Parametrização do paralelismo de stream em Benchmarks da suíte Parsec

• Ano de defesa: 2018
• Autor(a) principal: Maron, Carlos Alberto Franco
• Orientador(a): Fernandes, Luiz Gustavo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação
• Departamento: Escola Politécnica
• País: Brasil
• Palavras-chave em Português: Programação Paralela; Paralelismo de Stream; PARSEC
• Palavras-chave em Inglês: Parallel Programming; Stream Parallelism
• Área do conhecimento CNPq: CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/8556

The parallel software designer aims to deliver efficient and scalable applications. This can be done by understanding the performance impacts of the application’s characteristics. Parallel applications of the same domain use to present similar patterns of behavior and characteristics. One way to go for understanding and evaluating the applications’ characteristics is using parametrizable benchmarks, which enables users to play with the important characteristics when running the benchmark. However, the parametrization technique must be better exploited in the available benchmarks, especially on stream processing application domain. Our challenge is to enable the parametrization of the stream processing applications’ characteristics (also known as stream parallelism) through benchmarks. Mainly because this application domain is widely used and the benchmarks available for it usually do not support the evaluation of important characteristics from this domain (e.g., PARSEC). Therefore, the goal is to identify the stream parallelism characteristics present in the PARSEC benchmarks and implement the parametrization support for ready to use. We selected the Dedup and Ferret applications, which represent the stream parallelism domain. In the experimental results, we observed that our implemented parametrization has caused performance impacts in this application domain. In the most cases, our parametrization improved the throughput, latency, service time, and execution time. Moreover, since we have not evaluated the computer architectures and parallel programming frameworks’ performance, the results have shown new potential research investigations to understand other patterns of behavior caused by the parametrization.