Scalable and Energy Efficient Execution Methods for Multicore Systems

Colloq: Speaker: 
Dong Li
Colloq: Speaker Institution: 
Virginia Tech
Colloq: Date and Time: 
Fri, 2010-11-05 10:00
Colloq: Location: 
5100, Room 128 JICS Lecture Hall
Colloq: Host: 
Dr. Jeffrey Vetter
Colloq: Host Email:
Colloq: Abstract: 
Multicore architecture has the benefits of increasing thread level parallelism and better power efficiency than coupled multiple single core processors. In the mean time, multicore architecture imposes great pressure on resource management. The exploration spaces available for resource management are explosive increased, especially at large-scale high-end computing systems. The availability of abundant parallelism causes scalability concerns at all levels. Multicore architecture also imposes pressure on power management. Growth in the number of cores causes continuous growth in power. To maintain an affordable total cost of ownership (TCO), future exascale computing system equipped with the multicore requires a power efficiency much lower than the one achieved by the current supercomputer.In this talk, we introduce methods and techniques to enable scalable and energy efficiency execution of parallel applications on multicore architecture. In particular, we propose power-aware scalable models and algorithms for hybrid MPI/OpenMP applications. We investigate the interactions between MPI and OpenMP in a power-aware approach. Power-saving opportunities in both strong and weak scaling of hybrid applications in large scale are studied. We will also briefly discuss our research on power-aware MPI task aggregation prediction and scalable memory registration using helper threads. Our work yields substantial energy saving with negligible performance loss.
Colloq: Speaker Bio: 
Dong Li is a Ph.D. candidate in computer science at Virginia Tech. Dong's research area is high performance computing. In particular, he researches on programming and architectural supports for improving HPC applications' performance and energy efficiency. He is also interested in exploring emerging architectures and investigating their impact on HPC.