Plenary

Beowulfs and Beyond: Past Accomplishments and Future Directions


Thomas Sterling (California Institute of Technology, USA)
 

Abstract: Beowulf-class PC cluster computing and other PC cluster systems have been in evolution for approximately 6 years driven by application requirements, technology opportunities, and experience with earlier workstation clusters and cycle harvesting on local area networks of desk tops. PC cluster computing is in transition from its inchoate phase of separate installed system implementation to shared community-wide hardware and software systems with industry support. When once the primary challenge was implementation of the hardware and programming end applications on small or moderate sized systems, today the dominant challenge is scalability beyond a thousand processors and the development of robust system management software for large multiuser, multitasking Beowulfs. The future directions for PC clusters will be, in part, driven by advances and changes in trends in hardware technologies including system-on-a-chip, processor-in-memory, and SMP-on-a-chip as well as new optical technologies for inter-node communications. But not all trends in technology will be conducive to future Beowulf-class systems and computing. The movement to turnkey consumer boxes such as set-top boxes, PDAs, and gameboys, are all inaccessible to being harnessed in general purpose clusters. This talk will summarize the range of accomplishments over the last half dozen years and will project the likely opportunities, directions, paths, and degree of advances for the next half dozen years. The talk will also reflect on the content and implications of the presentations comprising the IWCC technical agenda.

Presentation Slides of The Closing Plenary Address to IWCC'99: Beowulfs and Beyond (click on this)