• Presentation: (PowerPoint format)

David Cheriton is a Professor of Computer Science and Electrical Engineering at Stanford University. His research includes the areas of high-performance scalable distributed systems, Internet architecture and protocols, and hardware-software interaction, particularly at the operating system level. He was the chief designer of the V Distributed system, the VMTP protocol, and the ParaDiGM scalable multiprocessor architecture. Prof. Cheriton was co-founder of Granite Systems, a leading developer of Gigabit Ethernet products, acquired lock, stock, and barrel by Cisco Systems, where he is now a (part-time) technical advisor. Prof. Cheriton received his Ph.D. in Computer Science from the University of Waterloo in 1978. For the past 22 years, he has been at Stanford. Most recently, he is a technical advisor with Google.com, VMware, Kleiner-Perkins, Caufield and Byers, and a few stealth efforts.

Today we have a cornucopia of high-speed networking solutions -- Myrinet, Quadrics, InfiniBand, SCI, and 10-Gigabit Ethernet. But do these networking technologies meet the needs of large-scale scientific applications, such as astrophysics, bioinformatics, and global climate modeling? With end-to-end latency approaching the sub-microsecond range and end-to-end throughput approaching 10 Gb/s in a system-area network, one might think that this problem is solved. However, although bandwidth and latency are important performance metrics, our experiences with application scientists, particularly after the launch of our Green Destiny cluster (http://sss.lanl.gov), indicate that their requirements are more extensive than what system researchers think, e.g.,

• Out-of-box performance that matches what a network wizard can achieve.
• Performance metrics that encompass more than bandwidth and latency, e.g., price/performance ratio, reliable QoS, automatic adaptation when migrating from a dedicated network to a shared network.
• A consistent network interface that does not require an application to be re-coded to use in different network environments.
• Tools that allow large-scale applications to self-configure based on processor speed, memory bandwidth, disk I/O, and network speed.

My talk will elaborate on the challenges of achieving these requirements across a multitude of networking environments.

• Presentation: (PDF format)

Dr. Wu-chun Feng is a technical staff member and team leader of Research & Development in Advanced Network Technology (RADIANT) in the Computer & Computational Sciences Division at Los Alamos National Laboratory (LANL) and an adjunct professor at the Ohio State University. He is also a fellow of the Los Alamos Computer Science Institute and the founder and director of the Advanced Summer Curriculum for Emerging Network Technologies (ASCENT).

Dr. Feng joined LANL in 1998, where he has been conducting research in high-performance networking and computing. In that span of time, he has established a respected record of over 50 journal and conference publications and has given over 20 invited talks and colloquia. Most recently, CNN and the New York Times reported on his revolutionary new supercomputer dubbed Green Destiny, and several news media outlets covered his team's breaking of the Internet2 Land Speed Record with a single TCP/IP stream.

He received a B.S. in Electrical & Computer Engineering and Music (Honors) and an M.S. in Computer Engineering from the Pennsylvania State University in 1988 and 1990, respectively. He earned a Ph.D. in Computer Science from the University of Illinois at Urbana-Champaign in 1996.