GreenDPA: Thermal-aware execution of Data Parallel Applications

Parallel execution of programs improves performance substantially, at the cost of more power consumption, and higher temperature that threatens system reliability, decreases both transistor age and transition speed and increases the leakage current. In this paper, we propose a method for thermal-aware parallel execution of data parallel programs. Our scheme utilizes an advanced resource predictor to estimate the CPU frequency and number of cores considering the desired temperature threshold. The experimental results indicate that our proposed method outperforms standard OpenCL matrix multiplication in reducing average and peak temperatures with an increase of runtime execution without any violation from temperature threshold.