Prediction Engine¶
The
com.atc.simulator.prediction_service.engine
package forms the core of the Prediction Service.
Diagrams
Class Design
PredictionEngineThread¶
Within this package, the PredictionEngineThread is a thread which
runs continuously, waiting for updates to the com.atc.simulator.flightdata.system_state_database.SystemStateDatabase
to occur. Each update results in the creation of a new
PredictionWorkItem. The new work item is then placed in a queue to
later be assigned to one of the Workers belonging to the prediction
engine.
PredictionEngineListener¶
For classes wishing to interact with the output of the prediction engine, there
is the PredictionEngineListener. The primary example of this is the
PredictionFeedServerThread which waits for prediction work items
to complete in the prediction engine, and broadcasts the predictions to its
clients.
PredictionWorkItem¶
The PredictionWorkItem class is designed to be a self contained
unit of work for the Workers to work on to produce predictions using
various Algorithms. It is used to keep track of what work is being done
and by whom, and stores a complete thread-safe copy of the data required for
the worker to perform the work required to complete this work item.
The PredictionWorkItem is also used to help track of the realtime status of work items and when they are complete This gains useful information on the performance of the system in order to accurately identify any major performance flaws if and when they occur.
Workers¶
Workers do work on PredictionWorkItem for the prediction engine. There are currently two types of workers in the project.
Java Prediction Worker¶
The workers.JavaPredictionWorkerThread is a pure java thread which performs
the job of a worker for the prediction engine. Its algorithm implementations
are stored here: com.atc.simulator.prediction_service.engine.algorithms
OpenCL Prediction Worker¶
The workers.OpenCLPredictionWorkerThread is a java thread, which uses the
OpenCL api to perform its work item
computations using specialised parallel compute hardware. The implementation
is currently optimised for use on an NVIDIA GTX760 graphics processing unit.
The front end to its algorithm implementations are stored here:
com.atc.simulator.prediction_service.engine.algorithms.opencl
The back end to its algorithm implementations, the OpenCL C code is stored in the project assets directory: /assets/opencl/