NB504: Cisco Updates Licensing, Branding with Wi-Fi 7; Google Cloud To Require MFA
44:43 Share

Take a net k break. I'm drew conry murry.

I'm john a. Johnson. Have yourself.

Have to a virtual donas. We try to set a new land speed record for the fastest technologies podcast. Today, we're going to talk about new cisco wifi seven ap and a new licensing approach, google cloud moving to enforce multifactorial th nician aviatrix rolling out a pas offering, this goes q one financial results and more.

We have a tech bites after the news. Today we're talking with sponsor mini o about how A I is authoring the data infrastructure landscape and my organizations are looking to build A I infrastructure on prem will also dig in to mini OS recently launched A I store. This is a software only distributed object store that offers simplicity, scalability and performance for a infrastructure, another high performance store use cases. And just to reminder, the pack pushed network has even more podcast goodness, including and is for networking, network automation, nerds and jha year other show heavy strategy.

Yes, thank you for the proof there.

absolutely. And spread the wealth. So this is a self follow up, a self, if you, I guess, johna. As last week, we have talked about new firewall from sonic wall, the t eighty.

I think the most everything about IT was that they were offering a warranty potentially to two hundred thousand dollars worth of a warranty if there is a security breach and financial loss. Um I mentioned having reached out to sonic while to get more details about how that warrantee might be claimed. Here's what I got.

First, customers need to purchase a qualified firework that is the the T A T M B registered in myonians al account after november first two thousand and twenty four qualified firewalls must be configured according to the sonic well configuration guide and run current software and meet specified requirements. And I didn't get any more details on what those specified requirements are at the warranty offers compensation for covered losses that lead to business interruption from three types of events, one non volume metric, d dos attacks to unauthorized remote access and three software exploitation. So john, I feel like we ve got a little bit more detail, but not everything. Yeah.

i'm trying to figure out what did not attack is of its non volume trc and unauthorized remote access of like what now the firewall or devices behind the firewall and similarly software exploit of the firewall or devices behind the firewall. My gut says that unequal was actually thought this through and probably has meaningful answers to those questions. But it's a little frustrating that they didn't specify them.

Yeah, I think so um I was just working with you know a peer of person who probably went to them and said I need some details and they just you know give us the top line stuff. Onic was working with a third party called since ance. And i'm sure insurance has you know a lot of detail on this because you know it's in the insurance company's interest to not necessarily have to pay out.

So again, I think we mentioned this in in our last conversation. There's going to a lot of fine print uh, and you want to dig into those details like even the first one non volume meta ddos attacks. I it's reading on that. And yes, there are different types of data attacks that could be qualified as non value metric. But whose deciding is IT do you get to decide the sanawar get to decide as some third party involved.

So a lot of, well, sins since you did the digging, what is an non volume metric ddos attack because that seems to be like non wet .

water in my mind yeah so um I guess they would call IT sort of a resource exhAusting attack like a sin flood. So trying to use the open as many sin connections as you can in the in the device essentially topples over, as opposed to just gaming as much traffic into the pipes you can.

that actually makes a lot of sense. Thanks for that distinction and .

glad you duggin IT. I did. But again, who's deciding that right? So yeah, if you if you're planning on you interested in in signing up, its on equal and being in line for this warranty.

Uh again, a lot of fine print h make you check the details. I appreciate the sa qua for sharing what they would with us, and i'm sure there are reasons why they don't want to share more. But this is to have some it's going to result in some conversations if if you're deciding .

to engage this way.

let's get into the new sisco has announced two new access points to support wifi seven. This is the seventy one seventy eight and C W ninety one seventy six ap till start shipping in december twenty twenty four and you may be wear wifi seven offers new features to improve performance and through put clean things like multiple link Operation or ml, which less devices connected on two bani multi ously.

Other upgrades include wider channels that you can stuff more data into for each transmission. At the big new series that cisco is putting, my the wifi seven line under a single brand called cisco wireless now also changing the licensing so that wifi seven, a peace, could be managed via the cloud or via an unprepared roller. And customers can switch between both Operating models without needing to change out the hardware or the switch license is on age. Jeff.

thoughts on this. Well, I think it's actually interesting both the branding and the licensing model. Essentially, what you're getting is unified for support for both your cisco wifi devices.

Andrew morarji, whether managed from the cloud, from on premises both and the new access points incorporate a blend of cisco morarji management and security features that previously quired quote, significant integration work from the customer or a third party and the quote. So I think this goes doing the right thing here, doing the immigration of of post acquisition, although its very post acquisition in this case. But you know they're giving you the best of both world's management. They're giving you an upgrade in terms of um wifi seven. And you know that that mo is actually quite, quite interesting, certainly something to dig in to and the fact that you can manage you ve got the best ball of worlds when IT comes to management security, net net, assuming at all you plays out as advertized, is actually pretty compelling.

Yeah, obviously, you know for any vendor to be running multiple product lines that you know just multiple cost for them are. So I am curious to see starts to apply uh, for I mean, assume am I going to do this going forward? I don't know what that means for previous generations of these products like right now, this mute license capability do IT on p do IT with a controller or in the cloud only a to the new wife, I seven, a PS.

Yes, something has to be aware of is that cisco s bondam its cisco spaces capabilities into its new wifi seven. If you're not familiar with that, zecca spaces is essentially is providing like location base services. You can do things like invitation tracking or providing way finding to guests or do environmental monory connecting I O T and sensors and stuff.

Just a question on that. But does that also does that also play into one nine, one one?

That's a good question. I am not sure. Yeah, I would have to look that up. I don't know. yes. So if you're .

listening to this, that would be definitely something to check to see how that effect or in nine one one capabilities. But yeah but yeah, my assumption .

is that does not because you know e nine one is probably a separate thing although he does involve location data. Yeah um but I will say with this this bundle um there are sort of taking a page uh from broken m uh in that what they're saying is we're providing you a simplified license. You there's two options with this license, essentially option a and option b.

Both of them include cisco o spaces and whether intend you intend the user or not, you're going to be paying for cisco spaces. So on the one hand, they are simplifying the licensing. On the other hand, the trade office you might be paying for features you don't plan to use.

yeah. But I mean, I think just having good location information is, in the end, good to be something useful. And I don't disagree with them for sort of pushing this on focus in the interest of simplicity. And one other thing I yeah and one .

other thing I want to notice that I think extreme um has done this already what you know this sort of like multiple personality or profiles for the same hardware um because extreme has a couple wireless companies under its um banner, essentially zebra arrive and others so they can say what you want sort of the zebra flavor. Okay, that's fine. It's it's the same A P the same management. So this is school also taking a page from extreme on this.

I think, yeah yeah I think there they're definitely we fallowing extremes lead. Uh, well moving on. Uh, ab A R, rics is now edging into network platforms of service they have up till now offered to A D Y I approach to creating a uniform cloud networking environment with tools, the customers who are downloaded, installed and then managed in their own I S environments in the order to bring to each a customer set of interfaces.

Now avia tricks is offering its initial as a service networking platform, which is fully managed by avia tricks, which customers connect to from the I S. environments. This is in preview in december and will be generally available in january and will be initially focused on cloud parameter security use cases, with additional use cases coming along later. What are your thoughts on this group?

I think a big sense if you're not familiar with the a va tricks, they essentially build themselves as sort of a common Operating a management layer for a public of networking. Are you deploy aviator software, you in a vpc or A V O or whatever. And then that that software, what's call in network gateway, they use the native constructs of the public cloud to set up a manager routing your VPN ingress and egress controlled and so on.

And you do all that from a single council. So your networking looks the same, whether you're trying to do in obs or as you a google or even connect all those, that was the benefit. But now what they are saying is, yeah, like you mentioned, you had to set all that up itself.

You had to, you know make sure you had you were picking the right infrastructure. You are signing the right memory and CPU and so on. You were doing all the software k grades with this PaaS approach. Aviators one now do that for you. So essentially, it's taking advantage of the cloud for this cloud networking capability if they've had make total sense.

yeah. And I mean, if you can look at the space, you can watch folks like OCR n AV tricks battled out in terms of user experience and ease of use and all that fun off. But net, net, I think the whole network is a service of movement has started to come of age with these recent announcements this fall.

I'm just two things to note though. Um first, this past capability will only apply to new customers. So if you're an existing aviator c customer, you may have to wait to get this past capability h and there may be A H services engagement with A V A tricks to make that transition. And second, IT only does apply to its cloud egress firework capabilities, not to its network gateway software. Aviatrix told me that they would be rolling out other capabilities ah in two thousand and twenty five, but right now just getting the ball rolling with their cloud egress capability.

And for existing customers that want to move to pass, they have to engage with the ab atrix professional services right drop. That's today's said, yeah, okay.

yeah. So it's for me it's like evy choices, making a baby step pair, testing the water, seeing how works. And I think there also probably has to be some significant work for them, other hood to develop that passed capability. So again, olympic roll out as they they get their feet with providing this capability or this service. Yeah.

that makes sense. The only thing is I don't actually agreed with the idea of making IT available to non customers first because that's giving them you're not treating the non customers Better than you treating customers, which I would question. But hey.

you mean the existing customers?

Where is the new customers? Yeah, yeah.

yeah. Well, need to get to start somewhere. I guess moving on bit, sticking with cloud versus networks. They provide a secure access service or si they've announced versus one which they say provide si for the when the cloud, the data center and campus.

Uh, I got a brief in from versa and this announcement kind of mostly about positioning versus competing as a conversion networking and security platform, a one stop shop for your network and security needs. One thing I learned from the briefing is that versa has a line of campus and land switches. I didn't know that they are fully intending to compete against traditional land vendors in this space.

Um so bringing that sai capability write down into your campus as opposed which is being sort of a from your S D wan or your S S C uh into the cloud ah and I am calling this out here and we put this in the show because I think i'm seeing a trend of vendors moving into this conversion network and security platform where they want to essentially be the one stop shop for you and cover most of the networks you care about your campus and land your branch office, your wan and your public cloud. Uh so versus putting its stake down here in this converge area. Um you'll hear from other folks like for net palo alto networks and others are looking to be your everything yeah .

it's it's really no in a venosa from a vago stic perspective, you're looking at the convergence of si, which is really still a architecture sd wan and zero trust york access, which are all sort of coming together in one big common thing. And you're seeing this what's interesting is you're seeing is from people who sell on prime gear or who started life selling on prime gear, people who mostly are selling cloud services, it's sort across the board.

So one of the things we're tracking very closely to see at what point network and security teams are going to begin to converge in enterprises right now based on our research, that hasn't happened yet, right? But we expect to see that starting to happen. And not sure how how much of a trend is going to be simply because there is still deep subject matter expertise in networking that's not necessarily security permanent and vice versa. So i'm going to keep an eye that you know suppose convergence of the Operations teams. And certainly, if you're listening and you want to give us a shout and tell us what's gone on in your environment, we'd love to hear that in the follow up.

yes. So the big pitcher obviously is beyond one vender. One need to sell you as much stuff as possible. The benefit to you is supposed to be a common Operating model, a common interface to manage your network and fix in your security policies.

And one OS that runs on all the hardware that the companies going to sell use to beat a firewall as which N S D V gateway at set a instead of you having to manage a bunch of different Operating systems and devices and kind of do all that, uh, stitching the other that engineers tend have to do. The premise is that one O S uh one interface will streamline your Operations and make things easier. But john, that sounds like you know, as you're looking at uh, talking to companies about whether it's going to happen on the the problem .

is for all this actually work out as plan, you'd have to have a sock knock convergence and that is just flat out not happening. We've been asking that question for years. Are you Operating separate knocks and socks and the answers? Yes, and they planned to do so for the foreseeable future.

More importantly, we haven't seen yet any meaningful correlation between converging your sock and knock and improving your security stance, although you can improve your network performance somewhat. So it's not entirely clear that the you in the sense that you just read their drew, that you there is a common Operating interface for you. My problem is who is that you? Because right now that you is two different people in two different places, and it's not entirely clear that, that there's an advantage to them both working on the same interface. That said, I could change. And so that's why i'm sort of reaching out and saying if anybody has thoughts on this and wants to talk about the poison coins of converging knock and sock or at least converging the tools and the management interface definitely we .

want to hear from ah yeah pack but does not .

flash if you are right um well moving right long. Paralo has warned users of potential for unauthentic ated remote core execution in its management interface. Up now just to be clear, let's let's lead with carefully reassuring customers that there are no known exploits in the wild while also admitting itself.

But IT does not yet know all the details of the flaw. Paula said customers should take steps to reduce the risk associated with possible remote code execution vulnerability in panorama, which is its management interface. The companies main recommendations are restricted access to the management interface to trusted internal networks and or trust I P addresses.

Always good practice anyway. Um polo alto users not up to date on patching probably want to get up to date on patching. So basically polo alto and the sister are urging everyone to get patched because there are ongoing, ongoing attacks, exploited other vulnerabilities. You know bottle bottom wine, restricted access, updated patches and watch this space. So there ago.

yeah, i'm just looking at the the button from pale of the networks. They're rating this themselves a severity of nine point three at a ten. So that is of the highest emergency. Uh, so well, that may be somewhat reassuring that they aren't seeing attacks in the wild yet. A does sound like this is serious and if your own panorama you need to go attend to IT yeah well.

I mean, only thing that that gives me a little confidence in this and I don't want to go too easy on on polar to, is the fact that essentially you can restrict limit risk by following basic security hydrant practices. So that's a good thing .

at least right for sure. Yeah right. And we have those links in the show. Note if you want to go check IT out. Moving on, google cloud has announced it's gona require multi factor authentication for all google cloud users. The companies taking a three phase approach of transition phase one, which is starting now, google essentially gonna nicely.

Will you please implement A M F A? It's gonna you some hints on how to do that? Uh, starting early twenty twenty five and face two, google will require M, F, A for users that logging with the password google rights quote.

You'll see notifications and guidance across the google cloud console, firebase console, g cloud and other platforms to continue using these tools, you'll need to enroll in mfa. So plan on that being a absolute requirement starting in early twenty twenty five if you want to use those things. Um that's the end of the asking nicely phase and phase three by the end of twenty twenty five M, F, A requirements will extend to users who federate, who federate, authenticate into google cloud. Google says it's working with identity providers to ensure the standards in place for a smooth and dove.

And the reason for this is, as google says, it's taking the step because fishing attacks and stone credentials are eight top attack vector. No surprise there. So yeah, this seems like a limited but perfectly reasonable stance for google to be taking, I guess.

yes. And I like they're doing the s nicely face first, uh but there will be a hard requirement, uh, pretty soon because the the calendar is winning and time is running out. yes.

And I think you had an observation when we looked at all our security related items for today, like what the hacker are we still talking about some of the basics here after all these years and all these new capabilities like zero trust we're still talking about, hey, don't forget to patch. Hey, restrict access. Hey, uma, like we could have been having this conversation in twenty fourteen and said we're having in twenty twenty .

four and will probably be having IT in twenty thirty four. So which keeps us in business.

right, I guess so .

I guess so so johna, we've been you know keeping eye on sort of the developments in in the nuclear space uh, because of the insane demand ah for power. Now h in data tenners, I found a story about a nuclear power start up oko or oko. I'm not sure how to pronounce that.

They say they have a letter of intent from two unnamed data center providers for OK loader provide up to seven hundred and fifty megawatts of power at data center sites in the united states. Uh, oko aims to build an Operate small modular reactors, which are essentially many nuclear power plants that could be built on site or near to locations such as data tres that require a massive amount of electricity. Uh, Apollo says its powerhouses come in fifteen and fifty megawatt units that can be deployed in phases.

You can sort of start small and then build out of the company else says its power plants, says that powers its plants with nuclear waste from existing reactors. This all sounds good, but there are still major federal regulatory hurdles overcome and municipal laws and codes. And uh according to a story in the register oklahoman, sn actually brought a single power plant online yet so what hopes you by the end of the decade?

Yeah, I think there's a handful of things to notice here. First and foremost, the wrapped up energy requirements, largely driven by AI, but not exclusively and will also be in place for quantum as we sort of roll onto that. That's going to be key.

I think it's also interesting to that people are getting louder about the fact that the demand is going to be there. So okay, is is saying, hey, we've got letters from two unnamed customers or future customers. I still am my jaw is still hanging open at h microsoft, saying that he wants to reactivate three my island.

Not because it's a bad idea, not because i'm disagree with them, but quite simply, the negative branding that it's risking associating there tells you how dire the need for energy really has to be before you, before you go and do that. And I think you know one of the one of the cool things is with the us. Administration change looking to happen, it's quite possible that this could all get accelerated very quickly.

Traditionally, republicans tend to be anti regulation and somewhat more pro nuclear power than democrats, although, you know, i'm not sure really goes by by party but that there's that and then there's also whole bunch of silicon valley folks who have deep interests and power hungry technologies like AI and crypto. And they are getting very visible and prominent places in the incoming administration. So here we are, but I don't know true. What do you think .

yeah I want I love the idea of being able to take you know nuclear waste and we're using IT and if okay, was actually figured out how to do that. That's very cool. Um but there's also a reason regulations are in place around nuclear power.

This is not the place where you move fast and break things ah so you know on one hand, yes you know innovation can be timing by regulation. On the other hand, there are often good reasons why you need that regulation and silicon valley. And I feel like the incoming administration both tend to be very capable here. So I guess I hear you and .

i'm sort of torn because on the one hand, one of the things immediately thought I was, gosh, modular nuclear power units could be really useful for small, small communities, for example and then I realized I can say things like that because I actually live right down the street from a nuclear power plant so basically not really using my risk much. It's just a new different flavor and um but then i'm imagining you .

know my nights hood where you know getting a new apartment complex built requires years of fighting a dropping a small modular reactor and and that never can happen here.

Yeah exactly. So I guess listening the regulations and maybe a little bit of a good thing. But there as you said, there is such a thing is going a little too far. So we'll watch and weight, but it's definitely an area that's interesting and worth keeping an ion.

Interesting times as as the schools. Yes, art dialog wrap up, the news with cisco are reporting their q one twenty twenty five earnings are for the quarter. Revenues were a thirteen point eight billion, down six percent year over year.

And C, N, B, C notes that this is, this goes fourth straight order of revenue decline. Net income was two point seven billion, down twenty five percent compared to this time last year. By business unit networking revenue, which counts for the bulk of discos, revenue was down twenty three percent.

Collaboration was down three percent. There were some bright spots. Their security revenue doubled versus this time last year for a total of just over two billion for the quarter and observably revenue was up thirty six percent to two hundred and fifty eight million.

Um despite this decline, this goes predicting or forecasting a full year uh twenty twenty five revenues of fifty five to fifty six billion, which would put IT slightly above the revenues that did in two and twenty four. So not a great start to the quarter, but this good things maybe it's got some things that can do um in in the future, particularly around A I C O. Chuck Robin is predicting a billion dollars worth of orders of ai future gear in financial twenty twenty five, and he said customers have already ordered three hundred million in A I inference cure from cisco for the first .

quarter yeah so that means that only another seven hundred million ago so which no IT really isn't that much in the scheme of things. yeah. I think the other thing that's interesting is when you note security revenue doubled this versus this time last year or for a total of just over two billion for the quarter, meaning that's that's a run rate of eight billion.

As adam also, who was the sea of a tt for a while, put IT at one point, sometimes you can hide a highly successful security company inside a larger company that may not look as successful. If you peel that out and had IT at its own company, IT would be a rockstar success company. So I have to tip my hat to disco on this. And i've been very skeptical of them in the security space, but good on them for driving so much security revenue. Let's hope the equality continues to be there.

And with that, C M, B C, note that this cisco s four straight quarter of revenue to decline. I went and looked back at cisco's, you know fully your revenues for the less four, five years they've bent up and downs, but they are generally you know somewhere in the ballpark of fifty five billion a year. So this go seems to be sort of like it's not a growth company anymore, but it's managing its finance as well as sort of a mature traditional or legacy vendors as they say. So I guess I am not worried for first this goes fortunes going forward.

Well, yeah, I would also add that what has happened in the past, you know, couple of years is covered. And so we saw a lot of craziness where people IT went nuts, ordering stuff with supply chain issues. And then I took a while for the pig in the python to get digested. So a lot of infrastructure vendors are blaming covet for recent slowdowns and looking optimistically towards Better times ahead. So hey.

i'm gonna save. Everybody's bottom line is the tea of that i'm reading.

You have that exactly IT well.

that reps up the news portion of the shows stick around for our tech bites conversation. We're talking with min I O and about how they are altering the data infrastructure landscape for A I that's come and read up today on the tech bites podcast. We welcome back sponsor min I O to talk about how A I is altering the data infrastructure landscape are also going to dig into.

Mini was recently launched A I store. This is a software only distributed object store that offers simplicity, scalability and performance for infrastructure and other high performance stored use cases. Our guests is artigue C T O at mini o. Uh are walk into the podcast and can you give us some quick background on mini o and how you're different? Shaking yourself in the objects to space.

I is a drop in replacement for A W S S, A P S as well as the full stack services. And the big difference for mini is we can run anywhere else. So we always h started the company to be the alternative to A W S S services.

Outside A W D S ecosystem. Weather is on the edge. Different clouds, different bar native services mean I O could be running on everywhere else. Simplice high performance and the best in class support a are the key of meanie.

For the last ten years, we are open source uh company and this has been the most important part of the reputation minani le has. And it's been quite popular with developers. And over the last five to six years, we have been getting into the I. T. As well as data use cases in the enterprise.

Now here we've been talking a lot on packet pushers about how artificial intelligence is affecting data networks. So we've talked about what A D, C bd looks like these days and speed requirements and concerns about buffing a packet and that kind of stuff. Ed, now mini s got a storage perspective on this. So how is A I from mini OS perspective affecting data infrastructure? And and i'm really curious to know if it's different from from big data, which we've been talking about forever.

The way IT worked in the past was the foundation of the A I wave, the A I transat and the way of A I all percent A I application are trying to handle data has always been built by the data lake houses, the data lake houses architecture or the modern data lakes, or different marketing names that different vendors has been giving them to. The architectures has been built, but the big data was or other hadoop p going back twenty years.

Hadoop started this um trying to organized the data in a way that the scale, performance and security is gone to be handled by the solutions. But none of them in the last twenty years has been quite successful trying to get them under one umbrella. A I expense this picture live IT further, because you have to have A H checking inter M L floor of q floor type of an activities, some kind of a pipeline, you have to have vector data basis, you have to have hugging a face and, you know, a repository for all the models and data sets.

So IT expands this picture into various areas. And what we do is mini o is the foundation for all of these factor. Databases started at a later age, and they have to use object storage because they started with object storage. And to your specific questions, when you go look at the networking and the data chAllenges is all about the scale and performance. The data networks of hadoop of one gig a bit is not the network that you usually hundred IT, is the networking that we use with A I infrastructure.

All of our customers that are large scale A I deployments and using for a distributed training, they have to have hundred to a bit that data net force and then N V M me on the service side supporting IT to feed their GPU because there's this problem in the industrial called hungry GPU problem. And in order to feed those GPU, you need the lowest state attention, least amount of hops and the simplistic networking with the highest band with possible. We can talk about what we are uh going to do with our A I store product and the technologies that we are investing in.

For example, s over D M A is critical for the next two, three years because when you go from hundred bit internet, you are going to go to four hundred and than eight hundred. When you get to those levels, the a number of packets that you have to process at the storage h at the CPU and networking layer tremendously increases. And that requires different protocols than T C P I P.

And we are looking at R D M A over eternel. Eternel is the base, but rocky, or R, D, M, A over internet is what we are investing on. And we are having an history over our D M.

A. Essentially, we're gonna be one of the few. People who are focusing on this solely, and that's going to help tremendously for the A I infrastructure A I use cases.

And to answer your question, the data infrastructure will tremendously change when you go to those higher speed for the next one to two. Two years is great because hundred gigabit, when you combine IT with two times hundred gigabit, will be totally fine, even for the largest use cases. But there two years out, three year out, we see this changing.

So can you give us sort of what a sample architecture might look like for a infrastructure built around a large data lake and using object storage?

Yeah, absolutely. So it's not different than predicting A I has been done for the last fifteen years. We talk about big data, and hadoop started that effort and moving things to big data, everything that you know off with a modern data lake and a warehousing architecture.

The two layers, which is core engine. The only big change was the core engines were getting separated. So location of storage and compute, if you will. So a tino, a drama type of accurate engine, is natively as compatible, and they can easily use object storage.

So you have different engines, curry engine requesting data from the modern data x, which is the raw data, which we have been replacing HDFS very successfully in the field. People have been using adobe has written S A connectors and it's been up there for many years. And mini o has been using S A to replace the three copies of hadoop and reducing the cost of IT to about one point two to one point for depending on the age of coding you are choosing.

So is a tremendous cost saving just to go from my HDFS p replacement of that in a hardup system and inject uh object storage or mini o there. You are saving tremendous amount of money. And we have uh, a few customers that has done this in a large scale and they have been very successful in terms of the simplistic environment, how many people are running this environment and how they simplify their infrastructure.

So IT would be just a modern object store like mini o career engines that would be the modern architecture. But when you go to the A I infrastructure, you expand this picture into two, three other sections. Um what what I call about the M L flow to type up of a pelle any type of pipeline.

Some people like open shift, for example, has open shift day I, which is their open source components as the pipeline. And at the bottom you need vector D B S. Most of the airline work in the enterprise today is going with l plus rag.

And you have to have a Better database. Milk was lsd. B, if you look at their documentation, they all recommend mini o. They started using mini o when they were, and they recommend that to their customers.

So you need that vector, D, B, because when you have the corporate data, the the data corpus of what you are doing for L M plus reg type of an environment, you need insert them to effect D B and take the mouse for prompting and other activities. And then you need model and data sets, which is early days of github, and give lab for the enterprise. Everything is on pram.

So you need a version of uh, hugging face on prem. And this is what we have done with A I store. Mean I oh only commercial product. We have built a on prem hugging face equivalent, basically on buckets that we have on mini o can be enabled with the hugging phase A P I, so developer can easily use without a change of code.

They can use take uh, data sets and the models that they're using in a hugger face, but put them on prem and use their on prem copy. So the same fashion that we have replaced A P S. And services for unprompted, we are doing the same thing for A I infrastructure with hugging face data sets and data models. They can just take same models, make IT between their four walls, store them on a bucket on object storage from mini o part of the A I store uh product that we are launching for our customers and we are providing to our customers.

Yeah I think this a pretty good time that is kind of drill in a little bit more on A I store. What what is IT that you're building um why my companies building out A I infrastructure wanted take a look at IT.

So the storage for A I represented specific chAllenges. We covered some of them. The scale is very important because the huge man with capacity needs because of the distributed training and modifying the models, training in distributed inference, a distributed training and then influence parts, they all, especially with the influence, you have to have the check points that required more and more.

So the scale was the problem. The other part was recovered with the data, uh, networking. The chAllenge was the performance, the high o benefit between the lost top or closer to the G, P, U.

So there is no hungry GPU problem. Do we cover that part as well? That was the performance chAllenge. And then protecting the data security. Most of our customers are nearly all of our customers, enterprise customers, they all wanted to control their data because of l lamps is commodity nowaday. But L O M plus reg or injecting injecting their a property information into the models or prompting with those models, they didn't feel comfortable putting something out there.

Are there meaning using a cloud services or using and other services that are on available publicly, they wanted to do IT themselves because that property data is the really important peace when IT comes to training the models, modifying the models and increasing the accuracy, also making them relevant for their customers. So if think about J P Morgan Morgan stanly, they all anus their A I chatbot um that's internal to their customer, internal to their employees where they can just check research papers or they can do ask certain things about a financial product. And all of those were done using their own proposal data.

Same with blooming burg, they have tremendous amount of data and they announced their AI initiatives. They all leveraging their own internal data. So security or protecting the data or controlling the data, this is not the security we are talking about.

Cloud is not secure enough. We should have everything on prime type of fair activity. See, when we talk about security chAllenge, they needed data to be on prem because they when they inject this data to A L M that is outside, they didn't want to leakage, they didn't want to share their data.

All of their vate data IT can be the training for somebody .

else is product or IT can be the data, the source data or you know, there can be many different ways to look at IT. Because of those things, we have come to the conclusion that we always did really great with object storage, high performance distributed object storage, best in class simplicity. But then we have seen over the last five to six years, we have seen our enterprise customers asking us for cashing, for example, that data for red uh cashing, they ask for observably.

They do know we were publishing all of our data for prometheus and graphical a to build dashpots, but our customers wanted to not install another product they want or alc or spunk to do observably for all the logs for things to be built our own observably, but for the internals of the desks and how to discard doing to audit logging, to error logging, everything lord balancing, they wanted to control things. So we built all these feature sets on top of the core of the high performance distributed object stories we we already have. And this is also aligned together with some of these feature step that we just mentioned, whether it's prompt, uh which we haven't talked too much about IT.

But hugging face A P S and having the data sets and the data models being store red down the all of them put together is the A I store. And this is our commercial product. Mini o will have A I store with all of these features built available to the to the customers and for specifically the use ACE of A I infrastructure.

So I think of point to make here, this is not just a rebranding exercise. You can get on A A I train. You've actually done something with the problem.

This is not the new bottle old wine. This is, this is, this is not responding. This is not marketing. I can go through, I think seven, eight, eight different uh feature sets.

And if you look at the in the legacy software world, they would be their own products if you were to go into those kind of details. So we built tremendous amount of this is not a this has been going on for the last one and a half years of development cycle. So IT has rebuilt all of these feature set and we packed them into one product.

So going forward, the life uh, a commercial life of mini o is very straight forward. Its single product focus is the A I infrastructure. And all of these things that we are talking about is going to be tremendously helpful for for those folks who are doing on prem large scale A I distributed training or any kind of a framed cure. The one feature .

you hinted that that you felt like I was unique to A I store or something compromised object. Can you um just a little over view .

then I I believe that the cool piece that in the whole stack there are many other h features. I from the we we build our own game. As for example, for encryption, we were work.

We always work work with a hash for the moto others and integrated into that. We even did some cashing type of an activity called K E S. But we build our own K M S part of this. We build our own catalogue, which is like a graph Q L in memory database, where we can and take all the names space and object store and curry those things but relevant to the AI questions that you had, prop object is basically an introduction of a new S P A P I.

We can do these things because we are the market leader when IT comes to following the a uh product and story spects and prom object basically enables uh users developers to do and a basically we have on structure objects or on structure data. All the objects can be JPEG files that can be CSV files, different type of files. But thinking about the image um you are just doing put off a received that is from a restaurant in california and you are putting another receipt from a restaurant in new york, say two different restaurants and all of the addresses and what's been uh charge in the receipts are there.

These are two objects you are putting them into the mini al object store, and with using prompt object, you can pass two N, L, M. You can pass A. Basic prot, you can say, what's the address on that receipt that gets passed through this API? We can extract the address within that object.

But this is going one step further. This is going inside the object, basically looking at that image, extracting the address. And how could would be the address of that restaurant in california?

So I can take an image file and get text out of, I don't need a text file to .

get and you you can not only x text file is the easy, you can get context out of IT. You can mean that the person can exponentially expand these capabilities to do for M R I scans. They can do K Y, C for a financial analysis that you can get context.

You can not only get to take you the quote about this when we do the demos, uh to our prospects and customers, you can compare these two images, the new york proceed and the california received, and then see what kind of restaurants were they so you can easily say, what's the difference between the two objects or two recedes and you will get, oh, this one is a mexican restaurant. The received has tackles. And the other one is a vegetarian restaurant in new ork.

And X, Y, Z. And you can get this context that's powerful. And having this type of a, we were always doing, put an object into an object store and get for a bian application. With this we can do a put, but do a prompt from meaning, get the context out of that object, understand the um what's inside the object. Basically in this simple example, IT was too recedes.

But IT can be an M R S, can, IT can be A K, Y, C application and an image of a person or or an I, D, or you can have an image of a person and see how many of the is this image in any of these videos. You can do tremendous among those things and it's one line code and is just natural language processing. Basically, you are just asking the question just like you would ask to anybody .

or I feel like we scratched just the surface I store and what I can do um if folks want to dig around and get more details, find out more information or even reach out to um uh men I O to get more details, where did they go?

Uh best place would be medi o um we have all the blocks, all the resources and materials that uh all the things that we are talking about here, h is available there. You can reach to us and then we can jump on a call with you. Uh, anybody who's interested to go into the details, technical details of all these things we talked about anta .

tic that was min di o and IT right therein the name mini o thanks for being here and thanks to min I O for being a sponsor and thanks to you for listening. You can find this and many more fine free technical podcast along with our community blog at all at packet precious that net can follow us on linked here is on spotify and read us on apple podcast and less at least remember that too much infrastructure would never be enough.