Speaker
Description
URL(s) for further info
www.physiovalues.org
Wider impact and conclusions
The system is developed with focus on the domain of human physiology and modeling technology based on Modelica language. However the protocol and data might be used in any modeling domain and modeling tecnology.
Description of work
The web application for calibration of model integrates several technologies. The computational demanding process is designed as master worker. The identification algorithm employs the well known genetic algorithm implemented in MATLAB environment and exported as DLL library. The server (master) module manages the identification algorithm and distributes single simulation tasks to workers. Workers provides a REST interface to receive simulation task requests. We exported the models of Human physiology from Modelica langugage into standardized FMU package which is in fact DLL library with standardized API on Windows platform. The client application is developed in HTML5 utilizing AJAX technology to communicate with server services, collect data and request computation.
The parameter identification of the complex model spends some time in parallel simulation and some time in communication and synchronization procedures.
If the majority of the time is spent on simulation, then significant speedup can be achieved if the computation (workers) is distributed into remote capacity e.g. within a cloud or grid. We got about 96x speedup when identified a parameter of complex model Hummod (http://physiome.cz/hummod) on about 100 virtual CPUs of several virtual machines in the scientific cloud provided by Czech NGI CESNET.
On the other hand, the parameter identification of simpler models converges to a highly parallel computation system where time spent in parallel simulation is moreless same as in communication and synchronization procedures. Such computation is better to distribute to single machine with e.g. high number of CPU cores.
