A Unifying Theory of Distributed Processing (Or, The Chutzpah One Should Expect When You Invite a Microarchitect into Your Sandbox)

Yale N. Patt · 2005

Summary form only given. In this talk the author proposes to give us his understanding of distributed processing, how it differs fundamentally from non-distributed processing, and show that in this world of continually increasing complexity, it is the only paradigm that makes sense. Not surprisingly, the effective (and ineffective) uses of distributed processing are all around us. Several enlightening examples: an effective military, organized religion, the American game of football, and university education. The author discusses each. The author submits that these examples have a unifying theory that could be relevant to microarchitecture. The author describes it as a two-phased approach: preparation and execution. The microarchitect historically has designed central processing units. But to keep pace with current and future design points, including processing power that device technology continues to provide, distributed processing technology becomes essential. Along the way, the author introduces and explains the relevance of my levels of transformation, multichip processors, the refrigerator, and data flow.

Read the paper · More papers on PaperTik