Data-aware Task Scheduling On Heterogeneous Hybrid Memory Multiprocessor Systems

Today,the mass data processing under the era of big data is an important research contents in the heterogeneous multiprocessor systems of high performance computing area.The execution time of tasks and the speed of data access are the main property indicators in the heterogeneous multiprocessor system.The realization of performan-ce indicators need reasonable task scheduling and data assignment algorithm to allocate the effective resource.At the same time,with the development of high performance computing,heterogeneous multiprocessing systems can handle the problem more and more complicated.These changes caused some problems,such as more energy consumption increased the operation cost of overall system,Reliability decline and environmental pollution.Therefore,high performance and low power consumption is an important requirements in the high performance computing of big data area.In this paper,the main research work focuses on the problem of task scheduling and data assignment in the heterogeneous multiprocessor systems.This work improved the overall performance of the system and reduced energy consum-ption.One of the important problem is how to reasonable scheduling real-time application tasks to processors and assigning data to the memories.The hybrid memory consists of DRAM and SSD.According to the current practical applications of the heterogeneous multipro-cessor systems,we present a task scheduling and data assignment strategy.The traditional in-memory data assignment strategy is not suitable the hybrid memory consist of DRAM and SSD.It will cut down the life of SSD if forced to use.Hence,we have done the following works,in order to reasonable use the advantage of hybrid memory.First,we proposed an integer linear programming(ILP)and two heuristic algorithms to solve the tasks scheduling and data assignment in the heterogeneous multiprocessor hybrid memory systems.According to the characteristics of current system,these algorithms to real-time task scheduling and data assignment.ILP algorithm is based on the ability of the system and the relationship between tasks and data.ILP builds a mathematical model to solve out the optimal solution of task scheduling and data assignment in the heterogeneous multiprocessor hybrid memory system.And the two algorithms are Improvement Greedy(IG)algorithm and the Data Assignment According to the Task Scheduling(DAA_TS)algorithm,which generate a near-optimal solution for real-time applications in polynomial time.And then,the two heuristic algorithms will be compared with ILP and greedy algorithm in this problem.We need to do a lot of experiments to evaluate the performance indicators.According to the result of experiment,we found that the algorithm is very effective,especially in the aspect of saving energy.Finally,we adopt ILP to solve the optimal solution,and compare with IG algorithm and DAA_TS algorithm,respectively.IG algorithm results with integer linear programming algorithm to deal with the gap between the result of more than 20%,and the processing of DAA TS algorithm is integer linear programming algorithm and the results of the gap between the results of more than 10%.