KubeCon + CloudNativeCon Europe 2024 | Day 1 Keynote Analysis

Clips
News
More from KubeCon + CloudNativeCon Europe 2024

Dustin Kirkland

VP of Engineering

Chainguard

Rob Strechay

Dir./Principal Analyst & Host

theCUBE Research

AI leads the agenda as CNCF and Nvidia executives highlight need for Kubernetes in training models

Key leaders in the Cloud Native Computing Foundation or CNCF community picked up where they left off at last year’s KubeCon + CloudNativeCon Europe by focusing on the role of Kubernetes in artificial intelligence during keynote presentations today in Paris.“Every segment was about AI and machine learning,” said Dustin Kirkland (pictured, left), analyst for theCUBE Research at theCUBE, SiliconANGLE Media’s livestreaming studio, during a discussion of the conference’s opening presentations. “The CNCF is trying to make sure that Kubernetes is the home for AI training and inferencing

play_circle_outline Exploring the Thrilling KubeCon + CloudNativeCon EU in Paris: Keynote Spotlight on AI and AIML

play_circle_outline Discussion on resource allocation and GPU utilization

play_circle_outline Challenges in monitoring, observability, and connectivity

Info

Day 1 Keynote Analysis

Dustin Kirkland

VP of Engineering Chainguard

Rob Strechay

Dir./Principal Analyst & Host theCUBE Research

HOST

search

(bright upbeat music) (logo whooshing)

>> Hello, and welcome to KubeCon + CloudNativeCon EU in Paris on theCUBE. We're going to have wall to wall coverage this week, bringing you all of the things that you need to know about Cloud Native applications and how they're really changing. Today, I'm joined by Dustin Kirkland who's going to help me break down the keynote that just got out a few minutes ago. Welcome.

>> Bonjor (speaking foreign language) Rob. (speaking foreign language)

>> Ah, I know that's about the extent of- (speaking foreign language)

>> There we go.

>> Yeah, yes. But well, I think it's great to be on with you again. I know you have a lot of background here. Again-

>> Love being here.

>> You know, really bring the developer perspective, the engineering perspective to this, and it's going to be a great week, and I think there's a lot going on. Let's kind of break down some of the stuff that was going on this morning and yesterday in the day zero events.

>> Yep.

>> Where we had a whole number of things from Red Hat Commons to Backstage Day and Observability Day, and a whole number of other different things going on, plus the keynotes today, which really AI was front and center.

>> Yeah, I mean, just start with the size of this place. I don't think I've seen the total numbers yet, but I mean, this expo hall, we're in expo hall seven of I don't know how many there are, eight or nine here in Paris, the lines for, you know, badge check-in, security. There's a lot of people here, Rob. I mean, it's a tremendous conference.

>> I think they were saying it was over 12,000.

>> Wow.

>> The largest KubeCon so far.

>> In Europe? Or in total?

>> In total.

>> Wow. >> In total. So I think Amsterdam was somewhere around 10 is where they capped it. So being over 12, I think they were expecting to be somewhere between 12 and 15,000.

>> It shows. I mean, energy is palpable for sure.

>> Yes, it is. And so let's jump in. I think what was really interesting yesterday and today was there was a lot of different sessions on AI, no matter if you were in the AI, you know, Kubernetes AI, Cloud Native AI. Things like Kubeflow or what have you. But everything talked about AI.

>> For sure.

>> And this morning's keynote was no different from that perspective. What really stood out to you from the keynote this morning?

>> Well, yes, AI certainly was the open, the middle, and the close. Every segment of the keynote, well orchestrated, but every segment of the keynote, all about AIML and really about what I think the CNCF is trying to do here is ensure that the Cloud Native Computing Foundation, Kubernetes and that family of applications really are the home for AIML. You know, training, inferencing, fine tuning, all the different types of workloads that, you know, need to run at this point.

>> Yeah, I think that was one of the keys that I heard. And even back to yesterday when I went to Observability Day, and I had been at the Red Hat Commons, there was a lot of talk about how the Kubernetes infrastructure needs to really evolve to take on model serving, to take on resource allocation. I think that was a big theme-

>> Yeah, DRA dynamic resource allocation, saw a major upgrade in Kubernetes 1.30; 1.30. And I think that's just the tip of the iceberg. NVIDIA had a keynote this morning where they talked about a couple of different ways to drive better GPU utilization sharing. They showed four different ways of sharing GPUs and that in fact those can be layered on top of one another. And I think, you know, one of the sort of paradoxes here is that there's both this worldwide GPU shortage. I mean, you can see that just in the prices of GPUs, either new or used if you're buying them, you know, slightly used. There's this shortage of hardware in part created by some supply chain issues. But at the same time, there's also underutilized GPUs in clouds. We hear that from CERN also, this morning in the keynote.

>> Yeah, and I thought what was interesting from CERN and just building off of what he was saying, is that CERN was very involved in previous open source groups and around resource allocation all the way back to what was the global grid form, the open grid forum, and then open stack. And there is prior art there to be able to do this. I think what was interesting was a lot of the discussion was not about how you train foundation models. It was about how you deploy these foundation models for inference.

>> Yep.

>> And how you fine tune them. And there was a lot of talk about, you know, the actual ARM-based processors and running inference on ARM-based processors. There was, Oracle made an announcement, the last KubeCon, that they were actually going to provide $3 million a year in free to the projects so that they can go out and do this. I think that we're going to see a lot around this. One of the interesting pieces also that was a thread through yesterday in the data on Kubernetes or Doc went to that session as well, was that it's around how do you actually get the data to the models or the models to the data. And there's a lot of, "What's going to happen out at the edge?" And I think that was a really interesting dichotomy was, "Hey, containers have been great. We've been talking about are they, you know, persistent or non persistent data?" And there was a lot of discussion about how you do persistency with like Postgres and SQL and stuff, but what are you seeing as you look out and see that people are just interested in AI? But there's a whole lot of pieces that go along with that.

>> Yeah, I actually don't think we heard all that much about the edge actually in that keynote here today. That's something that I hope we hear a lot more about. Looking at the guest list we have coming on theCUBE this week, I think there should be some interesting discussions there. I would be very interested in some of the talk tracks as well, around edge. You're spot on. I mean, the amount of data that it takes to train a model, moving that across networks is just, it's unrealistic; it's infeasible. And so just driving some of that actual training, GPU, CPU compute as close as possible to the actual data that it needs to consume. It's an important problem that I think we need to hear a lot more about.

>> Yeah, and I think it's a very interesting task at hand for the CNCF with the fact that they keep bringing in different groups and keep on having more in Sandbox. As of, I think it was as of March there was 113 in Sandbox.

>> Yeah.

>> There's 37 in incubating. There's 26 in graduated. And-

>> Is that total projects or are those AI related products?

>> Well, they all have AI now in their name I'm sure, but to get people in. But when you start to look at the 113 Sandbox, some of those are already being used in production.

>> Yeah.

>> By companies out there or there are commercial applications being built out. And I think one of the interesting things is there was a whole track yesterday for startups. And about how to build a business on top of open source and how to do it in a way that doesn't, you know, violate some of the agreements with the CNCF. What are you seeing from organizations out there? Do you see a lot more of people trying to figure those models out?

>> Yeah, for sure. I mean, things have come a long way. The first startup I was associated with, we were trying to raise money in 2009, '10, '11, built around an open source encrypted file system that I had co-authored and co maintained. The VC pitches in that timeframe. You know, '09, '10 and '11, we were explaining open source to the VCs themselves and many of the VCs ran the other way. We've come a long way in the last 15 years now, and open source is, it's table stakes for most startups, you know, talking to to VCs. They do want to see a business model, obviously, but there's some tried and true ones at this point. I did find it notable. I heard it; I counted at least three times in the keynote this morning. Specifically Apache 2 and MIT licenses were mentioned. I don't know if it was a nudge or encouragement, but, you know, Priyanka and it seems like CNCF is certainly, you know, pushing for, asking for Apache and MIT licensed software.

>> Yeah.

>> And you know, it's a little different than the GPL that drove Linux. There's a lot of comparisons between Linux the Kernel, Linux the OS, and Kubernetes. So, but you know, I think in part, the business models surrounding Apache and MIT license software are a little better understood. And maybe a little more tried and true at this point than-

>> Yeah, I think we're seeing it. And I did some briefings before we got here and there was a lot about, "Hey, this part of it's Apache two. This is GPL or it's a custom license." And how they're trying to actually create business value around the Apache 2 software, the MIT license software, which is I think, the right way to go about doing it. People like Dbt Labs have done that for years. And I think they actually have figured out a pretty successful model around that and said that, "Hey, you know, we're not going that direction." I think what will be interesting is that there was a half day open tofu day yesterday. Open tofu if for those who don't know, is once HashiCorp changed their license, really-

>> For Terraform.

>> For Terraform. This really exploded. And I think what was interesting coming into this was there's now rumors on the street that Hashi may be for sale as a public company. So we'll have to stay tuned to that this week because I'm trying to understand, you know, what could happen if somebody bigger bought them and actually went and changed the licenses back or did something with it that made it a more agreeable open source license versus where they went. So I think it's going to be really interesting.

>> Yeah, for sure. >> And I would say also one of the big things is right now some of the crew, Dave and John are out at GTC in San Jose.

>> That's right.

>> Meeting with Jensen and talking-

>> You think they're talking to AI right now?

>> I think they might be. I think they were talking about George Lucas being there and a whole number of others that I saw A-listers coming out. So AI has like jumped the shark when you have, you know, George Lucas going, and I can't remember if it was Xzibit or one of the other rappers going to the actual keynote for Jensen.

>> Wow.

>> It's crazy. But I think what I expect for the rest of the week is that we'll get a heavy dose of AI throughout and interweave through all of the different discussions that we have here, which I think will be great. I also think that, you know, it's 10 years of Kubernetes coming up. We're coming up on the, you know, the anniversary. Do you think most of the major things that the Kubernetes level are solved for and it's really about Cloud Native?

>> Yes and no. I think that question was posed to Clayton in the keynote as well, you know, what's left to be done? And he said, "What's not left to be done?" I mean, there's still plenty of work. The monitoring and observability still leaves a little bit to be desired. It's a complicated ecosystem of ways, you know, other things you have to plug into Kubernetes to get the logging and metrics and observability that you ultimately need. Especially when now you're talking about these AIML you know, workloads that are finding their way deep into the heart of Kubernetes, the scheduler, the resource allocation and so forth. There's a lot still left to do there, you know, so I think that's one key piece that we still have a ways to go, Rob.

>> I think so. And I think that's a place we're definitely dive deep with a number of the different companies that are part of that ecosystem. I think another one that's going to be interesting to see is really that service mesh and Istio and what's gone on there.

>> Connectivity. Yeah, absolutely.

>> Connectivity and networking 'cause I think it's still tough, and I think it's still overly complicated to actually have a mesh of applications talking to each other. And I think that organizations are looking to have that simplified, and I think a lot of that's going to be what we see over the coming years is like, how do we get some of these other things that, okay, yes. Maybe the containers and the format and how they get deployed is kind of solved for, but actually the Cloud Native aspects and observability is definitely one of the ones that and the networking and all of these meshes-

>> Security too.

I mean. >> Oh yeah.

>> You know, just the security requirements in an enterprise, you know, however they're leveraging first the Kubernetes pieces, second the rest of the Cloud Native ecosystem. And then third, trying to take on this whole new world of, you know, AI enablement of that app.

>> We'll actually have the GM from OpenSSF on tomorrow. And we'll actually have also Melinda Marks, who's the director, practice director for security over at Enterprise Strategy Group on with us to kind of wrap on that stuff as well tomorrow afternoon. So I think you're right, dead on that this is going to be a lot going on, and we have a lot of different and interesting things that we can dig into. So, you know, thank you for coming on board and-

>> Absolutely. Looking forward to it.

>> This is going to be an awesome time. And thank you for joining us here at KubeCon + CloudNativeCon EU in Paris on theCUBE, the leader in high tech analysis and news. Keep it tuned here. We'll be right back. (inspiring music)

Day 1 Keynote Analysis

search

(bright upbeat music) (logo whooshing)

>> Hello, and welcome to KubeCon + CloudNativeCon EU in Paris on theCUBE. We're going to have wall to wall coverage this week, bringing you all of the things that you need to know about Cloud Native applications and how they're really changing. Today, I'm joined by Dustin Kirkland who's going to help me break down the keynote that just got out a few minutes ago. Welcome.

>> Bonjor (speaking foreign language) Rob. (speaking foreign language)

>> Ah, I know that's about the extent of- (speaking foreign language)

>> There we go.

>> Yeah, yes. But well, I think it's great to be on with you again. I know you have a lot of background here. Again-

>> Love being here.

>> You know, really bring the developer perspective, the engineering perspective to this, and it's going to be a great week, and I think there's a lot going on. Let's kind of break down some of the stuff that was going on this morning and yesterday in the day zero events.

>> Yep.

>> Where we had a whole number of things from Red Hat Commons to Backstage Day and Observability Day, and a whole number of other different things going on, plus the keynotes today, which really AI was front and center.

>> Yeah, I mean, just start with the size of this place. I don't think I've seen the total numbers yet, but I mean, this expo hall, we're in expo hall seven of I don't know how many there are, eight or nine here in Paris, the lines for, you know, badge check-in, security. There's a lot of people here, Rob. I mean, it's a tremendous conference.

>> I think they were saying it was over 12,000.

>> Wow.

>> The largest KubeCon so far.

>> In Europe? Or in total?

>> In total.

>> Wow. >> In total. So I think Amsterdam was somewhere around 10 is where they capped it. So being over 12, I think they were expecting to be somewhere between 12 and 15,000.

>> It shows. I mean, energy is palpable for sure.

>> Yes, it is. And so let's jump in. I think what was really interesting yesterday and today was there was a lot of different sessions on AI, no matter if you were in the AI, you know, Kubernetes AI, Cloud Native AI. Things like Kubeflow or what have you. But everything talked about AI.

>> For sure.

>> And this morning's keynote was no different from that perspective. What really stood out to you from the keynote this morning?

>> Well, yes, AI certainly was the open, the middle, and the close. Every segment of the keynote, well orchestrated, but every segment of the keynote, all about AIML and really about what I think the CNCF is trying to do here is ensure that the Cloud Native Computing Foundation, Kubernetes and that family of applications really are the home for AIML. You know, training, inferencing, fine tuning, all the different types of workloads that, you know, need to run at this point.

>> Yeah, I think that was one of the keys that I heard. And even back to yesterday when I went to Observability Day, and I had been at the Red Hat Commons, there was a lot of talk about how the Kubernetes infrastructure needs to really evolve to take on model serving, to take on resource allocation. I think that was a big theme-

>> Yeah, DRA dynamic resource allocation, saw a major upgrade in Kubernetes 1.30; 1.30. And I think that's just the tip of the iceberg. NVIDIA had a keynote this morning where they talked about a couple of different ways to drive better GPU utilization sharing. They showed four different ways of sharing GPUs and that in fact those can be layered on top of one another. And I think, you know, one of the sort of paradoxes here is that there's both this worldwide GPU shortage. I mean, you can see that just in the prices of GPUs, either new or used if you're buying them, you know, slightly used. There's this shortage of hardware in part created by some supply chain issues. But at the same time, there's also underutilized GPUs in clouds. We hear that from CERN also, this morning in the keynote.

>> Yeah, and I thought what was interesting from CERN and just building off of what he was saying, is that CERN was very involved in previous open source groups and around resource allocation all the way back to what was the global grid form, the open grid forum, and then open stack. And there is prior art there to be able to do this. I think what was interesting was a lot of the discussion was not about how you train foundation models. It was about how you deploy these foundation models for inference.

>> Yep.

>> And how you fine tune them. And there was a lot of talk about, you know, the actual ARM-based processors and running inference on ARM-based processors. There was, Oracle made an announcement, the last KubeCon, that they were actually going to provide $3 million a year in free to the projects so that they can go out and do this. I think that we're going to see a lot around this. One of the interesting pieces also that was a thread through yesterday in the data on Kubernetes or Doc went to that session as well, was that it's around how do you actually get the data to the models or the models to the data. And there's a lot of, "What's going to happen out at the edge?" And I think that was a really interesting dichotomy was, "Hey, containers have been great. We've been talking about are they, you know, persistent or non persistent data?" And there was a lot of discussion about how you do persistency with like Postgres and SQL and stuff, but what are you seeing as you look out and see that people are just interested in AI? But there's a whole lot of pieces that go along with that.

>> Yeah, I actually don't think we heard all that much about the edge actually in that keynote here today. That's something that I hope we hear a lot more about. Looking at the guest list we have coming on theCUBE this week, I think there should be some interesting discussions there. I would be very interested in some of the talk tracks as well, around edge. You're spot on. I mean, the amount of data that it takes to train a model, moving that across networks is just, it's unrealistic; it's infeasible. And so just driving some of that actual training, GPU, CPU compute as close as possible to the actual data that it needs to consume. It's an important problem that I think we need to hear a lot more about.

>> Yeah, and I think it's a very interesting task at hand for the CNCF with the fact that they keep bringing in different groups and keep on having more in Sandbox. As of, I think it was as of March there was 113 in Sandbox.

>> Yeah.

>> There's 37 in incubating. There's 26 in graduated. And-

>> Is that total projects or are those AI related products?

>> Well, they all have AI now in their name I'm sure, but to get people in. But when you start to look at the 113 Sandbox, some of those are already being used in production.

>> Yeah.

>> By companies out there or there are commercial applications being built out. And I think one of the interesting things is there was a whole track yesterday for startups. And about how to build a business on top of open source and how to do it in a way that doesn't, you know, violate some of the agreements with the CNCF. What are you seeing from organizations out there? Do you see a lot more of people trying to figure those models out?

>> Yeah, for sure. I mean, things have come a long way. The first startup I was associated with, we were trying to raise money in 2009, '10, '11, built around an open source encrypted file system that I had co-authored and co maintained. The VC pitches in that timeframe. You know, '09, '10 and '11, we were explaining open source to the VCs themselves and many of the VCs ran the other way. We've come a long way in the last 15 years now, and open source is, it's table stakes for most startups, you know, talking to to VCs. They do want to see a business model, obviously, but there's some tried and true ones at this point. I did find it notable. I heard it; I counted at least three times in the keynote this morning. Specifically Apache 2 and MIT licenses were mentioned. I don't know if it was a nudge or encouragement, but, you know, Priyanka and it seems like CNCF is certainly, you know, pushing for, asking for Apache and MIT licensed software.

>> Yeah.

>> And you know, it's a little different than the GPL that drove Linux. There's a lot of comparisons between Linux the Kernel, Linux the OS, and Kubernetes. So, but you know, I think in part, the business models surrounding Apache and MIT license software are a little better understood. And maybe a little more tried and true at this point than-

>> Yeah, I think we're seeing it. And I did some briefings before we got here and there was a lot about, "Hey, this part of it's Apache two. This is GPL or it's a custom license." And how they're trying to actually create business value around the Apache 2 software, the MIT license software, which is I think, the right way to go about doing it. People like Dbt Labs have done that for years. And I think they actually have figured out a pretty successful model around that and said that, "Hey, you know, we're not going that direction." I think what will be interesting is that there was a half day open tofu day yesterday. Open tofu if for those who don't know, is once HashiCorp changed their license, really-

>> For Terraform.

>> For Terraform. This really exploded. And I think what was interesting coming into this was there's now rumors on the street that Hashi may be for sale as a public company. So we'll have to stay tuned to that this week because I'm trying to understand, you know, what could happen if somebody bigger bought them and actually went and changed the licenses back or did something with it that made it a more agreeable open source license versus where they went. So I think it's going to be really interesting.

>> Yeah, for sure. >> And I would say also one of the big things is right now some of the crew, Dave and John are out at GTC in San Jose.

>> That's right.

>> Meeting with Jensen and talking-

>> You think they're talking to AI right now?

>> I think they might be. I think they were talking about George Lucas being there and a whole number of others that I saw A-listers coming out. So AI has like jumped the shark when you have, you know, George Lucas going, and I can't remember if it was Xzibit or one of the other rappers going to the actual keynote for Jensen.

>> Wow.

>> It's crazy. But I think what I expect for the rest of the week is that we'll get a heavy dose of AI throughout and interweave through all of the different discussions that we have here, which I think will be great. I also think that, you know, it's 10 years of Kubernetes coming up. We're coming up on the, you know, the anniversary. Do you think most of the major things that the Kubernetes level are solved for and it's really about Cloud Native?

>> Yes and no. I think that question was posed to Clayton in the keynote as well, you know, what's left to be done? And he said, "What's not left to be done?" I mean, there's still plenty of work. The monitoring and observability still leaves a little bit to be desired. It's a complicated ecosystem of ways, you know, other things you have to plug into Kubernetes to get the logging and metrics and observability that you ultimately need. Especially when now you're talking about these AIML you know, workloads that are finding their way deep into the heart of Kubernetes, the scheduler, the resource allocation and so forth. There's a lot still left to do there, you know, so I think that's one key piece that we still have a ways to go, Rob.

>> I think so. And I think that's a place we're definitely dive deep with a number of the different companies that are part of that ecosystem. I think another one that's going to be interesting to see is really that service mesh and Istio and what's gone on there.

>> Connectivity. Yeah, absolutely.

>> Connectivity and networking 'cause I think it's still tough, and I think it's still overly complicated to actually have a mesh of applications talking to each other. And I think that organizations are looking to have that simplified, and I think a lot of that's going to be what we see over the coming years is like, how do we get some of these other things that, okay, yes. Maybe the containers and the format and how they get deployed is kind of solved for, but actually the Cloud Native aspects and observability is definitely one of the ones that and the networking and all of these meshes-

>> Security too.

I mean. >> Oh yeah.

>> You know, just the security requirements in an enterprise, you know, however they're leveraging first the Kubernetes pieces, second the rest of the Cloud Native ecosystem. And then third, trying to take on this whole new world of, you know, AI enablement of that app.

>> We'll actually have the GM from OpenSSF on tomorrow. And we'll actually have also Melinda Marks, who's the director, practice director for security over at Enterprise Strategy Group on with us to kind of wrap on that stuff as well tomorrow afternoon. So I think you're right, dead on that this is going to be a lot going on, and we have a lot of different and interesting things that we can dig into. So, you know, thank you for coming on board and-

>> Absolutely. Looking forward to it.

>> This is going to be an awesome time. And thank you for joining us here at KubeCon + CloudNativeCon EU in Paris on theCUBE, the leader in high tech analysis and news. Keep it tuned here. We'll be right back. (inspiring music)