upon this model.

3 Architecture

The oBIX architecture is based on the following principles:

Object Model: a concise object model used to define all oBIX information.
Encoding: a set of rules for representing the object model in certain common formats.
URIs: URIs are used to identify information within the object model.
REST: a small set of verbs is used to access objects via their URIs and transfer their state.
Contracts: a template model for expressing new oBIX “types”.
Extendibility: providing for consistent extendibility using only these concepts.

3.1 Object Model

All information in oBIX is represented using a small, fixed set of primitives. The base abstraction for these primitives is cleverly called object. An object can be assigned a URI and all objects can contain other objects.

There are ten special kinds of value objects used to store a piece of simple information:

bool: stores a boolean value - true or false;
int: stores an integer value;
real: stores a floating point value;
str: stores a UNICODE string;
enum: stores an enumerated value within a fixed range;
abstime: stores an absolute time value (timestamp);
reltime: stores a relative time value (duration or time span);
date: stores a specific date as day, month, and year;
time: stores a time of day as hour, minutes, and seconds;
uri: stores a Universal Resource Identifier;

Note that any value object can also contain sub-objects. There are also a couple of other special object types: list, op, feed, ref and err.

3.2 Encoding

A necessary facet of oBIX is set of simple syntax rules to represent the underlying object model. XML is a widely used language with well-defined and well-understood syntax that maps nicely to the oBIX object model. The rest of this document will use XML as the example encoding, because it is easily human-readable, and serves to clearly demonstrate the concepts presented. The syntax used is normative. Implementations using an XML encoding MUST conform to this syntax and representation of elements.

When encoding oBIX objects in XML, each of the object types map to one type of element. The value objects represent their data value using the val attribute. All other aggregation is simply nesting of elements. A simple example to illustrate:

</obj>

</list>

</obj>

Note in this simple example how the href attribute specifies URI references which may be used to fetch more information about the object. Names and hrefs are discussed in detail in Section 5.

3.3 URIs

No architecture is complete without some sort of naming system. In oBIX everything is an object, so we need a way to name objects. Since oBIX is really about making information available over the web using XML, it makes to sense to leverage the venerable URI (Uniform Resource Identifier). URIs are the standard way to identify “resources” on the web.

Often URIs also provide information about how to fetch their resource - that’s why they are often called URLs (Uniform Resource Locator). From a practical perspective if a vendor uses HTTP URIs to identify their objects, you can most likely just do a simple HTTP GET to fetch the oBIX document for that object. But technically, fetching the contents of a URI is a protocol binding issue discussed in later chapters.

The value of URIs are that they come with all sorts of nifty rules already defined for us. For example URIs define which characters are legal and which are illegal. Of great value to oBIX is URI references which define a standard way to express and normalize relative URIs. Plus most programming environments have libraries to manage URIs so developers don’t have to worry about nitty gritty normalization details.

3.4 REST

Many savvy readers may be thinking that objects identified with URIs and passed around as XML documents is starting to sound a lot like REST – and you would be correct. REST stands for REpresentational State Transfer and is an architectural style for web services that mimics how the World Wide Web works. The WWW is basically a big web of HTML documents all hyperlinked together using URIs. Likewise, oBIX is basically a big web of XML object documents hyperlinked together using URIs. Because REST is such a key concept in oBIX, it is not surprising that a REST binding is a core part of the specification. The specification of this binding is defined in the oBIX REST Binding document.

REST is really more of a design style, than a specification. REST is resource centric as opposed to method centric - resources being oBIX objects. The methods actually used tend to be a very small fixed set of verbs used to work generically with all resources. In oBIX all network requests boil down to four request types:

Read: an object
Write: an object
Invoke: an operation
Delete: an object

3.5 Contracts

In every software domain, patterns start to emerge where many different object instances share common characteristics. For example in most systems that model people, each person probably has a name, address, and phone number. In vertical domains we may attach domain specific information to each person. For example an access control system might associate a badge number with each person.

In object oriented systems we capture these patterns into classes. In relational databases we map them into tables with typed columns. In oBIX we capture these patterns using a concept called contracts, which are standard oBIX objects used as a template. Contracts are more nimble and flexible than strongly typed schema languages, without the overhead of introducing new syntax. A contract document is parsed just like any other oBIX document. In geek speak contracts are a combination of prototype based inheritance and mixins.

Why do we care about trying to capture these patterns? The most important use of contracts is by the oBIX specification itself to define new standard abstractions. It is just as important for everyone to agree on normalized semantics as it is as on syntax. Contracts also provide the definitions needed to map to the OO guy’s classes or the relational database guy’s tables.

3.6 Extendibility

We want to use oBIX as a foundation for developing new abstractions in vertical domains. We also want to provide extendibility for vendors who implement oBIX across legacy systems and new product lines. Additionally, it is common for a device to ship as a blank slate and be completely programmed in the field. This leaves us with a mix of standards based, vendor based, and even project based extensions.

The principle behind oBIX extendibility is that anything new is defined strictly in terms of objects, URIs, and contracts. To put it another way - new abstractions don’t introduce any new XML syntax or functionality that client code is forced to care about. New abstractions are always modeled as standard trees of oBIX objects, just with different semantics. That doesn’t mean that higher level application code never changes to deal with new abstractions, but the core stack that deals with networking and parsing shouldn’t have to change.

This extendibility model is similar to most mainstream programming languages such as Java or C#. The syntax of the core language is fixed with a built in mechanism to define new abstractions. Extendibility is achieved by defining new class libraries using the language’s fixed syntax. This means you don’t have to update the compiler every time someone adds a new class.

4 Object Model

The oBIX specification is based on a small, fixed set of object types The oBIX object model is summarized in the following illustration. Each box represents a specific object. Each object type also lists its supported attributes. The object types are described in the subsequent subsections. The rules for usage and interpretation of the oBIX object model are defined in these subsections. Additional rules defining complex behaviors such as naming and contract inheritance are described in Sections 5 and 6. These sections are essential to a full understanding of the object model.

Figure 1. The oBIX primitive object hierarchy

4.1 obj

The root abstraction in oBIX is object, modeled in XML via the obj element. Every XML element in oBIX is a derivative of the obj element. Any obj element or its derivatives can contain other obj elements. The attributes supported on the obj element include:

name: defines the object’s purpose in its parent object (discussed in the Section 5);
href: provides a URI reference for identifying the object (discussed in the Section 5);
is: defines the contracts the object implements (discussed in Section 6);
null: support for null objects (discussed in Section 4.17 and in Section 6.4);
facets: a set of attributes used to provide meta-data about the object (discussed in Section 4.18);
val: an attribute used only with value objects (bool, int, real, str, enum, abstime, reltime, date, time and uri) to store the actual value. The literal representation of values map to [XML Schema Part 2: Datatypes] - indicated in the following sections via the “xs:” prefix.

The contract definition of obj:

4.2 bool

The bool object represents a boolean condition of either true or false. Its val attribute maps to xs:boolean defaulting to false. The literal value of a bool MUST be “true” or “false” (the literals “1” and “0” are not allowed). The contract definition:

An example:

4.3 int

The int type represents an integer number. Its val attribute maps to xs:long as a 64-bit integer with a default of 0. The contract definition:

An example:

4.4 real

The real type represents a floating point number. Its val attribute maps to xs:double as a IEEE 64-bit floating point number with a default of 0. The contract definition:

An example:

4.5 str

The str type represents a string of Unicode characters. Its val attribute maps to xs:string with a default of the empty string. The contract definition:

An example:

4.6 enum

The enum type is used to represent a value which must match a finite set of values. The finite value set is called the range. The val attribute of an enum is represented as a string key using xs:string. Enums default to null. The range of an enum is declared via facets using the range attribute. The contract definition:

An example:

In this example, the val attribute is specified, so the null attribute is implied to be false. See Section 6.4.3 for details on the inheritance of the null attribute.

4.7 abstime

The abstime type is used to represent an absolute point in time. Its val attribute maps to xs:dateTime, with the exception that the timezone is required. According to XML Schema Part 2 section 3.2.7.1, the lexical space for abstime is:

'-'? yyyy '-' mm '-' dd 'T' hh ':' mm ':' ss ('.' s+)? (zzzzzz)

Abstimes default to null. The contract definition:

An example for 9 March 2005 at 1:30PM GMT:

In this example, the val attribute is specified, so the null attribute is implied to be false. See Section 6.4.3 for details on the inheritance of the null attribute.

The timezone offset is required, so the abstime can be used to uniquely relate the abstime to UTC. The optional tz facet is used to specify the timezone as a zoneinfo identifier. This provides additional context about the timezone, if available. The timezone offset of the val attribute MUST match the offset for the timezone specified by the tz facet, if it is also used. See the tz facet section for more information.

4.8 reltime

The reltime type is used to represent a relative duration of time. Its val attribute maps to xs:duration with a default of 0sec. The contract definition:

An example of 15 seconds:

4.9 date

The date type is used to represent a day in time as a day, month, and year. Its val attribute maps to xs:date. According to XML Schema Part 2 section 3.2.9.1, the lexical space for date is:

'-'? yyyy '-' mm '-' dd

Date values in oBIX MUST omit the timezone offset and MUST NOT use the trailing “Z”. Only the tz attribute SHOULD be used to associate the date with a timezone. Date objects default to null. The contract definition:

An example for 26 November 2007:

In this example, the val attribute is specified, so the null attribute is implied to be false. See Section 6.4.3 for details on the inheritance of the null attribute.

The tz facet is used to specify the timezone as a zoneinfo identifier. See the tz facet section for more information.

4.10 time

The time type is used to represent a time of day in hours, minutes, and seconds. Its val attribute maps to xs:time. According to XML Schema Part 2 section 3.2.8, the lexical space for time is the left truncated representation of xs:dateTime:

hh ':' mm ':' ss ('.' s+)?

Time values in oBIX MUST omit the timezone offset and MUST NOT use the trailing “Z”. Only the tz attribute SHOULD be used to associate the time with a timezone. Time objects default to null. The contract definition:

An example for 4:15 AM:

In this example, the val attribute is specified, so the null attribute is implied to be false. See Section 6.4.3 for details on the inheritance of the null attribute.

The tz facet is used to specify the timezone as a zoneinfo identifier. See the tz facet section for more information.

4.11 uri

The uri type is used to store a URI reference. Unlike a plain old str, a uri has a restricted lexical space as defined by [RFC 3986] and the XML Schema xs:anyURI type. oBIX servers MUST use the URI syntax described by [RFC 3986] for identifying resources. oBIX clients MUST be able to navigate this URI syntax. Most URIs will also be a URL, meaning that they identify a resource and how to retrieve it (typically via HTTP). The contract:

An example for the oBIX home page:

4.12 list

The list object is a specialized object type for storing a list of other objects. The primary advantage of using a list versus a generic obj is that lists can specify a common contract for their contents using the of attribute. If specified the of attribute MUST be a list of URIs formatted as a contract list. The definition of list is:

An example list of strings:

</list>

Because lists typically have constraints on the URIs used for their child elements, they use special semantics for adding children. Lists are discussed in greater detail along with contracts in section 6.8.

4.13 ref

The ref object is used to create an out of document reference to another oBIX object. It is the oBIX equivalent of the HTML anchor tag. The contract definition:

A ref element MUST always specify an href attribute. A ref element SHOULD specify the type of the referenced object using the is attribute. A ref element referencing a list (is=”obix:list”) SHOULD specify the type of the objects contained in the list using the of attribute. References are discussed in detail in section 8.2.

4.14 err

The err object is a special object used to indicate an error. Its actual semantics are context dependent. Typically err objects SHOULD include a human readable description of the problem via the display attribute. The contract definition:

4.15 op

The op object is used to define an operation. All operations take one input object as a parameter, and return one object as an output. The input and output contracts are defined via the in and out attributes. The contract definition:

Operations are discussed in detail in Section 7.

4.16 feed

The feed object is used to define a topic for a feed of events. Feeds are used with watches to subscribe to a stream of events such as alarms. A feed SHOULD specify the event type it fires via the of attribute. The in attribute can be used to pass an input argument when subscribing to the feed (a filter for example).

Feeds are subscribed via Watches discussed in Section 11.

4.17 Null

All objects support the concept of null. Null is the absence of a value. Null is indicated using the null attribute with a boolean value. All objects default null to false with the exception of enum, abstime, date, and time (since any other default would be confusing). An example of a null abstime object is:

Null is inherited from contracts a little differently than other attributes. See Section 6.4.3 for details.

4.18 Facets

All objects can be annotated with a predefined set of attributes called facets. Facets provide additional meta-data about the object. The set of available facets is: displayName, display, icon, min, max, precision, range, and unit. Although oBIX predefines a number of facets attributes, vendors MAY add additional facets. Vendors that wish to annotate objects with additional facets SHOULD consider using XML namespace qualified attributes.

4.18.1 displayName

The displayName facet provides a localized human readable name of the object stored as a xs:string:

Typically the displayName facet SHOULD be a localized form of the name attribute. There are no restrictions on displayName overrides from the contract (although it SHOULD be uncommon since displayName is just a human friendly version of name).

4.18.2 display

The display facet provides a localized human readable description of the object stored as a xs:string:

There are no restrictions on display overrides from the contract.

The display attribute serves the same purpose as Object.toString() in Java or C#. It provides a general way to specify a string representation for all objects. In the case of value objects (like bool or int) it SHOULD provide a localized, formatted representation of the val attribute.

4.18.3 icon

The icon facet provides a URI reference to a graphical icon which may be used to represent the object in an user agent:

The contents of the icon attribute MUST be a URI to an image file. The image file SHOULD be a 16x16 PNG file. There are no restrictions on icon overrides from the contract.

4.18.4 min

The min facet is used to define an inclusive minimum value:

The contents of the min attribute MUST match its associated val type. The min facet is used with int, real , abstime, date, time, and reltime to define an inclusive lower limit of the value space. It is used with str to indicate the minimum number of Unicode characters of the string. It is used with list to indicate the minimum number of child objects (named or unnamed). Overrides of the min facet may only narrow the value space using a larger value. The min facet MUST never be greater than the max facet (although they can be equal).

4.18.5 max

The max facet is used to define an inclusive maximum value:

The contents of the max attribute MUST match its associated val type. The max facet is used with int, real, abstime, date, time, and reltime to define an inclusive upper limit of the value space. It is used with str to indicate the maximum number of Unicode characters of the string. It is used with list to indicate the maximum number of child objects (named or unnamed). Overrides of the max facet may only narrow the value space using a smaller value. The max facet MUST never be less than the min facet (although they MAY be equal).

4.18.6 precision

The precision facet is used to describe the number of decimal places to use for a real value:

The contents of the precision attribute MUST be xs:int. The value of the precision attribute equates to the number of meaningful decimal places. In the example above, the value of 2 indicates two meaningful decimal places: “75.04”. Typically precision is used by client applications which do their own formatting of real values. There are no restrictions on precision overrides.

4.18.7 range

The range facet is used to define the value space of an enumeration. A range attribute is a URI reference to an obix:Range object (see section 10.2 for the definition). It is used with the bool and enum object types:

The override rule for range is that the specified range MUST inherit from the contract’s range. Enumerations are funny beasts in that specialization of an enum usually involves adding new items to the range. Technically this is widening the enum’s value space, rather than narrowing it. But in practice, adding items into the range is what we desire.

4.18.8 status

The status facet is used to annotate an object about the quality and state of the information:

Status is an enumerated string value with one of the following values (ordered by priority):

disabled: This state indicates that the object has been disabled from normal operation (out of service). In the case of operations and feeds, this state is used to disable support for the operation or feed.
fault: The fault state indicates that the data is invalid or unavailable due to a failure condition - data which is out of date, configuration problems, software failures, or hardware failures. Failures involving communications should use the down state.
down: The down state indicates a communication failure.
unackedAlarm: The unackedAlarm state indicates there is an existing alarm condition which has not been acknowledged by a user – it is the combination of the alarm and unacked states. The difference between alarm and unackedAlarm is that alarm implies that a user has already acknowledged the alarm or that no human acknowledgement is necessary for the alarm condition. The difference between unackedAlarm and unacked is that the object has returned to a normal state.
alarm: This state indicates the object is currently in the alarm state. The alarm state typically means that an object is operating outside of its normal boundaries. In the case of an analog point this might mean that the current value is either above or below its configured limits. Or it might mean that a digital sensor has transitioned to an undesired state. See Alarming (Section 14) for additional information.
unacked: The unacked state is used to indicate a past alarm condition which remains unacknowledged.
overridden: The overridden state means the data is ok, but that a local override is currently in effect. An example of an override might be the temporary override of a setpoint from it’s normal scheduled setpoint.
ok: The ok state indicates normal status. This is the assumed default state for all objects.

Status MUST be one of the enumerated strings above. It might be possible in the native system to exhibit multiple status states simultaneously, however when mapping to oBIX the highest priority status SHOULD be chosen – priorities are ranked from top (disabled) to bottom (ok).

4.18.9 tz

The tz facet is used to annotate an abstime, date, or time object with a timezone. The value of a tz attribute is a zoneinfo string identifier. Zoneinfo is a standardized database sometimes referred to as the tz database or the Olsen database. It defines a set of time zone identifiers using the convention “continent/city”. For example “America/New_York” identifies the time zone rules used by the east coast of the United Stated. UTC is represented as “Etc/UTC”.

The zoneinfo database defines the current and historical rules for each zone including its offset from UTC and the rules for calculating daylight saving time. oBIX does not define a contract for modeling timezones, instead it just references the zoneinfo database using standard identifiers. It is up to oBIX enabled software to map zoneinfo identifiers to the UTC offset and daylight saving time rules.

The following rules are used to compute the timezone of an abstime, date, or time object:

If the tz attribute is specified, use it;
If the contract defines an inherited tz attribute, use it;
Assume the server’s timezone as defined by the lobby’s About.tz.

When using timezones, it is still required to specify the timezone offset within the value representation of an abstime or time object. It is an error condition for the tz facet to conflict with the timezone offset. For example New York has a -5 hour offset from UTC during standard time and a -4 hour offset during daylight saving time:

4.18.10 unit

The unit facet defines a unit of measurement. A unit attribute is a URI reference to a obix:Unit object (see section 10.5 for the contract definition). It is used with the int and real object types:

It is recommended that the unit facet not be overridden if declared in a contract. If it is overridden, then the override SHOULD use a Unit object with the same dimensions as the contract (it must measure the same physical quantity).

4.18.11 writable

The writable facet specifies if this object can be written by the client. If false (the default), then the object is read-only. It is used with all objects except operations and feeds:

The writable facet describes only the ability of clients to modify this object’s value, not the ability of clients to add or remove children of this object. If a server does not support addition or removal of object children through writes, it MUST return an appropriate error response (see Section 9 for details).

5 Naming

All oBIX objects have two potential identifiers: name and href. Name is used to define the role of an object within its parent. Names are programmatic identifiers only; the displayName facet SHOULD be used for human interaction. Naming convention is to use camel case with the first character in lowercase. The primary purpose of names is to attach semantics to sub-objects. Names are also used to indicate overrides from a contract. A good analogy to names is the field/method names of a class in Java or C#.

Hrefs are used to attach URIs to objects. An href is always a URI reference, which means it might be a relative URI that requires normalization against a base URI. The exception to this rule is the href of the root object in an oBIX document – this href MUST be an absolute URI, not a URI reference. This allows the root object’s href to be used as the effective base URI (xml:base) for normalization. A good analogy is hrefs in HTML or XLink.

Some objects may have both a name and an href, just a name, just an href, or neither. It is common for objects within a list to not use names, since most lists are unnamed sequences of objects. The oBIX specification makes a clear distinction between names and hrefs - clients MUST NOT assume any relationship between names and hrefs. From a practical perspective many vendors will likely build an href structure that mimics the name structure, but client software MUST never assume such a relationship.

5.1 Name

The name of an object is represented using the name attribute. Names are programmatic identifiers with restrictions on their valid character set. A name SHOULD contain only ASCII letters, digits, underbar, or dollar signs. A digit MUST NOT be used as the first character. Convention is to use camel case with the first character in lower case: “foo”, “fooBar”, “thisIsOneLongName”. Within a given object, all of its direct children MUST have unique names. Objects which don’t have a name attribute are called unnamed objects. The root object of an oBIX document SHOULD NOT specify a name attribute (but almost always has an absolute href URI).

5.2 Href

The href of an object is represented using the href attribute. If specified, the root object MUST have an absolute URI. All other hrefs within an oBIX document are treated as URI references which may be relative. Because the root href is always an absolute URI, it may be used as the base for normalizing relative URIs within the document. The formal rules for URI syntax and normalization are defined in [RFC 3986]. oBIX implementations MUST follow these rules. We consider a few common cases that serve as design patterns within oBIX in Section 5.3.

As a general rule every object accessible for a read MUST specify a URI. An oBIX document returned from a read request MUST specify a root URI. However, there are certain cases where the object is transient, such as a computed object from an operation invocation. In these cases there MAY not be a root URI, meaning there is no way to retrieve this particular object again. If no root URI is provided, then the server’s authority URI is implied to be the base URI for resolving relative URI references.

5.3 HTTP Relative URIs

Vendors are free to use any URI scheme, although the recommendation is to use HTTP URIs since they have well defined normalization semantics. This section provides a summary of how HTTP URI normalization should work within oBIX client agents. The general rules are:

If the URI starts with “scheme:” then it is an globally absolute URI
If the URI starts with a single slash, then it is server absolute URI
If the URI starts with a “#”, then it is a fragment identifier (discussed in next section)
If the URI starts with “../”, then the path must backup from the base

Otherwise the URI is assumed to be a relative path from the base URI

Some examples:

http://server/a + http://overthere/x ® http://overthere/x

http://server/a + /x/y/z ® http://server/x/y/z

http://server/a/b + c ® http://server/a/c

http://server/a/b/ + c ® http://server/a/b/c

http://server/a/b + c/d ® http://server/a/c/d

http://server/a/b/ + c/d ® http://server/a/b/c/d

http://server/a/b + ../c ® http://server/c

http://server/a/b/ + ../c ® http://server/a/c

Perhaps one of the trickiest issues is whether the base URI ends with a slash. If the base URI doesn’t end with a slash, then a relative URI is assumed to be relative to the base’s parent (to match HTML). If the base URI does end in a slash, then relative URIs can just be appended to the base. In practice, systems organized into hierarchical URIs SHOULD always specify the base URI with a trailing slash. Retrieval with and without the trailing slash SHOULD be supported with the resulting document always adding the implicit trailing slash in the root object’s href.

5.4 Fragment URIs

It is not uncommon to reference an object internal to an oBIX document. This is achieved using fragment URI references starting with the “#”. Let’s consider the example:

</list>

</obj>

In this example there are two objects with a range facet referencing a fragment URI. Any URI reference starting with “#” MUST be assumed to reference an object within the same oBIX document. Clients SHOULD NOT perform another URI retrieval to dereference the object. In this case the object being referenced is identified via the href attribute.

In the example above the object with an href of “onOff” is both the target of the fragment URI, but also has the absolute URI “http://server/whatever/onOff”. But suppose we had an object that was the target of a fragment URI within the document, but could not be directly addressed using an absolute URI? In that case the href attribute SHOULD be a fragment identifier itself. When an href attribute starts with “#” that means the only place it can be used is within the document itself:

…

…

6 Contracts

Contracts are a mechanism to harness the inherit patterns in modeling oBIX data sources. What is a contract? Well basically it is just a normal oBIX object. What makes a contract object special, is that other objects reference it as a “template object” using the is attribute.

So what does oBIX use contracts for? Contracts solve many problems in oBIX:

Semantics: contracts are used to define “types” within oBIX. This lets us collectively agree on common object definitions to provide consistent semantics across vendor implementations. For example the Alarm contract ensures that client software can extract normalized alarm information from any vendor’s system using the exact same object structure.
Defaults: contracts also provide a convenient mechanism to specify default values. Note that when serializing object trees to XML (especially over a network), we typically don’t allow defaults to be used in order to keep client processing simple.
Type Export: it is likely that many vendors will have a system built using a statically typed language like Java or C#. Contracts provide a standard mechanism to export type information in a format that all oBIX clients can consume.

Why use contracts versus other approaches? There are certainly lots of ways to solve the above problems. The benefit of the contract design is its flexibility and simplicity. Conceptually contracts provide an elegant model for solving many different problems with one abstraction. From a specification perspective, we can define new abstractions using the oBIX XML syntax itself. And from an implementation perspective, contracts give us a machine readable format that clients already know how to retrieve and parse – to use OO lingo, the exact same syntax is used to represent both a class and an instance.

6.1 Contract Terminology

In order to discuss contracts, it is useful to define a couple of terms:

Contract: is a reusable object definition expressed as a standard oBIX XML document. Contracts are the templates or prototypes used as the foundation of the oBIX type system.
Contract List: is a list of one or more URIs to contract objects. It is used as the value of the is, of, in and out attributes. The list of URIs is separated by the space character. You can think of a contract list as a type declaration.
Implements: when an object specifies a contract in its contract list, the object is said to implement the contract. This means that the object is inheriting both the structure and semantics of the specified contract.
Implementation: an object which implements a contract is said to be an implementation of that contract.

6.2 Contract List

The syntax of a contract list attribute is a list of URI references to other oBIX objects. It is used as the value of the is, of, in and out attributes. The URIs within the list are separated by the space character (Unicode 0x20). Just like the href attribute, a contract URI can be an absolute URI, server relative, or even a fragment reference. The URIs within a contract list may be scoped with an XML namespace prefix (see ”Namespace Prefixes in Contract Lists” in the oBIX Encodings document).

6.3 Is Attribute

An object defines the contracts it implements via the is attribute. The value of the is attribute is a contract list. If the is attribute is unspecified, then the following rules are used to determine the implied contract list:

If the object is an item inside a list or feed, then the contract list specified by the of attribute is used.
If the object overrides (by name) an object specified in one of its contracts, then the contract list of the overridden object is used.
If all the above rules fail, then the respective primitive contract is used. For example, an obj element has an implied contract of obix:obj and real an implied contract of obj:real.

Note that element names such as bool, int, or str are syntactic sugar for an implied contract. However if an object implements one of the primitives, then it MUST use the correct XML element name. For example if an object implements obix:int, then it MUST be expressed as <int/>, rather than <obj is="obix:int"/>. Therefore it is invalid to implement multiple value types - such as implementing both obix:bool and obix:int.

6.4 Contract Inheritance

6.4.1 Structure vs Semantics

Contracts are a mechanism of inheritance – they establish the classic “is a” relationship. In the abstract sense a contract allows us to inherit a type. We can further distinguish between the explicit and implicit contract:

Explicit Contract: defines an object structure which all implementations must conform with.
Implicit Contract: defines semantics associated with the contract. Usually the implicit contract is documented using natural language prose. It isn’t mathematical, but rather subject to human interpretation.

For example when we say an object implements the Alarm contract, we immediately know that will have a child called timestamp. This structure is in the explicit contract of Alarm and is formally defined in XML. But we also attach semantics to what it means to be an Alarm object: that the object is providing information about an alarm event. These fuzzy concepts can’t be captured in machine language; rather they can only be captured in prose.

When an object declares itself to implement a contract it MUST meet both the explicit contract and the implicit contract. An object MUST NOT put obix:Alarm in its contract list unless it really represents an alarm event. There isn’t much more to say about implicit contracts other than it is recommended that a human brain be involved. So now let’s look at the rules governing the explicit contract.

6.4.2 Overriding Defaults

A contract’s named children objects are automatically applied to implementations. An implementation may choose to override or default each of its contract’s children. If the implementation omits the child, then it is assumed to default to the contract’s value. If the implementation declares the child (by name), then it is overridden and the implementation’s value should be used. Let’s look at an example:

</obj>

</obj>

In this example we have a contract object identified with the URI “/def/television”. It has two children to store power and channel. Then we specify a living room TV instance that includes “/def/television” in its contract list via the is attribute. In this object, channel is overridden to 8 from its default value of 2. However since power was omitted, it is implied to default to false.

An override is always matched to its contract via the name attribute. In the example above we knew we were overriding channel, because we declared an object with a name of “channel”. We also declared an object with a name of “volume”. Since volume wasn’t declared in the contract, we assume it’s a new definition specific to this object.

6.4.3 Attributes and Facets

Also note that the contract’s channel object declares a min and max facet. These two facets are also inherited by the implementation. Almost all attributes are inherited from their contract including facets, val, of, in, and out. The href attribute is never inherited. The null attribute inherits as follows:

If the null attribute is specified, then its explicit value is used;
If a val attribute is specified and null is unspecified, then null is implied to be false;
If neither a val attribute or a null attribute is specified, then the null attribute is inherited from the contract;
If the null attribute is specified and is true, then the val attribute is ignored.

This allows us to implicitly override a null object to non-null without specifying the null attribute.

6.5 Override Rules

Contract overrides are required to obey the implicit and explicit contract. Implicit means that the implementation object provides the same semantics as the contract it implements. In the example above it would be incorrect to override channel to store picture brightness. That would break the semantic contract.

Overriding the explicit contract means to override the value, facets, or contract list. However we can never override the object to be in incompatible value type. For example if the contract specifies a child as real, then all implementations must use real for that child. As a special case, obj may be narrowed to any other element type.

We also have to be careful when overriding attributes to never break restrictions the contract has defined. Technically this means we can specialize or narrow the value space of a contract, but never generalize or widen it. This concept is called covariance. Let’s take our example from above:

In this example the contract has declared a value space of 2 to 200. Any implementation of this contract must meet this restriction. For example it would an error to override min to –100 since that would widen the value space. However we can narrow the value space by overriding min to a number greater than 2 or by overriding max to a number less than 200. The specific override rules applicable to each facet are documented in section 4.18.

6.6 Multiple Inheritance

An object’s contract list may specify multiple contract URIs to implement. This is actually quite common - even required in many cases. There are two topics associated with the implementation of multiple contracts:

Flattening: contract lists SHOULD always be flattened when specified. This comes into play when a contract has its own contract list (Section 6.6.1).
Mixins: the mixin design specifies the exact rules for how multiple contracts are merged together. This section also specifies how conflicts are handled when multiple contracts contain children with the same name (Section 6.6.2).

6.6.1 Flattening

It is common for contract objects themselves to implement contracts, just like it is common in OO languages to chain the inheritance hierarchy. However due to the nature of accessing oBIX documents over a network, we wish to minimize round trip network requests which might be required to “learn” about a complex contract hierarchy. Consider this example:

In this example if we were reading object D for the first time, it would take three more requests to fully learn what contracts are implemented (one for C, B, and A). Furthermore, if our client was just looking for objects that implemented B, it would difficult to determine this just by looking at D.

Because of these issues, servers are REQUIRED to flatten their contract inheritance hierarchy into a list when specifying the is, of, in, or out attributes. In the example above, the correct representation would be:

This allows clients to quickly scan Ds contract list to see that D implements C, B, and A without further requests.

Because complex servers often have a complex contract hierarchy of object types, the requirement to flatten the contract hierarchy can lead to a verbose contract list. Often many of these contracts are from the same namespace. For example:

To save space, servers MAY choose to combine the contracts from the same namespace and present the contract list with the namespace followed by a colon, then a brace-enclosed list of contract names:

Clients MUST be able to consume this form of the contract list and expand it to the standard form.

6.6.2 Mixins

Flattening is not the only reason a contract list might contain multiple contract URIs. oBIX also supports the more traditional notion of multiple inheritance using a mixin metaphor. Consider the following example:

</obj>

</obj>

</obj>

In this example ClockRadio implements both Clock and Radio. Via flattening of Clock and Radio, ClockRadio also implements Device. In oBIX this is called a mixin – Clock, Radio, and Device are mixed into (merged into) ClockRadio. Therefore ClockRadio inherits four children: serialNo, snooze, volume, and station. Mixins are a form of multiple inheritance akin to Java/C# interfaces (remember oBIX is about the type inheritance, not implementation inheritance).

Note that Clock and Radio both implement Device - the classic diamond inheritance pattern. From Device, ClockRadio inherits a child named serialNo. Furthermore notice that both Clock and Radio declare a child named volume. This naming collision could potentially create confusion for what serialNo and volume mean in ClockRadio.

In oBIX we solve this problem by flattening the contract’s children using the following rules:

Process the contract definitions in the order they are listed
If a new child is discovered, it is mixed into the object’s definition
If a child is discovered we already processed via a previous contract definition, then the previous definition takes precedence. However it is an error if the duplicate child is not contract compatible with the previous definition (see Section 6.7).

In the example above this means that Radio.volume is the definition we use for ClockRadio.volume, because Radio has a higher precedence than Clock (it is first in the contract list). Thus ClockRadio.volume has a default value of “5”. However it would be invalid if Clock.volume were declared as str, since it would not be contract compatible with Radio’s definition as an int – in that case ClockRadio could not implement both Clock and Radio. It is the server vendor’s responsibility not to create incompatible name collisions in contracts.

The first contract in a list is given specific significance since its definition trumps all others. In oBIX this contract is called the primary contract. It is recommended that the primary contract implement all the other contracts specified in the contract list (this actually happens quite naturally by itself in many programming languages). This makes it easier for clients to bind the object into a strongly typed class if desired. Contracts MUST NOT implement themselves nor have circular inheritance dependencies.

6.7 Contract Compatibility

A contract list which is covariantly substitutable with another contract list is said to be contract compatible. Contract compatibility is a useful term when talking about mixin rules and overrides for lists and operations. It is a fairly common sense notion similar to previously defined override rules – however, instead of the rules applied to individual facet attributes, we apply it to an entire contract list.

A contract list X is compatible with contract list Y, if and only if X narrows the value space defined by Y. This means that X can narrow the set of objects which implement Y, but never expand the set. Contract compatibility is not commutative (X is compatible with Y does not imply Y is compatible with X). If that definition sounds too highfalutin, you can boil it down to this practical rule: X can add new URIs to Y’s list, but never take any away.

6.8 Lists (and Feeds)

Implementations derived from list or feed contracts inherit the of attribute. Like other attributes we can override the of attribute, but only if contract compatible - a server SHOULD include all of the URIs in the contract’s of attribute, but it MAY add additional ones (see Section 6.7).

Lists and feeds also have the special ability to implicitly define the contract list of their contents. In the following example it is implied that each child element has a contract list of /def/MissingPerson without actually specifying the is attribute in each list item:

</list>

If an element in the list or feed does specify its own is attribute, then it MUST be contract compatible with the of attribute.

If an implementor wishes to specify that a list should contain references to a given type, then the server SHOULD include obix:ref in the of attribute. This MUST be the first URI in the of attribute. For example, to specify that a list should contain references to obix:History objects (as opposed to inline History objects):

In many cases a server will implement its own management of the URI scheme of the child elements of a list. For example, the href attribute of child elements may be a database key, or some other string defined by the server when the child is added. Servers will not, in general, allow clients to specify this URI during addition of child elements through a direct write to a list’s subordinate URI.

Therefore, in order to add child elements to a list which supports client addition of list elements, servers MUST support adding list elements by writing to the list URI with an object of a type that matches the list’s contract. Servers MUST return the written resource (including any server-assigned href) upon successful completion of the write.

For example, given a list of <real> elements, and presupposing a server-imposed URI scheme:

Writing to the list URI itself will replace the entire list if the server supports this behavior:

WRITE /a/b

</list>

returns:

</list>

Writing a single element of type <real> will add this element to the list.

WRITE /a/b

returns:

Note that if a client has the correct URI to reference a list child element, this can still be used to modify the value of the element directly:

WRITE /a/b/3

returns:

7 Operations

Operations are the things that you can “do” to an oBIX object. They are akin to methods in traditional OO languages. Typically they map to commands rather than a variable that has continuous state. Unlike value objects which represent an object and its current state, the op element merely represents the definition of an operation you can invoke.

All operations take exactly one object as a parameter and return exactly one object as a result. The in and out attributes define the contract list for the input and output objects. If you need multiple input or output parameters, then wrap them in a single object using a contract as the signature. For example:

</obj>

Objects can override the operation definition from one of their contracts. However the new in or out contract list MUST be contract compatible (see Section 6.7) with the contract’s definition.

If an operation doesn’t require a parameter, then specify in as obix:Nil. If an operation doesn’t return anything, then specify out as obix:Nil. Occasionally an operation is inherited from a contract which is unsupported in the implementation. In this case set the status attribute to disabled.

Operations are always invoked via their own href attribute (not their parent’s href). Therefore operations SHOULD always specify an href attribute if you wish clients to invoke them. A common exception to this rule is contract definitions themselves.

8 Object Composition

A good metaphor for comparison with oBIX is the World Wide Web. If you ignore all the fancy stuff like JavaScript and Flash, basically the WWW is a web of HTML documents hyperlinked together with URIs. If you dive down one more level, you could say the WWW is a web of HTML elements such as <p>, <table>, and <div>.

What the WWW does for HTML documents, oBIX does for objects. The logical model for oBIX is a global web of oBIX objects linked together via URIs. Some of these oBIX objects are static documents like contracts or device descriptions. Other oBIX objects expose real-time data or services. But they all are linked together via URIs to create the oBIX Web.

Individual objects are composed together in two ways to define this web. Objects may be composed together via containment or via reference.

8.1 Containment

Any oBIX object may contain zero or more children objects. This even includes objects which might be considered primitives such as bool or int. All objects are open ended and free to specify new objects which may not be in the object’s contract. Containment is represented in the XML syntax by nesting the XML elements:

</list>

</obj>

In this example the object identified by “/a” contains “/a/b”, which in turn contains “/a/b/c”. Child objects may be named or unnamed depending on if the name attribute is specified (Section 5.1). In the example, “/a/b” is named and “/a/b/c” is unnamed. Typically named children are used to represent fields in a record, structure, or class type. Unnamed children are often used in lists.

8.2 References

Let’s go back to our WWW metaphor. Although the WWW is a web of individual HTML elements like <p> and <div>, we don’t actually pass individual <p> elements around over the network. Rather we “chunk” them into HTML documents and always pass the entire document over the network. To tie it all together, we create links between documents using the <a> anchor element. These anchors serve as place holders, referencing outside documents via a URI.

An oBIX reference is basically just like an HTML anchor. It serves as placeholder to “link” to another oBIX object via a URI. While containment is best used to model small trees of data, references may be used to model very large trees or graphs of objects. As a matter fact, with references we can link together all oBIX objects on the Internet to create the oBIX Web.

As a clue to clients consuming oBIX references, the server SHOULD specify the type of the referenced object using the is attribute. In addition, for the list element type, the server SHOULD use the of attribute to specify the type of objects contained by the list. This allows the client to prepare the proper visualizations, data structures, etc. for consuming the object when it accesses the actual object. For example, a server might provide a reference to a list of available points:

8.3 Extents

When oBIX is applied to a problem domain, we have to decide whether to model relationships using either containment or references. These decisions have a direct impact on how your model is represented in XML and accessed over the network. The containment relationship is imbued with special semantics regarding XML encoding and eventing. In fact, oBIX coins a term for containment called an object’s extent. An object’s extent is its tree of children down to references. Only objects which have an href have an extent. Objects without an href are always included in one or more of their ancestors extents.

</obj>

</obj>

In the example above, we have five objects named ‘a’ to ‘e’. Because ‘a’ includes an href, it has an associated extent, which encompasses ‘b’ and ‘c’ by containment and ‘d’ and ‘e’ by reference. Likewise, ‘b’ has an href which results in an extent encompassing ‘c’ by containment and ‘d’ by reference. Object ‘c’ does not provide a direct href, but exists in both the ‘a’ and ‘b’ objects’ extents. Note an object with an href has exactly one extent, but can be nested inside multiple extents.

8.4 XML

When marshaling objects into an XML, it is required that an extent always be fully inlined into the XML document. The only valid objects which may be referenced outside the document are ref element themselves.

If the object implements a contract, then it is required that the extent defined by the contract be fully inlined into the document (unless the contract itself defined a child as a ref element). An example of a contract which specifies a child as a ref is Lobby.about (Section 9.3).

8.5 Alternate Hierarchies

Servers MAY present alternate hierarchies of an object’s extent. A server MUST use the semicolon character (;) to indicate an alternate hierarchy. For example, a server might present tag metadata from tag dictionary d1 in presenting a particular object in its system:

</real>

</obj>

The metadata SHOULD be presented using the ref element, so this additional information can be skipped during normal encoding. If a client is able to consume the metadata, it SHOULD ask for the metadata by requesting the metadata hierarchy.

9 Networking

The heart of oBIX is its object model and associated encoding. However, the primary use case for oBIX is to access information and services over a network. The oBIX architecture is based on a client/server network model:

Server: software containing oBIX enabled data and services. Servers respond to requests from client over a network.
Client: software which makes requests to servers over a network to access oBIX enabled data and services.

There is nothing to prevent software from being both an oBIX client and server. However, a key tenet of oBIX is that a client is NOT REQUIRED to implement server functionality which might require a server socket to accept incoming requests.

9.1 Request / Response

All network access is boiled down into the following request / response types:

Read: return the current state of an object at a given URI as an oBIX object.
Write: update the state of an existing object at a URI. The state to write is passed over the network as an oBIX object. The new updated state is returned in an oBIX object.
Invoke: invoke an operation identified by a given URI. The input parameter and output result are passed over the network as an oBIX object.
Delete: delete the object at a given URI.

Exactly how these request/responses are implemented between a client and server is called a protocol binding. The oBIX specification defines two standard protocol bindings: HTTP Binding (see OBIX-REST-v1.0) and SOAP Binding (see OBIX-SOAP-v1.0). However all protocol bindings must follow the same read, write, invoke, and delete semantics discussed next.

9.1.1 Read

The read request specifies an object’s URI and the read response returns the current state of the object as an oBIX document. The response MUST include the object’s complete extent (see Section 8.3). Servers may return an err object to indicate the read was unsuccessful – the most common error is obix:BadUriErr (see Section 9.2 for standard error contracts).

9.1.2 Write

The write request is designed to overwrite the current state of an existing object. The write request specifies the URI of an existing object and its new desired state. The response returns the updated state of the object. If the write is successful, the response MUST include the object’s complete extent (see Section 8.3). If the write is unsuccessful, then the server MUST return an err object indicating the failure.

The server is free to completely or partially ignore the write, so clients SHOULD be prepared to examine the response to check if the write was successful. Servers may also return an err object to indicate the write was unsuccessful.

Clients are not required to include the object’s full extent in the request. Objects explicitly specified in the request object tree SHOULD be overwritten or “overlaid” over the server’s actual object tree. Only the val attribute should be specified for a write request (outside of identification attributes such as name). The null attribute MAY also be used to set an object to null. If the null attribute is not specified and the val attribute is specified, then it is implied that null is false. A write operation that provides facets has unspecified behavior. When writing int or reals with units, the write value MUST be in the same units as the server specifies in read requests – clients MUST NOT provide a different unit facet and expect the server to auto-convert (in fact the unit facet SHOULD NOT be included in the request).

9.1.3 Invoke

The invoke request is designed to trigger an operation. The invoke request specified the URI of an op object and the input argument object. The response includes the output object. The response MUST include the output object’s complete extent (see Section 8.3). Servers MAY instead return an err object to indicate the invoke was unsuccessful.

9.1.4 Delete

The delete request is designed to remove an existing object from the server. The delete request specifies the URI of an existing object. If the delete is successful, the server MUST return an empty response. If the delete is unsuccessful, the server MUST return an err object indicating the failure.

9.2 Errors

Request errors are conveyed to clients with the err element. Any time an oBIX server successfully receives a request and the request cannot be processed, then the server SHOULD return an err object to the client. Returning a valid oBIX document with err SHOULD be used when feasible rather than protocol specific error handling (such as an HTTP response code). Such a design allows for consistency with batch request partial failures and makes protocol binding more pluggable by separating data transport from application level error handling.

A few contracts are predefined for common errors:

BadUriErr: used to indicate either a malformed URI or a unknown URI;
UnsupportedErr: used to indicate an a request which isn’t supported by the server implementation (such as an operation defined in a contract, which the server doesn’t support);
PermissionErr: used to indicate that the client lacks the necessary security permission to access the object or operation.

The contracts for these errors are:

If one of the above contracts makes sense for an error, then it SHOULD be included in the err element’s is attribute. It is strongly encouraged to also include a useful description of the problem in the display attribute.

9.3 Lobby

All oBIX servers MUST provide an object which implements obix:Lobby. The Lobby object serves as the central entry point into an oBIX server, and lists the URIs for other well-known objects defined by the oBIX specification. Theoretically all a client needs to know to bootstrap discovery is one URI for the Lobby instance. By convention this URI is “http://server/obix”, although vendors are certainly free to pick another URI. The Lobby contract is:

</obj>

The Lobby instance is where vendors SHOULD place vendor specific objects used for data and service discovery.

The discovery of which encoding to use for communication between a client and a server is a function of the specific binding used. Clients and servers MUST be able to support negotiation of the encoding to be used according to the binding’s error message rules. Clients SHOULD first attempt to request communication using the desired encoding, and then fall back to other encodings as required based on the encodings supported by the server.

9.4 About

The obix:About object is a standardized list of summary information about an oBIX server. Clients can discover the About URI directly from the Lobby. The About contract is:

</obj>

The following children provide information about the oBIX implementation:

obixVersion: specifies which version of the oBIX specification the server implements. This string MUST be a list of decimal numbers separated by the dot character (Unicode 0x2E). The current version string is “1.1”.

The following children provide information about the server itself:

serverName: provides a short localized name for the server.
serverTime: provides the server’s current local time.
serverBootTime: provides the server’s start time - this SHOULD be the start time of the oBIX server software, not the machine’s boot time.

The following children provide information about the server’s software vendor:

vendorName: the company name of the vendor who implemented the oBIX server software.
vendorUrl: a URI to the vendor’s website.

The following children provide information about the software product running the server:

productName: with the product name of oBIX server software.
productUrl: a URI to the product’s website.
productVersion: a string with the product’s version number. Convention is to use decimal digits separated by dots.

The following children provide additional miscellaneous information:

tz: specifies a zoneinfo identifier for the server’s default timezone.

9.5 Batch

The Lobby defines a batch operation which is used to batch multiple network requests together into a single operation. Batching multiple requests together can often provide significant performance improvements over individual round-robin network requests. As a general rule, one big request will always out-perform many small requests over a network.

A batch request is an aggregation of read, write, and invoke requests implemented as a standard oBIX operation. At the protocol binding layer, it is represented as a single invoke request using the Lobby.batch URI. Batching a set of requests to a server MUST be processed semantically equivalent to invoking each of the requests individually in a linear sequence.

The batch operation inputs a BatchIn object and outputs a BatchOut object:

The BatchIn contract specifies a list of requests to process identified using the Read, Write, or Invoke contract:

</uri>

</uri>

The BatchOut contract specifies an ordered list of the response objects to each respective request. For example the first object in BatchOut must be the result of the first request in BatchIn. Failures are represented using the err object. Every uri passed via BatchIn for a read or write request MUST have a corresponding result obj in BatchOut with an href attribute using an identical string representation from BatchIn (no normalization or case conversion is allowed).

It is up to vendors to decide how to deal with partial failures. In general idempotent requests SHOULD indicate a partial failure using err, and continue processing additional requests in the batch. If a server decides not to process additional requests when an error is encountered, then it is still REQUIRED to return an err for each respective request not processed.

Let’s look at a simple example:

</uri>

</list>

</list>

In this example, the batch request is specifying a read request for “/someStr” and “/invalidUri”, followed by a write request to “/someStr”. Note that the write request includes the value to write as a child named “in”.

The server responds to the batch request by specifying exactly one object for each request URI. The first read request returns a str object indicating the current value identified by “/someStr”. The second read request contains an invalid URI, so the server returns an err object indicating a partial failure and continues to process subsequent requests. The third request is a write to “someStr”. The server updates the value at “someStr”, and returns the new value. Note that because the requests are processed in order, the first request provides the original value of “someStr” and the third request contains the new value. This is exactly what we would expect had we processed each of these requests individually.

10 Core Contract Library

This chapter defines some fundamental object contracts that serve as building blocks for the oBIX specification.

10.1 Nil

The obix:Nil contract defines a standardized null object. Nil is commonly used for an operation’s in or out attribute to denote the absence of an input or output. The definition:

10.2 Range

The obix:Range contract is used to define a bool or enum’s range. Range is a list object that contains zero or more objects called the range items. Each item’s name attribute specifies the identifier used as the literal value of an enum. Item ids are never localized, and MUST be used only once in a given range. You may use the optional displayName attribute to specify a localized string to use in a user interface. The definition of Range:

An example:

</list>

The range facet may be used to define the localized text of a bool value using the ids of “true” and “false”:

</list >

10.3 Weekday

The obix:Weekday contract is a standardized enum for the days of the week:

</list>

</enum>

10.4 Month

The obix:Month contract is a standardized enum for the months of the year:

</list>

</enum>

10.5 Units

Representing units of measurement in software is a thorny issue. oBIX provides a unit framework for mathematically defining units within the object model. An extensive database of predefined units is also provided.

All units measure a specific quantity or dimension in the physical world. Most known dimensions can be expressed as a ratio of the seven fundamental dimensions: length, mass, time, temperature, electrical current, amount of substance, and luminous intensity. These seven dimensions are represented in SI respectively as kilogram (kg), meter (m), second (sec), Kelvin (K), ampere (A), mole (mol), and candela (cd).

The obix:Dimension contract defines the ratio of the seven SI units using a positive or negative exponent:

</obj>

A Dimension object contains zero or more ratios of kg, m, sec, K, A, mol, or cd. Each of these ratio maps to the exponent of that base SI unit. If a ratio is missing then the default value of zero is implied. For example acceleration is m/s², which would be encoded in oBIX as:

</obj>

Units with equal dimensions are considered to measure the same physical quantity. This is not always precisely true, but is good enough for practice. This means that units with the same dimension are convertible. Conversion can be expressed by specifying the formula required to convert the unit to the dimension’s normalized unit. The normalized unit for every dimension is the ratio of SI units itself. For example the normalized unit of energy is the joule m²·kg·s^-2. The kilojoule is 1000 joules and the watt-hour is 3600 joules. Most units can be mathematically converted to their normalized unit and to other units using the linear equations:

unit = dimension · scale + offset

toNormal = scalar · scale + offset

fromNormal = (scalar - offset) / scale

toUnit = fromUnit.fromNormal( toUnit.toNormal(scalar) )

There are some units which don’t fit this model including logarithm units and units dealing with angles. But this model provides a practical solution for most problem spaces. Units which don’t fit this model SHOULD use a dimension where every exponent is set to zero. Applications SHOULD NOT attempt conversions on these types of units.

The obix:Unit contract defines a unit including its dimension and its toNormal equation:

</obj>

The unit element contains a symbol, dimension, scale, and offset sub-object:

symbol: The symbol element defines a short abbreviation to use for the unit. For example “°F” would be the symbol for degrees Fahrenheit. The symbol element SHOULD always be specified.
dimension: The dimension object defines the dimension of measurement as a ratio of the seven base SI units. If omitted, the dimension object defaults to the obix:Dimension contract, in which case the ratio is the zero exponent for all seven base units.
scale: The scale element defines the scale variable of the toNormal equation. The scale object defaults to 1.
offset: The offset element defines the offset variable of the toNormal equation. If omitted then offset defaults to 0.

The display attribute SHOULD be used to provide a localized full name for the unit based on the client’s locale. If the display attribute is omitted, clients SHOULD use symbol for display purposes.

An example for the predefined unit for kilowatt:

</obj>

</obj>

Automatic conversion of units is considered a localization issue.

11 Watches

A key requirement of oBIX is access to real-time information. We wish to enable clients to efficiently receive access to rapidly changing data. However, we don’t want to require clients to implement web servers or expose a well-known IP address. In order to address this problem, oBIX provides a model for client polled eventing called watches. The watch lifecycle is as follows:

The client creates a new watch object with the make operation on the server’s WatchService URI. The server defines a new Watch object and provides a URI to access the new watch.
The client registers (and unregisters) objects to watch using operations on the Watch object.
The client periodically polls the Watch URI using the pollChanges operation to obtain the events which have occurred since the last poll.
The server frees the Watch under two conditions. The client may explicitly free the Watch using the delete operation. Or the server may automatically free the Watch because the client fails to poll after a predetermined amount of time (called the lease time).

Watches allow a client to maintain a real-time cache for the current state of one or more objects. They are also used to access an event stream from a feed object. Plus, watches serve as the standardized mechanism for managing per-client state on the server via leases.

11.1 WatchService

The WatchService object provides a well-known URI as the factory for creating new watches. The WatchService URI is available directly from the Lobby object. The contract for WatchService:

</obj>

The make operation returns a new empty Watch object as an output. The href of the newly created Watch object can then be used for invoking operations to populate and poll the data set.

11.2 Watch

Watch object is used to manage a set of objects which are subscribed and periodically polled by clients to receive the latest events. The contract is:

</obj>

</obj>

</uri>

</obj>

Many of the Watch operations use two contracts: obix:WatchIn and obix:WatchOut. The client identifies objects to add and remove from the poll list via WatchIn. This object contains a list of URIs. Typically these URIs SHOULD be server relative.

The server responds to add, pollChanges, and pollRefresh operations via the WatchOut contract. This object contains the list of subscribed objects - each object MUST specify an href URI using the exact same string as the URI identified by the client in the corresponding WatchIn. Servers are not allowed to perform any case conversions or normalization on the URI passed by the client. This allows client software to use the URI string as a hash key to match up server responses.

11.2.1 Watch.add

Once a Watch has been created, the client can add new objects to watch using the add operation. This operation inputs a list of URIs and outputs the current value of the objects referenced. The objects returned are required to specify an href using the exact string representation input by the client. If any object cannot be processed, then a partial failure SHOULD be expressed by returning an err object with the respective href. Subsequent URIs MUST NOT be affected by the failure of one invalid URI. The add operation MUST never return objects not explicitly included in the input URIs (even if there are already existing objects in the watch list). No guarantee is made that the order of objects in WatchOut matches the order in of URIs in WatchIn – clients must use the URI as a key for matching.

Note that the URIs supplied via WatchIn may include an optional in parameter. This parameter is only used when subscribing a watch to a feed object. Feeds also differ from other objects in that they return a list of historic events in WatchOut. Feeds are discussed in detail in Section 11.4.

It is invalid to add an op’s href to a watch, the server MUST report an err.

If an attempt is made to add a URI to a watch which was previously already added, then the server SHOULD return the current object’s value in the WatchOut result, but treat poll operations as if the URI was only added once – polls SHOULD only return the object once. If an attempt is made to add the same URI multiple times in the same WatchIn request, then the server SHOULD only return the object once.

Note: the lack of a trailing slash can cause problems with watches. Consider a client which adds a URI to a watch without a trailing slash. The client will use this URI as a key in its local hashtable for the watch. Therefore the server MUST use the URI exactly as the client specified. However, if the object’s extent includes child objects they will not be able to use relative URIs. It is RECOMMENDED that servers fail-fast in these cases and return a BadUriErr when clients attempt to add a URI without a trailing slash to a watch (even though they may allow it for a normal read request).

11.2.2 Watch.remove

The client can remove objects from the watch list using the remove operation. A list of URIs is input to remove, and the Nil object is returned. Subsequent pollChanges and pollRefresh operations MUST cease to include the specified URIs. It is possible to remove every URI in the watch list; but this scenario MUST NOT automatically free the Watch, rather normal poll and lease rules still apply. It is invalid to use the WatchInItem.in parameter for a remove operation.

11.2.3 Watch.pollChanges

Clients SHOULD periodically poll the server using the pollChanges operation. This operation returns a list of the subscribed objects which have changed. Servers SHOULD only return the objects which have been modified since the last poll request for the specific Watch. As with add, every object MUST specify an href using the exact same string representation the client passed in the original add operation. The entire extent of the object SHOULD be returned to the client if any one thing inside the extent has changed on the server side.

Invalid URIs MUST never be included in the response (only in add and pollRefresh). An exception to this rule is when an object which is valid is removed from the URI space. Servers SHOULD indicate an object has been removed via an err with the BadUriErr contract.

11.2.4 Watch.pollRefresh

The pollRefresh operation forces an update of every object in the watch list. The server MUST return every object and it’s full extent in the response using the href with the exact same string representation passed by the client in the original add. Invalid URIs in the poll list SHOULD be included in the response as an err element. A pollRefresh resets the poll state of every object, so that the next pollChanges only returns objects which have changed state since the pollRefresh invocation.

11.2.5 Watch.lease

All Watches have a lease time, specified by the lease child. If the lease time elapses without the client initiating a request on the Watch, then the server is free to expire the watch. Every new poll request resets the lease timer. So as long as the client polls at least as often as the lease time, the server SHOULD maintain the Watch. The following requests SHOULD reset the lease timer: read of the Watch URI itself or invocation of the add, remove, pollChanges, or pollRefresh operations.

Clients may request a different lease time by writing to the lease object (requires servers to assign an href to the lease child). The server is free to honor the request, cap the lease within a specific range, or ignore the request. In all cases the write request will return a response containing the new lease time in effect.

Servers SHOULD report expired watches by returning an err object with the BadUriErr contract. As a general principle servers SHOULD honor watches until the lease runs out or the client explicitly invokes delete. However, servers are free to cancel watches as needed (such as power failure) and the burden is on clients to re-establish a new watch.

11.2.6 Watch.delete

The delete operation can be used to cancel an existing watch. Clients SHOULD always delete their watch when possible to be good oBIX citizens. However servers MUST always cleanup correctly without an explicit delete when the lease expires.

11.3 Watch Depth

When a watch is put on an object which itself has children objects, how does a client know how “deep” the subscription goes? oBIX requires watch depth to match an object‘s extent (see Section 8.3). When a watch is put on a target object, a server MUST notify the client of any changes to any of the objects within that target object’s extent. If the extent includes feed objects they are not included in the watch – feeds have special watch semantics discussed in Section 11.4. This means a watch is inclusive of all descendents within the extent except refs and feeds.

11.4 Feeds

Servers may expose event streams using the feed object. The event instances are typed via the feed’s of attribute. Clients subscribe to events by adding the feed’s href to a watch, optionally passing an input parameter which is typed via the feed’s in attribute. The object returned from Watch.add is a list of historic events (or the empty list if no event history is available). Subsequent calls to pollChanges returns the list of events which have occurred since the last poll.

Let’s consider a simple example for an object which fires an event when its geographic location changes:

<obj>

</obj>

We subscribe to the moved event feed by adding “/car/moved” to a watch. The WatchOut will include the list of any historic events which have occurred up to this point in time. If the server does not maintain an event history this list will be empty:

</list>

</obj>

</list>

</obj>

Now every time we call pollChanges for the watch, the server will send us the list of event instances which have accumulated since our last poll:

<obj>

</obj>

<obj>

</obj>

</feed>

</list>

</obj>

Note the feed’s of attribute works just like the list’s of attribute. The children event instances are assumed to inherit the contract defined by of unless explicitly overridden. If an event instance does override the of contract, then it MUST be contract compatible. Refer to the rules defined in Section 6.8.

Invoking a pollRefresh operation on a watch with a feed that has an event history, SHOULD return all the historical events as if the pollRefresh was an add operation. If an event history is not available, then pollRefresh SHOULD act like a normal pollChanges and just return the events which have occurred since the last poll.

12 Points

Anyone familiar with automation systems immediately identifies with the term point (sometimes called tags in the industrial space). Although there are many different definitions, generally points map directly to a sensor or actuator (called hard points). Sometimes the concept of a point is mapped to a configuration variable such as a software setpoint (called soft points). In some systems point is an atomic value, and in others it encapsulates a whole truckload of status and configuration information.

The goal of oBIX is to capture a normalization representation of points without forcing an impedance mismatch on vendors trying to make their native system oBIX accessible. To meet this requirement, oBIX defines a low level abstraction for point - simply one of the primitive value types with associated status information. Point is basically just a marker contract used to tag an object as exhibiting “point” semantics:

This contract MUST only be used with the value primitive types: bool, real, enum, str, abstime, and reltime. Points SHOULD use the status attribute to convey quality information. The following table specifies how to map common control system semantics to a value type:

bool	digital point	<bool is="obix:Point" val="true"/>
real	analog point	<real is="obix:Point" val="22" unit="obix:units/celsius"/>
enum	multi-state point	<enum is="obix:Point" val="slow"/>

12.1 Writable Points

Different control systems handle point writes using a wide variety of semantics. Sometimes we write a point at a specific priority level. Sometimes we override a point for a limited period of time, after which the point falls back to a default value. The oBIX specification doesn’t attempt to impose a specific model on vendors. Rather oBIX provides a standard WritablePoint contract which may be extended with additional mixins to handle special cases. WritablePoint defines write as an operation which takes a WritePointIn structure containing the value to write. The contracts are:

</obj>

</obj>

It is implied that the value passed to writePoint match the type of the point. For example if WritablePoint is used with an enum, then writePoint MUST pass an enum for the value.

13 History

Most automation systems have the ability to persist periodic samples of point data to create a historical archive of a point’s value over time. This feature goes by many names including logs, trends, or histories. In oBIX, a history is defined as a list of time stamped point values. The following features are provided by oBIX histories:

History Object: a normalized representation for a history itself;
History Record: a record of a point sampling at a specific timestamp
History Query: a standard way to query history data as Points;
History Rollup: a standard mechanism to do basic rollups of history data;
History Append: ability to push new history records into a history;

13.1 History Object

Any object which wishes to expose itself as a standard oBIX history implements the obix:History contract:

</obj>

Let’s look at each of History’s sub-objects:

count: this field stores the number of history records contained by the history;
start: this field provides the timestamp of the oldest record. The timezone of this abstime MUST match History.tz;
end: this field provides the timestamp of the newest record. The timezone of this abstime MUST match History.tz;
tz: standardized timezone identifier for the history data (see Section 4.18.9 )
formats: this field provides a list of strings describing the formats in which the server can provide the history data.
query: the query object is used to query the history to read history records;
feed: used to subscribe to a real-time feed of history records;
rollup: this object is used to perform history rollups (it is only supported for numeric history data);
append: operation used to push new history records into the history

An example of a history which contains an hour of 15 minute temperature data:

</list>

</obj>

13.2 History Queries

Every History object contains a query operation to query the historical data. A client MAY invoke the query operation to request the data from the server as an obix:HistoryQueryOut. Alternatively, if the server is able to provide the data in a different format, such as CSV, it SHOULD list these additionally supported formats in the formats field. A client MAY then supply one of these defined formats in the HistoryFilter input query.

13.2.1 HistoryFilter

The History.query input contract:

</obj>

These fields are described in detail:

limit: an integer indicating the maximum number of records to return. Clients can use this field to throttle the amount of data returned by making it non-null. Servers MUST never return more records than the specified limit. However servers are free to return fewer records than the limit.
start: if non-null this field indicates an inclusive lower bound for the query’s time range. This value SHOULD match the history’s timezone, otherwise the server MUST normalize based on absolute time.
end: if non-null this field indicates an inclusive upper bound for the query’s time range. This value SHOULD match the history’s timezone, otherwise the server MUST normalize based on absolute time.
format: if non-null this field indicates the format that the client is requesting for the returned data. If the client uses this field the server MUST return a HistoryQueryOut with a non-null dataRef URI, or return an error if it is unable to supply the requested format. A client SHOULD use one of the formats defined in the History’s formats field when using this field in the filter.
compact: if non-null and true, this field indicates the client is requesting the data in the compact format described below. If false or null, the server MUST return the data in the standard format compatible with the 1.0 specification.

13.2.2 HistoryQueryOut

The History.query output contract:

</obj>

Just like History, every HistoryQueryOut returns count, start, and end. But unlike History, these values are for the query result, not the entire history. The actual history data is stored as a list of HistoryRecords in the data field. Remember that child order is not guaranteed in oBIX, therefore it might be common to have count after data. The start, end, and data HistoryRecord timestamps MUST have a timezone which matches History.tz.

When using a client-requested format, the server MUST provide a URI that can be followed by the client to obtain the history data in the alternate format. The exact definition of this format is out of scope of this specification, but SHOULD be agreed upon by both the client and server.

13.2.3 HistoryRecord

The HistoryRecord contract specifies a record in a history query result:

</obj>

Typically the value SHOULD be one of the value types used with obix:Point.

13.2.4 History Query Examples

Let’s examine an example query from the “/outsideAirTemp/history” example above.

13.2.4.1 History Query as oBIX objects

First let’s see how a client and server interact using the standard history query mechanism:

Client invoke request:

INVOKE http://x/outsideAirTemp/history/query

</obj>

Server response:

</list>

</obj>

Note in the example above how the data list uses a document local contract to define facets common to all the records (although we still have to flatten the contract list).

13.2.4.2 History Query as Preformatted List

Now let’s see how this might be done in a more compact format. The server in this case is able to return the history data as a CSV list.

Client invoke request:

INVOKE http://x/outsideAirTemp/history/query

</obj>

Server response:

</obj>

Client then reads the dataRef URI specified and gets:

2005-03-16T14:00:00-05:00,40

2005-03-16T14:15:00-05:00,42

2005-03-16T14:30:00-05:00,43

2005-03-16T14:45:00-05:00,47

2005-03-16T15:00:00-05:00,44

Note that the server response is NOT an oBIX document, but just arbitrarily formatted data as requested by the client – in this case text/csv. The server and client are free to use any agreed-upon format, for example, one where the timestamps are inferred rather than repeated, for maximum brevity.

13.2.5 Compact Histories

When a server contains a large number of history records, it is important to be as concise as possible when retrieving the records. The HistoryRecord format is fine for small histories, but it is not uncommon for servers to contain thousands, or tens of thousands, of data points, or even more. To allow a more concise representation of the historical data, a client MAY request that the server provide the query output in a “compact” format. This is done by setting the compact attribute of the HistoryFilter contract to true. The server MUST then respond with a CompactHistoryQueryOut if it supports compact history reporting for the referenced History, or an error if it does not.

The CompactHistoryQueryOut contract is:

</obj>

Note that the data element is narrowed to require the CompactHistoryRecord type, which is defined as:

The CompactHistoryRecord contract narrows the HistoryRecord contract to the str element type. The semantic requirements of the contract allow for a more compact representation of the record as an oBIX object, although with some restrictions:

· The timestamp and value child elements MUST be null when encoded. These are determined from the val attribute.

· The val attribute of the CompactHistoryRecord MUST be a string containing a delimited list of entities matching the record definition. The delimiter MUST be included using the delimiter element of the CompactHistoryQueryOut.

· The record definition MUST be provided in an accessible URI to the client. The record definition SHOULD be provided in a document-local contract defining the type of each item in the record, as well as any facets that apply to every record’s fields.

· The CompactHistoryRecord MUST be interpreted by inserting each item in the delimited list contained in the val attribute into the respective child element’s val attribute.

· For histories with regular collection intervals, the timestamp field MAY be left empty, if it can be inferred by the consumer. If the timestamp field is left empty on any record, the server MUST include the interval element in the HistoryQueryOut. Consumers MUST be able to handle existence or non-existence of the timestamp field. Note that this only applies when the timestamp matches the expected value based on the collection interval of the history. If a record exists at an irregular time interval, such as for skipped records or COV histories, the timestamp MUST be included in the record.

· The interpretation of the CompactHistoryRecord MUST be identical to the interpretation of a HistoryRecord with the same list of values described as child elements.

· A consumer of the CompactHistoryRecord MAY skip the actual internal conversion of the CompactHistoryRecord into its expanded form, and use a ‘smart’ decoding process to consume the list as if it were presented in the HistoryRecord form.

13.2.5.1 CompactHistoryRecord Example

Let’s look at the same scenario as in 13.2.4, expressed using CompactHistoryRecords. The server is providing additional information with certain elements; this is reflected in the record definition at the end.

Client invoke request:

INVOKE http://x/outsideAirTemp/history/query

</obj>

Server response:

</list>

</obj>

13.3 History Rollups

Control systems collect historical data as raw time sampled values. However, most applications wish to consume historical data in a summarized form which we call rollups. The rollup operation is used to summarize an interval of time. History rollups only apply to histories which store numeric information. Attempting to query a rollup on a non-numeric history SHOULD result in an error.

13.3.1 HistoryRollupIn

The History.rollup input contract extends HistoryFilter to add an interval parameter:

</obj>

13.3.2 HistoryRollupOut

The History.rollup output contract:

</obj>

The HistoryRollupOut object looks very much like HistoryQueryOut except it returns a list of HistoryRollupRecords, rather than HistoryRecords. Note: unlike HistoryQueryOut, the start for HistoryRollupOut is exclusive, not inclusive. This issue is discussed in greater detail next. The start, end, and data HistoryRollupRecord timestamps MUST have a timezone which matches History.tz.

13.3.3 HistoryRollupRecord

A history rollup returns a list of HistoryRollupRecords:

</obj>

The children are defined as:

start: the exclusive start time of the record’s rollup interval;
end: the inclusive end time of the record’s rollup interval;
count: the number of records used to compute this rollup interval;
min: specifies the minimum value of all the records within the interval;
max: specifies the maximum value of all the records within the interval;
avg: specifies the mathematical average of all the values within the interval;
sum: specifies the summation of all the values within the interval;

13.3.4 Rollup Calculation

The best way to understand how rollup calculations work is through an example. Let’s consider a history of meter data where we collected two hours of 15 minute readings of kilowatt values:

</list>

<obj>

</obj>

If we were to query the rollup using an interval of 1 hour with a start time of 12:00 and end time of 14:00, the result should be:

<obj>

<abstime name="start" val="2005-03-16T12:00:00+04:00"

tz="Asia/Dubai"/>

<abstime name="end" val="2005-03-16T13:00:00+04:00"

tz="Asia/Dubai"/>

</obj>

<obj>

<abstime name="start" val="2005-03-16T13:00:00+04:00"

tz="Asia/Dubai"/>

<abstime name="end" val="2005-03-16T14:00:00+04:00"

tz="Asia/Dubai"/>

</obj>

</list>

</obj>

If you whip out your calculator, the first thing you will note is that the first raw record of 80kW was never used in the rollup. This is because start time is always exclusive. The reason start time has to be exclusive is because we are summarizing discrete samples into a contiguous time range. It would be incorrect to include a record in two different rollup intervals! To avoid this problem we always make start time exclusive and end time inclusive. The following table illustrates how the raw records were applied to rollup intervals:

Interval Start (exclusive)	Interval End (inclusive)	Records Included
2005-03-16T12:00	2005-03-16T13:00	82 + 90 + 85 + 81 = 338
2005-03-16T13:00	2005-03-16T14:00	84 + 91 + 83 + 78 = 336

13.4 History Feeds

The History contract specifies a feed for subscribing to a real-time feed of the history records. History.feed reuses the same HistoryFilter input contract used by History.query – the same semantics apply. When adding a History feed to a watch, the initial result SHOULD contain the list of HistoryRecords filtered by the input parameter (the initial result should match what History.query would return). Subsequent calls to Watch.pollChanges SHOULD return any new HistoryRecords which have been collected since the last poll that also satisfy the HistoryFilter.

13.5 History Append

The History.append operation allows a client to push new HistoryRecords into a History log (assuming proper security credentials). This operation comes in handy when bi-direction HTTP connectivity is not available. For example if a device in the field is behind a firewall, it can still push history data on an interval basis to a server using the append operation.

13.5.1 HistoryAppendIn

The History.append input contract:

</obj>

The HistoryAppendIn is a wrapper for the list of HistoryRecords to be inserted into the History. The HistoryRecords SHOULD use a timestamp which matches History.tz. If the timezone doesn’t match, then the server MUST normalize to its configured timezone based on absolute time. The HistoryRecords in the data list MUST be sorted by timestamp from oldest to newest, and MUST not include a timestamp equal to or older than History.end.

13.5.2 HistoryAppendOut

The History.append output contract:

</obj>

The output of the append operation returns the number of new records appended to the History and the new total count, start time, and end time of the entire History. The newStart and newEnd timestamps MUST have a timezone which matches History.tz.

14 Alarming

The oBIX alarming feature specifies a normalized model to query, watch, and acknowledge alarms. In oBIX, an alarm indicates a condition which requires notification of either a user or another application. In many cases an alarm requires acknowledgement, indicating that someone (or something) has taken action to resolve the alarm condition. The typical lifecycle of an alarm is:

Source Monitoring: algorithms in a server monitor an alarm source. An alarm source is an object with an href which has the potential to generate an alarm. Example of alarm sources might include sensor points (this room is too hot), hardware problems (disk is full), or applications (building is consuming too much energy at current energy rates)
Alarm Generation: if the algorithms in the server detect that an alarm source has entered an alarm condition, then an alarm record is generated. Every alarm is uniquely identified using an href and represented using the obix:Alarm contract. Sometimes we refer to the alarm transition as off-normal.
To Normal: many alarm sources are said to be stateful - eventually the alarm source exits the alarm state, and is said to return to-normal. Stateful alarms implement the obix:StatefulAlarm contract. When the source transitions to normal, we update normalTimestamp of the alarm.
Acknowledgement: often we require that a user or application acknowledges that they have processed an alarm. These alarms implement the obix:AckAlarm contract. When the alarm is acknowledged, we update ackTimestamp and ackUser.

14.1 Alarm States

Alarm state is summarized with two variables:

In Alarm: is the alarm source currently in the alarm condition or in the normal condition. This variable maps to the alarm status state.
Acknowledged: is the alarm acknowledged or unacknowledged. This variable maps to the unacked status state.

Either of these states may transition independent of the other. For example an alarm source can return to normal before or after an alarm has been acknowledged. Furthermore it is not uncommon to transition between normal and off-normal multiple times generating several alarm records before any acknowledgements occur.

Note not all alarms have state. An alarm which implements neither StatefulAlarm nor the AckAlarm contracts is completely stateless – these alarms merely represent event. An alarm which implements StatefulAlarm but not AckAlarm will have an in-alarm state, but not acknowledgement state. Conversely an alarm which implements AckAlarm but not StatefulAlarm will have an acknowledgement state, but not in-alarm state.

14.1.1 Alarm Source

The current alarm state of an alarm source is represented using the status attribute. This attribute is discussed in Section 4.18.8. It is recommended that alarm sources always report their status via the status attribute.

14.1.2 StatefulAlarm and AckAlarm

An Alarm record is used to summarize the entire lifecycle of an alarm event. If the alarm implements StatefulAlarm it tracks transition from off-normal back to normal. If the alarm implements AckAlarm, then it also summarizes the acknowledgement. This allows for four discrete alarm states:

alarm	acked	normalTimestamp	ackTimestamp
true	false	null	null
true	true	null	non-null
false	false	non-null	null
false	true	non-null	non-null

14.2 Alarm Contracts

14.2.1 Alarm

The core Alarm contract is:

</obj>

The child objects are:

source: the URI which identifies the alarm source. The source SHOULD reference an oBIX object which models the entity that generated the alarm.
timestamp: this is the time at which the alarm source transitioned from normal to off-normal and the Alarm record was created.

14.2.2 StatefulAlarm

Alarms which represent an alarm state which may transition back to normal SHOULD implement the StatefulAlarm contract:

</obj>

The child object is:

normalTimestamp: if the alarm source is still in the alarm condition, then this field is null. Otherwise this indicates the time of the transition back to the normal condition.

14.2.3 AckAlarm

Alarms which support acknowledgement SHOULD implement the AckAlarm contract:

</obj>

</obj>

</obj>

The child objects are:

ackTimestamp: if the alarm is unacknowledged, then this field is null. Otherwise this indicates the time of the acknowledgement.
ackUser: if the alarm is unacknowledged, then this field is null. Otherwise this field should provide a string indicating who was responsible for the acknowledgement.

The ack operation is used to programmatically acknowledge the alarm. The client may optionally specify an ackUser string via AckAlarmIn. However, the server is free to ignore this field depending on security conditions. For example a highly trusted client may be allowed to specify its own ackUser, but a less trustworthy client may have its ackUser predefined based on the authentication credentials of the protocol binding. The ack operation returns an AckAlarmOut which contains the updated alarm record. Use the Lobby.batch operation to efficiently acknowledge a set of alarms.

14.2.4 PointAlarms

It is very common for an alarm source to be an obix:Point. A respective PointAlarm contract is provided as a normalized way to report the value which caused the alarm condition:

</obj>

The alarmValue object SHOULD be one of the value types defined for obix:Point in Section 12.

14.3 AlarmSubject

Servers which implement oBIX alarming MUST provide one or more objects which implement the AlarmSubject contract. The AlarmSubject contract provides the ability to categorize and group the sets of alarms a client may discover, query, and watch. For instance a server could provide one AlarmSubject for all alarms and other AlarmSubjects based on priority or time of day. The contract for AlarmSubject is:

</obj>

</obj>

</obj>

The AlarmSubject follows the same design pattern as History. The AlarmSubject specifies the active count of alarms; however, unlike History it does not provide the start and end bounding timestamps. It contains a query operation to read the current list of alarms with an AlarmFilter to filter by time bounds. AlarmSubject also contains a feed object which may be used to subscribe to the alarm events.

14.4 Alarm Feed Example

The following example illustrates how a feed works with this AlarmSubject:

</obj>

The server indicates it has two open alarms under the specified AlarmSubject. If a client were to add the AlarmSubject’s feed to a watch:

</obj>

</uri>

</list>

</obj>

</obj>

</obj>

</feed>

</list>

</obj>

The watch returns the historic list of alarm events which is two open alarms. The first alarm indicates an out of bounds condition in AirHandler-2’s return temperature. The second alarm indicates that the system has detected that the front door has been propped open.

Now let’s fictionalize that the system detects the front door is closed, and alarm point transitions to the normal state. The next time the client polls the watch the alarm would show up in the feed list (along with any additional changes or new alarms not shown here):

</obj>

</feed>

</list>

</obj>

15 Security

Security is a broad topic, that covers many issues:

Authentication: verifying a user (client) is who he says he is;
Encryption: protecting oBIX documents from prying eyes;
Permissions: checking a user’s permissions before granting access to read/write objects or invoke operations;
User Management: managing user accounts and permissions levels;

The basic philosophy of oBIX is to leave these issues outside of the specification. Authentication and encryption is left as a protocol binding issue. Privileges and user management is left as a vendor implementation issue. Although it is entirely possible to define a publicly exposed user management model through oBIX, this specification does not define any standard contracts for user management.

15.1 Error Handling

It is expected that an oBIX server will perform authentication and utilize those user credentials for checking permissions before processing read, write, and invoke requests. As a general rule, servers SHOULD return err with the obix:PermissionErr contract to indicate a client lacks the permission to perform a request. In particularly sensitive applications, a server may instead choose to return BadUriErr so that an untrustworthy client is unaware that a specific object even exists.

15.2 Permission-based Degradation

Servers SHOULD strive to present their object model to a client based on the privileges available to the client. This behavior is called permission based degradation. The following rules summarize effective permission based degradation:

If an object cannot be read, then it SHOULD NOT be discoverable through objects which are available.
Servers SHOULD attempt to group standard contracts within the same privilege level – for example don’t split obix:History’s start and end into two different security levels such that a client might be able to read start, and not end.
Servers SHOULD NOT include a contract in an object’s is attribute if the contract’s children are not readable to the client.
If an object isn’t writable, then the writable attribute SHOULD be set to false (either explicitly or through a contract default).
If an op inherited from a visible contract cannot be invoked, then the server SHOULD set the null attribute to true to disable it.

16 Conformance

An implementation is conformant with this specification if it satisfies all of the MUST and REQUIRED level requirements defined herein for the functions implemented. Normative text within this specification takes precedence over normative outlines, which in turn take precedence over the XML Schema [XML Schema Part 1, Part 2] and WSDL [WSDL 1.1] descriptions, which in turn take precedence over examples.

An implementation is a conforming oBIX Server if it meets the conditions described in Section 16.1. An implementation is a conforming oBIX Client if it meets the conditions described in Section 16.2. An implementation is a conforming oBIX Server and a conforming oBIX Client if it meets the conditions of both Section 16.1 and Section 16.2.

16.1 Conditions for a Conforming oBIX Server

An implementation conforms to this specification as an oBIX Server if it meets the conditions described in the following subsections. oBIX servers MUST implement the oBIX Lobby object.

16.1.1 Lobby

A conforming oBIX server MUST meet the following conditions to satisfy the Lobby Conformance Clause:

1. oBIX Servers MUST have an accessible object which implements the obix:Lobby contract.

2. The Lobby MUST provide a <ref> to an object which implements the obix:About contract.

3. The Lobby MUST provide a <ref> to an object which implements the obix:WatchService contract.

4. The Lobby MUST provide an <op> to invoke batch operations using the obix:BatchIn and obix:BatchOut contracts.

16.1.2 Bindings

An implementation MUST support one of the bindings defined in the companion documents to this specification that describe oBIX Bindings.

16.1.3 Encodings

An implementation MUST support one of the encodings defined in the companion document to this specification, Encodings for oBIX An implementation SHOULD support the XML encoding, as this encoding is used by the majority of oBIX implementations. An implementation MUST support negotiation of the encoding to be used with a client according to the mechanism defined for the specific binding used.

An implementation MUST return values according to the rules defined in Section 4. For example, an implementation MUST encode bool objects’ val attribute using the literals “true” and “false” only.

16.1.4 Contracts

An implementation MUST flatten contract hierarchies when reporting them in an oBIX document, according to Section 6.6.1.

16.2 Conditions for a Conforming oBIX Client

An implementation conforms to this specification as an oBIX Client if it meets the conditions described in the following subsections.

16.2.1 Encoding

An implementation MUST support one of the encodings defined in this specification. An implementation SHOULD support the XML encoding, as this encoding is used by the majority of oBIX implementations. An implementation MUST support negotiation of which encoding to use in communicating with an oBIX server using the mechanism defined for the binding being used.

16.2.2 Naming

An implementation MUST be able to interpret and navigate URI schemes according to the general rules described in section 5.3. An implementation SHOULD be able to interpret and navigate HTTP URIs, as this is used by the majority of oBIX Server implementations.

16.2.3 Contracts

An implementation MUST be able to consume and use oBIX contracts defined by oBIX Server implementations with which it interacts.

Appendix A. Acknowledgments

The following individuals have participated in the creation of this specification and are gratefully acknowledged:

Participants:

Ron Ambrosio, IBM

Brad Benson, Trane

Ron Bernstein, LonMark International*

Ludo Bertsch, Continental Automated Buildings Association (CABA)

Chris Bogen, US Department of Defense

Rich Blomseth, Echelon Corporation

Anto Budiardjo, Clasma Events, Inc.

Jochen Burkhardt, IBM

JungIn Choi, Kyungwon University

David Clute, Cisco Systems, Inc.*

Toby Considine, University of North Carolina at Chapel Hill

William Cox, Individual

Robert Dolin, Echelon Corporation

Marek Dziedzic, Treasury Board of Canada, Secretariat

Brian Frank, SkyFoundry

Craig Gemmill, Tridium, Inc.

Matthew Giannini, Tridium, Inc.

Harald Hofstätter, Institute of Computer Aided Automation

Markus Jung, Institute of Computer Aided Automation

Christopher Kelly, Cisco Systems

Wonsuk Ko, Kyungwon University

Perry Krol, TIBCO Software Inc.

Corey Leong, Individual

Ulf Magnusson, Schneider Electric

Brian Meyers, Trane

Jeremy Roberts, LonMark International

Thorsten Roggendorf, Echelon Corporation

Anno Scholten, Individual

John Sublett, Tridium, Inc.

Dave Uden, Trane

Ron Zimmer, Continental Automated Buildings Association (CABA)*

Robert Zach, Institute of Computer Aided Automation

Rob Zivney, Hirsch Electronics Corporation

Markus Jung, Vienna University of Technology

Appendix B. Revision History

Revision	Date	Editor	Changes Made
wd-0.1	14 Jan 03	Brian Frank	Initial version
wd-0.2	22 Jan 03	Brian Frank
wd-0.3	30 Aug 04	Brian Frank	Move to Oasis, SysService
wd-0.4	2 Sep 04	Brian Frank	Status
wd-0.5	12 Oct 04	Brian Frank	Namespaces, Writes, Poll
wd-0.6	2 Dec 04	Brian Frank	Incorporate schema comments
wd-0.7	17 Mar 05	Brian Frank	URI, REST, Prototypes, History
wd-0.8	19 Dec 05	Brian Frank	Contracts, Ops
wd-0.9	8 Feb 06	Brian Frank	Watches, Alarming, Bindings
wd-0.10	13 Mar 06	Brian Frank	Overview, XML, clarifications
wd-0.11	20 Apr 06	Brian Frank	10.1 sections, ack, min/max
wd-0.11.1	28 Apr 06	Aaron Hansen	WSDL Corrections
wd-0.12	22 May 06	Brian Frank	Status, feeds, no deltas
wd-0.12.1	29 Jun 06	Brian Frank	Schema, stdlib corrections
obix-1.0-cd-02	30 Jun 06	Aaron Hansen	OASIS document format compliance.
obix-1.0-cs-01	18 Oct 06	Brian Frank	Public review comments
wd-obix.1.1.1	26 Nov 07	Brian Frank	Fixes, date, time, tz
wd-obix.1.1.2	11 Nov 08	Craig Gemmill (from Aaron Hansen)	Add iCalendar scheduling
wd-obix-1.1.3	10 Oct 09	Brian Frank	Remove Scheduling chapter Rev namespace to 1.1 Add Binary Encoding chapter
wd-obix-1.1.4	12 Nov 09	Brian Frank	MUST, SHOULD, MAY History.tz, History.append HTTP Content Negotiation
oBIX-1-1-spec-wd05	01 Jun 10	Toby Considine	Updated to current OASIS Templates, requirements
oBIX-1-1-spec-wd06	08 Jun 10	Brad Benson	Custom facets within binary encoding
oBIX-1-1-spec-wd07	03 Mar 2013	Craig Gemmill	Update to current OASIS templates, fixes
oBIX-1-1-spec-wd08	27 Mar 2013	Craig Gemmill	Changes from feedback
obix-v1.1-wd09	23 Apr 2013	Craig Gemmill	Update to new OASIS template Add of attribute to obix:ref Define additional list semantics Clarify writable w.r.t. add/remove of children Add deletion semantics Add encoding negotiation
obix-v1.1-wd10	08 May 2013	Craig Gemmill	Add CompactHistoryRecord Add preformatted History query Add metadata for alternate hierarchies (tagging)
obix-v1.1-wd11	13 Jun 2013	Craig Gemmill	Modify compact histories per TC feedback
obix-v1.1-wd12	27 Jun 2013	Craig Gemmill	Add delimiter, interval to compact histories
obix-v1.1-wd13	8 July 2013	Toby Considine	Replaced object diagram w/ UML Updated references to other oBIX artifacts