[CS] 2016-10-27 Sarita Adve
From Mike Donohoe
views
comments
“Coherence, Consistency, and
Déjà vu: Memory Hierarchies in the Era of Specialization”
The 20th century saw hardware designers make heroic attempts to optimize coherence and communication in the memory hierarchy for homogeneous multiprocessors. Deliberately or inadvertently, these optimizations were often exposed to software and led to ever more complex memory consistency models. The 21st century saw heroic attempts to hide this complexity in more software-centric models centered on the data-race-free model. Alas, the resulting models for popular languages such as Java and C++ still contain open issues and remain an enigma to most. A clear conclusion emerged from these efforts – the data-race-free model was central to the solution, but retrofitting it with legacy hardware-centric design decisions exposed difficulties that are still unresolved.
As we now enter the brave new world of heterogeneous computing and specialization to enable performance increases past the end of Moore's law, it is déjà vu again. I will describe how hardware designers are again on a trajectory of exposing too much hardware to software, resulting in even more complex memory hierarchies and programming models. I will then draw on results from the DeNovo project to show that this trajectory is neither necessary nor effective. With appropriately designed coherence, we can reap the efficiency benefits of specialization without further complicating the memory model or abandoning the benefits of a global address space.
Speaker: Sarita Adve is the Richard T. Cheng Professor of Computer Science at the University of Illinois at Urbana-Champaign. Her research interests are in computer architecture, parallel computing, and resilient systems. She co-developed the memory models for the C++ and Java programming languages based on her early work on data-race-free models. She is a recipient of the Anita Borg Institute Women of Vision award in innovation, the ACM SIGARCH Maurice Wilkes award, and an Alfred P. Sloan Research Fellowship. She is a fellow of the ACM and the IEEE and was named a University Scholar by the University of Illinois. She is currently the chair of ACM SIGARCH and on the board of the Computing Research Association. She received the Ph.D. in Computer Science from Wisconsin in 1993 and a B.Tech. in Electrical Engineering from IIT-Bombay in 1987.
The 20th century saw hardware designers make heroic attempts to optimize coherence and communication in the memory hierarchy for homogeneous multiprocessors. Deliberately or inadvertently, these optimizations were often exposed to software and led to ever more complex memory consistency models. The 21st century saw heroic attempts to hide this complexity in more software-centric models centered on the data-race-free model. Alas, the resulting models for popular languages such as Java and C++ still contain open issues and remain an enigma to most. A clear conclusion emerged from these efforts – the data-race-free model was central to the solution, but retrofitting it with legacy hardware-centric design decisions exposed difficulties that are still unresolved.
As we now enter the brave new world of heterogeneous computing and specialization to enable performance increases past the end of Moore's law, it is déjà vu again. I will describe how hardware designers are again on a trajectory of exposing too much hardware to software, resulting in even more complex memory hierarchies and programming models. I will then draw on results from the DeNovo project to show that this trajectory is neither necessary nor effective. With appropriately designed coherence, we can reap the efficiency benefits of specialization without further complicating the memory model or abandoning the benefits of a global address space.
Speaker: Sarita Adve is the Richard T. Cheng Professor of Computer Science at the University of Illinois at Urbana-Champaign. Her research interests are in computer architecture, parallel computing, and resilient systems. She co-developed the memory models for the C++ and Java programming languages based on her early work on data-race-free models. She is a recipient of the Anita Borg Institute Women of Vision award in innovation, the ACM SIGARCH Maurice Wilkes award, and an Alfred P. Sloan Research Fellowship. She is a fellow of the ACM and the IEEE and was named a University Scholar by the University of Illinois. She is currently the chair of ACM SIGARCH and on the board of the Computing Research Association. She received the Ph.D. in Computer Science from Wisconsin in 1993 and a B.Tech. in Electrical Engineering from IIT-Bombay in 1987.
- Tags
-