Dynamic Adjustments Of Threshold Temperatures

Abstract

This research work addresses the issues of thermal emergencies in real-time embedded system using scheduling based algorithms. The algorithms considers execution of periodic task sets including both independent and dependent tasks over homogeneous multi-core processors without violating deadlines. We propose two algorithms in this work. The first one adjusts the thermal threshold values based on the processor utilization to reduce thermal peaks and thermal spatial gradients for independent periodic task sets. The second algorithm is a DVFS enabled load-balancing algorithm for execution of dependent periodic task sets to reduce the temperature peaks and spatial gradients. The proposed algorithms are evaluated in a simulation environment using a scheduling and thermal model for synthetic and bench marks techniques. The used hardware platform comprises four and eight cores. The results show that GTAI with 74ºC threshold reduces average temperature up to 6ºC and maximum temperature spatial gradient was reduced by 33% as compared to benchmarks techniques. The DVFS enabled DFS was able to decrease peak temperature by 31% and temperature spatial gradients by almost 50% in a 4-Core system.