Study Of Improving Energy Efficiency By Phase-Grained Migration For Heterogeneous Multicore

Heterogeneous multicore architecture that consists of multiple cores with different power and performance features has the potential for improving the energy efficiency.Migrating tasks with different characteristics to appropriate cores can reap the full benefits of the heterogeneity to improve energy efficiency.In this paper,the energy efficiency is the performance-to-power ratio of the processor that is running a task.However,to fully tap the potential of heterogeneous multi-core processors needs to overcome some challenges.Firstly,it is necessary to capture the task execution characteristics on-line,because the execution characteristics of the task change with the execution of the task.Secondly,the traditional estimation model that estimate the performance-to-power ratio of task in different types of cores has lower accuracy.Finally,it is necessary to design a migration scheduling algorithm to schedule all the tasks that are running to the core according to the acquired energy efficiency information.This paper focusses on how improve energy efficiency of heterogeneous multicore by tasks' phase-grained migration.The main work includes:1).Providing a way that classify execution phases of tasks based on instruction type vector.This paper improve the hardware architecture of instructions type vector classifying tasks' execution phases due to the relatively low number of event counters in the processor core,which can save the hardware resources and reduce the amount of data that has to be processed.According to the basic idea of the classifying phase of instruction type vector,this paper proposes a classifying phase algorithm.The experimental results show that the algorithm can effectively classify the execution phases of the tasks.2).Establishing a performance-to-power ratio estimation model.This paper creates a linear estimation model for estimating the energy efficiency of task's each phase on each core type.In this paper,we firstly use correlation analysis to get the impact of these events on the processor performance and power consumption.Then,obtain several of the most influential events by the principal component analysis method.Last,we regard these events as variables to establish the estimation model.The estimation model is simple and beautiful.The experimental result shows that accuracy of the model is very high.3).Constructing an energy efficiency optimization model based on tasks migration to guide tasks' migration.The optimization model is used to abstract the energy efficiency optimization problem.Based on the Kuhn-Munkres(KM)algorithm,the optimization model is used to determine the migration path with the highest energy efficiency.The experimental results show that the whole scheme can improve the energy efficiency of the processor.