2009 Poster Sessions : Embracing Heterogeneity -- Parallel Programming for Changing Hardware with the Merge Framework

Student Name : Michael Linderman
Advisor : Teresa Meng
Research Areas: Computer Systems
The exponential growth in digital data is driving similarly exponential growth in the computational demands of important information processing (informatics) applications in the life and physical sciences, finance, entertainment and security. Heterogeneous multi-core computers -- systems that couple specialized accelerators such as graphics processing units (GPUs) or field programmable gate arrays (FPGAs) to general-purpose CPUs -- can provide the scalable performance and energy efficiency these applications require, and that traditional CPUs alone have difficulty delivering. However, developing applications for these accelerators is difficult and is exposing limitations in existing compiler and OS infrastructure. The Merge framework, a general-purpose programming model, improves programmer productivity by insulating the programmer from the complexity of heterogeneous systems. Instead of trying to define "the" language abstraction for heterogeneous systems, an approach that excludes many current and future accelerators; Merge is designed from the bottom-up to be inclusive of the broadest range of architectures and language models. Its library metaprogramming-based backbone supports programming at multiple levels of abstraction, from low-level, expert-created, "to-the-metal" modules that can immediately exploit new hardware features, to sophisticated parallel compilers and domain-specific auto-tuners that can automatically leverage those modules to make the capabilities of heterogeneous computing readily available to experts and non-experts alike.

Michael Linderman is an Engineering Research Associate at Stanford University in the Electrical Engineering department. His research focuses on programming models and processor architectures for heterogeneous multi-core computers. He is interested in applying these technologies to the inference of biological signaling networks, and other important information processing applications in finance, security and the life and physical sciences. Michael earned a Ph.D. and M.S. in electrical engineering from Stanford, and a B.S. in engineering from Harvey Mudd College.