Starting in HELICS v3.3 Callback federates were added as way to entirely inline the operation of a federate, allowing significant increases in the number of federates that could partake in a co-simulation for a given system size. Callback federates are a specific type of federate that adds some additional requirements when implementing a federate. This includes a couple new callbacks and properties not available in a regular federate.
Expected Use Cases¶
Callback federates allow the potential for a large number of small federates to execute in a core, up to 131,072 in each core. Trying this with normal federates rapidly exceeds the thread capacity of a typical system, so in the cases where the federate can be defined in terms of callbacks and has only a few simple connections and minimal computation the callback federate will be much more efficient. Since all callback federates execute serially internally to the core, the callback federate is not appropriate for federates with significant computation load as it will block the execution of the other callback federates.
The Callback Federate adds a few additional callbacks necessary for operation.
The initialize callback’s purpose is to allow a callback federate to attempt to enter executing mode iteratively if desired. The return value is an IterationRequest enumeration. If the callback is not specified for a callback federate NO_ITERATIONS is assumed.
void setInitializeCallback(std::function<IterationRequest()> initializeCallback); void helicsCallbackFederateInitializeCallback(HelicsFederate fed, HelicsIterationRequest (*initialize)(void* userdata), void* userdata, HelicsError* err);
The purpose of the following callbacks is to allow a user to specify what the next time request should be. This request can be a simple time value (integer) or an iteration_time if iterations are needed. If the iteration time callback method is specified it overrides the simpler callback. The callback must be cleared if desired to revert either through a NULL object or the
void setNextTimeCallback(std::function<Time(Time)> nextTimeCallback) void helicsCallbackFederateNextTimeCallback(HelicsFederate fed, HelicsTime (*timeUpdate)(HelicsTime time, void* userdata), void* userdata, HelicsError* err);
void setNextTimeIterativeCallback( std::function<std::pair<Time, IterationRequest>(iteration_time)> nextTimeCallback); void helicsCallbackFederateNextTimeIterativeCallback( HelicsFederate fed, HelicsTime (*timeUpdate)(HelicsTime time, HelicsIterationResult, HelicsIterationRequest* iteration, void* userdata), void* userdata, HelicsError* err);
There is also a method to clear the time callbacks.
All callback are optional.
Callback federates define an additional property
HELICS_PROPERTY_TIME_MAXTIME. This property allows the callback federate to finalize once a time returned is greater or equal to the specified maxtime.
Other key properties¶
Specifying a period
HELICS_PROPERTY_TIME_MAXTIME will allow the callback federate to automatically compute the next time request with no additional callback for the time computation.
Specifying that the federate is event-driven will send the next time request as
HELICS_TIME_MAXTIME which will grant only when there is some new data to process. This can specified with
Callback federates are Combination federates so any operation that works on Combination federates works on Callback federates. Callback federates behave identically to regular federates until a call is made to
enterInitializingMode(). In regular federates this call is blocking; in Callback federates this call is an asynchronous call and will transfer control of the federate to the HELICS core. All other blocking calls in HELICS will produce an error. All callbacks defined in a regular federate are available for use and are expected to be used. The additional callbacks defined here are to control the next time step and iterations which are directly user controlled in normal federates.