Rob, Hello, I'm Rob Hirschfeld, CEO and co founder of RackN and your host for the cloud 2030 podcast. In this episode, we do some 2025 predictions and dig into why I think this year is going to be both boring and terrifying for a lot of enterprise IT leaders, and that, of course, spans Amazon, reinvent storage, VMware, AI, agentic, AI, we run the gamut on what is coming and why this is actually going to be a very challenging year. From an IT perspective,
I actually it's funny, because I don't, in some ways, I think this is going to be a boring and terrifying year simultaneously for most tech providers that we deal
with, really, why
the VMware pieces, I think, are just really, really problematic. And then we were looking at the analysis from Gartner, and there is no alternative. The closest alternative is actually just as expensive as VMware.
What do they consider the alternative? The closest
alternative is
Nutanix. Yeah. Okay, so here's a thought. Why wouldn't you do, yeah, why wouldn't you do an agent to take both out of the picture, use AI tooling. Yeah. And create an agentic version of VMware or new tennis. Oh.
So the fundamental, the fundamental problem, to me, is not the hypervi, the hypervisor or lack of it. It's actually the enterprise support, right? Okay, the thing that's really causing the heartburn here is that, you know, these, the enterprises that we deal with, they they want a supported hypervisor or supported platform. Regard doesn't matter. Like we'll talk we actually, that's what I was assuming. I'm glad you're here. We can talk about my the topic I was thinking of for next time, which is containers versus VMs as the as the core of this and rich, this comes back to the question, I think this is going to be both a terrifying and a boring year for for it, right? It's, you know, we know what the problems are going to be, and the solutions are not going to emerge easily or quickly. Which is where I which is where I was, was going. So the, you know, the thing we saw from Gartner really clearly, was that there isn't a good VMware alternative. The closest was just Nutanix was is just as expensive and pretty differently architected. So it's, it's not easy to adopt. And then the the next alternatives down are all open source, no enterprise capability or no scale capability. Ironically, scale computing, which has a nice, nice solution, is not really an enterprise. You know, bra, it's good for the edge. It's good for small, but it's not a Not, not, yet. I don't know if it'll ever be enterprise. And then, like Proxmox is support not, not doesn't have a corporate support structure. Open Stack doesn't really have a com corporate support structure. The thing that people were getting excited about, the outliers are Microsoft Hyper V in part because people already own it and it has better controls than people give it credit for, but we just, we don't see it. We, I mean, we support it, but we just don't see it getting that adoption, or Oracle's KVM has some, has some traction, but, you know, it traction? I think traction is really hard to measure in this case. So it's, there's, there's no yeah, when they say that, you. Uh, Oracle, when you say Oracle has traction, is it? Are they somewhat specialized use cases? No. I mean, if you're looking for an enterprise KVM, yeah, solution, then the idea that Oracle, right, Oracle's KVM solution is, you know, is probably the, the most enterprise supported KVM solution that you're going to find. Open Stack has a lot of baggage on top of KVM and Proxmox is not, you know, just doesn't have the enterprise brand. But I don't, I don't see that kicking up that broadly still strikes me as a as a pretty far reach, because, you know, Oracle is not not the trusted brand compared to VM from a VMware migration. So quick question. I mean, you're, you're, you're quite focused, and I think appropriately focused on virtual machines, the virtual machine environment, and everything goes around that, right? Specifically, you were making a case. Your starting point was it's going to be both a boring and a terrifying year for it. Maybe not specifically regarding the VM where diaspora, as it as it works out, are you coming to any other conclusions about any other kind of sink holes and and major tar pits that need to that are likely to kind of get everybody kind of wrapped around the axle this year. No, that's what's weird. I mean, I think that the Kubernetes, which we see as the virtualization alternative, in its own weird way, is pretty well locked in. So I don't see that being upset. I think the TerraForm, all the TerraForm drama from last is sort of this point. People are just they don't care, although, right, HashiCorp acquisition has not been finalized yet. No, they're having problem. They're having problems with they're having problems getting getting that organized. And I'm not sure what the source of the problems are, but they were, they were advertising like a standalone company. You know, I think it created a lot of drama for HashiCorp, but it ultimately let them shake, shake out their competitors from that perspective. And so they're, they're now, you know, competing on their value proposition rather than on, you know, fighting with, you know, open source competitors. So, you know, not, you know, I, I don't, I don't like, so these, you know that this was a lot of the drama stuff. I don't think, you know, we don't see open source is that much of a driver in enterprise at the moment? Yeah, there's going to be concerns about SAS and South SAS sovereignty and data linkage. I think AI is going to be, you know, it's a huge deal. But I think I actually am coming to conclude one other, one conclusion that is another one of the other at yet, as yet, kind of unrealized, least unadvertised, serious issues that's going to come up is the orchestration of workflows, data data pipelines, which are so so locked in and so built around structured data. You know, it's sure manage your as manage your SQL, right? You know, ETLs and everything. And, you know, we've had since, you know, the last seven years. We've had the, you know, kind of the whole modern data stack and and that whole, that whole deal, which again, is predicated on structured data. Did now, good, good. Now, I think there's a there's a kind of a realization that's dawning on a number of people. Tool. And that is the these tools are not that kind of fragmented group of tooling. You know, the Legos. Legos being kind of smashed together to manage that whole thing is not going to be sufficient, and companies are going to get pushed to do a lot more, AI, a lot more analysis of unstructured, semi structured data. And you know, those tools, those packages, are not, are not appropriate, not set up, not ready for it. And the whole notion of orchestration, as it's com been kind of come to be known with kind of the workflows and data pipelines doesn't, doesn't fit. It doesn't fit. So there's nothing, there's nothing structured or organized about these, this kind of next generation, this next generation of orchestration, and that is going to cause a lot of heart heartburn is, that's my that's my take on you know what's what's going to happen? One of the things that's going to happen this year? Well,
if you recall, sorry, that's fine. If you're this, this has always been my argument about the difference between choreography and orchestration. Choreography is that next level, where you do have unstructured you do have more la flexibility and latitude to be able to put the pipelines together, where it's structured and unstructured and semi structured, and whatever this is where the vector databases come in. This is where all of the other stuff comes to play. But to me, the idea of composition and choreography will be the watch words for 2025 and orchestration is kind of composition
will be composition of what? And yes, I mean choreography as a as a concept. I'm, I'm, I'm all for it. I'm all with you. What I'm not seeing is realistic being put into production in the enterprise I have. I'm not seeing real choreography showing up to the party. It they're kind of little experiments and nibbles at the edge. But I'm, you know, so I'm, I'm wondering how much of the realization is going to happen this year. It's going to be significant. I think that the enterprise, you know, it estate, is going to be put on notice. Hey, you need to make good on some of this AI stuff you've been harping at for a long time. We need access to a lot of this structured data that you know, is sitting there, not being used correctly. And I don't think enterprise is going to have a lot of success finding packaged, packageable solutions to the problem. It's going to be kind of, I've been describing that as missing reference architectures? No, this is, this is the what, what we're we're in the middle of this really interesting disruption, because the last 10 years have been a VMware driven reference architecture. People knew what to bought by. They knew the architecture stack. They knew all these pieces. And now there's no new reference architecture putting all these pieces together, I you know, and I think that's and you can't just take Amazon's cloud mate and say, Oh, my reference architecture is Amazon. Even Amazon's reference architectures are a mess. And this is, this is not to say that, you know, those of us who are kind of tuned into it aren't absolutely correct about the need for it. Can kind of envision what it might eventually look at look like, right? But, you know, someone comes to me today and says, This is what I need. And you know, I almost say, you know, there are possibilities for you, but they're almost all precision, kind of built for purpose. Those kinds of solutions, and you can either DIY it, or you can get some, somebody who knows enough to kind of build it for you. But it's, it's a solution. It's, these are going to be solution sales. They're not going to be, not going to be, they're not products, yeah, no, this is, this is, I mean, we saw this, you know, across the board on the AI cluster pieces on, everybody's off reinventing the wheel. There's, there's no repeatable process for anybody. Jo Anne certainly has a, has an alternative, or at least a duplicate, on it. Come on, Joanne, I uh,
agents
where rich came in. Go ahead,
there are eight different design patterns that I have for agentic that would allow you to create the agents and do the choreography in such a way that you could leverage containers and micro services in those containers that allow you to do the the choreography that is, that is the design pattern. It is the choreography in the architecture that gives you the flexibility because, to your point rich, yeah, it could be one of over and over and over again. And how do you, how do you productize that to be a standard? But you could take the eight different patterns that exist for agent tech and leverage those as your baseline, because they all allow you to have the flexibility that you need. The notion of composability and or and choreography is, from my perspective, anyway, on the agent side, I can have my agents collaborate. I can delegate one agent to another agent. I can use all those design patterns. So to me, if I choose any of the eight or the most flexible of the eight patterns, and I say, This is my baseline, I could then create standardization around that pattern and use that as my architectural reference. When, when
you describe that as a pattern, though, I mean, what? What are people buying for that? I mean, this, this is the challenge with the reference architecture, is it? There's, there's controlling things, but there's also, like, I go by this standard set of something and it, I know what it I know how it's going to do it, and I know what it's going to work. You know how it's going to work? The agent, yeah, I'm sorry. I was going to say the, going back to, going back to the, the, the sorry, Rob, to the issue of of, you know, the next generation of ISVs delivering to, you know, to your to your Kubernetes, you know, clusters, yeah, are I we're not at a point where I can see anybody doing exactly that for choreography. I I'm not arguing your point that Joanne, that you don't use, you know, different design patterns. And there's a, there has to be a, you know, kind of a design language that you use and to express those paths, you know, to express them and but it's, it's not quite at the point of products, packaged products, you know, deliverables that that an enterprise is going to buy and install and make use of the way they have kind of learned to in the in the last, you know, 20 years. And I think that's the that's kind of where I think we're kind of in this this funny, you know, I kind of know what I need, but I can't find it anywhere. Situation. But
I think that that's rapidly changing because the companies that are coming to market, whether it's the new my new one, composable, or cohere, or any of the others that are playing in the agentic, even c3 AI, to a certain extent, they're all coming at it from this similar point of view. So there are packages that are emerging that either support the the notion of, here's the design patterns you can use, or that they're. Prefixed, right? If you look at the way composable is, what I argue with with cancer, about is you've taken a very defined set of capabilities and you've put them together for a pipeline in the standard, traditional way of pipelines. And I disagree with that only because the point that he's eliminated, the flexibility that they're trying to create goes away, because you put all this in structured form, when, in fact, you have a myriad of structured, unstructured and semi structured forms. Yeah.
And I think we're dealing with, we're dealing with the horseless carriage syndrome, which is, well, right? I need, I need that. I need this kind of structure. And this I have step one, step two, step three, four, a, 4b, so forth. Yeah, I get it. Okay,
so cohere, on the other hand, took a slightly different take, and they're coming much more from the unstructured side, but the design patterns still work. And so my argument is because the way everything has always been built in pipelines around workflow, not around data,
right, right? Yeah, that's the
paradigms that's shifting. Is that now we're building around data, not around around workflow.
So we should build around data instead of work, yeah, oh, because there's a lot of different workflows rather than data, okay?
Because, no, there's a lot of different data types versus workflows that are designed for structured data.
So this is where one of the questions that I was, I was I wanted to ask is, you know, a big announcement from Amazon reinvent that was sort of quiet. Was this s3 parquet support. So out of s3 they added like automatic indexing for parquet data. And you can do queries like they build the if you, if you drop parquet in, I think you have to turn it on too. But if you build parquet into an s3 bucket, they will give you querying capabilities to pull parquet to query that that, you know, I don't know if you consider parquet structured or semi structured data out and the statistic that drove that was fascinating, the lead of s3 said that parquet was like 100 times the most downloaded format by format, and basically there you're there. It's absolutely true. That's absolutely and, but what it says, but actually, have you gone through the list of announcements that of over the course of the last maybe two months, and kind of leading up to reinvent them, and then at reinvent the announcements around s3 and what they've added, what they've added to it, because they're there, it looked like they just did a major region re They did a pretty major rewrite. I mean, they've talked about it being like, six generations of s3 buckets of s3 so this is, it's still the foundational technology for Amazon, absolutely. But they've added all kinds of stuff. You're they they got, they made some very interesting announcements about supporting glacier, you know, that was that was also very much involved. So they're attending to some of the aspects, again, in hopes of retaining their enterprise customer base, keeping them, you know, keeping them local, and paying, paying their s3 bills. But this is, but this is fascinating, right? To Joan's point, right? What we're talking about is a upcoming new data storage revolution. I still think there's a missing architecture, right? Where, from an enterprise perspective, they were like, Well, wait, I don't need to buy sand anymore. I don't actually need NAS anymore. I need this, you know, and park is out of the Hadoop, right world. But I need this, this, you know, object, this, block storage, or, I'm sorry, object search, I need, I need fast and flexible object store, right? And that's critical for the training and inference, yeah, but
it's also in memory. Think about that. That's important well, because it's in memory, storage, and that means. They're trying to optimize towards AI and towards the agentic side of AI, or alternatively. And I came across this, not randomly, but in one of the searches, there's a whole movement around symbolistic AI that nobody's talking about except very few of us, and that is where I think Amazon is aiming, because on the symbology side, basically all you're doing Rob is you're taking and you're representing the prompts as sets of data in an algebraic format, so you can take a very complex problem and break it down, okay? And it's not only faster, less token consumption and less storage heavy, but it is memory heavy without the GPU, right? So they're running this weird balance, you know, or this is what I'm taking from the s3 announcements as well. But I'm looking at this and I'm saying, Okay, I think back rich to that paper that you sent me on the one megabit.
Ai, yeah. One bit. LLM, in the
yeah, sorry, one bit. LLM, right, right. It's, it's, it's down that road. But this is where, this is, what's informing some of my points of view about the agentic side and about the need for blob or ORM even, is probably closer in what s3 is, is releasing, remove, but that's a data centric view. That's not a workflow centric view, and that's the influx.
I'm not arguing. I'm not in disagreement. I'm just saying the, you know, the organization that needs to solve the problem still has nowhere, nowhere to go for something that's more product oriented or product productized for enterprise you, you know, manage you're, you're either doing DIY and trying to train up yourselves on how to, how to play that out, or you're going to go to GS eyes and other kinds of, you know, specialty houses to to build it for you. And I'm just saying that there's going to be a there's going to be an air, there's going to be a timeline here. There's going to be a gap between the haves and the have have nots, the people who either know how to do it or know where to buy, where to get the skills to buy it. But almost all of these solutions, all of these are going to be solutions, as opposed to products.
Well, all I can say is I'm hopeful that we're going to have that product by the middle of q2 and I'm looking at it from the point of view of leveraging what's already out there and what we just finished working on the last few days. But it's the it's also the difference between a vertical stack and a horizontal play. And that horizontal play, I think you're going to see more and more horizontals than you are verticals.
Horizontals from an agentic perspective, or or horizontals, and what, what do you you mean as people put together
to cover the bases of the enterprise play? Ah, you're gonna see a horizontal like more a platform play than a product play.
I agree with you in that, right? What we what I think the enterprises are looking for is a, you know, they need a replacement platform for for VMware. Yep, right. The only obvious one that we see at the moment is open shift, Kubernetes, generally. But open shift actually has more bits and pieces into it. It's more, a little bit more opinionated, but I do think you're highlighting something that's missing that Amazon is is very aware of, and we shouldn't ignore, is that, you know, s3 is a dumping place for AI data. They're like, they're like, you know, they're seeing a. Shift, you know, reading the tea leaves, they're seeing a shift of pretty mammoth proportions of people dumping data into s3 and using it as you know, Joanna, exactly like you're describing. It's this common source. I don't think it's a particularly good common source or safe common source, but, you know, it's easily accessible by multiple parties. There's relatively good controls on it, and with the optimizations they're putting in, pulling data out of it is increasingly optimized. I do think enterprises. I'm surprised. I'm surprised that Microsoft isn't showing any signs that I'm aware of to follow suit with its blob storage. That is, that is sort of strange, very expensive to cross platform, from that perspective, to pull, pull the data out of s3 so it still feels like Amazon, like just dominates the market. From an s3 perspective, there are a couple of, you know, people trying to compete on pure cost. Yeah, right.
And, and don't forget, you still have a competitor out there that people have kind of abandoned to a certain extent in the last two years, that's making a very strong comeback. And that's IBM, okay, okay. And when it comes to blobs, and when it comes to anything that's even closely call it arm's length, related to a mainframe, right? And Rich to your point, that's the only company that I see people suddenly taking interest in, whether it's the Kindred side or the IBM side, around granite and Watson X and whatever, whatever they're perfectly primed for exactly this shift. This is what they've been doing for a while, and that's going to be the Amazon competitor to s3 not Microsoft. In my view, IBM
has a chance. IBM definitely has the chance to show up at that party again. No argument there, um.
And I'll tell you who else is going to be part and parcel of that picture, big database, because database is now going to take second stage behind everything, AI, so what database you put in as your foundation? If it's not able to do all the various data forms and all the various data types, they're going to be knocked off, and all of them, and I mean all of them, including SAP, are scrambling around. Oh, bleep, we just got taken out. If we don't do the AI, and if we don't do the agentic AI, we are going to lose big time. So that shift also figures into where the s3 play is. It's not just about the storage, it's about what you're going to put in behind the scenes. And even though you know a database is a database is a database. No, they're not all created equal. Number one and number two, if you're moving to Blub
or or, yeah, you're, you've, you've, no the Absolutely. And this kind of goes back to my earlier point that, you know, modern data stack was built on the premise that, you know, everybody's using an SAP or using, you know, a SQL, SQL, RDBMS and so forth, yeah, and so in the back room, you're absolutely right the big data, the big data solution is going to be something that has the capability of managing data, that is, you know, down to the type that, you Know, key, value, pair, you know, object, as isn't and conversation doesn't just isn't. This just translated into the phrase data lake for that has because, okay, no, there's, I mean, data lake is maybe kind of a sloppy first, first attempt at it. But no, I think, I think we're talking about, you know, companies that are built on Cassandra, for example, have a leg up. Um. So it's the data stacks. It's, you know, foundation dB. You know that Apple is using exclusively and and at scale in both those cases, um, based on advertising and money flowing into these conferences, I would say, you know, Cassandra is definitely in a very good position, and they know it, and data stacks too, right? Yeah, well, data stacks is, is built on a Cassandra, right? That's, yeah, so, yeah, yeah, there. So it's I, I agree that the choice of your big, your Big Data, Data Store. Database is going to be critical as there's so there are, there are, there are a few that have a leg up. And so what we're, what we're getting to to me, because I'm going back to reference architecture and Joanna, I agree with you on the need for the agentic piece. But when, when I think about what the enterprises are figuring out, what they have to buy, they're they're starting to come back and say, All right, what's the foundation of my data center look like? It's not a SAN it's not virtualization. It's Cassandra, right? Or some Cassandra storage that's highly, high, very fast for object retrieval, that includes indexing around parquet systems, so I can dump a whole bunch of data into it without organizing it, and then pull data out. And then I have a layer on top where I actually have to run analysis. I'm going to need a big containerized work engine, right? A Kubernetes system on top of that, which makes sense, right? I'm going to have to have and you. And then there's also all that architecture like they don't know what to buy to build that Right, right? And historically, enterprises have said, I don't need to know what I have to buy. I'm just going to buy VMware architecture with very expensive sands and very expensive blades and very expensive platform and all of that stuff then guarantees what I need, what I need to build, what what? I think we're, you know, what? So they're, they're in this market where they're like, they know they don't want to keep buying that, but they don't know what they do need to buy,
right? That's a show and show them the way argument into the market for the data dogs, the Cassandras and the other things. But I would say that there's a layer that you didn't address. Oh, okay, because I think the storage is going to need a I think we're at, I think we're at an inflection point where you need another layer between the analytics and the storage. That's not necessarily the DB layer. It's more like either a semantic layer or a parsing layer or a something, and that's going to dictate also what you what tool you buy. So I think that's infrastructure. But
where I'm saying, you know, in a sense, that's kind of to Rob's point. I don't know if somebody came and asked me what? Okay, he's told me what I need here, between between my store and store in the and the and the analytics, where do I get it? And I, you know, I have to sit there and kind of go, Well, this is now, you guys got to build it. I'm sorry. Well, what I'm wondering is, if you can say, Look, this is what your storage layer looks like. This is what your hardware layer looks like with GPUs in it. They don't have to be the latest generation. I just need, we know we're going to need some, right? You don't want to do it in memory. I know. I'm going to, that's all going to end up, you know, those are, those are applications running in a containerized system. So it can be so I know, right? So, so the safe bet is, I build object oriented store, object storage. I build container at Kubernetes platforms. I build fast interconnect between them, because I know that's going to be critical. And then it says, it says also that your this at this shim, this layer in between has also got to live inside, you know, inside Kubernetes and inside containers, correct? That's why it gets delivered by the next generation of ISV, because they're going to be, they're going to be basically evolving this stuff at at light speed, right? And of those people, the right, Red Hat has the best opportunity to form an ecosystem because of the way they've they're doing open, open shift as a layer. Yeah, is, sorry, what I think, always think IBM, I think Ivy to Joanne's point, IBM has a, has a has a story there to tell that I think, is absolutely, there's a, there's a lot left to be it's. Said about smart management, kind of not a data lake house, but there's an enormous amount of governance, data hygiene that has to be done at various points, and you can't just throw it all in, you know, to the to the quote data lake. There's part of the job of this layer is going to be not just the semantic, symbolic, the whatever you decide is the right approach, it's going to have to it's going to have to be attentive to the to the stuff you're putting in there for everything from de duplication to, you know, data licenses to, you know, you pick it. Let me, I'll show you. I'll show you how we're talking about this. Okay, so this is, this is a part of what we're we're looking at this. This was a really interesting discussion with between me and our CTO, because I had the blue side, and then we added this other piece. So there's, there's a layer in here, which is reference architecture e, in that this is not about the number of VMs. It's actually about VM via virtualization first platforms, right? We lost. We lost at least I lost the picture there. Oh, hmm, let me stop sharing and redo it. Okay, okay, I'm gonna make a little smaller too. That helps. I recognized I had a higher, larger than 10 ADP rich. Is that better? Oh, I see it now. Yeah, good. Okay, so, so what we're this is about, right? Because I'm back to reference architectures. So this is a virtualization first platform, VMware, right? And we see this as having a dramatic decline. They're expensive, they're not fit to purpose, right? What we just we just spent 45 minutes talking about they have no VMs are adding no value in those architectures, right, right? What my CTO added, that I think is actually reasonable, is people still need the VMs, and so what, what we're gonna what we're gonna see is a significant amount of of, and there's a, there's an interesting twist in here, but that the companies are going to say, I don't want a virtualization first platform. But if I took my container first platform and made it do VMs, well, then that would be very attractive to me. And so what, what we were, what we're what we're seeing, what we're expecting is in this platform that we're going to remove virtualization and but the VMs are going to migrate. We're going to, we're not going to, we're going to reduce a virtualization platform without reducing the amount of virtualization as significantly, which is an interesting nuance in it. And then, then, right? So this, this gets layered into what we're saying is from a reference architecture, what do I buy? Perspective, we're seeing containerized workloads from, you know, they're already high, but they're going to dramatically transform. On top of this is, is the piece. But the thing that he would, that my CTO points out, and I agree here, is that network segmentation is a major problem. What, what y'all are describing is something that Kubernetes right now doesn't do very well without having dedicated clusters. And so, so you're we're in a weird position where you were describing all these layers, but those layers are not actually well commingled in a cluster. They're actually going to end up being cluster one, cluster two, cluster three, type of designs which just sort of brittle. Why? Because I don't know that you have to segment, but there's going to be, I believe that there's a degree of desire to have network segmentation between some of these more sensitive workloads. So if you're building a big containerized infrastructure, you don't, you don't want to give it access to everything in the planet, which is the way things are right now. So there's, there's reasonable the needs to say this workload needs to have, you know, in this VLAN, on this segmented traffic, and I don't want it commingled with my other my other network, just like we used to take data store and write VMs, people are people are used to like, oh, wait, the storage layer for my VMs is actually not accessible from the VMs. You have to have a V. M that attaches to that storage layer, and be the traffic cop, and be the traffic right? It's, it's, you know, especially with these tools, they're so powerful you could take an agent. This is, right, this is, I think, what, which should terrify people about what we're building with s3 which is, you're, you're, you've now turned on parquet indexing for s3 buckets. And so now you could take an agent and say, hey, I want you to find this whatever. And it doesn't even have to open all the files. It can just query s3 and say, I'm looking for this information. And s3 will go here. Here's the summary. Have a great day. Here's the summary. The other thing that's going to be kind of strange when it's when you're doing content based retrieval, basically when the content itself is used as the means of generating that the address in the in the address space where you're going to put so think about it. You know, when you've got chunks, when you've got vectors, and you know they're, they're already hashed, you've basically got a a unique, you've got a unique ID, right? Well, if you use that same unique ID as the address space that could be managed, you basically say, All right, I'm looking for something that is exactly this or close to this on some similarity measure. Bam. Yeah. I mean, you're, it's, it's a, it's basically, it addresses the whole issue of scale, you know, I can, I can, you know, I can put enormous amounts of data in here, and it's really not going to affect the, the the the retrieval performance right of that, of that system, and it can be even
more interesting if it's a rag performance that we're talking about. I'm sorry if it's
a rag mag. Oh, performance,
right? Because I mean to your point, rich, a Okay, so, so we've been playing around a lot with the architecture for Federation and how that, you know, how we would federate, not only federating, but we actually threw some stuff into Xeno to see how it would perform with the Kubernetes. Okay? And it was quite interesting, because it made the performance about six times faster.
Into federated design with Kubernetes.
Federated design with Kubernetes and using a protocol called Zeno, okay, which is basically
protocol that you were using it as a network protocol between, yeah,
we were using it as a network protocol and also as an object protocol to see whether we could do, how much we could do with it, from a point of view of not only network packets, but also objects or agents, right? If you make the agent composable, or, you know, using choreography, and the agents are so minuscule in terms of what they actually contain, that's when you can start making them dance. Because you can say, you go do this, you go do that, and pick up the broker and use the protocol notion,
with the notion of of doing replications across
data stores, with notions of pull it back quickly, right? A mixed, if we were using cradle back end, and it's fixed type retrieval, okay, right? Retrieval, it's mixed type data. Oh, here it is, an agent that does the actual query and comes back. You have the the layer that,
yeah, I see where you're going there. No, we, we're, we're, we're dealing with a form of Federation that is, is independent of the underlying protocol. But the the the idea is to minimize to, I mean, truly minimize replication, but still, still give the impression for the purpose of, you know, any of the major functions, you know, set theoretic kinds functions. You know have no you know, have no problem federating, federating multiple data stores on the basis of kind of establishing. Treaties among them. And then, from the applications point of view, they don't know they're using a federated, correct, a federated infrastructure. It just all looks like, you know, one addressable collection. No, I get, I see, I see how that might be very, very valuable, simply because I completely agree with the approach. It's still, let's just say, you know, this is not, you know, a flat pack IKEA, you know, roll it in and and, you know, make it work for your enterprise. There's, there's, there's, I think my point has been that there's just a lot of education that needs to go along. There's going to be a lot of gonna be a lot of bullshit. There's gonna be a lot of bad there's gonna be a lot of bad product that's sold. There's gonna be a lot of bad, bad kind of Systems Integration and consulting that's gonna be sold as well. That's why I think it's going to be this year, and then well into the next year that it's going to be for the enterprise, it's going to be messy. So let me just add, this is a crazy question, and then Rob, you're the you're the guy to answer this one with all that we've just said and and, you know, acknowledge, notwithstanding, the role of Kubernetes is, is there A potential new role for container other forms of containerization, including, you know, everything from, well, let's just say that you know the one that first, at least for smaller organizations, Docker may have a Docker Swarm or, I mean, there's Cooper, Nomad is still a valid I mean, does that, does that play at all? Does that have a does that have a ring of truth, or is that just somebody's pipe dream? Yeah, I don't, I don't know. There's definitely other container managers that are that are on the market. I think if I was building a dedicated application in my own containers and just trying to deploy it right then, I think one of those platforms would be a fine choice. But I think for anybody who's looking for an ecosystem play, I think it's going to be really hard, a really hard sell to to make and make a statement about it, because we're describing a really ecosystem plays and multi, multi machine. And I mean machine. So you're building a, you're building an, you're delivering an application as to run across multiple machines. And while Kubernetes has a lot of overhead for that, it also has a lot of structures for it. And the thing that makes open shift, open shift is actually the CR used. They use the CRD pattern very heavily, which I don't think has good management layers on top of it. I think, you know, one of the things, if, if I was looking to build a an interesting startup, I would build a CRD manager at the moment, because I don't think, I don't think, I don't think the ops teams are prepared to really manage CRDs, and I think open shift is doing this better. But the thing that that, you part of the magic to me that is making opens, our Kubernetes more popular than people think about is it's not just that I can manage containers and I have a CLI for it, but there's a way to register inside of Kubernetes that I'm providing you a service, and then you have this declarative pattern that says, Oh, I can interact with my service through this standardized, declarative pattern. This was, this was the real brilliance of, darn it. I'm blanking on their name. The there was a Kubernetes coming in the cup bought by Red Red Hat that was doing the they were doing their own thin Linux. They had a whole bunch of stuff based out of Seattle. I can't believe I'm forgetting their name, but they they were, they were. They had a vision for for sort of this, this pattern. But if you look at you're not talking about mirandas, no, it's not. At this. It's they, they're the ones who are behind container Linux. And if once they don't come back to me, the Yeah, because they, their, their their branding was, was slipped it slipped through and some of this stuff. And they did, like, rocket, like they were really innovating on, like they had some really good thinking with this or debate they were, they were very tied into the Kubernetes, whole Kubernetes mindset, the but the ability to not just deliver container management, but to deliver the control mechanism. On top of that is, I think, is, is actually a differentiator for Kubernetes. It's really important. And you'd either have to add it as a separate product, cross planes, doing some nice work. On top of this, they're using the same pattern, right? It's, this is the thing I think people forget with Kubernetes. I know I do. It's not just about container management. It actually has a control, a service control plane built into it, right? And so, you know, if you were going to, if you were going to escape with Nomad into a competitor for Kubernetes, which I think is not unreasonable, you still have to have somebody do the service control plane somewhere. And that's, that's a missing that's a missing piece for that. It's, it's been part of it same time. It's, it. There's a lot of overhead in making Kubernetes work. And from an ops perspective that the teams that we see, right, this is the thing that hasn't happened yet. And this is a 2025, thing, with VMware not being interesting right now. The way the companies deploy Kubernetes is they let the Kubernetes team, which is mostly developers, get VMs, run VMs, and the operations teams have that as an abstraction that's not going to be as interesting, but those operations teams aren't ready to run Kubernetes. It's not designed for them. They don't know how to do it. They don't have a reference architecture. Doing it on bare metal is incredibly complex because the abstractions don't work the same way. So it's, all you have to do is start adding variety of different sources, including, you know, streams and of various different sorts and audio, videos, all this, it's usually, and this is, this is why I think it's terrifying and boring, right? I think people know where they need to go and who's terrified, Rob, it's, it's sort of the same it you ever sat down and said, I know exactly what I have to do, and it's going to be a whole bunch of grind, and it's going to be mess, and I'm going to have to, like, you know, it's going to be a fight. I'm going to sit down every day, and I'm going to be fighting towards a predictable outcome. I'll be kicking the cat on a regular basis. Yeah, and that's, that's how I think people are. That's what I think this year is going to look like from an IT perspective. I think Joanna, you know, the agents are an obvious answer to a ton of things, but nobody knows how to do them. So it's going to be like, I'm going to be stumbling down blind alleys. I'm going to be building
stuff I disagree with that, Rob, I'm sorry. Okay. I don't think that people are stumbling around. I think they're looking at they're looking at the agentic frameworks, right, and trying to pick and choose, not realizing that you can use a variety of frameworks in the same basic setup, depending on how you want to parse them. You don't have to use a single framework to do a group of agents. You can use a group of frameworks to create specialized agents that as long as they collaborate, as long as they can task each other and or in a secure environment, they're good to go. No, you know, I know Rich, you're skeptical of that, but there's a couple of articles I'll send you where people are actually doing this in no pun intended, in real time or in real life where, and they're not small companies. The issue, to
me, isn't the plausibility. The issue is the predictability for the for the the not the early adopters. But the early, you know, the right, it's, I think even early adopters are the wave of adoption where a lot of it, you know, they're used to, they used they want, they want a predictable formula before, before they commit millions of dollars into something. They want a predictable formula. And so there's a lot of. It right? Companies that are that are not ready to get there yet, that they but they know they're going to have to fight through it.
Well, they know they're going to have to fight through it. But the one shift that I am seeing, and I've seen this, it's probably the August, September, time frame, is companies realize this is a big kahuna that they have to deal with, but they must become purposeful about it. Their whole AI strategy has to become very purposeful. It's not just you know, they may have done their experimentation with Gen AI, or whatever they're now realizing that you can't always get from here to there with a Gen AI, that there are other kinds of ML, and there are other kinds of AI that you need to use, and you have to have the purpose. What am I using this for? Right? And now what am I using this for is now the shift that is the differentiator between 24 and 25 because, as they become
available, data product the data product mindset, right? But once again, we're working back from that right now, I think most folks are going to run into, wow, I don't know. I don't know where to find that. And it's complex, and it looks very purpose built. It looks bespoke like this. The The answer is a bespoke solution, and at this point in time, it is kind of that the question of, you know, whether the companies you're referring to can package it up, put it into templates, templated forms, you know, sell it, have a license model, be on the supported list. You know that it's, I guess, I think that's, that's kind of where I think, you know, there's a mindset that says that's what I need in order to proceed with this, because otherwise my ops organization is going to, you know, basically flip me the burden and tell me to, you know, drop that, drop that keyboard and back away slowly. That's, that's the, that's the challenge, right? That you
like, that's the IB that's where the IBM site comes in. Agreed, agreed, yeah,
I but I don't think they have, I don't think they have a reference architecture yet. And that's, that's, that'll come. It'll come. I mean, once it comes, it's going to be, it's going to be a boon. But, yeah, I mean, that's where our investments going. No, I think the big question is, you know, how long will it take and and, you know, is it going to show up this year? And even if it shows up this year, how rapidly will people, you know, embrace it and adopted. Well, it's the challenge with, with any of these mass market things, is that right, that you have to have time to prove it for right? So there's a, there's a built in lag cycle, and it, I don't think it can happen faster than a year. So this year is going to be, you know, people trying, you know, trying, figuring out what to do. People are going to be trying bunch of stuff out. The ones are the constitution to do it. Yeah, yeah. And argues for, you know, having a, having a, you know, a place where you can walk in and, you know, I uh, sample the goods, try it out. Correct for that's the first the first phase is, I need to be able to have a pilot that comes up in a reliable way without needing a huge amount of custom support and bespoke whatever. Because, and that's not going to accelerate the market. Yeah, there will also still be a lot of people who are, are going to be wedded to this notion of a pipe, of a workflow and and pipelines, and want to structure their solutions that way. And they're going to, they're going to have, they're gonna, they're they gave us serious bit of indigestion, I think over the course of this year, I agree with you, garger on that one, you all for the lovely conversation. Next week, I am not available to host, but the week after, we should be back. And if you all have a topic you want to dig into, we covered, we covered my January 13 topic just now so well, this was a good this is a good conversation. Thank you for starting, kicking it off because I I had a lot of time. Okay, not a lot of time. I've had more time over the course of the holidays to be digging into some of this stuff. And think I would, I would be interested in this simple symbolistic Ai, because we touched on that. But I I'm interested. So, yeah, there are a couple of, you know, symbolistic ai, the semantic, you know, semantic layer, kinds of stuff, which is more old school, or it comes more out of the, you know, kind of the the ETL community. There are also some agents, agents tech that that's actually generating, basically generates code and then fixes itself, fixes its code runs into problem. So it's not just delivering, you know, JSON based responses back to, you know, the various players, and I saw one of them two days ago, and got to play with it, And it was pretty interesting. So we'll see.
Wow, it's great to start the new year on a fantastic footing, really identifying some key issues and challenges and then decomposing how these reference architectures are going to shape up in ways that will drive our conversations for the rest of the year. Part of the thing that makes us fun is the conversation. So if you are enjoying these podcasts, and if you're listening to me now you are, then please consider joining us. Come in, be part of the conversation. We would love to hear from you. Thank you for listening to the cloud 2030 podcast. It is sponsored by RackN, where we are really working to build a community of people who are using and thinking about infrastructure differently, because that's what RackN does. We write software that helps put operators back in control of distributed infrastructure, really thinking about how things should be run, and building software that makes that possible. If this is interesting to you, please try out the software. We would love to get your opinion and hear how you think this could transform infrastructure more broadly, or just keep enjoying the podcast and coming to the discussions and laying out your thoughts and how you see the future unfolding. It's all part of building a better infrastructure, operations community. Thank you. Applause.