Transitioning Data Flow-Based Visualization Software to Multi-core Hybrid Parallelism

Many of today´s programs for supercomputers are designed for distributed-memory parallelism, but not for shared-memory parallelism. As architectural trends for supercomputers suggest an ever-increasing number of cores per node, these projects must evaluate whether they can benefit from hybrid parallelism - parallelism that blends distributed- and shared-memory approaches - and whether the costs for migrating to new architectures are prohibitive. With this research effort, we explore whether a data flow-based visualization tool could be easily migrated to a hybrid parallel setting, what the challenges are, and what benefits hybrid parallelism achieves. For results, we find that only a relatively small investment was required to transition the tool and it significantly improves performance and also reduces memory usage and communication costs.