The dissertation from Tim Jammer covers the challenges of modernizing MPI usage in distributed memory computing. Despite advancements like RDMA, many applications still rely on older MPI operations, creating a cycle where new features remain underutilized and unoptimized. To address this, the research explores three key strategies: improving correctness tools, using compiler-based methods to modernize MPI code, and leveraging compiler insights to enhance performance. Contributions include MPI-BugBench, a benchmark for evaluating correctness tools, automatic code transformation for easier adoption of modern MPI features, and compiler-driven optimizations that improve efficiency without requiring manual code changes.

“Modernization and Optimization of MPI Code”