Research Of Thermal Aware Multi-core Real-time Scheduling Algorithm

Raising the processor's master frequency has been the main method to promote processor's performance, since the birth of micro-processor. However, master frequency's arise requires raising the amount of transistor. More transistors more power, coming with the power dissipation problem. Under the above background, multi-core and many-core processor gain people's attention. Multi-core processor consists of two or more executive units into one chip, so we can gain more processor performance with low master frequency. With the soaring of the execution core number, the power dissipation problem will come again. Thus, it is urgent to solve the temperature and energy problems of modern processors. Considering the hardware level thermal management method is harmful for the processor's performance, thermal management method in operating system level is becoming the research focus.Firstly, this work discusses the multi-core related technique, especially the technique related with the operating system. Then, it analyzes the technique that used by OS to support multi-core processor, including boot sequence of multi-core based OS etc. This paper proposes a dynamic thermal management mechanism for real-time tasks in the operating system level called LTCEDF (Low Thermal Contribution Early Deadline First). It gets the value of performance counters online and then calculates the cores'temperature which can be used to decide which core to run a new process. Based on those methods, we extend our algorithm by utilizing non-corporative game and general Tit-for-Tat method, and propose the method called GTTAS and GTFTTS, respectively.We construct our simulator, TASS, based on ATMI thermal model. Simulation results show that algorithms of this paper can schedule much better than other algorithms in terms of real-time and temperature properties.