Developing Custom HPC Data Services Using Mochi

Custom-tailored data services offer the potential to increase data-intensive scientific productivity and serve as test beds for innovation in distributed data algorithms. They accomplish this by targeting specific use cases, such as in situ analytics, experimental data refinement, or in-system distributed caching, using semantics and functionalities optimized for the task at hand. However, custom data services have often proven impractical to develop and maintain from scratch in intricate HPC runtime environments. This tutorial will equip attendees with tools and methodologies to rapidly compose specialized yet robust, high-performance, portable, and sustainable HPC data services using the state-of-the-art Mochi framework and its collection of foundational threading, data transfer, and storage building blocks. Developed by researchers at ANL, CMU, LANL, and THG, Mochi received an R&D100 award in 2021 and enjoys a growing user base in academia and industry. After an introduction to Mochi's core components, attendees will be invited to create their own microservice. Hands-on exercises will use accessible container images and guided development to help attendees unlock the potential of custom data services with a far lower barrier to entry than would be possible with traditional methods.