HyPE can be configured in four ways. First, modify the Static_Configuration of HyPE, which sets default values for all variables. Second, update the configuration at runtime. The class Runtime_Configuration provides methods to change all modifiable variables. Note that not all variables are modifiable during runtime. Third, create a configuration file 'hype.conf', and add the variables with their corresponding values. Note, that the structure of the file for each line is variable_name=value and one line may at most contain one assignment. Fourth, specify parameter values in environment variables.
- modify the hype::core::Static_Configuration of HyPE (requires recompilation)
- update the configuration at runtime using hype::core::Runtime_Configuration
- create a configuration file 'stemod.conf', and add the variables with their corresponding values
- help produce help message
- length_of_trainingphase set the number algorithms executions to complete training
- history_length set the number of measurement pairs that are kept in the history (important for precision of approximation functions)
- recomputation_period set the number of algorithm executions to trigger recomputation
- algorithm_maximal_idle_time set maximal number of operation executions, where an algorithm was not executed; forces retraining of algorithm
- retraining_length set the number of algorithm executions needed to complete a retraining phase (load adaption feature)
- ready_queue_length set the number of operators that are queued on a processing device, before scheduling decision stops scheduling new operators (The idea is to wait how the done scheduling decisions turn out and to adjust the scheduling accordingly)
- specify parameter values in environment variables:
- HYPE_LENGTH_OF_TRAININGPHASE set the number algorithms executions to complete training
- HYPE_HISTORY_LENGTH set the number of measurement pairs that are kept in the history (important for precision of approximation functions)
- HYPE_RECOMPUTATION_PERIOD set the number of algorithm executions needed to complete a retraining phase (load adaption feature)
- HYPE_ALGORITHM_MAXIMAL_IDLE_TIME set maximal number of operation executions, where an algorithm was not executed; forces retraining of algorithm
- HYPE_RETRAINING_LENGTH set the number of algorithm executions to trigger recomputation
- HYPE_READY_QUEUE_LENGTH set the number of operators that are queued on a processing device, before scheduling decision stops scheduling new operators (The idea is to wait how the done scheduling decisions turn out and to adjust the scheduling accordingly)