Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Effective visualisation of callgraphs for optimisation of parallel programs: a design study
Parallel programs are increasingly used to perform scientific calculations on supercomputers. Optimising parallel applications to scale well, and ensuring maximum parallelisation, is a challenging task. The performance of parallel programs is affected by a range of factors, such as limited network b...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Department of Computer Science
2019
|
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613291178622976 |
|---|---|
| access_status_str | Open Access |
| author | Mabakane, Mabule Samuel |
| author2 | Kuttel, Michelle |
| author_browse | Kuttel, Michelle Mabakane, Mabule Samuel |
| author_facet | Kuttel, Michelle Mabakane, Mabule Samuel |
| author_sort | Mabakane, Mabule Samuel |
| collection | Thesis |
| description | Parallel programs are increasingly used to perform scientific calculations on supercomputers. Optimising parallel applications to scale well, and ensuring maximum parallelisation, is a challenging task. The performance of parallel programs is affected by a range of factors, such as limited network bandwidth, parallel algorithms, memory latency and the speed of the processors. The term “performance bottlenecks” refers to obstacles that cause slow execution of the parallel programs. Visualisation tools are used to identify performance bottlenecks of parallel applications in an attempt to optimize the execution of the programs and fully utilise the available computational resources. TAU (Tuning and Analysis Utilities) callgraph visualisation is one such tool commonly used to analyse the performance of parallel programs. The callgraph visualisation shows the relationship between different parts (for example, routines, subroutines, modules and functions) of the parallel program executed during the run. TAU’s callgraph tool has limitations: it does not have the ability to effectively display large performance data (metrics) generated during the execution of the parallel program, and the relationship between different parts of the program executed during the run can be hard to see. The aim of this work is to design an effective callgraph visualisation that enables users to efficiently identify performance bottlenecks incurred during the execution of a parallel program. This design study employs a user-centred iterative methodology to develop a new callgraph visualisation, involving expert users in the three developmental stages of the system: these design stages develop prototypes of increasing fidelity, from a paper prototype to high fidelity interactive prototypes in the final design. The paper-based prototype of a new callgraph visualisation was evaluated by a single expert from the University of Oregon’s Performance Research Lab, which developed the original callgraph visualisation tool. This expert is a computer scientist who holds doctoral degree in computer and information science from University of Oregon and is the head of the University of Oregon’s Performance Research Lab. The interactive prototype (first high fidelity design) was evaluated against the original TAU callgraph system by a team of expert users, comprising doctoral graduates and undergraduate computer scientists from the University of Tennessee, United States of America (USA). The final complete prototype (second high fidelity design) of the callgraph visualisation was developed with the D3.js JavaScript library and evaluated by users (doctoral graduates and undergraduate computer science students) from the University of Tennessee, USA. Most of these users have between 3 and 20 years of experience in High Performance Computing (HPC). On the other hand, an expert has more than 20 years of experience in development of visualisation tools used to analyse the performance of parallel programs. The expert and users were chosen to test new callgraphs against original callgraphs because they have experience in analysing, debugging, parallelising, optimising and developing parallel programs. After evaluations, the final visualisation design of the callgraphs was found to be effective, interactive, informative and easy-to-use. It is anticipated that the final design of the callgraph visualisation will help parallel computing users to effectively identify performance bottlenecks within parallel programs, and enable full utilisation of computational resources within a supercomputer. |
| format | Thesis |
| id | oai:open.uct.ac.za:11427/30076 |
| institution | University of Cape Town (South Africa) |
| language | eng |
| last_indexed | 2026-06-10T12:33:48.261Z |
| license_str | Not specified — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from UCTD — University of Cape Town Open Access Repository |
| publishDate | 2019 |
| publishDateRange | 2019 |
| publishDateSort | 2019 |
| publisher | Department of Computer Science |
| publisherStr | Department of Computer Science |
| record_format | dspace |
| source_str | UCTD — University of Cape Town Open Access Repository |
| spelling | oai:open.uct.ac.za:11427/30076 Effective visualisation of callgraphs for optimisation of parallel programs: a design study Mabakane, Mabule Samuel Kuttel, Michelle Parallel programs are increasingly used to perform scientific calculations on supercomputers. Optimising parallel applications to scale well, and ensuring maximum parallelisation, is a challenging task. The performance of parallel programs is affected by a range of factors, such as limited network bandwidth, parallel algorithms, memory latency and the speed of the processors. The term “performance bottlenecks” refers to obstacles that cause slow execution of the parallel programs. Visualisation tools are used to identify performance bottlenecks of parallel applications in an attempt to optimize the execution of the programs and fully utilise the available computational resources. TAU (Tuning and Analysis Utilities) callgraph visualisation is one such tool commonly used to analyse the performance of parallel programs. The callgraph visualisation shows the relationship between different parts (for example, routines, subroutines, modules and functions) of the parallel program executed during the run. TAU’s callgraph tool has limitations: it does not have the ability to effectively display large performance data (metrics) generated during the execution of the parallel program, and the relationship between different parts of the program executed during the run can be hard to see. The aim of this work is to design an effective callgraph visualisation that enables users to efficiently identify performance bottlenecks incurred during the execution of a parallel program. This design study employs a user-centred iterative methodology to develop a new callgraph visualisation, involving expert users in the three developmental stages of the system: these design stages develop prototypes of increasing fidelity, from a paper prototype to high fidelity interactive prototypes in the final design. The paper-based prototype of a new callgraph visualisation was evaluated by a single expert from the University of Oregon’s Performance Research Lab, which developed the original callgraph visualisation tool. This expert is a computer scientist who holds doctoral degree in computer and information science from University of Oregon and is the head of the University of Oregon’s Performance Research Lab. The interactive prototype (first high fidelity design) was evaluated against the original TAU callgraph system by a team of expert users, comprising doctoral graduates and undergraduate computer scientists from the University of Tennessee, United States of America (USA). The final complete prototype (second high fidelity design) of the callgraph visualisation was developed with the D3.js JavaScript library and evaluated by users (doctoral graduates and undergraduate computer science students) from the University of Tennessee, USA. Most of these users have between 3 and 20 years of experience in High Performance Computing (HPC). On the other hand, an expert has more than 20 years of experience in development of visualisation tools used to analyse the performance of parallel programs. The expert and users were chosen to test new callgraphs against original callgraphs because they have experience in analysing, debugging, parallelising, optimising and developing parallel programs. After evaluations, the final visualisation design of the callgraphs was found to be effective, interactive, informative and easy-to-use. It is anticipated that the final design of the callgraph visualisation will help parallel computing users to effectively identify performance bottlenecks within parallel programs, and enable full utilisation of computational resources within a supercomputer. 2019-05-15T07:12:01Z 2019-05-15T07:12:01Z 2019 2019-05-14T11:37:01Z Doctoral Thesis Doctoral PhD http://hdl.handle.net/11427/30076 eng application/pdf Department of Computer Science Faculty of Science |
| spellingShingle | Mabakane, Mabule Samuel Effective visualisation of callgraphs for optimisation of parallel programs: a design study |
| thesis_degree_str | Doctoral |
| title | Effective visualisation of callgraphs for optimisation of parallel programs: a design study |
| title_full | Effective visualisation of callgraphs for optimisation of parallel programs: a design study |
| title_fullStr | Effective visualisation of callgraphs for optimisation of parallel programs: a design study |
| title_full_unstemmed | Effective visualisation of callgraphs for optimisation of parallel programs: a design study |
| title_short | Effective visualisation of callgraphs for optimisation of parallel programs: a design study |
| title_sort | effective visualisation of callgraphs for optimisation of parallel programs a design study |
| url | http://hdl.handle.net/11427/30076 |
| work_keys_str_mv | AT mabakanemabulesamuel effectivevisualisationofcallgraphsforoptimisationofparallelprogramsadesignstudy |