Font Size: a A A

Research On Communication Characteristic Analysis Technology For High-performance Applications

Posted on:2021-05-14Degree:MasterType:Thesis
Country:ChinaCandidate:C Y LiFull Text:PDF
GTID:2518306122464214Subject:Computer technology
Abstract/Summary:PDF Full Text Request
High performance computing is accompanied by the growing situation of large-scale data processing.Its scientific and technological research and all aspects of the national life economy have massive application needs and wide application prospects.Along with this,the computing speed of high-performance computer systems is constantly refreshing new heights,and high-performance applications are constantly developing and improving.In the face of high-performance applications of massive computing,how to more intuitively analyze program performance status and find out program performance bottlenecks is imminent.At present,domestic and foreign research institutions and research institutes have carried out a lot of work and accumulated Many experiences and research on program performance monitoring tools are also representative.Although they are dev eloped for different applications and the analysis methods used are different,they all monitor and analyze the performance of the program.For the performance an alysis of high-performance applications,this paper studies and analyzes the communication cha racteristics of high-performance applications,and designs and implements a communication characteristic analysis tool.This tool implements real-time acquisition of application performance data through a data collection interface,which provides a higher Practicality,portability and ease of use.The main work of this article is as follows:(1)Design and implement communication data collection methods for high-performance computing.This article mainly determines the types of communication data to be collected and the means of collection based on different network layers.On this basis,a reasonable data collection scheme is designed.Through code insertion technology,the correct and automatic collection of data is automatically and effectively achieved without affecting high-performance applications.Performance parameters and communication behavior characteristic data.(2)Design and implement analysis of communication characteristics for high-performance applications.Embed the data collection interface on multiple levels of the application,run the program by setting different parameter co mbinations,record the communication data of the application,and then extract the communication characteristics of the program from the collected data,and then extract the key parameters of the application communication characteristics,such as Said the start time,end time of the communication function,the amount of data transmitted,etc.,and finally use the communication feature analysis technical report for subsequent performance analysis.In this article,the data collected for different levels and different nodes have different formats,extract and rearrange the data according to time sequence,calculate the communication volume for global communication,integrate these communication data and g enerate reports,and carry out the communication characteristics obtained Analysis,which is convenient for users to optimize the application.(3)Realization and example analysis of feature analysis technology in Linux system.In this paper,the key points of communication feature analysis technology for high-performance applications are tested in Linux system,and the communication feature analysis of parallel program performance data is carried out through examples,which proves that the communication feature analysis technology can be realized in a real environment.At the same time,the performance data is displayed in an intuitive and semantic way,such as program structure visualization,data distribution visualization and calculation visualization.I n this paper,an application example is used to analyze the communication characteristics of the application,and different data are used for comparison.The results show that the technology is extremely convenient for users to find the performance bottlen eck of the application.Therefore,the communication feature analysis technology for high-performance applications can efficiently optimize the program.In the coming E-level computing era,it can easily collect parallel information,data distribution info rmation and communication information,and judge the perfo rmance of the application more accurately.Let us accordingly allocate the computing load to the processor more reasonably,so as to achieve the processor load balance under large-scale computing,improve the efficiency of the calculation and use the resou rces more reasonably.
Keywords/Search Tags:High-performance computing, data collection, feature analysis, visualization
PDF Full Text Request
Related items