Date & Time: 

2022 / 03 / 04 (Fri) 14:20 - 16:20

Location: 

Delta Building R216, NTHU

Speaker: 

田伯隆教授 (Prof. Po-Lung  Tien)

國立陽明交通大學 電機系 (Dept. of Electrical and Computer Engineering, NYCU)

Topic: 

Optical Networks Technology for HPC Systems

Abstract: 

High-Performance Computing (HPC) network infrastructure is required to flexibly deliver massive bandwidth and ultra-low latency for 5G/beyond 5G killer applications, such as machine learning and AR/VR. Existing electrical-switching-based infrastructure, however, has been shown incapable of meeting such requirements. In this talk, I will present an innovative optical network architecture and prototype testbed, referred to as intelligence-defined Optical Tunnel Network System (OPTUNS). OPTUNS consists of a set of optical switching subsystems that operate collectively to facilitate packet transport through logical wavelength-based optical tunnels. These optical tunnels are governed by an SDN-based intelligent tunnel control system, in a proactive manner. As such, optical tunnels are always made available whenever needed. In essence, OPTUNS boasts several crucial features, including high scalability, massive wavelength reuse (yielding high bandwidth), proactive optical tunnel control (yielding ultra-low latency), and fault tolerance. We have built an OPTUNS testbed including 30 optical switching subsystem prototypes that interconnect a total of 25 racks (400 servers). Benchmarking results show that OPTUNS achieves 82.6% power saving compared to electrical spine-leaf networks. Further, experimental results show that OPTUNS invariably achieves mean and p99 end-to-end latencies of less than 17us, regardless of traffic load and locality. 

Bio:

Po-Lung Tien received the B.S. degree in Applied Mathematics, the M.S. degree in Computer and Information Science, and the Ph.D. degree in Computer Science and Information Engineering, from the National Chiao Tung University (NCTU), Taiwan, in 1992, 1995, and 2000, respectively. From 2001 to 2005, he was with the department of Computer Science and Information Engineering in NCTU, where he was a research assistance professor. In 2005, he joined NCTU, where he is currently an associate professor of the department of Electrical and Computer Engineering. He has served on the technical program committee of many international conferences including IEEE GLOBECOM, WCNC, and HPSR. His current research interests include machine learning, parallel computing, network optimization, data center networking, and optical networking.