This workshop aims to connect researchers, developers, and Python practitioners to share their experiences scaling Python applications and codes on supercomputers. The goal is to provide a platform for topical discussion of best practices, hands-on demonstrations, and community engagement via open-source contributions to new libraries, runtimes, and frameworks. Based on invited talks that survey and summarize the best practices and recent success stories, panel sessions that discuss details of implementation and live demo sessions for hands-on enthusiasts – the workshop will serve as a requirements gathering exercise for the future of Python in HPC and science.
In this full-day tutorial we will explore several object storage technologies and discuss both old and new approaches to getting maximum performance from them. With a mix of lectures and hands-on exercises, attendees will
learn about object storage design and usage. We will also tackle how the classic I/O software stack has evolved to make use of object stores, as well as provide attendees with the knowledge to know when filesystems and object stores are the most appropriate storage approach for their applications. By the end of the day, attendees will know a bit more about what these new storage systems are doing while at the same time learning how to use libraries and tools to hide the details.
AI has had a profound impact on the supercomputing ecosystem, with dramatic advances in technology, size, and scale of supercomputers being deployed. While there is no doubt that AI software is very different from scientific computing applications, underneath the software, how different are the supercomputers powering the AI revolution? Have AI workloads fundamentally changed the technology used to build supercomputers that are the foundation of scientific computing, or have AI workloads just accelerated in time the path that scientific computing was already on? In this panel, our experts will examine this question, looking at the hardware and software that power the world’s largest supercomputers for both HPC and AI, and discuss the implications this has on organizations who rely on these systems and how this affects the next generation of professionals starting out in a supercomputing field that is changing faster than it has in its 60+ year history.
Bandwidth, not capacity, drives the power draw of modern storage systems, and flash has made this performance-capacity ratio a genuinely open design choice rather than one fixed by economics. Every architect must now decide how much bandwidth per petabyte to power, and those watts come out of the compute budget. Both directions are expensive: too little bandwidth costs productivity as applications contend with each other, and too much forfeits the GPUs those watts could have powered. The capacity-optimized end carries an added reliability risk, because ultra-high-capacity SSDs demand data protection techniques that scale with them. We map this design space using two production systems built from the same architecture and components: one providing over 900 PB of usable capacity, and the other 10x less capacity but significantly higher performance. From this, we demonstrate that bandwidth requirements now warrant scrutiny proportional to the costs they impose on system design.
