Performance Analysis in Parallel Programming Education: A Comparative Usability Study
Anna-Lena Roth, David James, Jonas Posner, Michael Kuhn · 2026
Read Performance Analysis in Parallel Programming Education: A Comparative Usability Study by Anna-Lena Roth, David James, Jonas Posner, Michael Kuhn (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-31458.txt.
Introduction
Roth and three colleagues study how students learn performance analysis for parallel programs, and test a tool built to lower the barrier. Courses on message passing with MPI need cluster practice, yet professional analysis tools assume deep knowledge of clusters, resource managers, and common parallel faults. Students often stall before they learn anything about speed.
Their EduMPI tool gives a simple graphical interface that runs student programs on a cluster for them. It draws MPI communication almost in real time, placed by where each process sits in the machine, so load imbalance and bottlenecks show up visually. A user study against established professional tools finds EduMPI lowers entry barriers and builds an intuitive feel for parallel performance, raising its teaching value.
Catalog record
| CREATOR | Anna-Lena Roth, David James, Jonas Posner, Michael Kuhn |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.31458
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Performance Analysis in Parallel Programming Education: A Comparative Usability Study Anna-Lena Roth, David James, Jonas Posner, Michael Kuhn Abstract Parallel programming curricula encompass not only the development of parallel code and algorithm design but also emphasize efficiency, optimization, and performance analysis. To equip students with the skills necessary for writing efficient parallel code using message passing with MPI, practical experience on HPC environments is essential. Performance analysis tools assist in identifying issues such as load imbalances or bottlenecks. Despite their use by experienced developers, these tools’ complexity and required knowledge of cluster architectures, resource management, MPI, and common parallel issues hinder their educational integration. To address these barriers, we developed EduMPI, a learning support tool designed to simplify cluster usage and performance analysis for students. EduMPI offers an intuitive GUI that automates program execution on clusters and delivers near-real-time visualizations of MPI communication. This enables students to track process communication according to their physical placement within the cluster and detect performance problems interactively. This paper presents a user study comparing EduMPI with established professional performance analysis tools, demonstrating that EduMPI lowers entry barriers and fosters an intuitive understanding of parallel program performance, thereby enhancing its educational value. [email protected] [email protected] [email protected] [email protected] keywords Parallel Programming MPI Performance Analysis Education Usability Study 1 Introduction In parallel programming education, it is essential not only to teach students how to write parallel code but also to provide practical experience with real high-performance computing (HPC) environments and to develop efficient code that makes optimal use of system resources. To achieve this, concepts for avoiding performance problems, such as load imbalances, bottlenecks, and excessive communication overhead, must be addressed. Performance analysis is central to identifying such inefficiencies. Several working groups behind the European Master for HPC Curriculum [ 3 ] , the NSF/IEEE-TCPP Curriculum Initiative [ 14 ] , and the ITiCSE Reports [ 15 ] recommend making performance analysis and visualization an integral part of parallel programming curricula. Realization of this recommendation is challenging due to a lack of dedicated HPC practice environments and educational tools [ 15 , 22 ] . For many students, parallel programming courses are their first significant exposure to threads, processes, and cluster computing. Learning communication standards such as the Message Passing Interface (MPI) is already demanding. Performance analysis requires knowledge of code instrumentation and measurement, using appropriate tools such as Score-P [ 11 ] and Scalasca [ 7 ] . Most performance analysis tools work post-mortem , collecting metrics at runtime for visual analysis after program termination. Measurement must be repeated after every code change, causing long delays, especially on small educational clusters. The complexity and depth of detail involved in the subsequent visual analysis with tools such as CUBE [ 20 ] , TAU [ 21 ] , or Vampir [ 10 ] can be overwhelming for students, as they often require a comprehensive interpretation of multiple views and metrics. Without in-depth prior knowledge, performance issues often remain hidden initially. This makes performance analysis in an educational environment considerably difficult. At Fulda University of Applied Sciences, the course Parallel Programming was theory-heavy due to the lack of access to real HPC environments. The introduction of an educational cluster posed initial challenges, as many students lacked prior knowledge of SSH, Bash, and SLURM [ 24 ] , requiring substantial support. Performance analysis was often limited to simple runtime comparisons, since professional tools were perceived as complex [ 17 ] . To address this, we developed EduMPI [ 18 ] , a GUI-based learning tool that simplifies cluster access and, unlike common professional tools, provides near-real-time, intuitive visualizations of MPI communication without manual instrumentation or measurement. Performance data is collected and displayed automatically during program execution, enabling seamless student engagement with performance analysis. In this work, we conduct a controlled laboratory study to evaluate EduMPI’s impact on students’ ability to interpret runtime data and identify performance issues, as well as its usability compared to the tools CUBE and TAU. Results show that EduMPI lowers barriers to performance analysis and fosters an intuitive understanding of parallel programs. 2 Related Work In a previous study conducted during the winter semester 2023/24, we investigated to what extent the automation and accessibility of EduMPI motivate students to conduct performance analyses [ 17 ] . The focus was on the measurement process, comparing the user experience of EduMPI with the conventional approach of running performance measurements via the terminal using Score-P and Scalasca. Results showed that students working with EduMPI were more motivated to perform multiple measurements, compare results, required less support, and worked more efficiently overall. The study in this work shifts the focus to the visualization and interpretation of performance analysis results. Several studies highlight the importance of real HPC environments and practical analysis in education [ 8 , 23 , 2 ] , and others explore the integration of professional tools into teaching [ 4 , 12 ] . However, specialized educational tools for parallel programming are still rare [ 15 ] . Zhang et al. developed a web-based tool for simplified parallel execution and measurement [ 25 ] , while Oden et al. offer easy analysis workflows within Jupyter Notebooks [ 13 ] . Atala et al. introduced a visualization tool for collective MPI algorithms aimed at students, though it relies on simulations rather than real measurements [ 1 ] . All these tools follow a post-mortem approach, showing performance data only after termination, while EduMPI offers near-real-time, beginner-friendly visualizations bridging the gap between professional complexity and educational accessibility. 3 EduMPI EduMPI consists of three core components: EduMPI GUI can be used as an AppImage without requiring installation. MPI programs can be executed automatically on the cluster through the GUI, and MPI communication data can be interactively analyzed in near-real-time. On the cluster side, EduMPI is installed as a fork of Open MPI [ 6 ] and, in addition to standard Open MPI functionality, includes a data measurement system that collects MPI event-based communication data. This data is immediately processed by the third component, a time-series database, which enables fast queries, even when handling large volumes of data simultaneously. A detailed description of EduMPI’s functionality and GUI visualization features is provided in [ 18 ] . EduMPI was developed to simplify the visualization of MPI communication in educational use. It reduces the abstraction needed to understand process distribution across nodes/cores and their communication. Multiple students can use EduMPI simultaneously, are exposed to performance data from the start, can identify errors or bottlenecks early, and cancel programs via the GUI to avoid blocking cluster resources. Through custom data collection in a dedicated Open MPI fork, internal MPI data, such as point-to-point communication behind collectives, can be visualized. This benefits both teaching and experienced developers by enabling insight into collective communication algorithms [ 16 ] . Overall, EduMPI supports MPI understanding on real clusters, helps identify issues such as bottlenecks or late senders/receivers, and serves as an educational entry point to parallel performance analysis. 4 Usability Study We conducted a lab-based usability study in the summer semester 2025 to evaluate how well EduMPI supports students in identifying performance problems in MPI programs and how it compares to the professional tools TAU and CUBE,