For decades, the computing industry has been fueled by the exponential growth in microprocessor performance. The ability to translate Moore's Law cost savings so effectively into computing performance increases has caused even some in the architecture community to confuse what exactly Moore wrote. There is, however, clear evidence that the exponential growth in microprocessor performance is now over. Moreover, a compelling case can be made that further improvements in single-core performance will be few and far between. This is creating the most significant change in the architecture of a desktop computer since their conception. Industry has responded by turning to multicore architectures. The fact that the masses really have no idea how to write parallel desktop software means our entire industry is venturing into highly risky territory. Success or failure of desktop parallel computing, either way, it is the most exciting time to be in this field.

This talk is divided into two parts. In the first part, I review how we got into this mess and what the broader commercial and research communities are doing about it. In the second part, I describe WaveScalar, which is a research architecture developed at the University of Washington. WaveScalar is the first tagged-token dataflow machine capable of executing imperative language software. Being dataflow, it efficiently executes any granularity of thread-level parallelism by converting it to instruction-level parallelism. Moreover, the instruction set architecture enables efficient low-complexity hardware implementations. I will describe the architecture, microarchitecture, and software support necessary for execution. No research project is flawless, however (anyone who claims otherwise suffers from hubris). So, in the last part of this talk I'll describe what went wrong; i.e. the lessons we learned from designing WaveScalar "version one" and how we might apply those lessons to a "version two" design.

Host: Peter Varman

Biography of Mark Oskin:
Mark Oskin is an Assistant Professor in the Computer Science and Engineering Department at the University of Washington. He received his PhD in 2001 from the University of California at Davis. His research interests span a wide range from systems architecture, to quantum computing, to synthetic biology. For amusement he enjoy tooling around Puget Sound at high speeding his boat and riding his motorcycle through the mountains of the Pacific Northwest.