Its All about the Memory (again)
Christos KozyrakisAssociate Professor of Electrical Engineering & Computer Science
Friday, February 17, 2012|
10:30am - 12:00pm
1690 Beyster Bldg.
Add to Google Calendar
About the Event
While it has been twenty years since the term "memory wall" became popular, memory systems have not changed much in the meantime. As a result, memory is a major determinant of the performance, cost, and energy consumption for computers of all sizes. In this talk, we will review memory system challenges and identify opportunities for major improvements in efficiency. We will start with warehouse-scale data centers, where memory is used for fast distributed storage, shifting the focus from high bandwidth to low latency. This allows us to use mobile memories (in the sort term) and emerging memory technologies (in the long term) to implement cost effective and energy proportional memory systems. Next, we will discuss the memory system of the multi-core chips used in all client devices. Even if ample parallelism is available, the performance and energy efficiency is limited by the frequency and cost of remote memory accesses. This motivates scheduling approaches that optimize for locality using domain or application specific information about memory usage. Finally, we will review opportunities to introduce new functionality in the memory system in order to minimize communication, assist with synchronization, enforce quality of service requirements, or provide security features. While several factors make it easier to introduce new functionality in the memory system now rather than twenty year ago, we need to identify simple primitives that are easy to use and manage in order to justify the hardware changes.
Christos Kozyrakis is an Associate Professor of Electrical Engineering & Computer Science at Stanford University. He works on architectures, runtime environments, and programming models for parallel computing systems. At Berkeley, he developed the IRAM architecture, a novel media-processor system that combined vector processing with embedded DRAM technology. At Stanford, he led the Transactional Coherence and Consistency (TCC) project at Stanford that developed hardware and software mechanisms for programming with transactional memory. He also led the Raksha project, that developed practical hardware support and security policies to deter high-level and low-level security attacks against deployed software. Dr. Kozyrakis is currently working on hardware and software techniques for next-generation data centers. He is also a member of the Pervasive Parallelism Lab at Stanford, a multi-faculty effort to make parallel computing practical for the masses.
Contact: Lauri Johnson-Rafalski
Sponsor: Tom Wenisch
Open to: Public