Energy Efficient High Performance Computing: Theoretical Foundation and Enabling Technology

Rong Ge
Assistant Professor and Director of the High Performance Computing Laboratory
Department of Mathematics, Statistics and Computer Science at Marquette University
SERC 306
Monday, February 2, 2015 - 11:00
Highly efficient scalable computing systems are crucial to scientific discovery and technology innovation. However, the scalability of HPC systems is increasingly constrained by the power requirement and the necessity to limit the power density of components and server rooms. Comprising millions of components, today’s HPC systems already consume megawatts of power; to meet an insatiate demand for performance from mission-critical applications, future systems will consist of even more components and consume more power. To resolve the conflicting needs of scaling performance and limiting power, we need to develop theoretical foundation and enabling technology for efficient and scalable computing on emerging and future computer systems.
In this talk, I will introduce an experimental system approach to energy efficient high performance computing. With this approach, we first identify the performance and power inefficiency of applications on real HPC systems through prototyping, profiling and benchmarking. We then derive analytical models based on system abstraction and experimental data to describe the impact of hardware/software parameters on performance, power, and energy efficiency. Guided by experimental findings and model projections, we design algorithms, middleware, and system software to interact with hardware for better efficiency, scalability, and reliability. I will discuss several models and tools developed with this approach, including power-aware speedup, performance and energy aware cooperative hybrid computing, and cross-layer power-bounded HPC scheduling, and summarize relevant results and their implications on emerging HPC architecture, algorithm and system design.