Analysis of the extreme value theory on the estimation of probabilistic wcet

• Ano de defesa: 2021
• Autor(a) principal: Machado, Guilherme Isaias Debom
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul; Escola Politécnica; Brasil; PUCRS; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Worst-Case Execution Time (WCET); WCET Estimation; Measured-Based Probabilistic Timing Analysis (MBPTA); Extreme Value Theory (EVT); Block Maxima (BM); Peak Over Threshold (POT); Aplicação de Tempo Real; Real-Time Application; ENGENHARIAS
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/10022

Real Time systems developed for critical applications require a proper execution time as important as the correct computed outcome. Owing to this, the maximum execution time of a Real Time System shall be determined by design. Estimate the execution time of complex systems affects time and analysis cost directly during the system development. In this sense, this dissertation aims to assess the possibility of applying MBPTA (Measured-Based Probabilistic Timing Analysis) based on EVT (Extreme Value Theory) to estimate the pWCET (Probabilistic Worst-Case Execution Time) of a given application in a given hardware platform. With this purpose, this dissertation makes use of the MIPS processor executing two algorithms as case studies: Bubble Sort and FIR Filter. These algorithms have WCET estimated by SDTA (Static Deterministic Timing Analysis). This work applies MBPTA with two different approaches: Block Maxima (BM) and Peak Over Threshold (POT), which are combined with EVT to estimate the final WCET. Then it compares the obtained WCET values by MBPTA/BM and MBPTA/POT with EVT to reference values, obtained by the SDTA method. The obtained results suggest that the BM approach presents a reduced complexity implementation as compared to the POT approach. Nevertheless, besides the POT higher complexity, this approach is more accurate than the BM, especially when input data values are not repeated.