2012 Poster Sessions : Legion: Expressing Independence and Locality with Logical Regions

Student Name : Mike Bauer
Advisor : Alex Aiken
Research Areas: Computer Systems
Modern parallel architectures have both heterogeneous processors and deep, complex memory hierarchies. We present Legion, a programming model, type system, and runtime system for programming these machines. Legion is organized around logical regions, which express both locality and independence of program data. Legion also enables explicit, programmer controlled movement of data through the memory hierarchy and placement of tasks based on locality information via a novel mapping interface. Running on a 4 node cluster with 8 total GPUs and 4 levels of memory hierarchy, our implementation of Legion achieves a 5.9X speedup over a single CPU-GPU node on real-world applications.

Michael Bauer is a fourth year PhD candidate working with Professor Aiken on the design and implementation of new programming systems for targeting heterogeneous processors and deep memory hierarchies.