Research On Multi-core Processor Scheduling Strategy Under The Influence Of Process Variation

With the further development of semiconductor manufacturing processes,the integration of integrated circuits is becoming higher and higher.The number of ICs that can be accommodated per unit area has increased dramatically,and the scale of integrated circuits is shrinking.At the same time,with the advancement of semiconductor technology,there exist some problems.Since the manufacturing of transistors with smaller dimensions requires more manufacturing difficulties,the manufacturing difficulty increases exponentially with the change of transistor size,and the consistency of different parts of the same chip is more difficult to control.Some small errors in the fabrication of transistors can cause large differences in performance between integrated chips.Compared with single-core processors,multi-core processors have the advantages of high parallelism and good performance,and have become mainstream processors.At present,the research on task scheduling in multi-core environment rarely considers the influence of process variation on multi-core processors.The research on homogeneous multi-core processors almost always thinks that the frequency of different cores is the same,which is very inaccurate that will cause loss of processor performance.Therefore,it is necessary to design a reasonable task scheduling algorithm considering the frequency inconsistency between cores due to the manufacturing process in the production process of homogeneous multi-core processors.This paper expounds the multi-core processor from different angles,and introduces the multi-core processor task scheduling in detail,and deeply analyzes the problems of task scheduling in multi-core environment.Considering the process variation,a task scheduling strategy based on critical path for homogeneous multi-core processors is proposed.Considering load balancing,a task stealing strategy is proposed.In order to solve the problem of processor performance loss in multi-core environment under the influence of process variation in practical application,the paper considers the influence of process variation on the core frequency of homogeneous multi-core processor,and divides the cores with the same computing speed into fast cores and slow cores.Fastcores handle tasks on critical paths,and slow cores handle tasks on non-critical paths.In this way,the advantages of fast cores are exerted,and the limitation of slow cores on processor performance is reduced.Secondly,considering the situation that the critical path may change with the execution of the tasks,a strategy for dynamically detecting the critical path is proposed.The existing critical path-based task scheduling strategies cannot solve the problem of critical paths' changing caused by task execution.The paper determines whether the critical path has changed by recording the task that was last determined to be critical and the priority of the currently ready task.After the critical path changes,the critical path can be re-detected for subsequent task scheduling.In the case where the task priority graph is deep or wide,there are two two situations.One situation is that there exist many tasks on the critical path while very few tasks on the non-critical path.Another situation is the number of tasks on the critical path is small,while the number of tasks on the non-critical path is large.At this point,there may be situations where a kind of cores have finished the tasks and is idle for a long time while another kind of cores are overloading.In order to solve the problems,a task stealing strategy is proposed.In the strategy,the idle cores will help busy cores perform tasks.Finally,the performance of the algorithm is verified by experiments.By randomly generating task priority graph with different number of tasks,experiments are carried out under different core numbers,and compared with other two algorithms in terms of running time and algorithm robustness,it is found that the proposed algorithm reduces the running time and is more stable.It is more suitable for task scheduling in a multi-core environment with process differences in the actual production process.