Applying the ATLAS Predictor to HPC Applications

The operation of computers demands a substantial amount of energy, with the CPU being a pivotal contributor to energy consumption. Blindly setting the CPU core frequencies to their maximum values for the idle cores can result unnecessary energy wastage. Addressing this issue is crucial for improving energy efficiency during MPI parallel computing processes. Therefore, the main objective of this thesis is to optimize load balancing and reduce CPU energy consumption in MPI applications based on prediction outcomes obtained from ATLAS's LLSP (Linear Least Squares Prediction) predictor. In most of our tests, our program achieved significant improvements in load balancing and CPU energy consumption.

(Master Defense)