Rafael Keller Tesser
Personal Web Page

I developed my PhD in the Post-Graduate Program on Computer Science of the Federal University of Rio Grande do Sul, in Brazil. The chosen research field was Parallel and Distributed Computing.

In my thesis, we developed and validated a low-cost simulation workflow to evaluate the performance of dynamic load balancing. This workflow was focused on dynamic load balancing based on domain over-decomposition and process migration applied to iterative parallel applications.

Initially, we implemented the support for dynamic load balancing into the Ondes3D geophysics application. Besides evaluating its performance with different load balancing heuristics, we did an extensive analysis of the possible causes for its imbalance. This application was used as study case in the evaluation of our simulation workflow.

Our simulation workflow was built upon the SMPI interface of the SimGrid simulator, combining its sequential emulation and trace-replay modes. This way, we can use traces from a single emulated MPI execution to simulate different configurations of the load balancing mechanism. Besides, our workflow requires a single computer node. We also developed two application execution modeling techniques to speed-up the sequential emulation process.

This PhD resulted in two scientific paper publications in international conferences, and one in a international journal. Besides, as member of GPPD, I participated in several research projects.

Languages: C, R, shell-script

Tools, frameworks, libraries, etc.: MPI, Adaptive MPI (AMPI), SimGrid, GNU R, Linux, Emacs/org-mode (to keep a lab notebook)

My PhD thesis is available here.

Master's in the field of Parallel and Distributed Computing, on the subject of trace visualization and monitoring for analyzing the behavior of parallel and distributed systems and applications.

We developed and online visualization mechanism for TRIVA, which is a tool that implements visualization techniques to analyse the behavior of distributed systems and applications. Online here meaning that the visualization is updated at execution time. This amounted to (1) parallelizing TRIVA, making it asynchronous, and implementing the capacity of receiving input data at runtime; and (2) developing and implementing an online distributed data collection model. Through an analysis of the state-of-the-art, we decided to create a model with and hierarchical structure and using a publish-subscribe data forwarding mechanism.

Besides the scientific work, I participated in various research projects and on the organization of the CLCAR 2010 conference (3rd Latin-American Conference on High Performance computing).

This Master's resulted in a scientific paper published in a recognized international conference. Besides, as member of GPPD, I participated in several research projects and on the organization of the 3^rd Latin-American Conference on High-performance Computing (CLCAR 2010).

• Languages: Objective-C, C, shell script
• Tools, frameworks, libraries, etc.: TRIVA, Ganglia, GNUStep, Linux
• Other: I employed in Objective-C's own multi-threading and remote method invocation mechanisms.

Besides attending classes, I performed the following activities:

• Member of the Tutorial Education Program (PET) (March 2004 - March 2007): As member of PET I participated on the organization and execution of lectures and short courses. Besides, I was part of the organizing committee of a multidisciplinary lecture cycle organized by all the PET groups, from different undergraduate courses from UFSM.
• Monitor (teaching assistant) of the Data Structures course (2^nd semester of 2005): This activity consisted in being available a few hours a week to answer students questions and helping with the correction of practical (programming) assignments. This course was taught using the C programming language.
• Course conclusion monograph: For my conclusion monograph, I developed and evaluated the use of assembly language instructions in an attempt to improve the libRastro parallel program instrumentation library. The implementation was done in C, using PThreads, and incorporating x86 assembly language code sections.
  • Languages: C (with embedded assembly code), shell-script
  • Tools, frameworks, libraries, etc.: PThreads, Linux

1. TESSER, R. K.; SCHNORR, L. M. ; LEGRAND, A. HEINRICH, F-C.; DUPROS, F.; NAVAUX, P. O.A. “Performance modeling of a geophysics application to accelerate over-decomposition parameter tuning through simulation” In: Concurrency and Computation-Practice and Experience, 2018.
   Available online: http://dx.doi.org/10.1002/cpe.5012
   Pre-print: https://hal.inria.fr/hal-01891416

1. TESSER, R. K.; SCHNORR, L. M.; LEGRAND, A.; DUPROS, F.; NAVAUX, P O. A. “Using Simulation to Evaluate and Tune the Performance of Dynamic Load Balancing of an Over-Decomposed Geophysics Application” In: Lecture Notes in Computer Science. 1 ed. Cham : Springer International Publishing, 2017, v.10417, p. 192-205.
   Available online: http://dx.doi.org/10.1007/978-3-319-64203-1_14
   Pre-print: https://hal.archives-ouvertes.fr/hal-01567792/
2. TESSER, R. K.; PILLA, L. L.; DUPROS, F.; NAVAUX, P. O. A.; MEHAUT, J-F.; MENDES, C. “Improving the performance of seismic wave simulations with dynamic load balancing” In: Euromicro International Conference on Parallel Distributed and network-based Processing, 2014, Turim, Itália. 2014 22nd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing. IEEE Computer Society Conference Publishing Services, 2014. p.196 - 203.
   Available online: http://dx.doi.org/10.1109/pdp.2014.37
3. TESSER, R. K.; NAVAUX, P. O. A. “DIMVHCM: An On-line Distributed Monitoring Data Collection Model” In: Euromicro International Conference on Parallel, Distributed and Network-Based Processing, Munich. In: 2012 20th Euromicro International Conference on Parallel, Distributed and Network-based Processing. IEEE, 2012. p.37 - 41.
   Available online: http://dx.doi.org/10.1109/PDP.2012.83
4. TESSER, R. K.; SCHNORR, L. M.; NAVAUX, P. O. A. “On-line Collection of Monitoring Data in Distributed Systems Using a Hierarchical Structure” In: Latin American Conference On High Performance Computing, 2010, Gramado/RS- Brazil. In: Proceedings/3rd Latin American Conference in High Performance Computing. Porto Alegre/RS, Brazil: Editora Evangraf, 2010. p.133 - 140.
   Download link (PDF)

The HPC4E project aims to apply the new exascale HPC techniques to energy industry simulations, customizing them, and going beyond the state-of-the-art in the required HPC exascale simulations for different energy sources: wind energy production and design, efficient combustion systems for biomass-derived fuels (bio-gas), and exploration geophysics for hydrocarbon reservoirs.

• Status: Concluded
• Funding: EU H2020 Programme and MCTI/RNP-Brazil.

The LAGClima project continues the previous STIC AMSUD project, called AMSUB-GBRAMS. It aims to build upon the previous results, using the Grid between the international partners to perform the computation of a meso-scale climatology of the Río de La Plata.

• Status: Concluded
• Funding: Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES

The international cooperation project HPC-GA, which aims to develop and port applications of the geophysics field for heterogeneous environments equipped with multicore processors and and GPU accelerators. Besides, propose new mechanisms and techniques for the efficient scheduling of processes/threads and for intelligent data distribution in such platforms. This project is a collaboration between several institutions: UFRGS (Brazil), INRIA (France), BCAM (Spain), UNAM (Mexico).

• Status: Concluded
• Funding: EU FP7 Framework Programme

Evaluation of the impact of the climatic change in its various aspects: action of the sea on the coast of the state of Rio Grande do Sul, analysis of the precipitation regimen and of the hydrological cycle, impact on the main agricultural crops and a study of the dispersion of pollutants under different climatological conditions. The analyses will result on an long term integration (30, 60, and 100 years) under three scenarios with variations of the minor gases on the atmosphere, considering the same emission rates registered in the last years, moderate increment of the emission rates, an severe increment of the industrial emission rates.

• Status: Concluded
• Funding: Financiadora de Estudos e Projetos - FINEP

The Atmosfera Massiva (Massive Atmosfere) project has the general objective of studying the impact of multicore architectures and multilevel parallelism in meteorological an environmental models.

• Status: Concluded
• Funding: Conselho Nacional de Desenvolvimento Científico e Tecnológico - CNPq

This project develops a Java system for environments with multi-clusters and a Web Server based manager for launching (executing) applications. This project was a cooperation between UFRGS and COPPE/UFRJ.

• Status: Concluded
• Funding: Financiadora de Estudos e Projetos – FINEP