A review of ISC25 

Birds of a Feather sessions

A major part of the ISC25 programme was the Birds of a Feather sessions (BoFs). These were a great environment to learn about new developments and the future roadmap for a number of projects, as well as connect with new HPC communities. 

One such example was the Open OnDemand (OOD) community. OOD is a web portal for HPC that I have been exploring as part of my work here at EPCC. Therefore, it was great to attend the OOD user group meeting, which offered a chance to hear from the developers and gain valuable insights into how OOD is being used across different organisations. Use cases ranged from single-site setups to more complex configurations, such as a single OOD portal providing access to systems across Europe. There were also some engaging discussions about the recent version 4 release, as well as questions about security considerations and application development. I look forward to connecting with this community more in the future and putting some of what I learned into practice. 

There was also a lot of talk at the conference about what comes next. We have a number of Exascale systems either online or about to come online - how do we make good use of these resources? And what might be the next innovation or new technology?

One answer to this could be quantum technologies, which had a large presence at the conference. However, a BoF on the "Dynamic Use of Resources for Affordable Exascale and Beyond" caught my eye. This was a very interesting discussion on how we can go beyond the static resource allocation used in classic batch scheduling systems. Instead, real-time system monitoring could optimise the amount of resources allocated to a job to balance power consumption, I/O bandwidth, and compute throughput - maybe even adjusting the resources available partway through a job. There are a lot of challenges to solve here, but ISC provided a useful forum to discuss these ideas and see where the HPC community may be heading. 

Tutorials 

The final day of ISC was dedicated to workshops and tutorials. I had the opportunity to attend two tutorials: "Portable GPU Acceleration of HPC Applications with Standard C++26" and "Efficient Distributed GPU Programming for Exascale". 

The answer to which approach to use for GPU programming has been "it depends" for a while now; nonetheless, I was excited to see how support for GPU programming is improving within Standard C++. The tutorial was a half-day session and mostly focused on parallelism in the standard algorithms library. However, the brief introduction to the new C++26 scheduler and task graph concepts caught my attention, and I’m looking forward to exploring them further. If ultimate performance on your specific hardware is not the goal, but performance portability really matters, I think the changes coming in C++26 (and beyond) are really promising! 

I found the tutorials an excellent way to develop my HPC skills and put questions to experts in their fields. I appreciated the chance to join some practical sessions alongside the rest of the ISC programme. 

Other highlights 

Outside the main sessions, there were countless opportunities to connect with other attendees in the exhibition hall. I enjoyed talking to the other exhibitors as well as spending time at our EPCC booth. We brought along our Raspberry Pi mini supercomputer "Wee Archie" and I had the chance to talk with visitors about our work at EPCC. 

Conclusions 

Overall, attending ISC25 was a valuable experience. There was so much going on! It was a great opportunity to learn more about the latest developments in the field, connect with others working on similar challenges, and get a better sense of where the industry is heading. I’m looking forward to applying what I’ve learned and continuing the conversations that started in Hamburg.