Progress Report on the Early - Career Research Project : Energy - Efficient Parallel Graph and Data Mining Algorithms

Data are fundamental sources of insight for experimental and computational sciences. The Department of Energy acknowledges the challenges posed by fast-growing scientific data sets and more complex data. The graph abstraction provides a natural way to represent relationships among complex fast-growing scientific data sets. On future exascale systems, power consumption is of primary concern yet existing graph algorithms consume too much energy per useful operation due to their high communication costs, lack of locality, and inability to exploit hierarchy. This project explores methods to increase the energy efficiency of parallel graph algorithms and data mining tasks. A new family of algorithms will be developed to drastically reduce the energy footprint and running time of the graph and sparse matrix computations that form the basis of various data mining techniques. This project will also exploit the well-known duality between graph and sparse matrices to develop communication-avoiding graph algorithms that consume significantly less power. This project is relevant to DOE mission-critical science including bioinformatics and genomics with particular emphasis on plant genomics that can result in better biofuels through efficient genetic mapping, climate science where recent graph-based methods show increased accuracy in hurricane predictions, and combustion science where graph search techniques are used to analyze extreme-scale simulation data.