An Elastic Energy-aware Algorithm For Parallel Tasks In Heterogeneous Environment

With the development of computer hardware equipment, large-scale computing systems rapid decline in costs and volume but substantial increase in computing power at the same time, massive computing power with different heterogeneous computing systems are widely used to form a large amount of energy consumption and environmental pollution. Energy consumption is mainly from the central server, and a cooling effect of the cooling system operation device, a fuel for power generation and produces carbon dioxide and other harmful gases. High performance computing systems to maintain normal life cycle cost of electricity needed to run beyond the cost of system hardware. How to design efficient high-performance computing cluster system has become an urgent problem, while the use of technical means to reduce the energy consumption of computer equipment, electronic products production and reduce pollution, improve system performance, low-carbon society, but also two types social goal. This thesis is through heterogeneous multi-core computer software designed scheduling algorithm to achieve the flexibility of the system energy efficiency, reduce energy consumption and improve system performance purposes.In this paper, based on the scheduling strategy for heterogeneous multi-core parallel task environment. The key of the algorithm is divided into three.First:in calculating the build tasks scheduling priority queue,in order to make the priority of the task in line with the task of real-time situation,using a dynamic method of recalculated the task priority after each task is completed.Second,when the task can be completed at the latest cut-off period,the minimum energy consumption by adjusting the voltage selector node scheduling to save energy.Third,in order to take into account the shorter scheduling time and avoid excessive consumption of resources for two purposes,to prepare a precursor replication tasks redundant tasks to determine whether,and if so then cancel the copy to avoid the consumption of resources. From the CCR (communication to computation ratio) value, the processor heterogeneous, heterogeneous EAPS network algorithm and the last HEFT (heterogeneous earliest finish time) three, HCPFD (heterogeneous critical parent and rapid replication) algorithm to analyze the scheduling algorithm and is expected to be completed at the time of the effect of the energy balance between.