Tuning genetic algorithms for real time systems using a grid

The use of locking caches has been recently proposed to ease the analysis of the performance and predictability of a cache when used in a real-time system. One promising method to adequately select the cache contents is the use of a genetic algorithm. However, this method requires the tuning of analysis parameters and this step requires a huge computational cost that can be reduced only if a massively parallel computing infrastructure is used. The work presented here analyses the specific requirements of the genetic algorithm tuning and the facilities provided by commercial grid software. Although the grid eases the resource management and job execution it lacks some communication link with submitted jobs, which is solved by the use of a specialized program called the Experiment Manager. This experiment manager supplements the grid and offers a completely automated environment for algorithm tuning to the researcher.