Research On Low-power Schedulingfor Heterogeneous Multi-core Critical Tasks Based On Genetic Algorithm

In recent years, the problem of energy consumption is more and more serious,and it causes the conversion of processor architecture from single-core to multi-core.Multi-core processor systems can effectively control the running speed and power ofprocessing engine with scheduling strategy and low power technology, and it becomesan effective way to reduce system’s energy consumption. However, the existinglow-power scheduling techniques mainly focus on single-core systems, and they aredifficult to adequately exploit multi-core processors’ effect of energy optimization.When the heterogeneous multi-core processors run tasks, the frequency and power ofevery processing engine is different, and it makes the problem of low-powerscheduling based on heterogeneous multi-core processors more complex and becomesa hot research topic of current academicLow-power task scheduling based on heterogeneous multi-core systems is atypical NP-complete problem. Existing algorithms solving this problem use two-stageheuristic strategy mostly. First, tasks are assigned to processing elements by currentalgorithms for heterogeneous multi-core platforms. And then, determine the tasks’execution order and processing engines’ voltage or frequency levels combininglow-power techniques and scheduling algorithms for the given tasks allocation.Current algorithms rely on random search strategy mainly, and their time complexityis high and the effect of task characteristics for low-power energy-saving is concernedrarely. For the problem that current algorithms run inefficiently and the effect ofenergy-saving is poor, an improved low-power tasks scheduling strategies forheterogeneous multi-core systems are proposed in this paper.An improved genetic algorithm is proposed in this paper for the tasks assignmentstrategie of heterogeneous multi-core low-power scheduling algorithms. Thisalgorithm improves population’ updating mechanism of genetic algorithm withmetropolis acceptance criteria of Simulated Annealing to expand the solution spaceand overcome the "premature" phenomenon that falls into local optimal solutioneasily for traditional genetic algorithm. Improved partitioning strategy encodes tasksallocation scheme into chromosomes, and search options of tasks assignment withrelevant operators of genetic algorithm and simulated annealing algorithm.For the algorithm’s low-power scheduling strategy, a low-power schedulingalgorithm combination with the current popular dynamic voltage scaling technology is proposed based on the critical tasks analysis on the basis of analyzing existingalgorithms in this paper. First, the influence of tasks’ real-time by critical tasks isanalyzed in this paper. The time-urgent task nodes are arranged preferentially underthe given strategy of tasks allocation, then adjust the voltages’ level of units tasksexecute on which for the purpose of minimizing system energy consumption meetingthe requirements of tasks’ deadline.A simulation system is designed according to low-power task scheduling modelbased on current heterogeneous multi-core platforms in order to verify the algorithm’performance, and the solutions proposed in this paper and several existing algorithmsare simulated and implemented based on this system, tasks partitioning andlow-power scheduling strategy of algorithm proposed in this paper are evaluated fromtwo aspects of energy-saving rate and time complexity.The experimental results showthat task partitioning strategy included in heterogeneous multi-core low-powerscheduling algorithms proposed in this paper expands the solution space and providesmore extensive task allocation scheme for low power scheduling. Low-powerscheduling strategy of algorithm proposed in this paper reduces energy consumptionand time complexity effectively at the same time.