CAREER: QoS-Aware, High-Performance, and Scalable Many-Core Memory Systems

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Computer science and engineering is undergoing a revolution. Many-core systems are rapidly becoming the foundation of computing systems that are well-integrated into every aspect of our lives and society. Their unpredictable performance and performance misbehavior adversely affects productivity, efficiency, and profit in all domains that make use of computers. Unfortunately, existing many-core systems are largely designed based on the assumptions made for single-core systems, i.e. there are no shared resources between cores, even though the memory system, a major performance and power bottleneck, is shared. As a result, many-core systems are severely vulnerable to denial of service, uncontrollable, unscalable, and low-performance. To enable the efficient and productive use of many-core systems, there is an urgent need to design them ensuring high quality-of-service (QoS), performance-robustness, and scalability. This research focuses on developing fundamental breakthroughs that enable scalable, controllable, and high-performance many-core memory systems. It aims to change the design paradigm of many-core processors to treat QoS and scalability in shared resources as first-class design goals, and educate future engineers to design systems with these goals as fundamental design objectives. The central approach is to develop hardware/software cooperative techniques to enable flexible QoS, partitioning, and performance mechanisms in memory systems and interconnects. The project develops fundamental techniques, targeting a very wide range of applications in cloud computing, data centers, client systems, mobile systems, and sensor environments. It is expected that research ideas developed in this project will enable controllable, robust, and therefore usable and efficient many-core systems, making our daily lives better and more productive, and taking a large step in making computing green.