Banca de QUALIFICAÇÃO: DEMÉTRIOS ARAÚJO MAGALHÃES COUTINHO

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
DISCENTE : DEMÉTRIOS ARAÚJO MAGALHÃES COUTINHO
DATA : 25/06/2019
HORA: 15:30
LOCAL: Auditório nPITI
TÍTULO:

Predicting Parallel Performance and Energy Efficiency Optimal Trade-Offs in Heterogeneous Multi-Processing Systems

PALAVRAS-CHAVES:

Energy-efficient software, Pareto frontier, Heterogêneous multi-processing systems.

PÁGINAS: 10
GRANDE ÁREA: Engenharias
ÁREA: Engenharia Elétrica
RESUMO:

This work proposes a novel methodology to predict the optimal performance and energy efficiency trade-off configurations of parallel applications running on Heterogeneous Multi-Processing (HMP) systems. The heterogeneous architectures are known for delivering significantly more performance and energy efficiency than their counterparts.  They offer more flexibility in the form of different core types and voltage and frequency pairings, defining a vast configuration space.  Therefore, the optimal configuration choice that is the desired number of cores, frequency, and voltage configurations for optimizing performance and energy consumption tailored to the needs of a given application is not straightforward to find it. The methodology is intended to be used by operating systems (OS) to make knowledgeable runtime decisions according to performance and energy consumption requirements of a given application and the system it is running on. For that, we propose analytic models for performance and power that are generated from off-line measurements. These models are then used to estimate the application's performance and energy consumption for the whole configuration space. Then,  we use these off-line predictions to choose Pareto-optimal configurations. We validated our methodology on an ODROID XU3 board on several PARSEC Benchmark and Phoronix Test Suite applications. The generated Pareto-optimal configuration represented a overall reduction of nearly 99% from the universe of all available configurations. Energy savings of up to 45.13% was observed when compared to the powersave Linux governor with better or similar performance.

MEMBROS DA BANCA:
Presidente - 1673543 - SAMUEL XAVIER DE SOUZA
Externo à Instituição - DANIELE DE SENSI
Externo à Instituição - Arthur Francisco Lorenzon