Research On Parallel Application Scheduling Algorithms For Energy Management In Heterogeneous Distributed Systems

Heterogeneous distributed system architectures have been widely adopted from embedded devices to large data centers.With the increase in system integration and performance,energy consumption has gradually increased and has become a major bottleneck in system design.Therefore,there is a need to coordinate the use of energy through effective energy management(both energy constrained and energy-efficient)aspects.Dynamic voltage and frequency regulation(DVFS)technology has become an important energy management technology by simultaneously reducing the power supply voltage and frequency of the processor to adjust the energy consumption.In this paper,parallel application scheduling algorithms for energy management in heterogeneous distributed systems are studied.Starting from general heterogeneous distributed systems,the performance and reliability optimization problems of parallel applications under energy constraint conditions are studied.Based on this,the work is extended to heterogeneous distributed embedded systems to study energy-efficient scheduling and optimization problems that meet different constraints.The main work and innovations of this article are summarized as follows:1.The high performance scheduling of energy constrained parallel applications is studied to solve the problem of minimizing the scheduling length of parallel applications with energy constraints on heterogeneous distributed systems.This problem is decomposed into two sub-problems in this paper,namely satisfying energy constraints and minimizing the length of the schedule.In this paper,Minimum Schedule Length with Energy Consumption Constraint(MSLECC)algorithm is proposed.First,the unallocated task is pre-allocated to the processor with the smallest energy consumption,and the applied energy constraint is transferred to the energy constraint of each task.The problem of satisfying the energy constrains is solved.Then,the second sub-problem is solved by heuristically scheduling each task with low time complexity,and choosing the smallest processor and frequency combination of EFT.Finally,experiments with real parallel application are carried out.The results show that compared with the classic HEFT and ECS algorithms,the proposed MSLECC algorithm not only makes the actual energy value of the application satisfy the given energy constraint,but also has a shorter scheduling length.2.The reliability enhancement scheduling of energy constrained parallel applications is studied to solve the problem of maximizing the reliability of the parallel application of energy constraints in heterogeneous distributed systems.This problem is decomposed into two sub-problems in this paper: meeting energy constraints and maximizing reliability.This paper proposes a Maximize Reliability with Energy Constraint(MREC)algorithm.First,the first sub-problem is solved by transferring the applied energy constraints to the energy constraints of each task.Then,the problem of maximizing reliability is solved by determining the energy of each task before task allocation and selecting the processor and frequency combination with the maximum reliability value while satisfying its energy constraints.Finally,experiments are conducted with real parallel application examples.The results show that compared with the excellent RMEC algorithm,the MREC algorithm proposed in this paper not only meets the given energy constraint,but also has a higher reliability value.3.The energy-efficient scheduling of real-time parallel applications is studied to solve the problem of energy consumption minimization in real-time parallel applications in heterogeneous distributed systems.First,the deadline relaxation algorithm is proposed.The algorithm introduces the concept of deadline relaxation to achieve efficient task allocation with minimum dynamic energy consumption without using DVFS,while satisfying the deadline constraints of the task as much as possible.Secondly,a Non-DVFS Energyefficient Scheduling(NDES)algorithm is proposed.The algorithm introduces the concept of variable deadline relaxation,and uses the deadline relaxation algorithm repeatedly to ensure that the deadline for the application is met,thereby reducing energy consumption.Once again,a Global DVFS-enabled Energy-efficient Scheduling(GDES)algorithm is further proposed.Under the condition of satisfying the priority constraints between tasks and the deadline constraints of applications,the algorithm migrates the tasks to the processor relaxation that generates the minimum dynamic energy consumption.Finally,simulation experiments and real platforms experiments are used to verify the results.The results show that the combination of NDES algorithm and GDES algorithm proposed in this paper(NDES&GDES algorithm)can save more energy than the current optimal energy-efficient scheduling algorithm.4.The energy-efficient fault-tolerant scheduling for reliable and parallel applications is studied to solve the problem of energy-efficient scheduling and energy-efficient faulttolerant scheduling for reliable and parallel applications in heterogeneous distributed embedded systems.First,a non-fault-tolerant,Energy-efficient Scheduling with a Reliability Goal(ESRG)algorithm is proposed to reduce energy consumption while satisfying the reliability objectives of parallel applications based on DAG-based heterogeneous embedded systems.Considering that using ESRG,if the reliability target exceeds a certain threshold,the reliability target of the application is unreachable.Therefore,an Energy-efficient Faulttolerant Scheduling with a Reliability Goal(EFSRG)algorithm is further proposed,and a fault-tolerant mechanism is used to ensure that the application's reliability goal is reachable.Finally,experiments on practical parallel applications are carried out on different application scales including Fast Fourier Transform and Gaussian elimination.Experimental results show that the reduced energy consumption of the EFSRG algorithm proposed in this paper is higher than that of other methods under the same scale.