Research On Task Scheduling Based On Multi-core Array

With the explosive growth of communication services, the future communication systems with limited resources will support complex data transfer, exchange and processing, which causes delays and hardware implementations difficult to meet the future demand for high speed and low power communication systems, so massive parallel processing technology is to replace traditional serial approaches. It needs an efficient task scheduling algorithm to develop parallel technology fully. This paper focuses on task scheduling in multi-core array environment, that is, by placing tasks to the appropriate processors to meet the system requirements.The traditional task scheduling algorithm aims to make scheduling length minimum, while the optimization target is more than one for multi-core arrays. Only optimizing energy consumption or delay performance, may lead to local overheating. Local overheating can affect the network throughput, latency and other system performance and can even cause node failure, resulting in system reliability decreased. Therefore, it is necessary to study the power, delay and heat distribution performance for task scheduling algorithm.Based on the network topology, routing algorithm, task scheduling and other related theories of the multi-core array, this paper proposes 4 heat balance assessment model for uneven heat distribution problem of multi-core array and selects one as heat balance assessment model. Last, on the basis of research about multi-objective optimization models, using the advantage of NSGA-II algorithm and list scheduling algorithm, this paper combines list scheduling and NSGA-II algorithm for multi-objective optimization solution,(1) the initial population of NSGA-II algorithm adds the list scheduling solution;(2) for avoiding the excessive of loss delay and power improving heat distribution, the fitness function adds the compensation function. With the flexible control over the compromise of power, delay and heat distribution, it enables the cost of energy consumption and latency in control while optimizing heat distribution;(3) meanwhile, on the basis of in-depth analysis of NSGA-II algorithm, improve elitist mechanisms and crowded computing strategy, thereby increasing the population diversity and preventing premature local convergence.Through using actual application samples as FFT, and Gaussian elimination task graph and random task graph produced by TGFF for performance simulation, simulation result shows that this paper algorithm compared to the existing algorithm for power and delay, the algorithm of this paper pays controllable delay and power cost while heat distribution gets a good improvement. And, compare and analyze the simulation results of the algorithm of this paper with the standard NSGA-II algorithm and weighting algorithm for optimizing delay, power and heat distribution. Experiment results shows the superiority of the algorithm of this paper. Finally, we get the temperature distribution of the algorithm of this paper and the existing algorithm for power and delay through the HotSpot tool, and verify the effectiveness of the thermal equilibrium evaluation model and the effectiveness of the algorithm of this paper.