KubeCon + CloudNativeCon NA 2024 | Anand Babu Periasamy, MinIO Inc. & Sanjeev Mohan, SanjMo

Clips
News
More from KubeCon + CloudNativeCon NA 2024

Anand Babu Periasamy

Co-Founder & CEO

MinIO, Inc.

Sanjeev Mohan

Principal

SanjMo

MinIO’s mission to streamline AI infrastructure: Inside AIStor

Artificial intelligence infrastructure increasingly focuses on streamlining AI data management solutions to meet enterprise demands.For software company MinIO Inc., this priority led to the creation of its AIStor, a commercial product designed to centralize data for AI applications, according to Anand Babu Periasamy (pictured), co-founder and chief executive officer of MinIO. Built on MinIO’s open-source platform, AIStor represents a shift toward delivering enterprise-scale solutions for data-driven AI needs.MinIO’s Anand Babu Periasamy talks with theCUBE and guest host Sanjeev Mohan about the company’s focus on simplifying AI infrastructure and how centralizing data advances enterprise AI

play_circle_outline Discussion on MinIO's new product AI Store and its evolution from data infrastructure

play_circle_outline MinIO's Progress in AI Store Integration and Future Outlook for AI Capabilities

play_circle_outline Unlocking the Power of AI: Company's Strategic Shift to Focus on Killer Use Case and AI Store

play_circle_outline Data-Driven Decision Making: The Key to Success in AI for Founders

Info
Transcript

Anand Babu Periasamy, MinIO Inc. & Sanjeev Mohan, SanjMo

Anand Babu Periasamy

Co-Founder & CEO MinIO, Inc.

Sanjeev Mohan

Principal SanjMo

KubeCon North America kicks off with AB discussing the launch of AI Store by MinIO, targeted towards AI workloads. Data infrastructure is key in modern AI practices, driving the need for solutions like AI Store. MinIO's success in various use cases, from large clusters to edge deployments, showcases their versatility and broad adoption. The focus on simplicity and scalability guides their product development, leveraging technologies like RDMA to optimize data access for AI workloads. The consumer-inspired approach to product development ensures accessibility ... Read more

explore Keep Exploring

What has been a significant milestone for MinIO and how has the evolution of their use cases impacted their market dominance and focus on specific business use cases? add

What is the importance of centralizing and building an AI data repository for the success of an AI practice? add

What should the business focus on as the one killer use case to bet the company on commercially? add

What is the advice to founders and anyone out there in terms of focusing on one thing and excelling at it? add

bolt Powered by CUBE AI

Anand Babu Periasamy, MinIO Inc. & Sanjeev Mohan, SanjMo

search

Savannah Peterson

>> Good morning, AI fans, and welcome back to Salt Lake City, Utah. We are here just getting started with our three days of coverage at KubeCon North America. My name's Savannah Peterson with my KubeCon wing man forever, Rob Strechay. Rob, this one's going to be fun.

Rob Strechay

>> This is going to be fun. I think the entire community's fun, but I think, again, being a data geek, I love being able to dig into the applicability and how it plays in the community.

Savannah Peterson

>> I know. This is really kind of in our nerd sweet spot, you and I. It's-

Rob Strechay

>> It is. This is very much in our sweet spot.

Savannah Peterson

>> Speaking of sweet nerds, AB, thank you so much for being here with us on the show. We're very delighted to have you. And Sanjeev, always a pleasure to have you coming and hanging out with us.

Sanjeev Mohan

>> Likewise. Although, I have to say, I'm hurt. You said AI fans, what about data fans and Kubernetes fans?

Rob Strechay

>> Yes.

Savannah Peterson

>> They're all a part of that.

Rob Strechay

>> They're all watching. They're all out there watching.

Savannah Peterson

>> And in fact, I think, all of those things play a role in AI as well, so they all get a shout-out throughout the course of the 32-

Sanjeev Mohan

>> It's been bushed.

Savannah Peterson

>> Trust me, Sanjeev, I just try and keep them on their toes over there on the other side of the camera, so it's good. AB, bumped into Jonathan earlier today. You know we're big MinIO fans over here. I've actually modeled your shirts. We have you on the show every KubeCon. I bumped into Jonathan today and he was very excited, telling me all about your announcement from yesterday. AI Store, give us the breakdown. It sounds super cool.

Anand Babu Periasamy

>> It is actually a huge milestone for MinIO because all of the deployments, if you see the past, MinIO is deployed all kinds of use cases, from really small to very large clusters, and the deployments are just... The number of deployments, we are the dominant player in this market by adoption, the largest player by numbers, but then when it came to business, you don't want to boil the ocean. You are better off winning one use case at a time and, as the product matured into the commercial market, we started seeing where is the most powerful business use case. It started out as data infrastructure, running analytics and database and machine learning, and that evolved into GenAI. Post-GenAI, every one of our customer is now restructuring their organization to put AI at the heart of their business and that leads to data. And now, that data, you can put it in public cloud or you can put it in private cloud. Private cloud, we are the choice and here we learned a great deal watching our other large scale customers, the kind of stuff that we did at hyperscale level, these are exabyte scale deployments, we took those improvements.

Savannah Peterson

>> Which is huge, for the record.

Anand Babu Periasamy

>> Yeah, and then created a commercial version of MinIO optimized for AI workloads and that's AI Store.

Rob Strechay

>> Yeah, I think, again, we were talking-

Savannah Peterson

>> Congrats....

Rob Strechay

>> before this and there was a lot of statistics about the amount of downloads and you were talking about...

Sanjeev Mohan

>> Yeah. Rob, it's so interesting. I decided to write one of my blogs on MinIO, which I will publish tomorrow on my Medium site, and so I started doing my own research and I thought MinIO had half a billion docker pools. It turns out there's two billion now. And then I find out that there's six million deployments and, actually, according to MinIO, there are more S3 deployments from MinIO than AWS itself, which I find mind-boggling because AWS owns it.

Savannah Peterson

>> Wow.

Rob Strechay

>> Yeah, I think that just talks to the power of-

Savannah Peterson

>> That's impressive...

Rob Strechay

>> open source.

Savannah Peterson

>> That is really impressive.

Sanjeev Mohan

>> Yeah.

Anand Babu Periasamy

>> It's the same way Kubernetes grew into the enterprise. They come through the back door and, as a shadow IT developers, they didn't listen to the analysts and CIO what technology we should bring. That's how it used to be, right? But when these developers brought in new technologies, that became the foundation for modern infrastructure, the same way Kubernetes grew, Kafka grew, Elastic grew, MinIO came through the shadow ID and now every one of these organizations are surprised to see at the scale we are operating. Data is the heart of their business and where is it sitting? Most often, it's on our technology. They're discovering that and they're coming to us to build a long-term partnership.

Savannah Peterson

>> One of the things I'm always really impressed with your company, and I haven't had the chance to tell you this specifically, is you're this humble, but wildly adopted everywhere. Every time I talk to Jonathan, we're talking about different users of your product and different use cases, and I'm thinking to myself, that is so impressive and embedded. You recruit from your customers, as well. I know we have Daniel on the show tomorrow, which is awesome. Since this is your baby, and we are seeing a different level of adoption right now, what are some of the use cases that get you most excited or some of your customers that have you personally most excited?

Anand Babu Periasamy

>> Yeah, even on the first part, we are everywhere, but you didn't know about us, or some people, if you see the business side, the CEOs might not have heard about us, but if you actually talk to the infrastructure people, we are everywhere, right? It was an intentional persona we created, learning from the consumer market. The consumer market first create a beautiful product, but average product still dies, right? Bad product dies quickly. You build a product that people fall in love with and then earn their trust and then it grows and grows and then you build the business relationship. Now, these customers come to us and they want to forge a long-term relationship that's starting to work and this is where the use cases-wise, I was like, we are running in all the most popular cruise ships you go, we are running there. 5G Towers. Even the same users who would run us at hundreds of petabytes to access scale on their day-time job, they go home and they would run MinIO on their Synology and QNAP Home NAS systems. Because MinIO has an R port, we saw all kinds of absolutely fun and crazy deployments. I would never imagine someone who you can use it like that. These are some large retail stores. I can't name them. They put MinIO inside the camera, and lots of cameras, and these are like 2000 cameras per store. They're trying to digitize and automate everything. I never intended that kind of use case. Originally, I built MinIO for very large deployments, not for these small deployments.

Rob Strechay

>> Not for the edge deployments and things of that...

Anand Babu Periasamy

>> Not for the edge deployments.

Rob Strechay

>> But you're seeing it. I think when we were talking earlier, and I know Sanjeev is seeing similar things, is that a lot of people are bringing the AI to the data. The data has weight, it has gravity, it's on-premise, it's in these edge locations. Talk to that, a little bit about what you've been seeing in your customer base as people grow that data and need these commercial related, I guess you could say advancements, that you've been learning from that open source community.

Anand Babu Periasamy

>> First, I think bringing AI to the data, now it has become very clear, if you brought 10,000 GPUs, if you don't have data, those GPUs are going to sit idle, right? That's where

Sanjeev Mohan

>> And they're expensive.

Anand Babu Periasamy

>> Expensive. Every hour counts. And this is where they realize that, unless you put data at the heart of your business, bring all of your data from different teams and centralize and build the AI data repository, you're not going to have an AI practice. AI starts with the data, so once they do that, then the AI will start spinning. Interestingly, for us, when we saw that we got pulled into the AI market because the data was sitting on our platform, public cloud object store is the foundation, private cloud you wouldn't go to San or NAS, so we were at the heart of it, but if AI is the key commercial use case we need to win, we can do more for the AI use case. If they took us to AI, why can't we take AI to MinIO product and that's what resulted in AI Store. We actually brought in cool AI capabilities inside the product itself. That way, if your data is sitting in our system, you don't need to download the data, you can directly prompt the data and talk to the data. That opened up a whole new set of possibilities for our users.

Savannah Peterson

>> That allows people to realize values so much faster. It's such a lower barrier to entry too. And it makes so much sense. Why would you do all of those things if you don't need to, if you only need to communicate? You mentioned communication there. You mentioned the word cool. Before we were live, you said something very cool to us about your product development and how you looked to Instagram and TikTok as inspo in your product roadmap. Tell us about that. That was very striking,

Anand Babu Periasamy

>> Most people don't know all the kind of stuff that's happening inside the company. What comes out is actually very small part of all the development that goes on, very much like you write a novel. Only one version of it you release, everything else is destroyed, right? You actually write code and throw it away and it keeps us sane doing this.

Savannah Peterson

>> I'm writing a book right now, so I feel seen and attacked and also very empathetic right now.

Anand Babu Periasamy

>> You know exactly what I mean?

Savannah Peterson

>> I know exactly what you mean.

Anand Babu Periasamy

>> All those variations are necessity, right? It's not a waste, right? It leads you to the right answer and the AI work that we have been doing, we were drowning in our own data and the more data you capture more data, you make it accessible, suddenly it's not accessible, meaning too much information is also useless. This is where we saw... Always we look up to the consumer world, they solve problems at that massive scale. When you have too much data, how can you make it consumable? Take the most exciting part of the data and surface it, it goes viral. And how do you do that when you have unstructured data? In the past, you could do that in a database, but if you're talking about unstructured data, you cannot do this. This is exactly where GenAI comes to help. We actually turned all kinds of information, initially to a rag model, then to a Q&A, automatic FAQ. Nobody consumed that information. Turn that into a viral podcast, automatically, even celebrities talking about all the things that's happening in the company over the last one month, everybody's involved. Then, these exciting innovations that we were doing, customers were excited too, so they wanted that. We took that and brought it inside the product in a consumable way so everybody can take advantage of it.

Savannah Peterson

>> That's so cool. It's really impressive.

Rob Strechay

>> And I think what's interesting is, again, as AI evolves, you're bringing out things, like you have your AI hub, you have, I think, new protocol support, RDMA. Again, I know there's InfiniBand out there, but Ethernet's going to win. Most organizations. Do you look at all of these enhancements and how are you again bringing that into your roadmap? Is it, hey, customers are saying to us, we have these HPC clusters, we're doing RDMA, we need to go to higher speeds. We need this protocol support. Is that really how you're pulling that in?

Anand Babu Periasamy

>> No, actually not. It's quite opposite. Again, learning from the consumer world, right? Consumer playbook is actually the playbook we follow for enterprise because a customer paying $10, $100, they expect a world of everything. In the enterprise world, when you pay a million dollar, what kind of service you deserve? You should expect a white glove service and a red carpet. And this is where, if you turn to more and more complex and larger systems, it's not worth it. Enterprise should not be complex and shiny and heavy. It should be simpler. Only simple systems scale, so the filter we always apply is, if you can take the bleeding edge, it could be RDMA, it could be a rag model, doesn't matter, if I cannot simplify it and I cannot operate it at scale, it's not going to scale in every way. That is the filter. Even RDMA. We have used it extensively since my previous and previous startup. I came from the super computing space, but if you see, RDMA never got the enterprise adoption. These are Java developers and they don't know how to do RDMA, but this is where you take this and hide it inside the protocol. It still looks like S3 API for you, but underneath, when you transfer data, it switches to the RDMA channel and make it consumable, make it easy that you don't need to understand what's under the hood. If you want to, the code is there, go read it, right? But simplifying everything is the only way to make it work.

Savannah Peterson

>> Go for it.

Rob Strechay

>> How do you see this playing out? Because again, you're an analyst, you see this perspective and you see a lot of different things.

Sanjeev Mohan

>> Simplification is something that every client I talk to wants and this is not just in object storage. It's true even for what's happening with Iceberg, Delta and all that. Simplification is... Reduction of complexity means reduction of cost. In a world where skills are so hard to get, it's inevitable that we need to move into simplification. Now, what I find interesting, and I want to get your thoughts on this, is I feel AI has a hunger problem. It needs data and if it doesn't... And, that too, contextual data, which is your own PDFs, your own videos, your audio, images, that are inside the organization's security firewall. What I really like, what I'm starting to see is from simplicity point of view, you have something called prompt object where I can now ask questions directly without having to do rag pipeline. The other thing that I'd like, that what you announced today, is that, by using RDMA and you just get rid of all this memory to memory copying on Ethernet that runs on 200 or 400 gigabit per second Ethernet, you are really saturating the pipe, so you're making that data, A, accessible through prompt object, B, available to the models, right? You're solving both the top front end and the back end. Is that how-

Anand Babu Periasamy

>> It is. We look at it in two parts. One, the data problem, if you look at the compute side, it has gone complete belly up. The old infrastructure was CPU and they went scale out, and it's a very different architecture, but in the new modern GenAI world, CPUs are like co-processors.

Sanjeev Mohan

>> Yes, correct.

Anand Babu Periasamy

>> GPUs are-

Sanjeev Mohan

>> GPUs are, yes....

Anand Babu Periasamy

>> are the part of the modern processing.

Sanjeev Mohan

>> Correct.

Anand Babu Periasamy

>> And GPUs, like the supercomputing I built in 2004, was the fastest in the US. Today, a single GPU can run circles around it. And that kind of density, a single node has eight GPUs, and then you have now thousands and thousands of GPUs, the whole compute infrastructure has changed and that put enormous tax on the data infrastructure. You cannot build modern AI data infrastructure on traditional San and NAS type infrastructure, right? This is where the networking, everything is changing. In the last 12 to 18 months, we are seeing a massive shift in the infrastructure architecture. That's why the AI Store was so critical for us, that it needs to adapt to the modern infrastructure. But end of the day, if you're an average web developer writing in Node.js and PHP, if you have to learn all this, AI is not going to work.

Sanjeev Mohan

>> Correct.

Anand Babu Periasamy

>> That is where I'm going to hide all the complexity underneath, but it's important for me that your compute side, GPU, has to be kept busy. If data is the bottleneck, GPUs cannot be busy. RDMA, stuff like that, are so critical. Network is the bottleneck because data and compute are desegregated and they need to be connected with a high bandwidth network and that is hidden underneath API that it looks so trivial from the application side, but all that magic, they don't even know. You know how to do get object and put object, you can do prompt object without knowing anything how it works. That is consumerization of AI.

Rob Strechay

>> No, I think that totally makes sense. I think, again, it's one of these things that you got to look at it... I'll ask your question, how about this? I love your question.

Savannah Peterson

>> Okay. You can have it today.

Rob Strechay

>> Can I go there?

Savannah Peterson

>> Yeah, you can go there. I do have another question for him, though, before we wrap this, but you can go ahead.

Rob Strechay

>> Okay, I'll let you go then.

Savannah Peterson

>> I think it's interesting. You started MinIO 10 years ago, same 10 year anniversary as Kubernetes right now. Community obviously an exciting spot. We're in a whole new world. I just want to know, as a founder, as a creator, as a builder, did you think you would be in this spot a decade ago when you embarked on this journey?

Anand Babu Periasamy

>> I picked this problem because of a very simple reason, that I want to pick a long-term problem. I don't want to pick some short-term problem, build a startup, get acquired, go home. I didn't do this for making money. You do this because you want to change the way, fundamentally, infrastructure works, and I picked this problem of data because I knew, 10 years from now, I ask the same question, it's going to be more true than ever. Today, if you see, because post-GenAI world, data is everything. I said this 10 years before, it's more true now, but we always pick problems that are worth solving, but it's so easy to articulate that why it is important, and then, every day you make improvement incrementally, and it's going to give you enormous advantage because you focused on just one problem. And so, we are here because we saw a long-term important big problem that, every day, you wake up to go solve it.

Savannah Peterson

>> Beautifully said.

Rob Strechay

>> I was going to say. That actually leads-

Savannah Peterson

>> You're so on it there....

Rob Strechay

>> right into her question that I'll ask this time.

Savannah Peterson

>> It's your question this interview, Rob. I love it. Let's go.

Rob Strechay

>> What do you hope that, when we have you on, either in London in six months or in Atlanta in a year, that you can say about AI, AI Store and how MinIO is moving forward here again?

Anand Babu Periasamy

>> Yeah. I think the path for us is clear. Literally, this launch is the second phase of the company. The first phase was about land grab and adoption. We need to be everywhere. We needed to create the market because, outside of the public cloud, the world looks so old school and incompatible with the cloud, so Kubernetes level, the ground on the compute side and object store, MinIO level, the ground on the storage side. The SAN and NAS and traditional VMs are irrelevant in the modern world. That is achieved now. Now, as a business, the second phase of the company, pick one killer use case and win it hands down. What is that one killer use case? The market has clearly answered our existing customer success, all the large scale deployments and the revenue, and the pipeline is everything pointing towards AI. Because it's just about AI. Then, commercially, let's bet the company only on data part of the AI. If you think AI, you're going to think data and we need to be the data part. And that part is the only commercial focus. Do one thing really, really well and that's the AI Store bet we are talking about. We are going to put all our muscle power behind this commercial product and it's going to take off.

Savannah Peterson

>> I love that. That's advice to founders and to anyone out there, man. Put all your focus on one thing and really smash it. AB, this has been extraordinarily inspiring. Thank you for joining us today and congratulations. Such a milestone for you all. And also 10 years is a legitimate amount of time to be crushing it like you are.

Anand Babu Periasamy

>> Yeah, it is. It is. And it feels short and it feels long too, right? But we think-

Sanjeev Mohan

>> In database world, we say it takes 10 years to build a good database.

Anand Babu Periasamy

>> Exactly.

Sanjeev Mohan

>> We are saying the same thing now with object store.

Anand Babu Periasamy

>> It's true for any enterprise product. It's a great honor to be here.

Rob Strechay

>> Thank you.

Savannah Peterson

>> That's exciting and we're excited we get to have the conversation. Sanjeev, thank you for joining us.

Sanjeev Mohan

>> Very welcome.

Savannah Peterson

>> Always love having you here.

Sanjeev Mohan

>> My pleasure.

Savannah Peterson

>> Excited to check out your Medium post tomorrow as well.

Sanjeev Mohan

>> Thank you.

Savannah Peterson

>> That sounds awesome. Rob, thanks for parting.

Rob Strechay

>> Absolutely.

Savannah Peterson

>> And for distributing my questions today. I like this. We can wear many hats over here.

Rob Strechay

>> Absolutely.

Savannah Peterson

>> It's awesome. And thank all of you for tuning in, wherever you might be on this beautiful rock. We're in Salt Lake City, Utah. Day one of our coverage here at KubeCon. My name's Savannah Peterson. You're watching theCUBE, the leading source for enterprise tech news.