Performance and energy efficiency on heterogeneous platforms by Katzalin Olcoz Herrero, Complutense University of Madrid
Event details
| Date | 08.05.2018 |
| Hour | 11:00 › 12:00 |
| Speaker | By Katzalin Olcoz Herrero, Associate Professor at the Department of Computer Architectures and Engineering in Complutense University of Madrid Katzalin Olcoz received a Ph.D. degree in Physics in 1997 from the Complutense University of Madrid (UCM). She has been Associate Professor in the Department of Computer Architecture and System Engineering of the Complutense University since 2000. She has been Head of the Department of Computer Architecture and System Engineering and she has also collaborated with the Spanish R&D Agency. Within the ARTECS group of the Complutense University, she has been involved in several projects in the field of computer architecture and design automation from high-level specifications. Her current research addresses emerging issues related to asymmetric processors, heterogeneous systems and energy-aware computing, with a special emphasis on the interaction between the system software and the underlying architecture. |
| Location | |
| Category | Conferences - Seminars |
Abstract
Over the last years, heterogeneity has become one of the main concepts explored by computer architects to improve performance while keeping power and energy in the face of continued technology miniaturization (as predicted by Moore's Law) and slowed supply voltage reduction (i.e. the end of Dennard scaling).
Heterogeneity ranges from heterogeneous configurations in datacenters to processors with on-chip hardware accelerators, hybrid CPU/GPU platforms, single-ISA asymmetric multicores and multi-ISA heterogeneous multicores.
Developing software for such systems in a way that all the different resources are efficiently used is complex and requires collaboration among the different levels of the software stack. We will focus on the collaboration among the programming models, runtimes and libraries.
In this talk I will show you some of my recent work on different ways of dealing with the heterogeneity of the system for the efficient execution of task parallel applications on ARM asymmetric big.LITTLE platforms. First, the library will take care of asymmetry, presenting a homogeneous system to the runtime. Then, it will be the runtime who will deal with the asymmetry of the cores.
Over the last years, heterogeneity has become one of the main concepts explored by computer architects to improve performance while keeping power and energy in the face of continued technology miniaturization (as predicted by Moore's Law) and slowed supply voltage reduction (i.e. the end of Dennard scaling).
Heterogeneity ranges from heterogeneous configurations in datacenters to processors with on-chip hardware accelerators, hybrid CPU/GPU platforms, single-ISA asymmetric multicores and multi-ISA heterogeneous multicores.
Developing software for such systems in a way that all the different resources are efficiently used is complex and requires collaboration among the different levels of the software stack. We will focus on the collaboration among the programming models, runtimes and libraries.
In this talk I will show you some of my recent work on different ways of dealing with the heterogeneity of the system for the efficient execution of task parallel applications on ARM asymmetric big.LITTLE platforms. First, the library will take care of asymmetry, presenting a homogeneous system to the runtime. Then, it will be the runtime who will deal with the asymmetry of the cores.
Practical information
- Expert
- Free
Organizer
- EcoCloud