You can deploy a model in simulation mode to run it against historical input data (such as Cumulocity IoT measurements). This allows testing the behavior of a newly developed model against historical data or fine-tuning an existing model. Or it allows testing a model against a set of historical data with known properties.

You use the model manager to deploy a model in simulation mode. See Deploying a model for more details.

When a model is deployed in simulation mode, it uses data from a virtual device (see also Virtual devices). Thus, a simulated model can run alongside other non-simulated models without interfering with them.

A simulated model runs as if it is running at the time of the historical data. The input data are processed in the order of their historical time. The simulated model also uses the historical time for the timestamps of the generated output.

Events, alarms and operations are created with a timestamp. However, with time there can be updates to these objects. For example, an alarm can be cleared or the status of an operation can be changed. As a history of changes to event, alarm and operation objects is not maintained, the object is only replayed at its initial timestamp, with the latest version of its properties. Thus, changes to these objects are not replayed and simulation mode is of limited use if your models depend on changes to objects.

When running a simulation, historical data is replayed into the Apama correlator from the Cumulocity IoT database. If there is a significant delay in the data being queried from the database or high load in the system, this can lead to dropping the input in exceptional circumstances. A simulated model processes input data at normal speed. For example, if the historical data entries are separated by one second, they are processed one second apart. This means that simulating a model with one hour of historical data will take approximately one hour of simulation time.