Welcome to Institute of Computer Networks, GU
Home>Seminars>Hai Jiang
美国阿肯色州立大学(Arkansas State University, USA)Hai Jiang教授学术报告通知
撰稿人:邢萧飞     发布时间:2018年07月05日 18:45
题目: Towards Constructing Application-Level GPU Computation States
时间: 2018年8月6日(星期一)上午9:00-10:30
地点: 广州大学行政西楼前座428(学院会议室)
报告人: Hai Jiang教授,美国阿肯色州立大学(Arkansas State University, USA)

GPU (Graphics Processing Unit) computing has been widely adopted since 2006. However, traditional operating systems treat GPU as an I/O device and only use batch-mode scheduling to dispatch GPU tasks. The actual scheduling algorithms are embedded in drivers provided by GPU vendors such as Nvidia and AMD. GPU task preemption is critical for fair GPU sharing and application fault tolerance. Although Nvidia has announced to support preemption in their latest GPU generations, no actual API has been provided to programmers for actual use. To work around the restriction of GPU drivers, preemption can be accomplished at application level. Pre-compiler and run-time support module can work together to reconstruct GPU computation states. Traditional data segment, stack and heap are duplicated in applications. The pre-compiler helps process the annotations in the original code. During checkpointing, the run-time module fetches application data from the hierarchical GPU memory system including local, shared and global memory units. Such computation states are saved back in system memory. To resume a paused application, its state will be loaded back to GPU memory and computation will continue from where it stopped before. Such application-level checkpointing scheme works across different GPU generations to support fair resource sharing and desired fault tolerance feature for long running scientific applications.

Hai Jiang received his B.S. degree from Beijing University of Posts and Telecommunications, China, M.A. and Ph.D. degrees from Wayne State University, Detroit, MI, USA. He is a Professor in the Department of Computer Science at Arkansas State University, USA. Before joining Arkansas State University, he has spent four years in State key Laboratory of Switching Technology and Telecommunication Networks, Beijing, China as a research fellow and five years in industry with Ford Motor Company, Dearborn, MI, USA, where he held various technical positions. His current research interests include Parallel & Distributed Systems, Cloud Computing, Big Data, Cryptography, Computer & Network Security, High Performance Computing and Communication, and Modeling & Simulation.

Dr. Jiang is a professional member of ACM and IEEE computer society. He has published three books and more than 100 papers in refereed journals, conference proceedings and book chapters. He has been involved in more than 100 conferences and workshops as a program/workshop chair or as a program committee member. He serves as an editor for International Journal of High Performance Computing and Networking, International Journal of Computational Science and Engineering, and International Journal of Embedded Systems as well as a guest editor for IEEE Systems Journal, Cluster Computing, and Concurrency and Computation: Practice and Experience. He also serves on the editorial board of International Journal of Big Data Intelligence, GSTF Journal on Social Computing, Open Journal of Internet of Things, and The Scientific World Journal. He also chaired IEEE TCSC PhD Dissertation Award Selection Committee, 2017.