Dynamic Voltage And Frequency Scaling Algorithm Based On Variable Frequency Update Interval

Embedded system has gradually integrated into human society in various fields.But the contradiction between energy efficiency and time efficiency in electronic devices is more and more seriously.DVFS technologies provide an opportunity to compromise between performance and power consumption of equipment,and DVFS has been one of the most significant energy optimization technologies,and the algorithm closely related to DVFS has also become a hot topic for research.DVFS technologies can optimize the energy consumption of systems to a certain degree,make devices more energy efficient.However,the theoretical model which DVFS technologies use has a poor adaptability to multi-phase tasks,and there is a large error when estimating the energy usage of systems.The purpose of this paper is to improve this theoretical model and the variable-interval DVFS algorithm,promote the accuracy of energy estimation by adding a DVFS overhead calculation in the energy consumption model,make the DVFS algorithm combine the advantages of different fixed-interval DVFS technologies,reduce the similarity requirements between tasks by improving the granularity of the variable interval,and to introduce the multi-phase model.The work detailed in this paper is as follows:The theoretical model is too simple,and is such that it restricts the progress of DVFS technologies;a single-phase task model restricts the variety of the tasks,and the simplified energy consumption estimation model has a large error in system energy calculation.There are some improvements proposed in this paper:1)Firstly,a multi-phase task model is taken to show the frequency characteristics of the tasks.This model divides the task phase by the frequency bound in the task,and it can reflect the complexity of different phases in a task;2)Secondly,the accuracy of the energy consumption estimate of a DVFS system is improved by adding a DVFS overhead model to the traditional energy consumption model.This paper calculates the average energy consumption of the DVFS module when it adjusts the frequency and voltage of a system,and constructs a simple DVFS overhead model.The data used all come from measurements taken in other papers.The new model constructed is then combined with the traditional energy consumption model and is used to calculate energy consumption statistics.Fixed-interval DVFS algorithms may cause energy wastage or make tasks time out.The introduction of variable-interval DVFS algorithms somewhat made up for these defects,but they have strict constraints on the similarity between tasks.The variable-interval DVFS algorithm in this paper uses an online strategy and takes a coarse-grained variable interval instead of a fine-grained one.It ensures the update rates of the frequency can meet the requirements of different phases in tasks by switching the update interval of the frequency.This algorithm reduces the amount of wasted energy,and it can not only meet the performance and energy consumption requirements of the system,but can also reduce the constraints on the similarity between the tasks.In order to verify the performance of the variable-interval DVFS algorithm,this paper realizes the prototype code of the algorithm and the corresponding experimental platform in VS2010.The performances of different algorithms with the same task sets in the experimental platform indicate that the algorithm in this paper can handle multi-phase tasks easily.It combines the advantages of different fixed intervals and operates at or near the best value in both the energy efficiency and response times of fixed-interval DVFS algorithms.