Publicado em: 27/02/2012

Dissertação de Mestrado em Computação Gráfica

UNIVERSIDADE FEDERAL DO RIO GRANDE DO SUL
INSTITUTO DE INFORMÁTICA
PROGRAMA DE POS-GRADUAÇÃO EM COMPUTAÇÃO
———————————————————
DEFESA DE DISSERTAÇÃO DE MESTRADO

Aluno: Daniel Köhler Osmari – Por videoconferência
Orientador: Prof. Dr. João Luiz Dihl Comba

Título: Visualization of Execution Traces of Dataflow Systems
Linha de Pesquisa: Computação Gráfica

Data: 29/02/2012
Hora: 13h30
Local: Sala INF (Auditório). Prédio 43413(67)

Banca Examinadora:
Profa. Dra. Carla Maria Dal Sasso Freitas (UFRGS)
Prof. Dr. Cláudio Fernando Resin Geyer (UFRGS)
Prof. Dr. Yi-Jen Chiang (Poly-NYU) – Por videoconferência

Presidente da Banca: Prof. Dr. João Luiz Dihl Comba

Abstract:
The computer industry relies heavily on parallel architectures to achieve improve- ments in performance. But the large diversity of software and hardware platforms make it very difficult for programmers to predict the performance behavior of parallel programs in the same way as if they were sequential. Many of the variables affecting the platform, such as bandwidth and latency, interact in complex ways, to the point where the only reli- able way of explaining the performance limits reached by certain program is by examining precisely how it’s executing. The inspection of performance of parallel programs has been studied for a long time, and has been shown to be a fairly complex task due mainly to the large volume and variety of data that needs to be analyzed. This work focuses on the anal- ysis of parallel dataflow systems, constituting a subset of parallel programs, through their execution traces. We propose the Smart Traces concept to aid creating visualizations for trace data, and use it to derive specialized visualizations that address questions relevant for dataflow systems.

Keywords:
Software visualization, graph and network data, time series data, coordinated multiple views, performance analysis, parallel systems, dataflow systems, execution trace.