Load monitoring module for multiprocessor performance optimization

Distributed sensor networks, heterogeneous multiprocessing systems, different kinds of mobile devices pose increasing need for efficient power/performance optimization methods. To cope with load variations and runtime uncertainties present in the most of real-time systems today online optimization methods are becoming increasingly popular. We propose architecture of hardware module for load monitoring which should enable easier implementation of online power/performance optimization algorithms. Proposed module enables automatic load measurement on each processing element and exchange of these statistics among processors in the system. Adapted communication interface module which supports flexible heterogeneous processor network is also presented. Load measurement can be controlled from application layer which provides flexibility in defining jobs of interest for optimization process. System architecture with centralized load statistics processing for high level optimization increases flexibility leaving low level optimization decisions to each processor locally. Moreover architecture of the load monitoring module is the same in all processors and only difference between master and slave nodes is in application layer leaving a room for easy reconfiguration. Proposed hardware load monitor was implemented on Cyclone II FPGA with Nios II as a processor and OS-II operating system.