Research Of Task Scheduling Algorithm For Soft Real-time System Based On Feedback Mechanism

Along with the development and application of the embedded system and relatedtechnology, a kind of soft real-time systems that runs on open and unpredictableenvironment is spreading quickly. The category of the soft real-time system researchis numerous, which bring new challenges to the research of real-time theories. Thetraditional real-time scheduling algorithm is mostly open-loop scheduling, and suchalgorithm has better performance in static or specific dynamic systems, but itsperformance will be sharply worsening in unpredictable dynamic systems. Therefore,feedback mechanisms applied to soft real-time scheduling algorithm have become ahot spot of scheduling algorithm.System overload is always the scheduling research topic, the overload rate isvery important for the validity and reliability of task scheduling research into softreal-time system; the equity between tasks is another important issue of soft real-timescheduling system, different tasks having different importance. How to allocateresources for them fairly still needs further research. In view of the aboverequirements, the paper funded by the national natural science funds-"research onadaptive real-time tasks scheduling mechanism and algorithm for heterogeneousmulti-core SoC"-was studied under the existing classic real-time scheduling algorithmand the traditional adaptive scheduling algorithm. We perfect and improve the systemoverload scheduling and task fairness scheduling. The main jobs of this paper are asfollows:Aiming at the system operation overload problems, the present study basic ishow to stop by feedback after overloading occurs in time, not solving the problemradically. This paper using the regression model and the inaccurate calculationdesigned a kind of early prevention system overload strategy, this strategy formultitasking real-time system traces each task, and adjusts before the sum of all thetask’s utilization is greater than system maximum utilization in advance, consequentlyavoiding overloading.Aiming at fairness of task scheduling, some existing researches did not considerthe fairness between tasks, adjusting the level of service for all tasks according tofeedback information, and it will lead the tasks run on either high or low servinggrade; or considering the fairness issue between the tasks, but not further considering different weights between tasks, treat them in the same way. This paper based onfeedback control describes a fair strategy for weight of task. This strategy added theweight factors of task, according to the relative weight of each task, adjusting servinglevels of all tasks by the feedback control in time, to ensure every task takes processorresources fairly, in order to bring task weight to real fairness.In order to verify the efficiency of the method, based on the existing imprecisecomputational task scheduling model, we realize a simulation system to simulate andcompare the proposed algorithm with the existing classical algorithms. Experimentaldata shows: overload avoiding algorithm is very good at guaranteeing the real time oftask when emergency occurs to the system load, and enhance the system’s adaptability;task fairness algorithm not only reasonably coordinates between system availabilityand missing rate, but also effectively improves the system task fairness.