Description: Processor speed has traditionally grown at a rate faster than that of communication speed in computer and supercomputer networks, and it is expected that this trend will continue even stronger, as we move into the exascale age in the upcoming decade. This has resulted in what is known as the "communication gap" for communication-bound HPC applications: their communication-to-computation time-ratio is so large, that the processors remain mostly sub-utilized, with lots of "disposable" FLOPS available. In the last few years, scientists have proposed to use these disposable FLOPS (which otherwise would be wasted idling) to compress and decompress the communicated data so to effectively speed up the underlying application. Although the idea bears tremendous potential, efforts in this direction have consistently rendered very poor results, with typical resulting speedups averaging below 1.5x. In this project, we identify the strongest reasons why traditional data compression has fallen short in terms of speedup performance for HPC, and propose novel techniques particularly crafted for groundbreaking performance within the HPC framework. Preliminary results show that these techniques break the 10x speedup markup consistently for a wide class of HPC applications of primary importance to NASA. We propose to develop the theory and methods behind these techniques, which ultimately will result into a library product for transparent acceleration of HPC communication platforms, such as MPI. Accelogic has already secured Phase III private capital in the amount of $1 million for the deployment of such potentially revolutionary product, following a successful Phase II.

Benefits: The proposed project tackles a wide spectrum of applications in the HPC field. Over 90% of the HPC applications in the different NASA areas use software libraries that can be accelerated with different compression techniques. Developing a tool that enables compression in a transparent manner, such as it is proposed in this SBIR, has the potential to impact all disciplines. Some of the most representative areas include the following: - Aerospace/automotive industries, computational fluid dynamics (CFD), defense, weather and climate research, computational molecular dynamics, combustion, computational electromagnetics, and weather forecasting. Nowhere is the role of supercomputing more important than at NASA. Given the magnitude of NASA's broad-based computer applications, where large-scale simulations are required ubiquitously, this organization has traditionally been at the vanguard of supercomputing advances. The proposed technology enables a significant leap in performance for applications at the core of NASA's mission, and we expect the commercial opportunities that arise from this project to be large and impacting.

Several HPC applications in industry and other Government Agencies apart from NASA also have the potential to be accelerated through novel compression techniques such as the ones proposed in this SBIR project. This is a reduced list of some of the most relevant disciplines that can be benefited directly by the proposed technology: - energy, finance, economic forecasting, computational chemistry, molecular biology, computational physics, civil and environmental engineering, digital content creation, digital media electronics, gaming, geophysics, image processing, information services, life sciences, media, medicine, semiconductors, and telecommunications. It is clear that the potential economical benefits of having such a tool readily available for the HPC industry are enormous. Several HPC application developers will grow an interest for such a product, which will give us the opportunity to have a viral penetration in industry and develop a successfully commercial product.