ISC is one of the two major international supercomputing conferences held every year. It is complementary to the SC conference.
My history at ISC
| Year | Conference | Location | Mine |
|---|---|---|---|
| 2018 | ISC18 | Frankfurt | 1st |
| 2019 | ISC19 | Frankfurt | 2nd |
| 2020 | ISC20 | Virtual Event | 3rd (remote) |
| 2021 | ISC21 | Virtual Event | 4th (remote) |
| 2022 | ISC22 | Hamburg | |
| 2023 | ISC23 | Hamburg | |
| 2024 | ISC24 | Hamburg | 5th (3rd in-person) |
| 2025 | ISC25 | Hamburg | 6th (4th in-person) |
I have been attending ISC for eight years, though two were remote-only and I missed two due to a career change. My perspective is nowhere near as complete as multi-decadal attendees’, but I think this gives me an appreciation for how the conference is today and how quickly it has evolved to meet the changing field in the past ten years.
ISC25
ISC’25 is the 40th anniversary of the conference. I was invited to offer my thoughts on the following questions for a panel:
Scientific innovation and advancement
What scientific innovations and advancements do you remember from past ISC events? Think about presentations, BOFs, poster sessions, TOP500, awards, etc.
Because ISC has single-track sessions where all attendees gather to hear about a topic, the advancements presented there are what stand out clearest to me. The technical content of these sessions has always been interesting if nothing else, and because all attendees were there, they often became the centerpieces of informal conversations throughout the week and discourse (via news or social media) with those who didn’t attend.
Top500 is specifically the most notable to me. Countless innovative or milestone systems have been announced at ISC, and a few that I distinctly remember are:
- Alps (ISC’24) - the biggest GH200 system, deployed first in Europe
- LUMI (ISC’23) - the biggest European system of its time, built on AMD GPUs
- Fugaku (ISC’20) - an incredible system on many dimensions
- Tianhe with Matrix-2000 (ISC’18) - first demonstration of Chinese-designed accelerators at scale
Fugaku and Tianhe+Matrix-2000 were particularly noteworthy because they debuted non-US silicon at massive scale, reminding the world that there is room for leadership and innovation in all corners of the globe.
If I’m not mistaken, the entire leapfrog-race to exascale also played out exclusively at ISC over the period of a decade. Every new #1 supercomputer between 2013 and 2024 debuted at ISC, taking us from a modest 34 PF with Tianhe-2A to 1.1 EF with Frontier:
- Tianhe-2A debuted at ISC’13, beating out OLCF Titan with Xeon Phi.
- TaihuLight debuted at ISC’16, beating out Tianhe-2A with Chinese-designed SW26010 coprocessors.
- Summit beat out TaihuLight and broke the 100 PF barrier at ISC’18, alighting the race to surpassing 1 EF.
- Fugaku surpassed Summit and broke the 400 PF barrier two years later at ISC’20
- Frontier surpassed Fugaku and broke the 1 EF barrier two years after that at ISC’22
In addition, Aurora broke the 1 EF barrier two years after that at ISC’24.
Top500 at ISC has captured the accelerated pace of innovation of the top-end of the HPC industry, with the three-year cadence of #1 systems coming down to two years.
How has ISC played a role in the development of major industry trends, such as clustering, cloud computing, big data, AI, quantum computing, or other areas?
I’ve found the workshops at ISC have always served as a focused, deeply technical, and intense way to capture the strongest currents of the HPC industry. Because they punctuate the week, the workshops capture a distillation of the most serious and dedicated members of the community. By the time they take place, the vendors, sponsors, and fly-by-night participants have moved on to their next event, leaving behind only discussions of the trends that have real staying power.
For example, the UK Met Office announced that it would shift from its long history of on-prem HPC into Microsoft Azure in early 2021, and ISC’21 was UKMO’s first opportunity to discuss the decision. There was no shiny new system at that point, so the news during the conference week was partly overshadowed by other participants’ big announcements. It wasn’t until the workshops that people from the UKMO could stand up and, with undivided attention, explain what led them to the bold step. It was at these workshops, and in the conversations between the sessions, that I began to grasp a foundational shift happening in the credibility of cloud in HPC. And just a year later, I had left my job in on-prem HPC to join a cloud provider.
Similarly, the HPC-IODC workshop at ISC is where I first learned about how CERN was tackling data management and massive-scale storage challenges—an area in which they continue to lead by finding innovative ways to merge open-source software with large-scale disk and tape deployments. The Arm HPC User Group Workshop (AHUG) at ISC is also where I began to understand how Arm was positioned to challenge the x86 hegemony in HPC in the years preceding Fugaku. There has never been a better place to gain direct access to the people at the forefront of innovation—especially innovation led outside of the US.
In what ways has ISC influenced the public perception of scientific computing? Does it still play that role, now and into the future?
ISC offers a strong, annual reminder that there is a world innovation in HPC outside of the US. Companies like Eviden, Fujitsu, NEC, Sugon/Dawning, Huawei, and Inspur all make incredible strides in building systems at scale. Companies like Thinkparq, SiPearl, and Graphcore operate in different parts of the HPC stack but are equally noteworthy in the novel technologies they have developed. ISC serves as showcase of these companies’ contributions to the state of the art, without as much of the glitz and glamor often seen at other major industry conferences in the US.
Sadly, the political climate in the US is becoming increasingly destabilized, and the internationally collaborative efforts endemic throughout HPC will need to recalibrate. As a result, ISC’s role as the focal point of global HPC innovation will become more important. It has often felt like the steadiest hand among HPC conferences, striking a thoughtful balance between the commercial interests of sponsors, blue-sky research, and practical, hands-on engineering. I expect ISC’s role in maintaining this even balance will only become more critical as the US position on science and technology becomes more mercurial.
Community and collaboration
In what ways is ISC, or similar events, important for the connecting our community?
Smaller HPC conferences are usually targeted at specific subsets of the community—whether system operators, application developers, hardware designers, or CS researchers. ISC is one of the few conferences that brings these groups together on equal footing: a researcher can have a conceptual idea critiqued by an end-user who understands the reality of implementation, and a system designer collaborate with a researcher to bring novel ideas into production.
This orthogonality manifests in many dimensions beyond specializations as well. Early-career individuals can accost community leaders in hallways or after sessions, and like-minded engineers from centers across Europe, Asia, and the Americas can compare notes on innovations and operational challenges. Even competitors in industry have the opportunity to come together in panels and BOF sessions, which serve as neutral territory for debate. These opportunities do not exist outside of conferences like ISC.
What community and collaboration efforts do you recall from past ISC events that influenced the course of your career?
ISC ranks among my favorite industry conferences because it is a concentration of people and organizations who are serious about HPC. I’ve found it easier to run into notable people in the halls of Messe Frankfurt or Congress Center Hamburg than anywhere else in the world. And when I do run into them, the odds of them being swamped is also lower. People seem more accessible, and conversations less hectic, at ISC than at elsewhere.
I say this because my most memorable community and collaborative moments have arisen from serendipitous encounters with an few industry greats. I’ve been amazed by how many accomplished figures have shared a kind word with me about a presentation I gave or something I wrote as I wandered through a social event or waited for a session to start. These encounters have built up my confidence over the years, and I no longer hesitate to reach out to these people when there’s opportunity to compare ideas or work together.
I started my career seeing distinguished names in news articles and being too intimidated to consider reaching out to them, but my experiences at ISC have emboldened me. Now, when I see a headline about a novel new system being deployed, I don’t hesitate to send a quick note of congratulations to the headliner. After all, we met at ISC, and their words of encouragement meant a lot to me. Why wouldn’t I reciprocate?
Feeling comfortable talking to leading figures has had a major impact on my career. I feel empowered to reach out and ask questions to better understand the news behind the headlines. I can get clarity on sticky problems that are only affecting a handful of massive supercomputers across the world, and we can amplify each others’ voices when proposing solutions to the hard technical and sociological problems we face. And I am regularly invited to contribute what I know to a variety of audiences across public and private sector on the basis that we’ve all had friendly conversations in the past at ISC.
To give a concrete example, I was involved in the HPC-IODC workshop at ISC for over five years, first as a presenter and then as a program committee member. I’ve built my credibility in the storage community there by presenting topics ranging from long-term storage strategy to tactical deep-dives into specific file systems. I’ve also amplified the voices of others doing great work on novel storage technologies. Becoming a part of this community has expanded my involvement over the years; for example, I’ve contributed ideas and talks to the IO500 community (e.g., the IO500 ten-node challenge was created in response to a concern that I voiced). I also strengthened the partnership between Microsoft and VAST Data by co-presenting our combined learnings on how LLM training uses storage at the workshop. I’ve often felt that the more time I invest in workshops, the more I get out of them, snd this holds true of the community at large as well.
How have community meetings changed over the last 40 years? How do you think they will continue to change?
Though I’ve been attending ISC for less than a decade, I have noticed a marked improvement in its broadening representation of the HPC community. Early in my attendance, it often felt like ISC organizers wanted to have well-known headliners for every single-track session. While this attracted attention and buzz, it also overrepresented many of the same old voices: largely senior leadership, largely from the USA, and largely male.
I have been heartened to see the conference make strides towards ensuring more voices are heard at these single-track sessions. More luminaries from Europe and Asia have come to the fore, and panels are not exclusively men. Reverence of the past—the gold old days of Meuer, Dongarra, and Simon, and the annual opining of Thomas Sterling—are being dialed back to give breath to a new generation of innovators. As ISC has grown and the community broadened, I am seeing the conference evolve to be more representative of all of us regardless of whether we’ve been on the entire 40-year journey of ISC or are attending for the first time. The only way for the conference to grow with the community is to fairly represent it in its entirety, and I’ve been proud to see how this has been playing out over the last few years.
A big challenge before us (and an area of change) will be how the conference and the community recenters itself in the new world of HPC dominated by AI. ISC is no longer the place where the largest supercomputers in the world debut, and the users of those supercomputers are more likely to be represented at conferences like NeurIPS than ISC. Early signs are showing that the conference is anchoring itself to its historical base—scientific computing at scale—and treating supercomputing in the context of cloud and AI as “them” rather than “us.” I would welcome a change in this line of thinking, though I see no way to do that overnight. It will take time, and it will require both inclusion of and willingness to contribute by stakeholders from scientific computing, cloud, and commercial AI.
Building the next generation
What people do you think of, whose careers you have watched grow from early stage to respected leader, over the arc of the ISC events you have attended?
I think I am still too early in my career to see this, and I am only now beginning to see a new generation of HPC practitioners begin to make waves. However, I may be a part of a rising tide myself, and I’ll share the following anecdote to illustrate:
In the earliest days of my involvement in the HPC-IODC workshop, one of its founders (Julian Kunkel) had just moved from a postdoctoral appointment at DKRZ to a junior professorship at University of Reading. I remember talking to him for the first time post-transition over lunch at ISC’18, at the communal dining tables at the back of the exhibit hall, hearing how his move from Germany to the UK (with his young family!) had gone.
At ISC’21, I then remember chatting with him (via Slack, since ISC was all virtual on account of the pandemic) throughout the HPC-IODC workshop. Due to the time zone difference, I recall the workshop beginning somewhere around midnight for me in California, and it was somewhere around 4:00am the following day that he told me, in confidence, that he received a professorship at the University of Göttingen. A month later it was official, and I recall feeling proud that his tireless work at Reading paid off.
By ISC’24—the first one I attended post-pandemic—our paths were no longer crossing at an I/O events. Rather, we were both talking about AI at the Workshop on Interactive and Urgent High-Performance Computing. His scope had broadened as he took on the title of Deputy Head of GWDG, and I had expanded from being a storage architect at NERSC to being a system architect for Microsoft’s largest AI clusters.
As the years go on, I look forward to seeing this progression play out with Julian’s graduate students and the entire next generation of ISC contributors.
How important is it still to provide a platform to discuss diversity and inclusion? How has ISC done this in the past, how is it doing now, and what should be done going forward?
Embracing diversity and being inclusive of people, perspectives, and ideas across many dimensions is essential for the health and growth of the international HPC community. Just as the “total addressable market” for HPC grows and evolves to include new domain sciences (such as the arts and humanities), new industries (such as generative AI for enterprises), and developing nations (India, Brazil, and others come to mind), so too must the community’s efforts to ensure that they all have a seat at the table of community discussion. Without this deliberate effort, the ISC community risks losing relevance. The next generation of leaders may not come up through the conventional path of PhD-to-postdoc-to-national-lab, and they are apt to build their own communities and conferences if the existing ones do not represent them.
ISC has shown clear growth in this area as I described above. In the earlier years of my attendance, a few of the single-track sessions had a distinct flavor of an “old boys’ club” to them, heavy on retrospectives and back-patting that felt exclusive of those of us who weren’t in the industry twenty or thirty years ago. I think the conference organizers were receptive to the voices that challenged this approach though. It wasn’t long before these keynote sessions began featuring distinguished speakers drawn from beyond the greybeards.
This question specifically asks about a “platform to discuss diversity and inclusion” though, and I hope to see ISC (and the community) begin to shift its focus beyond this concept in the coming years. I’ve often felt it a shame when an important talk or workshop conflicts with a session whose focus is being a platform to discuss diversity. Attendees are confronted with the hard choice between spending time deepening their technical expertise or spending time in a place where they feel better represented.
A question I always ask myself is: how do we bring the community to a place where this choice does not need to exist? I would much rather have underrepresented groups be a part of every panel, every technical session, and every BOF rather than ask them to prioritize one over the other. I would love for representation and inclusion to be an innate part of every conversation rather than a topic in and of itself.
To do this though, we must first collectively understand the reasons why ISC attendees prioritize platforms to discuss diversity and inclusion over other aspects of the conference. We then must make a serious and focused effort to ensure that addressing these factors are a necessary element of every session throughout the conference. There is no reason every panel or workshop cannot leave room to hear underrepresented voices.
What advice would you give people new to HPC as to how to use events like ISC to advance their career?
The amount of value that you can get out of a conference like ISC is proportional to the effort you put in. There is tremendous value in attending the technical program and listening to the brightest minds in HPC from around the world discuss the problems they’re trying to solve. But engaging with speakers adds a new dimension to the experience. Many presenters at technical sessions are early in their careers as well, and a thoughtful question might strike up a conversation that continues over lunch or at a coffee break. And as was the case with me and Prof. Julian Kunkel at HPC-IODC, it might turn into years of delightful discussion with someone who will always have firsthand experience with navigating the same career growth challenges as you.
ISC is also a fantastic place to meet notable figures in the community due to its manageable size but broad scope. A kind hello and a comment about their work might be well received or even reciprocated. I’ve met a few of my “HPC heroes” in-person for the first time at ISC, and I was sometimes amazed to know that they knew who I was or saw something I posted online. I firmly believe that most members of the HPC community are interested in the growth of others; after all, they wouldn’t have gotten far otherwise. Though it may be intimidating to break the ice with someone who you’ve only seen on a stage from afar, these individuals are usually less scary and more just busy. In the worst case, they politely excuse themselves and run off to their next obligation.
And even if the opportunity to approach someone you recognize does not present itself in an obvious way, there are innumerable opportunities to chat with strangers throughout the conference. Lines for lunch and coffee are natural places to strike up a conversation. Set yourself a goal of meeting at least one new person a day. Even if you find that you work in dissimilar parts of HPC, you may know someone in common. Or they may pull you into a later conversation with someone who they think you should meet. This is how professional networks form, and there is no better place to make connections than at events like ISC.
ISC24
ISC24 was a set of firsts:
- my post-pandemic ISC
- my first time in Hamburg
- my first ISC as a Microsoft employee
I accepted the Top500 award for Eagle on behalf of Microsoft that year.
ISC23
ISC’23 was also in Hamburg, but I did not attend. The Azure Managed Lustre service was announced and this was the ISC at which it debuted.
ISC22
ISC’22 returned to an in-person format and moved to Hamburg from Frankfurt.
I did not attend because I had just joined Microsoft a month before. I had booked travel to attend, but it just didn’t work out.
ISC21
ISC’21 was also a remote-only conference due to the pandemic. I presented the early results from the Perlmutter file system at the HPC-IODC’21 workshop.
ISC20
ISC’20 was a remote-only conference due to the pandemic, but I did my best to follow along online in the digital program. I also presented “Managing decades of scientific data in practice at NERSC,” a talk that reflected my then-new role as group lead of the storage operations team, at the HPC-IODC’20 workshop.
ISC19
ISC’19 was the second time I attended ISC. It was in Frankfurt, and I documented my conference attendance in a blog post. I attended to present how NERSC architected its all-flash parallel file system for the Perlmutter system at the HPC-IODC’19 workshop.
ISC18
ISC18 was the first time I attended ISC. It was in Frankfurt. I attended nominally to present on NERSC’s 10-year storage strategy at the HPC-IODC’18 workshop.