부산대학교 도서관

We investigate the OpenMP parallelization and optimizationof two novel data classification algorithms. The new algorithms arebased on graph and PDE solution techniques and provide significant accuracy and performance advantages over traditional data classificationalgorithms in serial mode. The methods leverage the Nystrom extensionto calculate eigenvalue/eigenvectors of the graph Laplacian and thisis a self-contained module that can be used in conjunction with othergraph-Laplacian based methods such as spectral clustering. We use performancetools to collect the hotspots and memory access of the serialcodes and use OpenMP as the parallelization language to parallelize themost time-consuming parts. Where possible, we also use library routines.We then optimize the OpenMP implementations and detail the performanceon traditional supercomputer nodes (in our case a Cray XC30),and test the optimization steps on emerging testbed systems based on Intel’sKnights Corner and Landing processors. We show both performanceimprovement and strong scaling behavior. A large number of optimizationtechniques and analyses are necessary before the algorithm reachesalmost ideal scaling.