Accelerating Memory-Access-Limited HPC Applications via Novel Fast Data Compression

A fast-paced continual increase on the ratio of CPU to memory speed feeds an exponentially growing limitation for extracting performance from HPC systems. Breaking this memory wall is one of the most important challenges that the HPC community faces today. In Phase I we introduced aggressive innovations enable the injection of unprecedented acceleration into vast classes of memory-access-bound HPC codes via ultra-fast software-based data compression. Groundbreaking speedup on a fully functional NPBCG prototype was delivered to NASA, thus validating the tremendous potential of our approach. The proposed approach is based on a revolutionary theory of compression spearheaded by Accelogic (Compressive Computing), which is able to provide enormous compressive gains for the typical floating point data of HPC applications. In Phase II we will build on our success with the NPBCG benchmark, and move boldly into tackling the acceleration of a real-life high-profile code, namely NASA?s Cart3D, improving its performance by a paradigm-shifting 2x to 4x end-to-end wall-clock time acceleration by the end of Phase II. Our firm has accumulated crucial know-how and has synthesized its expertise into a powerful industrial-quality process for software acceleration that will be used to ensure success on completing Phase II objectives. In Phase II we also plan on injecting a second NASA code with basic Compressive Computing techniques, and providing it with base levels of acceleration of ~1.3-2x. We will choose this second code from a pool of high-profile codes that have already signed up as early adopters for this project: FUN3D, USM3D, Enzo, and WRF. The work on a second NASA code will also serve as the ultimate field test of the broadness and ease-of-infusion of the proposed technology. We have secured complementary funds in the amount of $500,000 to increase resources and ensure that the proposed Phase II proposed will be successfully accomplished.