Research On Low Power Scheduling Algorithm For Multi-core Systems

Due to the rapid development of processing tool performance,energy consumption has become a problem to which people have to pay attention..Therefore,the low-power scheduling has become a hot research topic in the multi-core processor.There are the following three problems in above research.First,many scholars only considered dividing tasks into processors and then scheduling tasks in the tasks schedule model.They did not effectively combine low-power technologies DVFS and DPM.In addition,most scholars only considered dynamic energy consumption,static energy consumption,communication energy consumption.And they rarely took sleep energy consumption,voltage switching energy consumption,and sleep voltage switching energy consumption into account in energy consumption.Second,many scholars used heuristic algorithms such as genetic algorithms,particle swarm optimization,etc.These algorithms were easy to fall into local optimum.Especially in the scenario of a number of jobs,the algorithm was not effective.Third,most scholars only payed attention to the effects of the algorithm,and they showed little regard for time-consuming of the optimization algorithm.In fact,the optimization algorithm seriously consume time in the scenario of a large number of jobs.For the first problem,in the process of building the model in Chapter 2,the DVFS and DPM low-power technologies was combined,and the above six energy consumptions are comprehensively considered.For the second problem,first,the genetic algorithm and particle swarm optimization algorithm were implemented in the third chapter.Then the template matrix replacement optimization algorithm(TMR)was proposed.The algorithm was inspired by the morphological algorithm and had a very good ability of local search.Finally,the effect analysis of the above algorithm was carried out through comparative experiments.According to the experimental results,the average energy consumption reduction rate of the TMR algorithm in the light task scenario was 10%~15% higher than the particle swarm algorithm,and 15%~25% higher than the genetic algorithm.The average energy consumption reduction rate of TMR was 25%~30% higher than the particle swarm algorithm in the heavy task scenario,and 45%~50% higher than the genetic algorithm.However,experiments had also shown that the TMR algorithm took a lot of time in heavy task scenarios.The TMR algorithm takes much longer than the genetic and particle swarm algorithm in a number of jobs.Aiming at the third problem,the time-consuming problem of TMR was firstly improved by Q-Learning algorithm——Q-TMR.The Q-TMR algorithm was proposed to reduce the number of searches in the solution space.Then based on Q-TMR,the concept of breakpoint calculation was proposed to reduce the time consumption of calculation task scheduling time and energy consumption.Finally,the experimental results show that in the heavy task scenario,the improved algorithm reduced the energy consumption rate by 15.61% compared with the TMR algorithm,but the time consumption was reduced by 47.33%.