Peak Power Consumption Control Method for Heterogeneous Parallel Computing Nodes

Abstract

The invention discloses a control method for the peak value power consumption of irregularly structural parallel calculation nodes and aims to dynamically control the peak value power consumption of irregularly structural parallel nodes to reduce the influence of the power consumption to the execution performance of the nodes as much as possible. The method in the technical scheme comprises the steps of firstly establishing an irregularly structural parallel calculation node peak value power consumption control system formed by a node power consumption description file, a node power consumption budget setting subsystem, a coprocessor load monitoring subsystem and a processor power consumption setting subsystem; recording related power consumption parameters of the nodes into the node power consumption description file; using the node power consumption budget setting subsystem to set a node power consumption budget, using the coprocessor load monitoring subsystem to query the real-time operation load Wco of a coprocessor and using the processor power consumption setting subsystem to set the power consumption threshold value of a processor. By adopting the control method for the peak value power consumption of the irregularly structural parallel calculation nodes, the peak value power consumption of the calculation nodes can be enabled not to exceed the power consumption budget set by a user and has small influence to the node performance, so the power consumption efficiency of the nodes is improved.

Description

technical field [0001] The invention relates to a method for controlling peak power consumption of a node in the computer field, in particular to a method for controlling peak power consumption of a co-processor (Co-processor) heterogeneous parallel computing (Heterogeneous Parallel Computing) node. Background technique [0002] Large-scale parallel computing systems consume a lot of power, which leads to a series of problems such as power supply overhead, heat dissipation, reliability, and environmental pollution. These computing systems are composed of tens of thousands of computing nodes, and node power consumption control is the basis for realizing system power consumption control. Therefore, it is necessary to control the power consumption of computing nodes, that is, control the operating power consumption of computing nodes within a certain threshold. [0003] Heterogeneous parallel architecture within computing nodes is an important direction in the current high-per...

AI Technical Summary

Problems solved by technology

In heterogeneous parallel computing nodes, processors and coprocessors are important power consumption devices, and the real-time power consumption of coprocessors cannot be judged only based on the operating behavior of the processor itself, so that the available functions of the processor cannot be accurately judged. consumption, it is impossible to accurately adjust the power consumption threshold of the processor

Therefore, the current dynamic threshold method is not suitable for power consumption control of heterogeneous parallel computing nodes

Images