Presenter: Shrijeet Mukherjee, Silicon Graphics, Inc

Abstract:

The subject of this talk would be the options and evolution of architectures for large scale visualizations and the problems associated with them.

The hottest topic today is of course clusters for doing scalable graphics. We will look at some of the needs that a cluster needs to be endowed with to be a full-featured robust solution. Synchronization issues like Video and Buffer synchronization will be examined. We will take a quick look at the higher level issue of data distribution which plagues cluster solutions.

The second part of the talk, will focus on SGI's answer to some of the problems discussed, the SGI hierarchical cluster coupled with a Single-System-Image(SSI) and system wide cache/data coherency. The multiple synchronization options that it offers and the solutions it enables.

The final leg of the talk will be about algorithms that would be best benefited from this architecture, we will look at some uses of a SSI machine for implementing pipelined, streaming algorithms and usage of a coherent global memory. Needless to say, we will close with some of the real-world problems with getting efficiency out even seemingly efficient hardware.