Energy-efficient Strategy By DRAM Voltage Threshold Regulation In Storm

As a new fault-tolerant and distributed real-time computing system,big data stream processing is mainly embodied in distributed remote call,real-time analysis,online machine learning,continuous calculation,and data extraction,data conversion as well as data loading and other fields.Meanwhile,the efficient and simple operation of processing has meet with great favor.Apache Storm,one of the major platforms for big data stream processing,has the high energy consumption,which the energy-efficient strategy needs to be improved.Therefore,whether in order to cut down the operating costs of big data processing,or to reduce the energy consumption and to protect the environment,the researches on energy-efficient strategies of the Apache Storm platform have a broad application prospect.Ignoring the issues of energy consumption in Apache Storm,the problems of high energy consumption generated from data processing need to be solved.Aim at this problem,models of directed acyclic graph,parallelism of instance,resource allocation for task and critical path are set up based on the features of data stream processing and the structure of Storm cluster.Meanwhile,the DVTR-Storm(Energy-efficient Strategy by DRAM Voltage Threshold Regulation in Storm)is proposed according to the analysis of the critical path and the system performance,which includes two energy-efficient algorithms aiming at whether there are any work nodes executing on the non-critical path of a topology.If there exists any work nodes that execute a topology on the non-critical path,DVRNP(DRAM Voltage Regulation on Non-critical Path)is used to lower the DRAM voltage of the work nodes on the non-critical path to achieve the high energy efficiency without changing the critical path of the topology.If all the work nodes execute a topology on the critical path,DVRCP(DRAM Voltage Regulation on Critical Path)is used to lower the DRAM voltage of the work nodes on critical path to achieve high energy efficiency based on the evaluation of the performance and energy consumption of the system.Finally,the appropriate threshold values fit for the CPU utilization of work node and the volume of transmitted data are determined based on the data processing and transmission constraints of system to dynamically regulate the DRAMvoltage of the system..The experimental result shows that DVRNP can reduce the energy consumption by about 28.5%,and DVRCP can reduce the energy consumption by about 35.1%.Moreover,the fewer constraints of data processing and transmission the higher the energy efficiency is.