A Performance Visualization Tool for Parallel Systems

• Main Authors: Chen, Sou Pin, 陳少平
• Other Authors: Lin Woei
• Format: Others
• Language: zh-TW
• Published: 1996
• Online Access: http://ndltd.ncl.edu.tw/handle/56619451867699258257

碩士 === 國立中興大學 === 資訊科學學系 === 84 === Analyzing parallel algorithm is much harder than analyzing sequential algorithm since human being is accustomed to thinking sequentially.Text description is also sequential by its nature. Hence, it is difficult for programmer to understand a parallel algorithm described in text form. Conversely, graph is more user-friendly in representing a parallel algorithm because of its visibility. Events of parallel processing are converted into graphical objects to help programmers to analyze algorithms. A performance observation tool,PVTPS(Performance Visualiztion Tool for Parallel Systems), is a tool that simulates execution of parallel algorithmsin a graphical way. It visualizes run-time behavior and performance statistics of parallel algorithms. This will help programmers to debug parallel programs . In order to achieve easy maintenance in developing software and comply the trend of future operation system platform, it is appropriate to use a programming language with the features of " information hidding" and " module indepedence". Theerefore, an OOP(Object- Oriented Programming)language with cross-platform compiler is our choice. PVTPS supports several graphic modules that allows the user to select a proper observation as demanad. This tool uses different colors to representprocess status. From this cause, the user can observe process status condition by colors. Graphical tables are used to describe performance summary. With these versatile graphical facilities, the user can observe performance of parallel algorithms form different perspectives. For example, the user can easily identify computational bottlenecks and erroneous message transferring due to programming errors by the "Ganttcharts" function in PVTPS. Base on the observation, the user can improve parallel algorithms for achieving their peak performance in different execution environment system. In tihs thesis, we discuss the process of developing PVTPS.