Berg, T.W. van den
|Source:||2014 Fall Simulation Interoperability Workshops (SIW), 8-12 September 2014, Orlando, Florida, USA, 317-328|
Simulation · Cloud computing · Cost effectiveness · Data mining · Decision support systems · Distributed simulation environments · High level architecture · Identified parameter · Performance assessment · Performance measure · Run-time infrastructure · Simulation performance · Training applications · Parameter estimation · Defence Research · Defence, Safety and Security · Operations Modelling · MSG - Modelling Simulation & Gaming · ELSS - Earth, Life and Social Sciences
Performance assessment is a key factor in designing distributed simulation environments that are fit-forpurpose and cost-effective. Simulations used for training applications should provide the required level of responsiveness and interactivity. Simulations used for analysis or decision support should execute as fast as possible to enable quick results on large numbers of scenarios and variables. There are many parameters that have an impact on the performance of a typical High Level Architecture (HLA) federation. We describe a structured process for using data farming (experiment design, simulation, cloud computing, and data mining) in order to identify key parameters affecting the performance of HLA federations. A parameterized and instrumented federation is designed with parameters covering areas suspected of impacting HLA performance and instrumentation measuring key values of HLA performance. One key parameter set for HLA performance is the computing platform that hosts the federation. Varying such a platform in an efficient and cost effective manner is possible using cloud computing via Infrastructure-as-a- Service providers. Finally, the performance measures captured during federation execution are collated and analysed using data mining techniques to identify key parameters and their effect on the performance of the federation. By including the Runtime Infrastructure (RTI) as a parameter in the federation design, it may also be possible to identify where the HLA itself (instead of particular RTIs or computing platforms) is impacting on simulation performance. Initial results from this process are presented and future work, including the creation of a HLA performance model from identified parameters, is discussed.