We have many ways to store and process information: vector supercomputers, mainframe and microprocessor-based multiprocessors, and personal computers. Cluster technology is being used to lash these computers together for: scalability, cross-generational upgradeabity, reliability, and geographic distributability. Over the last decade all the performance metrics e.g. flops, accesses/sec, and memory size have improved by a factor of 1000. In the short term, personal computers are emerging as a cyberbrick from which nearly all other computers can be built using cluster technology. For the past 2 years, a cluster of 9,000 Pentium Pros has held the speed record of over a Teraflops. Will the PC cluster momentum drive out large scale multiprocessors, DSM, vector supers, SIMDs, multi-threading, functional parallelism and the plethora of architectural alternatives? Is there another factor of 1000 to get to Peta-x's for 2010?
The speaker's bio is found at http://www.research.microsoft.com/barc/gbell.
The speaker's bio is found at http://www.orl.co.uk/~mvw.
Carole Dulong graduated in 1982 from Institut Superieur d'Electronique de Paris. She started her computer architect career at Bull where she worked on a French supercomputer project. She was the chief architect of the ES1, the Evans and Sutherland parallel supercomputer. She joined Intel over 8 years ago. She was one of the architects of the MMX(TM) technology. Carole is the Merced 3D and Multimedia architect, and she manages the IA-64(TM) experimental compiler project. Carole holds 8 patents.