Podcast: Doug Kothe Looks back at the Exascale Computing Project Annual Meeting

Print Friendly, PDF & Email


&npsp;

Doug Kothe is Project Director for the Exascale Computing Project.

In this podcast, Doug Kothe from the Exascale Computing Project describes the 2019 ECP Annual Meeting.

The 2019 Exascale Computing Project (ECP) Annual Meeting will convene to highlight technical accomplishments that are being enabled by interactions and collaborations within the ECP community, which encompasses the ECP focus areas, U.S. Department of Energy high-performance computing facilities, and vendors. Networking opportunities will be emphasized and project participants asked to explore ways to leverage and share requirements with other ECP efforts. Key topics to be covered at the meeting are discussions of future systems, software stack plans, and interactions with facilities. Several parallel sessions are also planned throughout the meeting.

Transcript:

insideHPC: Welcome to the Rich Report, a podcast with news and information on high-performance computing. Today, my guest is Doug Kothe. He’s from the Exascale Computing Project. Doug, welcome to the show today.

Doug Kothe: Thanks for having me, Rich. It’s always good to talk to you.

insideHPC: Yes. I’m sure we’ll see each other soon at the HPC User Forum or something like that. But I thought, Doug, it’d be great if we could catch up. You guys had a user meeting for all the people preparing for Exascale. Can you tell me more about that?

Doug Kothe: Yes, I can. We actually call it our annual meeting, and we are in our third year of the Exascale Computing Project, or ECP, so this was our third annual meeting. It is a unique meeting in that it is really a working project meeting for everyone funded on the project to come together once a year and to tackle issues, challenges, to work integration. A very successful meeting. It was just a couple weeks ago—the week of January 14th in Houston. Very exciting meeting.

insideHPC: And Doug, what kind of scope are we talking about? Is this hundreds of people that are getting together and working on this?

Doug Kothe: Yes, it is. It is. In principle, everyone funded to work on ECP is invited. And so that includes a lot of DOE Laboratory staff, students, post-docs, university professors, US industry, and our HPC vendors are there as well. The agenda for the meeting– this isn’t a roll out of our great results, and believe me, we have a lot of them, but the agenda is developed bottom-up by the staff who are, in particular, wanting to tackle certain issues, understanding what they need to do to meet our project metrics. So the agenda is really many working breakouts, tutorials, sessions that are working integration issues across various aspects of ECP. Our HPC PathForward vendors are briefing ECP staff in terms of where they are in meeting their R&D milestones. So it’s a large meeting, but a working project meeting. So in that sense, it’s really become crucial for us in executing year by year.

insideHPC: I was watching Twitter as this was going on. It looked like a lot of collaboration and people meeting in person for the first time, even though they’ve spent years –maybe even working on the same piece of code.

Doug Kothe: Yes, that’s true, Rich. I mean, it’s incredible in that some of us who know each other by name or email have come together for the first time. In the applications area of ECP, you have 25 teams that are there. Many share a lot of the same experiences in background working with similar pieces of software but never come together to exchange lessons learned or best practices. And again, the agenda is really developed based on key issues we need to work. And so you’ll see apps teams integrating with software teams in a session. In fact, the way the meeting works, Rich, is the software stack, in particular, is kind of the nexus of the interactions for the applications and the hardware and the HPC vendors. So a lot of the breakouts and tutorials are in and around the software stack, how the apps can best use the stack to meet their performance goals, and how are the hardware vendors and the software stack interacting? So it really is kind of a week-long co-design effort.

insideHPC: Well Doug, we at—insideHPC we’ve been following your progress because this has a huge potential impact on the HPC community. Would you say that the project is executing as planned? How would you characterize it today?

Doug Kothe: Yes. I would say that, Rich. We have specific performance metrics that we’re targeting for the end of the project. Nominally 2023. And right now we’re on track to meet those metrics. That said, as you know, software development, application development, needs to be agile. And so there are known unknowns and unknown unknowns that you have to essentially plan for to the extent you can. So we’re on track. But yet we’re also working hard to make sure we maintain some agility in terms of making course corrections as we go. Right now things are looking pretty good as we’re in our third year. I guess later this summer we’ll kind of be mid-project. And right now all projections look very good for a successful completion.

insideHPC: Well, thanks, Doug. So one question I get from my readers is about your role at the ECP in terms of acquiring systems versus preparing software for eventual systems. Could you help clear this up for us? Because it seems to be an ongoing concern.

Doug Kothe: Yes. And I’ll do my best here, Rich. The ECP, the Exascale Computing Project, is a part of a larger initiative in the department of energy known as ECI or Exascale Computing Initiative. And this is a cross cut across the entire department that was started in 2016. So the ECP is essentially building the applications, the software stack, and funding the upstream vendor R&D for the actual first three mover exascale systems purchased by the Department of Energy. But those are separate projects. For the Argonne system, Aurora, the Oak Ridge system, Frontier, and the Livermore system, El Capitan, those three procurements are separate projects. But those three plus ECP comprise ECI. And there’s too many ECs here. I know [laughter]. But we’re building the apps and the software stack for the exascale systems. And the exascale systems will be procured by the three DOE labs I mentioned and operated as separate projects. But yet there is intimate codependency with regard to these projects.

insideHPC: Well, great. Doug, it seems like just yesterday this was 10 years out. And it’s three years from now. It’s just around the corner. So I guess you’ve got a lot of work to do. So I better let you go [laughter], Doug. Thanks for coming on the show today.

Doug Kothe: You bet, Rich. Let me just conclude by saying the ECP is truly honored to work in. The personnel we have are just the best and the brightest in their field. And meetings like this just energize us, realize that we have such a talented group. And we’re really motivated to succeed. And we feel like we’re on a good track. So really appreciate you asking about this annual meeting. It’s a very exciting event every year. Really, really energizes us to achieve our objectives. Thanks again, Rich.

insideHPC: Thank you. Well, that’s it for the Rich Report, folks. Stay tuned for more news and information in high-performance computing.

Download the MP3 * Subscribe on iTunes * Subscribe to RSS 

Sign up for our insideHPC Newsletter