A Statistical Approach to Temperature Estimation in Multi-Core Systems for Task Scheduling

In the past few decades, technologies such as cloud computing, IoT, blockchain and many more which were talked about only on paper a few years back, have now come to life. High performance and faster processors have played quintessential roles in the rapid development of these technologies. As the world moves from uni-core processors to multi-core processors, there is a dire need for efficient thermal management. One such thermal management scheme is that of thermal aware task schedulers. In this paper, we present a combination of two statistical models that will help to determine the temperature of individual cores in a multi-core system. Our linear regression model estimates the current temperature of a system from the commonly available parameters of CPU power, CPU usage, etc. Further, our ARIMA model, predicts the value of the temperature of the individual core for the next time instance. This acts like an input to the thermal aware task schedulers which can then allocate the task to the core with the minimum temperature. Such a two-step temperature estimation model can then help towards e?cient thermal management in multi-core systems.