Research On Server Energy Efficiency Benchmark Tools And Performance Evaluation

In recent years,with the large-scale increase in the number of cloud data centers,the energy consumption of servers has also increased day by day,and the increase in energy costs has forced the industry to focus its attention no longer on performance indicators.More and more enterprises try to focus on Arm architecture servers with low power consumption advantages.In order to give consideration to both performance and energy saving,it is necessary to provide quantitative indicators of energy efficiency.Using energy efficiency benchmarks is an important way to achieve energy efficiency assessment.However,the existing energy efficiency benchmark tools are outdated,single,do not conform to typical scenarios and high cost of energy data acquisition,which cannot effectively reflect the performance and energy efficiency test results of the server.This article designed and implemented CSBench,a server-oriented energy efficiency benchmarking tool,based on the Chauffeur framework,as a starting point for improving SERT.The main work of this paper is as follows:(1)In order to obtain the performance and energy efficiency index of the server more reasonably,this paper designs and implements new compression and decompression load,encryption and decryption load,sorting load,memory load,IO load and so on.In this paper,a new set of energy efficiency evaluation indicators is innovatively proposed.The realized compression and decompression load is a representative mainstream high-performance algorithm in the field of file compression and big data compression.Encryption and decryption load can reflect the most typical mixed encryption scenarios involving symmetry and asymmetry.The sort load is not dependent on the framework call of the language and has better generality.To extend the test coverage,the memory and IO loads provide test results for each sub-item of the mixed load.This article experimentally compares benchmark results from SERT with CSBench to demonstrate the effectiveness of the improved benchmark and the need to provide sub-test results in a mixed load.The proposed energy efficiency evaluation index includes the original energy efficiency score and the standardized energy efficiency score.The original energy efficiency value can evaluate the real energy efficiency of each benchmark subitem,while the standardized energy efficiency value can avoid the deviation of the comparison result caused by the energy efficiency value bias to a certain component,which can accurately evaluate the energy efficiency of the server from multiple dimensions.Finally,the effectiveness of the proposed energy efficiency evaluation index is verified by experiments.(2)Data collection of energy consumption is an essential part of the benchmark.This paper investigates the data collection technology of server energy consumption and analyzes their respective advantages and disadvantages.Based on the above research,this paper designed and implemented a more convenient and fast energy consumption data monitoring tool RFMonitor based on the Redfish standard,which can automatically start and record the energy consumption data of the server under test following CSBench during the test,and has the advantages of simple configuration and low cost.In this paper,RFMontior was compared with the data collected by the power meter,and the error rate was only 0.7%,which fully verified the accuracy of RFMonitor.(3)In view of the low power characteristics of Arm server and the research requirements of domestic CPU with independent Arm architecture in recent years,CSBench was used in the energy efficiency benchmark test of Kunpeng920 independently developed by Huawei based on Arm architecture and Xeon Gold 6248 server based on Intel's X86 architecture respectively.After obtaining the performance,power consumption and energy efficiency data of the two servers in different benchmarks,the reasons for the difference in their performance were compared and analyzed.This paper finds that the built-in acceleration instruction in Arm can effectively improve the performance of the server,and verifies that the extended encryption and decryption instruction in Arm V8 drives the hardware acceleration to achieve the extremely high performance and energy efficiency of SHA256.At the same time,this paper also analyzes the energy consumption stability of the two different computing architectures,and finds that the energy consumption of the Arm server has a high stability under the same type of load.This paper provides the latest reference for the performance and energy efficiency of Arm and X86 architecture servers,explores and verifies the characteristics of Arm,and fills part of the gap for the research of domestic CPU based on Arm architecture.