Stanford EE Computer Systems Colloquium

4:15PM, Wednesday, Oct 01, 2008
HP Auditorium, Gates Computer Science Building B01
http://ee380.stanford.edu

Issues in Mainstream Clusters

Richard Kaufmann
Hewlett-Packard HPC
About the talk:

What's a poor server designer to do? More functionality is being integrated into the processor, while processors are threatening to break the 200W barrier and graphics cards are pushing 300W. Dense server designs are easily pushing 40 Kilowatts in to a standard rack, while most datacenters only tolerate 8KW. Interconnect requirements range from very low (two tin cans and a string would do just fine) to insane (for some applications, 1 byte/floating point operation is required -- necessitating a 100GB/s network link in 2010!). New technologies, such as application accelerators, also threaten to fractionate the server market. This talk will present an overview of the technologies and user needs driving cluster designs over the next few years, and hopefully the listener will leave with a little sympathy for server designers!

Slides:

Download the slides for this presentation in PDF format.

About the speaker:

Richard is a Distinguished Technologist (read: old curmudgeon) at Hewlett Packard, and works in the Scalable Computing and Infrastructure group. He's been at HP since 1991 (actually via acquisition from Digital and Compaq), and at various times has managed compiler groups and been the technical director for high performance computing. Prior to HP he worked at an Ada compiler company (for way too long). He received a BA in Computer Science from UC San Diego in 1978, where he was part of the UCSD Pascal Project.

Contact information:

Richard Kaufmann
email: Richard.Kaufmann (at) hp.com