Logo of 52°North

Sensor Event Service (SES)

The Sensor Event Service (SES) is an enhancement of the Sensor Alert Service (SAS). Both are used to provide an publish / subscribe based access to sensor data and measurements. They also provide optional methods to dynamically add new sensors and send notifications to the service (see Figure 1). The SES specification is currently available as an OGC discussion paper (OGC 08-133). A comprehensive engineering report about the SWE Event Architecture elaborated in the OWS-6 project can be found here.

Figure 1: Overview of the SES

The major enhancements are the SOAP binding, the increased filter capabilities and the automatic unit conversion functionality.

SOAP Binding

The default binding of the SES is based upon SOAP and WS-Notification but the specification allows the definition of other bindings for instance an XMPP binding compatible to the SAS. In the basic version the SES acts as a notification producer (the mandatory part in Figure 1). Figure 2 shows the methods provided by the basic SES interface. The notification producer interface provides methods to subscribe for notifications and retrieve the latest notification. The basic SES interface adds common methods from OGC and SWE. These are for instance necessary for service discovery.

Figure 2: UML diagram of the basis SES interface (source: Echterhoff and Everding 2008)

The extended version of the SES encompasses also the optional part of Figure 1. The extended SES interface therefore implements the basic SES interface and two interfaces from the brokered notification package of WS-Notification, namely the notification consumer and the register publisher interface (see Figure 3).

Figure 3: UML diagram of the extended SES interface (source: Echterhoff, Everding 2008)

The additional methods provided by the extended SES interface allow to register new publishers at the SES and to send notifications to it. In terms of WS-Notification the extended SES is a notification broker. All requests sent to the SES are encoded as SOAP messages. Examples can be found in the SES specification.

Filtering

The notifications received by the SES are posted to topics. These topics define logical groups of notifications. They are declared following the WS-Topics standard. Figure 4 shows an overview of the topics pre-defined in the SES specification. Further topics can be added as needed. Sensor notifications are for instance posted to the measurements topic, notifications regarding the management of sensors are posted to the sensor management topic.

Figure 4: Pre-defined topics of the SES (source: Echterhoff, Everding 2008)

When subscribing for notifications at the SES one can filter the notifications by topic. A system administrator for instance could only subscribe for sensor management notifications since he is not interested in the actual measurements. In addition to the topic based filtering one can define content based filter criteria to further restrict the notifications. This was already possible at the SAS but has been extended since the filter capabilities of the SAS were far from optimal. The SAS only offered two spatial filters (within bounding box and at a location) and a small set of content filters (smaller than, greater than, equals, not equal to). More enhanced spatial filters, temporal filters as well as further comparison operators and logical operators are not supported by the SAS.

The SES addresses the content based filtering with a multi stage approach. Therefore the specification defines three filter levels each allowing more complex filters. It allows to easily extend the filter capabilities of the SES later on. The first filter level is the only mandatory one. It uses XPath expressions to define the filter statements. Using filter level 1 one can define content based filter criteria based on the syntax of a notification.

The filtering mechanism that comes closest to the SAS filtering is filter level 2. But instead of defining an own filter language this level relies on the OGC filter encoding. In the specification of the SES the filter encoding was also extended to address some shortcomings regarding temporal filtering. In future versions of the SES version 2 of the filter encoding should be used for filter level 2. Nonetheless both filter encodings eliminate the shortcomings of the SAS filter capabilities described above.

The third filter level of the SES uses Complex Event Processing (CEP) and Event Stream Processing (ESP) to derive information from the stream of incoming sensor notifications. The rules for CEP and ESP are formalized in so called event patterns encoded in an event pattern language. The SES makes use of the Event Pattern Markup Language (EML) which was developed in parallel. As the concepts of CEP and ESP are not obvious to understand they are further outlined in the next section.

Event Processing

Event Processing in common is creating, deleting, reading and editing of and reacting to events (Chandy and Schulte 2007). In this context events can be defined as "anything that happens or is contemplated as happening" (Luckham and Schulte 2008) which is a very broad definition. It also makes no distinction between an event as a real world phenomenon and an event object as the representation in a computer of a real world event. Following this definition the term event can have both meanings whereas the actual meaning is given by the context (Luckham, Schulte 2008). Other definitions often add further restrictions for instance on temporal aspects of events. As these restrictions do not influence the ideas of event processing they are not further discussed.

Complex Event Processing (CEP) is a specialization of Event Processing. In CEP relations between multiple events are used to derive higher level information. The relations may be temporal (e.g. A followed by B) spatial (e.g. A intersects B), causal (e.g. A is cause of B) or of any other relation type that can be applied to events. The derived information is usually stored in complex events. These are event objects representing a complex real world phenomenon which consists of multiple sub-phenomenon. Likewise a complex event object can store references to its sub-event objects. As an example the event "Buy a car" consists of the sub-events like "Select a favorite car", "Gather money for a new car", "Negotiate the price", "Pay the price" and so on. An event like "Old car broke down" may be the cause for "Buy a car" (causal relationship) and therefore also be referenced as sub-event.

EML

The Event Pattern Markup Language (EML) which is used in the third SES filter level is a means to describe rules for CEP and ESP so called event patterns. Its specification is available as OGC discussion paper (OGC 08-132).

The EML offers a set of different types of event patterns. Simple patterns are used for common event processing since they operate on only one event input (e.g. one sensor). They can be used to filter incoming events as done in SES filter level 2 but also allow deriving values like the average or the maximum from a set of events (e.g. sensor measurements). Complex patterns have two events streams as input thus allowing to apply CEP functionality. The inputs of complex patterns are always the results of other event pattern. Only simple patterns can be connected to an external event stream.

In addition to these two pattern types the EML offers two types for specialized tasks. Repetitive patterns can be used to count event satisfying a given criterion. Timer patterns can be used to generate events depending on the clock for instance one event per minute. In the following paragraphs only simple and complex patterns are taken into account because some statements do not apply for the specialized patterns.

An event pattern in EML defines a set of select functions, a view (see ESP above) and a guard. This can be compared to the standard SQL phrase "SELECT FROM WHERE". The select functions obviously are the SELECT part defining how the results of an event pattern are calculated (e.g. the maximum of the available events, an average, ...) and also what is done with the result. This can for instance be further use in other (complex) event patterns or the output for instance to a client subscribed at the SES. The view is the FROM part of the pattern. It contains a subset of the received events. This may for example be the events received in the last five minutes or the newest ten events. The guards are the WHERE part of an event pattern. In EML a guard is defined using the OGC filter encoding as in the SES filter level 2. This filter is applied to every incoming event.

The execution of an event pattern is performed in the opposite order. At first the actual events has to be available. For simple patterns this is always the case if a new event is received on the specified input whereas for complex patterns the defined relation between these events has to be present. If a new event or set of events are available the guard is executed. Only if the guard is satisfied the events are forwarded to the view for insertion (which might be rejected). Afterwards the view decides whether the select functions are triggered. If this is the case the select function are executed on the set of events provided by the view.

To demonstrate the benefits of using CEP a short example is given. In this example the temperature in a room is observed. When the temperature exceeds a given threshold an alarm shall be triggered. When using an SAS, this alarm would be triggered every time a measurements reports a too high temperature which may be very often (e.g. once per second). On the one hand this can be very annoying but on the other hand this is also not suitable to trigger an automatic notification of the fire department or other human user. Figure 5 shows a stream with temperature measurements and the SAS-filtered result stream.

Figure 5: Simple filtering of an event stream

When using SES and EML, one can subscribe for notifications when the threshold is crossed instead of just too high temperatures. Therefore one uses two simple patterns, one to filter too high measurements like the SAS does and one to filter measurements that do not exceed the threshold. The results are then combined in a complex pattern which matches on "a low temperature is followed by a too high temperature" and thus detecting when the threshold is crossed and publishing the result as one temperature warning. A second complex pattern can easily be added to detect when a too high temperature is followed by a low temperature to generate all clear messages. Figure 6 gives an overview of the event processing tanking place.

Figure 6: Complex filtering of an event stream

The results produced by the complex patterns described above can be used in a further step to build a fire detector. This fire detector would be based on temperature and smoke measurements an would trigger an alert if the temperature id too high and smoke is detected. Taking the results of the complex patterns above and smoke readings as inputs one can apply two simple patterns filtering temperature warnings and too much smoke and combine their results in a complex pattern. If both is found a fire alarm is triggered. This process is illustrated in Figure 7.

Figure 7: Fire detection using Complex Event Processing

Unit Conversion

Another new feature of the SES is the on the fly unit of measurement conversion. Internally the SES only operates on so called SI units (Systeme international d'unites) which is a set of base units for physical phenomena. Nearly every physical unit can be expressed as a combination of these SI units.

When receiving new measurements all values are automatically transferred to SI units. The same is done with values given in subscription requests. This allows users on the one hand to easily specify the desired output units used in notifications and on the other hand to subscribe also for sensor measurements which use different units.

For example if one is interested in notifications if temperatures below the freezing point are measured one could subscribe for notifications of measurements below 0 degree Celsius. If no unit conversion is used measurements using Fahrenheit as unit for temperatures would either have to be ignored or would produce incorrect results (e.g. 20 degree Fahrenheit is below 0 degree Celsius but would not result in a notification because 20 is larger than 0). In this case, the SES coverts all values to degree Kelvin which is the SI unit for temperatures. This way all measurements can be used in the filtering process, produce correct results and can be published in any compatible unit of measurement.

In order to specify the units of measurement it is recommended to use UCUM codes (Footnote: see also http://unitsofmeasure.org/). These codes can be interpreted by machines but also easily read by humans as they are based on common abbreviations for units such as m for meter. Also URNs can be used to specify unit of measurements for instance as defined in the OGC document 06-023r1.

Code Manager

  • Matthes Rieke

Contributors