Parallel Laboratory (ParLab)
*** THIS PROJECT IS NO LONGER ACTIVE ***
Dave Patterson, Ras Bodik, James Demmel, Kurt Keutzer, Koushik Sen, Kathy Yelick, Krste Asanović and John D. Kubiatowicz
Given the historic switch to parallelism, the greatest challenge facing computing is making it easy to write programs that execute efficiently on computing systems with hundreds of processors per chip. A multidisciplinary team is charting a novel approach based on targeting 13 "dwarfs" that will be the core of future applications instead of benchmarks, relying on "autotuners" instead of compilers, many simple processors instead fewer complex ones, and programming systems inspired more by principles of psychology than by hardware. The focus is on parallel applications, programming models, and testing, but we expect breakthroughs will also require advances in parallel architectures and operating systems.