Speaker: Carole-Jean Wu, Arizona State University
Title: Performance, Power, and Energy Efficiency Optimization from Cloud to Edge Computing
Computing touches almost all aspects of our modern-day lives – from real-time image and voice recognition to newsfeed and advertisement ranking, from video and web browsing to gene sequencing and powering the Internet. Personalized computing has gained increasing attention in the past decade. In this talk, I will first present our recent work on improving smartphone user experience by balancing performance and energy with probabilistic QoS guarantee. I will then share my vision for mobile performance quality, temperature and energy efficiency optimization with statistical performance prediction and with advanced cooling technologies. Our statistical modeling framework for application execution time are within 99.34% of the measured values. We show that, on an actual Qualcomm Snapdragon 8074 mobile chipset, our proposed energy efficiency controller achieves a 29% power saving over commonly-used Linux governors while maintaining an average web page load time of 2 seconds with a likelihood of 90%. I will also touch upon our temperature management techniques, considering advanced cooling technologies, to address the thermal challenges faced particularly by mobile platforms.
As machine learning inference is becoming increasingly important for the edge, delivering high performance quality poses a unique design challenge for the mobile production deployment. I will present the state-of-the-industrial practice of making inference at the edge, from billions of smartphones to augmented and virtual reality platforms. I will show how the statistical performance modeling techniques can be used to approximate performance variability in the field, influencing mobile design and evaluation strategies.
On the other hand, performance and energy efficiency are equally, if not more, important at the other end of the computing spectrum—high capacity computing in the cloud. Programmability advancement of high-performance accelerators, such as graphics processors, has enabled a large diverse set of general-purpose algorithms to enjoy performance acceleration on GPU-enabled heterogeneous systems. I will summarize our recent research efforts in this domain, addressing the strong performance and energy efficient scalability challenge faced in the post Moore’s law era. In particular, I will touch upon optimization techniques we design for heterogenous systems used for large-scale machine learning training and inference in the cloud
Carole-Jean Wu is an Associate Professor of Computer Science and Engineering in Arizona State University. She holds a Research Scientist position with Facebook’s AI Infrastructure Research. She is a senior member of both ACM and IEEE.
Prof. Wu works in the area of Computer and System Architectures. In particular, her research interests include high-performance and energy-efficient computer architecture through hardware heterogeneity, energy harvesting techniques for emerging computing devices, temperature and energy management for portable electronics. More recently, her research has pivoted into designing systems for machine learning. She is the recipient of the 2018 IEEE ITHERM Best Paper Award, the 2017 NSF CAREER Award, the 2017 IEEE Young Engineer of the Year Award, the 2014 IEEE Best of Computer Architecture Letter Award, the 2013 Science Foundation Arizona Bisgrove Early Career Scholarship, and the 2011-12 Intel Ph.D. Fellowship. Her research has been supported by both industry sources and the National Science Foundation.